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" |
fec577fb CM |
29 | #include "ctree.h" |
30 | #include "disk-io.h" | |
31 | #include "print-tree.h" | |
e089f05c | 32 | #include "transaction.h" |
0b86a832 | 33 | #include "volumes.h" |
53b381b3 | 34 | #include "raid56.h" |
925baedd | 35 | #include "locking.h" |
fa9c0d79 | 36 | #include "free-space-cache.h" |
3fed40cc | 37 | #include "math.h" |
fec577fb | 38 | |
709c0486 AJ |
39 | #undef SCRAMBLE_DELAYED_REFS |
40 | ||
9e622d6b MX |
41 | /* |
42 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
43 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
44 | * if we really need one. | |
45 | * | |
0e4f8f88 CM |
46 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
47 | * if we have very few chunks already allocated. This is | |
48 | * used as part of the clustering code to help make sure | |
49 | * we have a good pool of storage to cluster in, without | |
50 | * filling the FS with empty chunks | |
51 | * | |
9e622d6b MX |
52 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
53 | * | |
0e4f8f88 CM |
54 | */ |
55 | enum { | |
56 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
57 | CHUNK_ALLOC_LIMITED = 1, |
58 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
59 | }; |
60 | ||
fb25e914 JB |
61 | /* |
62 | * Control how reservations are dealt with. | |
63 | * | |
64 | * RESERVE_FREE - freeing a reservation. | |
65 | * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for | |
66 | * ENOSPC accounting | |
67 | * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update | |
68 | * bytes_may_use as the ENOSPC accounting is done elsewhere | |
69 | */ | |
70 | enum { | |
71 | RESERVE_FREE = 0, | |
72 | RESERVE_ALLOC = 1, | |
73 | RESERVE_ALLOC_NO_ACCOUNT = 2, | |
74 | }; | |
75 | ||
c53d613e | 76 | static int update_block_group(struct btrfs_root *root, |
f0486c68 | 77 | u64 bytenr, u64 num_bytes, int alloc); |
5d4f98a2 YZ |
78 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
79 | struct btrfs_root *root, | |
80 | u64 bytenr, u64 num_bytes, u64 parent, | |
81 | u64 root_objectid, u64 owner_objectid, | |
82 | u64 owner_offset, int refs_to_drop, | |
83 | struct btrfs_delayed_extent_op *extra_op); | |
84 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
85 | struct extent_buffer *leaf, | |
86 | struct btrfs_extent_item *ei); | |
87 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
88 | struct btrfs_root *root, | |
89 | u64 parent, u64 root_objectid, | |
90 | u64 flags, u64 owner, u64 offset, | |
91 | struct btrfs_key *ins, int ref_mod); | |
92 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
93 | struct btrfs_root *root, | |
94 | u64 parent, u64 root_objectid, | |
95 | u64 flags, struct btrfs_disk_key *key, | |
96 | int level, struct btrfs_key *ins); | |
6a63209f | 97 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 JB |
98 | struct btrfs_root *extent_root, u64 flags, |
99 | int force); | |
11833d66 YZ |
100 | static int find_next_key(struct btrfs_path *path, int level, |
101 | struct btrfs_key *key); | |
9ed74f2d JB |
102 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
103 | int dump_block_groups); | |
fb25e914 JB |
104 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
105 | u64 num_bytes, int reserve); | |
5d80366e JB |
106 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
107 | u64 num_bytes); | |
48a3b636 ES |
108 | int btrfs_pin_extent(struct btrfs_root *root, |
109 | u64 bytenr, u64 num_bytes, int reserved); | |
6a63209f | 110 | |
817d52f8 JB |
111 | static noinline int |
112 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
113 | { | |
114 | smp_mb(); | |
36cce922 JB |
115 | return cache->cached == BTRFS_CACHE_FINISHED || |
116 | cache->cached == BTRFS_CACHE_ERROR; | |
817d52f8 JB |
117 | } |
118 | ||
0f9dd46c JB |
119 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
120 | { | |
121 | return (cache->flags & bits) == bits; | |
122 | } | |
123 | ||
62a45b60 | 124 | static void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
125 | { |
126 | atomic_inc(&cache->count); | |
127 | } | |
128 | ||
129 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
130 | { | |
f0486c68 YZ |
131 | if (atomic_dec_and_test(&cache->count)) { |
132 | WARN_ON(cache->pinned > 0); | |
133 | WARN_ON(cache->reserved > 0); | |
34d52cb6 | 134 | kfree(cache->free_space_ctl); |
11dfe35a | 135 | kfree(cache); |
f0486c68 | 136 | } |
11dfe35a JB |
137 | } |
138 | ||
0f9dd46c JB |
139 | /* |
140 | * this adds the block group to the fs_info rb tree for the block group | |
141 | * cache | |
142 | */ | |
b2950863 | 143 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
144 | struct btrfs_block_group_cache *block_group) |
145 | { | |
146 | struct rb_node **p; | |
147 | struct rb_node *parent = NULL; | |
148 | struct btrfs_block_group_cache *cache; | |
149 | ||
150 | spin_lock(&info->block_group_cache_lock); | |
151 | p = &info->block_group_cache_tree.rb_node; | |
152 | ||
153 | while (*p) { | |
154 | parent = *p; | |
155 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
156 | cache_node); | |
157 | if (block_group->key.objectid < cache->key.objectid) { | |
158 | p = &(*p)->rb_left; | |
159 | } else if (block_group->key.objectid > cache->key.objectid) { | |
160 | p = &(*p)->rb_right; | |
161 | } else { | |
162 | spin_unlock(&info->block_group_cache_lock); | |
163 | return -EEXIST; | |
164 | } | |
165 | } | |
166 | ||
167 | rb_link_node(&block_group->cache_node, parent, p); | |
168 | rb_insert_color(&block_group->cache_node, | |
169 | &info->block_group_cache_tree); | |
a1897fdd LB |
170 | |
171 | if (info->first_logical_byte > block_group->key.objectid) | |
172 | info->first_logical_byte = block_group->key.objectid; | |
173 | ||
0f9dd46c JB |
174 | spin_unlock(&info->block_group_cache_lock); |
175 | ||
176 | return 0; | |
177 | } | |
178 | ||
179 | /* | |
180 | * This will return the block group at or after bytenr if contains is 0, else | |
181 | * it will return the block group that contains the bytenr | |
182 | */ | |
183 | static struct btrfs_block_group_cache * | |
184 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
185 | int contains) | |
186 | { | |
187 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
188 | struct rb_node *n; | |
189 | u64 end, start; | |
190 | ||
191 | spin_lock(&info->block_group_cache_lock); | |
192 | n = info->block_group_cache_tree.rb_node; | |
193 | ||
194 | while (n) { | |
195 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
196 | cache_node); | |
197 | end = cache->key.objectid + cache->key.offset - 1; | |
198 | start = cache->key.objectid; | |
199 | ||
200 | if (bytenr < start) { | |
201 | if (!contains && (!ret || start < ret->key.objectid)) | |
202 | ret = cache; | |
203 | n = n->rb_left; | |
204 | } else if (bytenr > start) { | |
205 | if (contains && bytenr <= end) { | |
206 | ret = cache; | |
207 | break; | |
208 | } | |
209 | n = n->rb_right; | |
210 | } else { | |
211 | ret = cache; | |
212 | break; | |
213 | } | |
214 | } | |
a1897fdd | 215 | if (ret) { |
11dfe35a | 216 | btrfs_get_block_group(ret); |
a1897fdd LB |
217 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
218 | info->first_logical_byte = ret->key.objectid; | |
219 | } | |
0f9dd46c JB |
220 | spin_unlock(&info->block_group_cache_lock); |
221 | ||
222 | return ret; | |
223 | } | |
224 | ||
11833d66 YZ |
225 | static int add_excluded_extent(struct btrfs_root *root, |
226 | u64 start, u64 num_bytes) | |
817d52f8 | 227 | { |
11833d66 YZ |
228 | u64 end = start + num_bytes - 1; |
229 | set_extent_bits(&root->fs_info->freed_extents[0], | |
230 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
231 | set_extent_bits(&root->fs_info->freed_extents[1], | |
232 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
233 | return 0; | |
234 | } | |
817d52f8 | 235 | |
11833d66 YZ |
236 | static void free_excluded_extents(struct btrfs_root *root, |
237 | struct btrfs_block_group_cache *cache) | |
238 | { | |
239 | u64 start, end; | |
817d52f8 | 240 | |
11833d66 YZ |
241 | start = cache->key.objectid; |
242 | end = start + cache->key.offset - 1; | |
243 | ||
244 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
245 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
246 | clear_extent_bits(&root->fs_info->freed_extents[1], | |
247 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
817d52f8 JB |
248 | } |
249 | ||
11833d66 YZ |
250 | static int exclude_super_stripes(struct btrfs_root *root, |
251 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 252 | { |
817d52f8 JB |
253 | u64 bytenr; |
254 | u64 *logical; | |
255 | int stripe_len; | |
256 | int i, nr, ret; | |
257 | ||
06b2331f YZ |
258 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
259 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
260 | cache->bytes_super += stripe_len; | |
261 | ret = add_excluded_extent(root, cache->key.objectid, | |
262 | stripe_len); | |
835d974f JB |
263 | if (ret) |
264 | return ret; | |
06b2331f YZ |
265 | } |
266 | ||
817d52f8 JB |
267 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
268 | bytenr = btrfs_sb_offset(i); | |
269 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
270 | cache->key.objectid, bytenr, | |
271 | 0, &logical, &nr, &stripe_len); | |
835d974f JB |
272 | if (ret) |
273 | return ret; | |
11833d66 | 274 | |
817d52f8 | 275 | while (nr--) { |
51bf5f0b JB |
276 | u64 start, len; |
277 | ||
278 | if (logical[nr] > cache->key.objectid + | |
279 | cache->key.offset) | |
280 | continue; | |
281 | ||
282 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
283 | continue; | |
284 | ||
285 | start = logical[nr]; | |
286 | if (start < cache->key.objectid) { | |
287 | start = cache->key.objectid; | |
288 | len = (logical[nr] + stripe_len) - start; | |
289 | } else { | |
290 | len = min_t(u64, stripe_len, | |
291 | cache->key.objectid + | |
292 | cache->key.offset - start); | |
293 | } | |
294 | ||
295 | cache->bytes_super += len; | |
296 | ret = add_excluded_extent(root, start, len); | |
835d974f JB |
297 | if (ret) { |
298 | kfree(logical); | |
299 | return ret; | |
300 | } | |
817d52f8 | 301 | } |
11833d66 | 302 | |
817d52f8 JB |
303 | kfree(logical); |
304 | } | |
817d52f8 JB |
305 | return 0; |
306 | } | |
307 | ||
11833d66 YZ |
308 | static struct btrfs_caching_control * |
309 | get_caching_control(struct btrfs_block_group_cache *cache) | |
310 | { | |
311 | struct btrfs_caching_control *ctl; | |
312 | ||
313 | spin_lock(&cache->lock); | |
314 | if (cache->cached != BTRFS_CACHE_STARTED) { | |
315 | spin_unlock(&cache->lock); | |
316 | return NULL; | |
317 | } | |
318 | ||
dde5abee JB |
319 | /* We're loading it the fast way, so we don't have a caching_ctl. */ |
320 | if (!cache->caching_ctl) { | |
321 | spin_unlock(&cache->lock); | |
11833d66 YZ |
322 | return NULL; |
323 | } | |
324 | ||
325 | ctl = cache->caching_ctl; | |
326 | atomic_inc(&ctl->count); | |
327 | spin_unlock(&cache->lock); | |
328 | return ctl; | |
329 | } | |
330 | ||
331 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
332 | { | |
333 | if (atomic_dec_and_test(&ctl->count)) | |
334 | kfree(ctl); | |
335 | } | |
336 | ||
0f9dd46c JB |
337 | /* |
338 | * this is only called by cache_block_group, since we could have freed extents | |
339 | * we need to check the pinned_extents for any extents that can't be used yet | |
340 | * since their free space will be released as soon as the transaction commits. | |
341 | */ | |
817d52f8 | 342 | static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
0f9dd46c JB |
343 | struct btrfs_fs_info *info, u64 start, u64 end) |
344 | { | |
817d52f8 | 345 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
346 | int ret; |
347 | ||
348 | while (start < end) { | |
11833d66 | 349 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 350 | &extent_start, &extent_end, |
e6138876 JB |
351 | EXTENT_DIRTY | EXTENT_UPTODATE, |
352 | NULL); | |
0f9dd46c JB |
353 | if (ret) |
354 | break; | |
355 | ||
06b2331f | 356 | if (extent_start <= start) { |
0f9dd46c JB |
357 | start = extent_end + 1; |
358 | } else if (extent_start > start && extent_start < end) { | |
359 | size = extent_start - start; | |
817d52f8 | 360 | total_added += size; |
ea6a478e JB |
361 | ret = btrfs_add_free_space(block_group, start, |
362 | size); | |
79787eaa | 363 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
364 | start = extent_end + 1; |
365 | } else { | |
366 | break; | |
367 | } | |
368 | } | |
369 | ||
370 | if (start < end) { | |
371 | size = end - start; | |
817d52f8 | 372 | total_added += size; |
ea6a478e | 373 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 374 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
375 | } |
376 | ||
817d52f8 | 377 | return total_added; |
0f9dd46c JB |
378 | } |
379 | ||
bab39bf9 | 380 | static noinline void caching_thread(struct btrfs_work *work) |
e37c9e69 | 381 | { |
bab39bf9 JB |
382 | struct btrfs_block_group_cache *block_group; |
383 | struct btrfs_fs_info *fs_info; | |
384 | struct btrfs_caching_control *caching_ctl; | |
385 | struct btrfs_root *extent_root; | |
e37c9e69 | 386 | struct btrfs_path *path; |
5f39d397 | 387 | struct extent_buffer *leaf; |
11833d66 | 388 | struct btrfs_key key; |
817d52f8 | 389 | u64 total_found = 0; |
11833d66 YZ |
390 | u64 last = 0; |
391 | u32 nritems; | |
36cce922 | 392 | int ret = -ENOMEM; |
f510cfec | 393 | |
bab39bf9 JB |
394 | caching_ctl = container_of(work, struct btrfs_caching_control, work); |
395 | block_group = caching_ctl->block_group; | |
396 | fs_info = block_group->fs_info; | |
397 | extent_root = fs_info->extent_root; | |
398 | ||
e37c9e69 CM |
399 | path = btrfs_alloc_path(); |
400 | if (!path) | |
bab39bf9 | 401 | goto out; |
7d7d6068 | 402 | |
817d52f8 | 403 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 404 | |
5cd57b2c | 405 | /* |
817d52f8 JB |
406 | * We don't want to deadlock with somebody trying to allocate a new |
407 | * extent for the extent root while also trying to search the extent | |
408 | * root to add free space. So we skip locking and search the commit | |
409 | * root, since its read-only | |
5cd57b2c CM |
410 | */ |
411 | path->skip_locking = 1; | |
817d52f8 | 412 | path->search_commit_root = 1; |
026fd317 | 413 | path->reada = 1; |
817d52f8 | 414 | |
e4404d6e | 415 | key.objectid = last; |
e37c9e69 | 416 | key.offset = 0; |
11833d66 | 417 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 418 | again: |
11833d66 | 419 | mutex_lock(&caching_ctl->mutex); |
013f1b12 CM |
420 | /* need to make sure the commit_root doesn't disappear */ |
421 | down_read(&fs_info->extent_commit_sem); | |
422 | ||
52ee28d2 | 423 | next: |
11833d66 | 424 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 425 | if (ret < 0) |
ef8bbdfe | 426 | goto err; |
a512bbf8 | 427 | |
11833d66 YZ |
428 | leaf = path->nodes[0]; |
429 | nritems = btrfs_header_nritems(leaf); | |
430 | ||
d397712b | 431 | while (1) { |
7841cb28 | 432 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 433 | last = (u64)-1; |
817d52f8 | 434 | break; |
f25784b3 | 435 | } |
817d52f8 | 436 | |
11833d66 YZ |
437 | if (path->slots[0] < nritems) { |
438 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
439 | } else { | |
440 | ret = find_next_key(path, 0, &key); | |
441 | if (ret) | |
e37c9e69 | 442 | break; |
817d52f8 | 443 | |
0a3896d0 | 444 | if (need_resched()) { |
589d8ade | 445 | caching_ctl->progress = last; |
ff5714cc | 446 | btrfs_release_path(path); |
589d8ade JB |
447 | up_read(&fs_info->extent_commit_sem); |
448 | mutex_unlock(&caching_ctl->mutex); | |
11833d66 | 449 | cond_resched(); |
589d8ade JB |
450 | goto again; |
451 | } | |
0a3896d0 JB |
452 | |
453 | ret = btrfs_next_leaf(extent_root, path); | |
454 | if (ret < 0) | |
455 | goto err; | |
456 | if (ret) | |
457 | break; | |
589d8ade JB |
458 | leaf = path->nodes[0]; |
459 | nritems = btrfs_header_nritems(leaf); | |
460 | continue; | |
11833d66 | 461 | } |
817d52f8 | 462 | |
52ee28d2 LB |
463 | if (key.objectid < last) { |
464 | key.objectid = last; | |
465 | key.offset = 0; | |
466 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
467 | ||
468 | caching_ctl->progress = last; | |
469 | btrfs_release_path(path); | |
470 | goto next; | |
471 | } | |
472 | ||
11833d66 YZ |
473 | if (key.objectid < block_group->key.objectid) { |
474 | path->slots[0]++; | |
817d52f8 | 475 | continue; |
e37c9e69 | 476 | } |
0f9dd46c | 477 | |
e37c9e69 | 478 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 479 | block_group->key.offset) |
e37c9e69 | 480 | break; |
7d7d6068 | 481 | |
3173a18f JB |
482 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
483 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
817d52f8 JB |
484 | total_found += add_new_free_space(block_group, |
485 | fs_info, last, | |
486 | key.objectid); | |
3173a18f JB |
487 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
488 | last = key.objectid + | |
489 | fs_info->tree_root->leafsize; | |
490 | else | |
491 | last = key.objectid + key.offset; | |
817d52f8 | 492 | |
11833d66 YZ |
493 | if (total_found > (1024 * 1024 * 2)) { |
494 | total_found = 0; | |
495 | wake_up(&caching_ctl->wait); | |
496 | } | |
817d52f8 | 497 | } |
e37c9e69 CM |
498 | path->slots[0]++; |
499 | } | |
817d52f8 | 500 | ret = 0; |
e37c9e69 | 501 | |
817d52f8 JB |
502 | total_found += add_new_free_space(block_group, fs_info, last, |
503 | block_group->key.objectid + | |
504 | block_group->key.offset); | |
11833d66 | 505 | caching_ctl->progress = (u64)-1; |
817d52f8 JB |
506 | |
507 | spin_lock(&block_group->lock); | |
11833d66 | 508 | block_group->caching_ctl = NULL; |
817d52f8 JB |
509 | block_group->cached = BTRFS_CACHE_FINISHED; |
510 | spin_unlock(&block_group->lock); | |
0f9dd46c | 511 | |
54aa1f4d | 512 | err: |
e37c9e69 | 513 | btrfs_free_path(path); |
276e680d | 514 | up_read(&fs_info->extent_commit_sem); |
817d52f8 | 515 | |
11833d66 YZ |
516 | free_excluded_extents(extent_root, block_group); |
517 | ||
518 | mutex_unlock(&caching_ctl->mutex); | |
bab39bf9 | 519 | out: |
36cce922 JB |
520 | if (ret) { |
521 | spin_lock(&block_group->lock); | |
522 | block_group->caching_ctl = NULL; | |
523 | block_group->cached = BTRFS_CACHE_ERROR; | |
524 | spin_unlock(&block_group->lock); | |
525 | } | |
11833d66 YZ |
526 | wake_up(&caching_ctl->wait); |
527 | ||
528 | put_caching_control(caching_ctl); | |
11dfe35a | 529 | btrfs_put_block_group(block_group); |
817d52f8 JB |
530 | } |
531 | ||
9d66e233 | 532 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 533 | int load_cache_only) |
817d52f8 | 534 | { |
291c7d2f | 535 | DEFINE_WAIT(wait); |
11833d66 YZ |
536 | struct btrfs_fs_info *fs_info = cache->fs_info; |
537 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
538 | int ret = 0; |
539 | ||
291c7d2f | 540 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
541 | if (!caching_ctl) |
542 | return -ENOMEM; | |
291c7d2f JB |
543 | |
544 | INIT_LIST_HEAD(&caching_ctl->list); | |
545 | mutex_init(&caching_ctl->mutex); | |
546 | init_waitqueue_head(&caching_ctl->wait); | |
547 | caching_ctl->block_group = cache; | |
548 | caching_ctl->progress = cache->key.objectid; | |
549 | atomic_set(&caching_ctl->count, 1); | |
550 | caching_ctl->work.func = caching_thread; | |
551 | ||
552 | spin_lock(&cache->lock); | |
553 | /* | |
554 | * This should be a rare occasion, but this could happen I think in the | |
555 | * case where one thread starts to load the space cache info, and then | |
556 | * some other thread starts a transaction commit which tries to do an | |
557 | * allocation while the other thread is still loading the space cache | |
558 | * info. The previous loop should have kept us from choosing this block | |
559 | * group, but if we've moved to the state where we will wait on caching | |
560 | * block groups we need to first check if we're doing a fast load here, | |
561 | * so we can wait for it to finish, otherwise we could end up allocating | |
562 | * from a block group who's cache gets evicted for one reason or | |
563 | * another. | |
564 | */ | |
565 | while (cache->cached == BTRFS_CACHE_FAST) { | |
566 | struct btrfs_caching_control *ctl; | |
567 | ||
568 | ctl = cache->caching_ctl; | |
569 | atomic_inc(&ctl->count); | |
570 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); | |
571 | spin_unlock(&cache->lock); | |
572 | ||
573 | schedule(); | |
574 | ||
575 | finish_wait(&ctl->wait, &wait); | |
576 | put_caching_control(ctl); | |
577 | spin_lock(&cache->lock); | |
578 | } | |
579 | ||
580 | if (cache->cached != BTRFS_CACHE_NO) { | |
581 | spin_unlock(&cache->lock); | |
582 | kfree(caching_ctl); | |
11833d66 | 583 | return 0; |
291c7d2f JB |
584 | } |
585 | WARN_ON(cache->caching_ctl); | |
586 | cache->caching_ctl = caching_ctl; | |
587 | cache->cached = BTRFS_CACHE_FAST; | |
588 | spin_unlock(&cache->lock); | |
11833d66 | 589 | |
d53ba474 | 590 | if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) { |
9d66e233 JB |
591 | ret = load_free_space_cache(fs_info, cache); |
592 | ||
593 | spin_lock(&cache->lock); | |
594 | if (ret == 1) { | |
291c7d2f | 595 | cache->caching_ctl = NULL; |
9d66e233 JB |
596 | cache->cached = BTRFS_CACHE_FINISHED; |
597 | cache->last_byte_to_unpin = (u64)-1; | |
598 | } else { | |
291c7d2f JB |
599 | if (load_cache_only) { |
600 | cache->caching_ctl = NULL; | |
601 | cache->cached = BTRFS_CACHE_NO; | |
602 | } else { | |
603 | cache->cached = BTRFS_CACHE_STARTED; | |
604 | } | |
9d66e233 JB |
605 | } |
606 | spin_unlock(&cache->lock); | |
291c7d2f | 607 | wake_up(&caching_ctl->wait); |
3c14874a | 608 | if (ret == 1) { |
291c7d2f | 609 | put_caching_control(caching_ctl); |
3c14874a | 610 | free_excluded_extents(fs_info->extent_root, cache); |
9d66e233 | 611 | return 0; |
3c14874a | 612 | } |
291c7d2f JB |
613 | } else { |
614 | /* | |
615 | * We are not going to do the fast caching, set cached to the | |
616 | * appropriate value and wakeup any waiters. | |
617 | */ | |
618 | spin_lock(&cache->lock); | |
619 | if (load_cache_only) { | |
620 | cache->caching_ctl = NULL; | |
621 | cache->cached = BTRFS_CACHE_NO; | |
622 | } else { | |
623 | cache->cached = BTRFS_CACHE_STARTED; | |
624 | } | |
625 | spin_unlock(&cache->lock); | |
626 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
627 | } |
628 | ||
291c7d2f JB |
629 | if (load_cache_only) { |
630 | put_caching_control(caching_ctl); | |
11833d66 | 631 | return 0; |
817d52f8 | 632 | } |
817d52f8 | 633 | |
11833d66 | 634 | down_write(&fs_info->extent_commit_sem); |
291c7d2f | 635 | atomic_inc(&caching_ctl->count); |
11833d66 YZ |
636 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
637 | up_write(&fs_info->extent_commit_sem); | |
638 | ||
11dfe35a | 639 | btrfs_get_block_group(cache); |
11833d66 | 640 | |
bab39bf9 | 641 | btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 642 | |
ef8bbdfe | 643 | return ret; |
e37c9e69 CM |
644 | } |
645 | ||
0f9dd46c JB |
646 | /* |
647 | * return the block group that starts at or after bytenr | |
648 | */ | |
d397712b CM |
649 | static struct btrfs_block_group_cache * |
650 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 651 | { |
0f9dd46c | 652 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 653 | |
0f9dd46c | 654 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 655 | |
0f9dd46c | 656 | return cache; |
0ef3e66b CM |
657 | } |
658 | ||
0f9dd46c | 659 | /* |
9f55684c | 660 | * return the block group that contains the given bytenr |
0f9dd46c | 661 | */ |
d397712b CM |
662 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
663 | struct btrfs_fs_info *info, | |
664 | u64 bytenr) | |
be744175 | 665 | { |
0f9dd46c | 666 | struct btrfs_block_group_cache *cache; |
be744175 | 667 | |
0f9dd46c | 668 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 669 | |
0f9dd46c | 670 | return cache; |
be744175 | 671 | } |
0b86a832 | 672 | |
0f9dd46c JB |
673 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
674 | u64 flags) | |
6324fbf3 | 675 | { |
0f9dd46c | 676 | struct list_head *head = &info->space_info; |
0f9dd46c | 677 | struct btrfs_space_info *found; |
4184ea7f | 678 | |
52ba6929 | 679 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 680 | |
4184ea7f CM |
681 | rcu_read_lock(); |
682 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 683 | if (found->flags & flags) { |
4184ea7f | 684 | rcu_read_unlock(); |
0f9dd46c | 685 | return found; |
4184ea7f | 686 | } |
0f9dd46c | 687 | } |
4184ea7f | 688 | rcu_read_unlock(); |
0f9dd46c | 689 | return NULL; |
6324fbf3 CM |
690 | } |
691 | ||
4184ea7f CM |
692 | /* |
693 | * after adding space to the filesystem, we need to clear the full flags | |
694 | * on all the space infos. | |
695 | */ | |
696 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
697 | { | |
698 | struct list_head *head = &info->space_info; | |
699 | struct btrfs_space_info *found; | |
700 | ||
701 | rcu_read_lock(); | |
702 | list_for_each_entry_rcu(found, head, list) | |
703 | found->full = 0; | |
704 | rcu_read_unlock(); | |
705 | } | |
706 | ||
e02119d5 | 707 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 708 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
709 | { |
710 | int ret; | |
711 | struct btrfs_key key; | |
31840ae1 | 712 | struct btrfs_path *path; |
e02119d5 | 713 | |
31840ae1 | 714 | path = btrfs_alloc_path(); |
d8926bb3 MF |
715 | if (!path) |
716 | return -ENOMEM; | |
717 | ||
e02119d5 CM |
718 | key.objectid = start; |
719 | key.offset = len; | |
3173a18f | 720 | key.type = BTRFS_EXTENT_ITEM_KEY; |
e02119d5 CM |
721 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, |
722 | 0, 0); | |
3173a18f JB |
723 | if (ret > 0) { |
724 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
725 | if (key.objectid == start && | |
726 | key.type == BTRFS_METADATA_ITEM_KEY) | |
727 | ret = 0; | |
728 | } | |
31840ae1 | 729 | btrfs_free_path(path); |
7bb86316 CM |
730 | return ret; |
731 | } | |
732 | ||
a22285a6 | 733 | /* |
3173a18f | 734 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
735 | * |
736 | * the head node for delayed ref is used to store the sum of all the | |
737 | * reference count modifications queued up in the rbtree. the head | |
738 | * node may also store the extent flags to set. This way you can check | |
739 | * to see what the reference count and extent flags would be if all of | |
740 | * the delayed refs are not processed. | |
741 | */ | |
742 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
743 | struct btrfs_root *root, u64 bytenr, | |
3173a18f | 744 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
745 | { |
746 | struct btrfs_delayed_ref_head *head; | |
747 | struct btrfs_delayed_ref_root *delayed_refs; | |
748 | struct btrfs_path *path; | |
749 | struct btrfs_extent_item *ei; | |
750 | struct extent_buffer *leaf; | |
751 | struct btrfs_key key; | |
752 | u32 item_size; | |
753 | u64 num_refs; | |
754 | u64 extent_flags; | |
755 | int ret; | |
756 | ||
3173a18f JB |
757 | /* |
758 | * If we don't have skinny metadata, don't bother doing anything | |
759 | * different | |
760 | */ | |
761 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) { | |
762 | offset = root->leafsize; | |
763 | metadata = 0; | |
764 | } | |
765 | ||
a22285a6 YZ |
766 | path = btrfs_alloc_path(); |
767 | if (!path) | |
768 | return -ENOMEM; | |
769 | ||
3173a18f JB |
770 | if (metadata) { |
771 | key.objectid = bytenr; | |
772 | key.type = BTRFS_METADATA_ITEM_KEY; | |
773 | key.offset = offset; | |
774 | } else { | |
775 | key.objectid = bytenr; | |
776 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
777 | key.offset = offset; | |
778 | } | |
779 | ||
a22285a6 YZ |
780 | if (!trans) { |
781 | path->skip_locking = 1; | |
782 | path->search_commit_root = 1; | |
783 | } | |
784 | again: | |
785 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, | |
786 | &key, path, 0, 0); | |
787 | if (ret < 0) | |
788 | goto out_free; | |
789 | ||
3173a18f | 790 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
791 | metadata = 0; |
792 | if (path->slots[0]) { | |
793 | path->slots[0]--; | |
794 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
795 | path->slots[0]); | |
796 | if (key.objectid == bytenr && | |
797 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
798 | key.offset == root->leafsize) | |
799 | ret = 0; | |
800 | } | |
801 | if (ret) { | |
802 | key.objectid = bytenr; | |
803 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
804 | key.offset = root->leafsize; | |
805 | btrfs_release_path(path); | |
806 | goto again; | |
807 | } | |
3173a18f JB |
808 | } |
809 | ||
a22285a6 YZ |
810 | if (ret == 0) { |
811 | leaf = path->nodes[0]; | |
812 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
813 | if (item_size >= sizeof(*ei)) { | |
814 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
815 | struct btrfs_extent_item); | |
816 | num_refs = btrfs_extent_refs(leaf, ei); | |
817 | extent_flags = btrfs_extent_flags(leaf, ei); | |
818 | } else { | |
819 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
820 | struct btrfs_extent_item_v0 *ei0; | |
821 | BUG_ON(item_size != sizeof(*ei0)); | |
822 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
823 | struct btrfs_extent_item_v0); | |
824 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
825 | /* FIXME: this isn't correct for data */ | |
826 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
827 | #else | |
828 | BUG(); | |
829 | #endif | |
830 | } | |
831 | BUG_ON(num_refs == 0); | |
832 | } else { | |
833 | num_refs = 0; | |
834 | extent_flags = 0; | |
835 | ret = 0; | |
836 | } | |
837 | ||
838 | if (!trans) | |
839 | goto out; | |
840 | ||
841 | delayed_refs = &trans->transaction->delayed_refs; | |
842 | spin_lock(&delayed_refs->lock); | |
843 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
844 | if (head) { | |
845 | if (!mutex_trylock(&head->mutex)) { | |
846 | atomic_inc(&head->node.refs); | |
847 | spin_unlock(&delayed_refs->lock); | |
848 | ||
b3b4aa74 | 849 | btrfs_release_path(path); |
a22285a6 | 850 | |
8cc33e5c DS |
851 | /* |
852 | * Mutex was contended, block until it's released and try | |
853 | * again | |
854 | */ | |
a22285a6 YZ |
855 | mutex_lock(&head->mutex); |
856 | mutex_unlock(&head->mutex); | |
857 | btrfs_put_delayed_ref(&head->node); | |
858 | goto again; | |
859 | } | |
860 | if (head->extent_op && head->extent_op->update_flags) | |
861 | extent_flags |= head->extent_op->flags_to_set; | |
862 | else | |
863 | BUG_ON(num_refs == 0); | |
864 | ||
865 | num_refs += head->node.ref_mod; | |
866 | mutex_unlock(&head->mutex); | |
867 | } | |
868 | spin_unlock(&delayed_refs->lock); | |
869 | out: | |
870 | WARN_ON(num_refs == 0); | |
871 | if (refs) | |
872 | *refs = num_refs; | |
873 | if (flags) | |
874 | *flags = extent_flags; | |
875 | out_free: | |
876 | btrfs_free_path(path); | |
877 | return ret; | |
878 | } | |
879 | ||
d8d5f3e1 CM |
880 | /* |
881 | * Back reference rules. Back refs have three main goals: | |
882 | * | |
883 | * 1) differentiate between all holders of references to an extent so that | |
884 | * when a reference is dropped we can make sure it was a valid reference | |
885 | * before freeing the extent. | |
886 | * | |
887 | * 2) Provide enough information to quickly find the holders of an extent | |
888 | * if we notice a given block is corrupted or bad. | |
889 | * | |
890 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
891 | * maintenance. This is actually the same as #2, but with a slightly | |
892 | * different use case. | |
893 | * | |
5d4f98a2 YZ |
894 | * There are two kinds of back refs. The implicit back refs is optimized |
895 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
896 | * back refs of this kind provide information about the block's owner tree | |
897 | * and the pointer's key. These information allow us to find the block by | |
898 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
899 | * referenced by their owner trees. The location of tree block is recorded | |
900 | * in the back refs. Actually the full back refs is generic, and can be | |
901 | * used in all cases the implicit back refs is used. The major shortcoming | |
902 | * of the full back refs is its overhead. Every time a tree block gets | |
903 | * COWed, we have to update back refs entry for all pointers in it. | |
904 | * | |
905 | * For a newly allocated tree block, we use implicit back refs for | |
906 | * pointers in it. This means most tree related operations only involve | |
907 | * implicit back refs. For a tree block created in old transaction, the | |
908 | * only way to drop a reference to it is COW it. So we can detect the | |
909 | * event that tree block loses its owner tree's reference and do the | |
910 | * back refs conversion. | |
911 | * | |
912 | * When a tree block is COW'd through a tree, there are four cases: | |
913 | * | |
914 | * The reference count of the block is one and the tree is the block's | |
915 | * owner tree. Nothing to do in this case. | |
916 | * | |
917 | * The reference count of the block is one and the tree is not the | |
918 | * block's owner tree. In this case, full back refs is used for pointers | |
919 | * in the block. Remove these full back refs, add implicit back refs for | |
920 | * every pointers in the new block. | |
921 | * | |
922 | * The reference count of the block is greater than one and the tree is | |
923 | * the block's owner tree. In this case, implicit back refs is used for | |
924 | * pointers in the block. Add full back refs for every pointers in the | |
925 | * block, increase lower level extents' reference counts. The original | |
926 | * implicit back refs are entailed to the new block. | |
927 | * | |
928 | * The reference count of the block is greater than one and the tree is | |
929 | * not the block's owner tree. Add implicit back refs for every pointer in | |
930 | * the new block, increase lower level extents' reference count. | |
931 | * | |
932 | * Back Reference Key composing: | |
933 | * | |
934 | * The key objectid corresponds to the first byte in the extent, | |
935 | * The key type is used to differentiate between types of back refs. | |
936 | * There are different meanings of the key offset for different types | |
937 | * of back refs. | |
938 | * | |
d8d5f3e1 CM |
939 | * File extents can be referenced by: |
940 | * | |
941 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 942 | * - different files inside a single subvolume |
d8d5f3e1 CM |
943 | * - different offsets inside a file (bookend extents in file.c) |
944 | * | |
5d4f98a2 | 945 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
946 | * |
947 | * - Objectid of the subvolume root | |
d8d5f3e1 | 948 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
949 | * - original offset in the file |
950 | * - how many bookend extents | |
d8d5f3e1 | 951 | * |
5d4f98a2 YZ |
952 | * The key offset for the implicit back refs is hash of the first |
953 | * three fields. | |
d8d5f3e1 | 954 | * |
5d4f98a2 | 955 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 956 | * |
5d4f98a2 | 957 | * - number of pointers in the tree leaf |
d8d5f3e1 | 958 | * |
5d4f98a2 YZ |
959 | * The key offset for the implicit back refs is the first byte of |
960 | * the tree leaf | |
d8d5f3e1 | 961 | * |
5d4f98a2 YZ |
962 | * When a file extent is allocated, The implicit back refs is used. |
963 | * the fields are filled in: | |
d8d5f3e1 | 964 | * |
5d4f98a2 | 965 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 966 | * |
5d4f98a2 YZ |
967 | * When a file extent is removed file truncation, we find the |
968 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 969 | * |
5d4f98a2 | 970 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 971 | * |
5d4f98a2 | 972 | * Btree extents can be referenced by: |
d8d5f3e1 | 973 | * |
5d4f98a2 | 974 | * - Different subvolumes |
d8d5f3e1 | 975 | * |
5d4f98a2 YZ |
976 | * Both the implicit back refs and the full back refs for tree blocks |
977 | * only consist of key. The key offset for the implicit back refs is | |
978 | * objectid of block's owner tree. The key offset for the full back refs | |
979 | * is the first byte of parent block. | |
d8d5f3e1 | 980 | * |
5d4f98a2 YZ |
981 | * When implicit back refs is used, information about the lowest key and |
982 | * level of the tree block are required. These information are stored in | |
983 | * tree block info structure. | |
d8d5f3e1 | 984 | */ |
31840ae1 | 985 | |
5d4f98a2 YZ |
986 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
987 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
988 | struct btrfs_root *root, | |
989 | struct btrfs_path *path, | |
990 | u64 owner, u32 extra_size) | |
7bb86316 | 991 | { |
5d4f98a2 YZ |
992 | struct btrfs_extent_item *item; |
993 | struct btrfs_extent_item_v0 *ei0; | |
994 | struct btrfs_extent_ref_v0 *ref0; | |
995 | struct btrfs_tree_block_info *bi; | |
996 | struct extent_buffer *leaf; | |
7bb86316 | 997 | struct btrfs_key key; |
5d4f98a2 YZ |
998 | struct btrfs_key found_key; |
999 | u32 new_size = sizeof(*item); | |
1000 | u64 refs; | |
1001 | int ret; | |
1002 | ||
1003 | leaf = path->nodes[0]; | |
1004 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
1005 | ||
1006 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1007 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
1008 | struct btrfs_extent_item_v0); | |
1009 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
1010 | ||
1011 | if (owner == (u64)-1) { | |
1012 | while (1) { | |
1013 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1014 | ret = btrfs_next_leaf(root, path); | |
1015 | if (ret < 0) | |
1016 | return ret; | |
79787eaa | 1017 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
1018 | leaf = path->nodes[0]; |
1019 | } | |
1020 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1021 | path->slots[0]); | |
1022 | BUG_ON(key.objectid != found_key.objectid); | |
1023 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
1024 | path->slots[0]++; | |
1025 | continue; | |
1026 | } | |
1027 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1028 | struct btrfs_extent_ref_v0); | |
1029 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
1030 | break; | |
1031 | } | |
1032 | } | |
b3b4aa74 | 1033 | btrfs_release_path(path); |
5d4f98a2 YZ |
1034 | |
1035 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1036 | new_size += sizeof(*bi); | |
1037 | ||
1038 | new_size -= sizeof(*ei0); | |
1039 | ret = btrfs_search_slot(trans, root, &key, path, | |
1040 | new_size + extra_size, 1); | |
1041 | if (ret < 0) | |
1042 | return ret; | |
79787eaa | 1043 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 1044 | |
4b90c680 | 1045 | btrfs_extend_item(root, path, new_size); |
5d4f98a2 YZ |
1046 | |
1047 | leaf = path->nodes[0]; | |
1048 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1049 | btrfs_set_extent_refs(leaf, item, refs); | |
1050 | /* FIXME: get real generation */ | |
1051 | btrfs_set_extent_generation(leaf, item, 0); | |
1052 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1053 | btrfs_set_extent_flags(leaf, item, | |
1054 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1055 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1056 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1057 | /* FIXME: get first key of the block */ | |
1058 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
1059 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
1060 | } else { | |
1061 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1062 | } | |
1063 | btrfs_mark_buffer_dirty(leaf); | |
1064 | return 0; | |
1065 | } | |
1066 | #endif | |
1067 | ||
1068 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
1069 | { | |
1070 | u32 high_crc = ~(u32)0; | |
1071 | u32 low_crc = ~(u32)0; | |
1072 | __le64 lenum; | |
1073 | ||
1074 | lenum = cpu_to_le64(root_objectid); | |
163e783e | 1075 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1076 | lenum = cpu_to_le64(owner); |
163e783e | 1077 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1078 | lenum = cpu_to_le64(offset); |
163e783e | 1079 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1080 | |
1081 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1082 | } | |
1083 | ||
1084 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1085 | struct btrfs_extent_data_ref *ref) | |
1086 | { | |
1087 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1088 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1089 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1090 | } | |
1091 | ||
1092 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1093 | struct btrfs_extent_data_ref *ref, | |
1094 | u64 root_objectid, u64 owner, u64 offset) | |
1095 | { | |
1096 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1097 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1098 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1099 | return 0; | |
1100 | return 1; | |
1101 | } | |
1102 | ||
1103 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
1104 | struct btrfs_root *root, | |
1105 | struct btrfs_path *path, | |
1106 | u64 bytenr, u64 parent, | |
1107 | u64 root_objectid, | |
1108 | u64 owner, u64 offset) | |
1109 | { | |
1110 | struct btrfs_key key; | |
1111 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1112 | struct extent_buffer *leaf; |
5d4f98a2 | 1113 | u32 nritems; |
74493f7a | 1114 | int ret; |
5d4f98a2 YZ |
1115 | int recow; |
1116 | int err = -ENOENT; | |
74493f7a | 1117 | |
31840ae1 | 1118 | key.objectid = bytenr; |
5d4f98a2 YZ |
1119 | if (parent) { |
1120 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1121 | key.offset = parent; | |
1122 | } else { | |
1123 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1124 | key.offset = hash_extent_data_ref(root_objectid, | |
1125 | owner, offset); | |
1126 | } | |
1127 | again: | |
1128 | recow = 0; | |
1129 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1130 | if (ret < 0) { | |
1131 | err = ret; | |
1132 | goto fail; | |
1133 | } | |
31840ae1 | 1134 | |
5d4f98a2 YZ |
1135 | if (parent) { |
1136 | if (!ret) | |
1137 | return 0; | |
1138 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1139 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1140 | btrfs_release_path(path); |
5d4f98a2 YZ |
1141 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1142 | if (ret < 0) { | |
1143 | err = ret; | |
1144 | goto fail; | |
1145 | } | |
1146 | if (!ret) | |
1147 | return 0; | |
1148 | #endif | |
1149 | goto fail; | |
31840ae1 ZY |
1150 | } |
1151 | ||
1152 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1153 | nritems = btrfs_header_nritems(leaf); |
1154 | while (1) { | |
1155 | if (path->slots[0] >= nritems) { | |
1156 | ret = btrfs_next_leaf(root, path); | |
1157 | if (ret < 0) | |
1158 | err = ret; | |
1159 | if (ret) | |
1160 | goto fail; | |
1161 | ||
1162 | leaf = path->nodes[0]; | |
1163 | nritems = btrfs_header_nritems(leaf); | |
1164 | recow = 1; | |
1165 | } | |
1166 | ||
1167 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1168 | if (key.objectid != bytenr || | |
1169 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1170 | goto fail; | |
1171 | ||
1172 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1173 | struct btrfs_extent_data_ref); | |
1174 | ||
1175 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1176 | owner, offset)) { | |
1177 | if (recow) { | |
b3b4aa74 | 1178 | btrfs_release_path(path); |
5d4f98a2 YZ |
1179 | goto again; |
1180 | } | |
1181 | err = 0; | |
1182 | break; | |
1183 | } | |
1184 | path->slots[0]++; | |
31840ae1 | 1185 | } |
5d4f98a2 YZ |
1186 | fail: |
1187 | return err; | |
31840ae1 ZY |
1188 | } |
1189 | ||
5d4f98a2 YZ |
1190 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
1191 | struct btrfs_root *root, | |
1192 | struct btrfs_path *path, | |
1193 | u64 bytenr, u64 parent, | |
1194 | u64 root_objectid, u64 owner, | |
1195 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
1196 | { |
1197 | struct btrfs_key key; | |
1198 | struct extent_buffer *leaf; | |
5d4f98a2 | 1199 | u32 size; |
31840ae1 ZY |
1200 | u32 num_refs; |
1201 | int ret; | |
74493f7a | 1202 | |
74493f7a | 1203 | key.objectid = bytenr; |
5d4f98a2 YZ |
1204 | if (parent) { |
1205 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1206 | key.offset = parent; | |
1207 | size = sizeof(struct btrfs_shared_data_ref); | |
1208 | } else { | |
1209 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1210 | key.offset = hash_extent_data_ref(root_objectid, | |
1211 | owner, offset); | |
1212 | size = sizeof(struct btrfs_extent_data_ref); | |
1213 | } | |
74493f7a | 1214 | |
5d4f98a2 YZ |
1215 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1216 | if (ret && ret != -EEXIST) | |
1217 | goto fail; | |
1218 | ||
1219 | leaf = path->nodes[0]; | |
1220 | if (parent) { | |
1221 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1222 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1223 | struct btrfs_shared_data_ref); |
1224 | if (ret == 0) { | |
1225 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1226 | } else { | |
1227 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1228 | num_refs += refs_to_add; | |
1229 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1230 | } |
5d4f98a2 YZ |
1231 | } else { |
1232 | struct btrfs_extent_data_ref *ref; | |
1233 | while (ret == -EEXIST) { | |
1234 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1235 | struct btrfs_extent_data_ref); | |
1236 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1237 | owner, offset)) | |
1238 | break; | |
b3b4aa74 | 1239 | btrfs_release_path(path); |
5d4f98a2 YZ |
1240 | key.offset++; |
1241 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1242 | size); | |
1243 | if (ret && ret != -EEXIST) | |
1244 | goto fail; | |
31840ae1 | 1245 | |
5d4f98a2 YZ |
1246 | leaf = path->nodes[0]; |
1247 | } | |
1248 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1249 | struct btrfs_extent_data_ref); | |
1250 | if (ret == 0) { | |
1251 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1252 | root_objectid); | |
1253 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1254 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1255 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1256 | } else { | |
1257 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1258 | num_refs += refs_to_add; | |
1259 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1260 | } |
31840ae1 | 1261 | } |
5d4f98a2 YZ |
1262 | btrfs_mark_buffer_dirty(leaf); |
1263 | ret = 0; | |
1264 | fail: | |
b3b4aa74 | 1265 | btrfs_release_path(path); |
7bb86316 | 1266 | return ret; |
74493f7a CM |
1267 | } |
1268 | ||
5d4f98a2 YZ |
1269 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
1270 | struct btrfs_root *root, | |
1271 | struct btrfs_path *path, | |
1272 | int refs_to_drop) | |
31840ae1 | 1273 | { |
5d4f98a2 YZ |
1274 | struct btrfs_key key; |
1275 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1276 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1277 | struct extent_buffer *leaf; |
5d4f98a2 | 1278 | u32 num_refs = 0; |
31840ae1 ZY |
1279 | int ret = 0; |
1280 | ||
1281 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1282 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1283 | ||
1284 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1285 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1286 | struct btrfs_extent_data_ref); | |
1287 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1288 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1289 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1290 | struct btrfs_shared_data_ref); | |
1291 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1292 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1293 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1294 | struct btrfs_extent_ref_v0 *ref0; | |
1295 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1296 | struct btrfs_extent_ref_v0); | |
1297 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1298 | #endif | |
1299 | } else { | |
1300 | BUG(); | |
1301 | } | |
1302 | ||
56bec294 CM |
1303 | BUG_ON(num_refs < refs_to_drop); |
1304 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1305 | |
31840ae1 ZY |
1306 | if (num_refs == 0) { |
1307 | ret = btrfs_del_item(trans, root, path); | |
1308 | } else { | |
5d4f98a2 YZ |
1309 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1310 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1311 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1312 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1313 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1314 | else { | |
1315 | struct btrfs_extent_ref_v0 *ref0; | |
1316 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1317 | struct btrfs_extent_ref_v0); | |
1318 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1319 | } | |
1320 | #endif | |
31840ae1 ZY |
1321 | btrfs_mark_buffer_dirty(leaf); |
1322 | } | |
31840ae1 ZY |
1323 | return ret; |
1324 | } | |
1325 | ||
5d4f98a2 YZ |
1326 | static noinline u32 extent_data_ref_count(struct btrfs_root *root, |
1327 | struct btrfs_path *path, | |
1328 | struct btrfs_extent_inline_ref *iref) | |
15916de8 | 1329 | { |
5d4f98a2 YZ |
1330 | struct btrfs_key key; |
1331 | struct extent_buffer *leaf; | |
1332 | struct btrfs_extent_data_ref *ref1; | |
1333 | struct btrfs_shared_data_ref *ref2; | |
1334 | u32 num_refs = 0; | |
1335 | ||
1336 | leaf = path->nodes[0]; | |
1337 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1338 | if (iref) { | |
1339 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1340 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1341 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1342 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1343 | } else { | |
1344 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1345 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1346 | } | |
1347 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1348 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1349 | struct btrfs_extent_data_ref); | |
1350 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1351 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1352 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1353 | struct btrfs_shared_data_ref); | |
1354 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1355 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1356 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1357 | struct btrfs_extent_ref_v0 *ref0; | |
1358 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1359 | struct btrfs_extent_ref_v0); | |
1360 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1361 | #endif |
5d4f98a2 YZ |
1362 | } else { |
1363 | WARN_ON(1); | |
1364 | } | |
1365 | return num_refs; | |
1366 | } | |
15916de8 | 1367 | |
5d4f98a2 YZ |
1368 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1369 | struct btrfs_root *root, | |
1370 | struct btrfs_path *path, | |
1371 | u64 bytenr, u64 parent, | |
1372 | u64 root_objectid) | |
1f3c79a2 | 1373 | { |
5d4f98a2 | 1374 | struct btrfs_key key; |
1f3c79a2 | 1375 | int ret; |
1f3c79a2 | 1376 | |
5d4f98a2 YZ |
1377 | key.objectid = bytenr; |
1378 | if (parent) { | |
1379 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1380 | key.offset = parent; | |
1381 | } else { | |
1382 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1383 | key.offset = root_objectid; | |
1f3c79a2 LH |
1384 | } |
1385 | ||
5d4f98a2 YZ |
1386 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1387 | if (ret > 0) | |
1388 | ret = -ENOENT; | |
1389 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1390 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1391 | btrfs_release_path(path); |
5d4f98a2 YZ |
1392 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1393 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1394 | if (ret > 0) | |
1395 | ret = -ENOENT; | |
1396 | } | |
1f3c79a2 | 1397 | #endif |
5d4f98a2 | 1398 | return ret; |
1f3c79a2 LH |
1399 | } |
1400 | ||
5d4f98a2 YZ |
1401 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1402 | struct btrfs_root *root, | |
1403 | struct btrfs_path *path, | |
1404 | u64 bytenr, u64 parent, | |
1405 | u64 root_objectid) | |
31840ae1 | 1406 | { |
5d4f98a2 | 1407 | struct btrfs_key key; |
31840ae1 | 1408 | int ret; |
31840ae1 | 1409 | |
5d4f98a2 YZ |
1410 | key.objectid = bytenr; |
1411 | if (parent) { | |
1412 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1413 | key.offset = parent; | |
1414 | } else { | |
1415 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1416 | key.offset = root_objectid; | |
1417 | } | |
1418 | ||
1419 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
b3b4aa74 | 1420 | btrfs_release_path(path); |
31840ae1 ZY |
1421 | return ret; |
1422 | } | |
1423 | ||
5d4f98a2 | 1424 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1425 | { |
5d4f98a2 YZ |
1426 | int type; |
1427 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1428 | if (parent > 0) | |
1429 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1430 | else | |
1431 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1432 | } else { | |
1433 | if (parent > 0) | |
1434 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1435 | else | |
1436 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1437 | } | |
1438 | return type; | |
31840ae1 | 1439 | } |
56bec294 | 1440 | |
2c47e605 YZ |
1441 | static int find_next_key(struct btrfs_path *path, int level, |
1442 | struct btrfs_key *key) | |
56bec294 | 1443 | |
02217ed2 | 1444 | { |
2c47e605 | 1445 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1446 | if (!path->nodes[level]) |
1447 | break; | |
5d4f98a2 YZ |
1448 | if (path->slots[level] + 1 >= |
1449 | btrfs_header_nritems(path->nodes[level])) | |
1450 | continue; | |
1451 | if (level == 0) | |
1452 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1453 | path->slots[level] + 1); | |
1454 | else | |
1455 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1456 | path->slots[level] + 1); | |
1457 | return 0; | |
1458 | } | |
1459 | return 1; | |
1460 | } | |
037e6390 | 1461 | |
5d4f98a2 YZ |
1462 | /* |
1463 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1464 | * to the address of inline back ref, and 0 is returned. | |
1465 | * | |
1466 | * if back ref isn't found, *ref_ret is set to the address where it | |
1467 | * should be inserted, and -ENOENT is returned. | |
1468 | * | |
1469 | * if insert is true and there are too many inline back refs, the path | |
1470 | * points to the extent item, and -EAGAIN is returned. | |
1471 | * | |
1472 | * NOTE: inline back refs are ordered in the same way that back ref | |
1473 | * items in the tree are ordered. | |
1474 | */ | |
1475 | static noinline_for_stack | |
1476 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1477 | struct btrfs_root *root, | |
1478 | struct btrfs_path *path, | |
1479 | struct btrfs_extent_inline_ref **ref_ret, | |
1480 | u64 bytenr, u64 num_bytes, | |
1481 | u64 parent, u64 root_objectid, | |
1482 | u64 owner, u64 offset, int insert) | |
1483 | { | |
1484 | struct btrfs_key key; | |
1485 | struct extent_buffer *leaf; | |
1486 | struct btrfs_extent_item *ei; | |
1487 | struct btrfs_extent_inline_ref *iref; | |
1488 | u64 flags; | |
1489 | u64 item_size; | |
1490 | unsigned long ptr; | |
1491 | unsigned long end; | |
1492 | int extra_size; | |
1493 | int type; | |
1494 | int want; | |
1495 | int ret; | |
1496 | int err = 0; | |
3173a18f JB |
1497 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
1498 | SKINNY_METADATA); | |
26b8003f | 1499 | |
db94535d | 1500 | key.objectid = bytenr; |
31840ae1 | 1501 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1502 | key.offset = num_bytes; |
31840ae1 | 1503 | |
5d4f98a2 YZ |
1504 | want = extent_ref_type(parent, owner); |
1505 | if (insert) { | |
1506 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1507 | path->keep_locks = 1; |
5d4f98a2 YZ |
1508 | } else |
1509 | extra_size = -1; | |
3173a18f JB |
1510 | |
1511 | /* | |
1512 | * Owner is our parent level, so we can just add one to get the level | |
1513 | * for the block we are interested in. | |
1514 | */ | |
1515 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1516 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1517 | key.offset = owner; | |
1518 | } | |
1519 | ||
1520 | again: | |
5d4f98a2 | 1521 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1522 | if (ret < 0) { |
5d4f98a2 YZ |
1523 | err = ret; |
1524 | goto out; | |
1525 | } | |
3173a18f JB |
1526 | |
1527 | /* | |
1528 | * We may be a newly converted file system which still has the old fat | |
1529 | * extent entries for metadata, so try and see if we have one of those. | |
1530 | */ | |
1531 | if (ret > 0 && skinny_metadata) { | |
1532 | skinny_metadata = false; | |
1533 | if (path->slots[0]) { | |
1534 | path->slots[0]--; | |
1535 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1536 | path->slots[0]); | |
1537 | if (key.objectid == bytenr && | |
1538 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1539 | key.offset == num_bytes) | |
1540 | ret = 0; | |
1541 | } | |
1542 | if (ret) { | |
1543 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1544 | key.offset = num_bytes; | |
1545 | btrfs_release_path(path); | |
1546 | goto again; | |
1547 | } | |
1548 | } | |
1549 | ||
79787eaa JM |
1550 | if (ret && !insert) { |
1551 | err = -ENOENT; | |
1552 | goto out; | |
492104c8 JB |
1553 | } else if (ret) { |
1554 | err = -EIO; | |
1555 | WARN_ON(1); | |
1556 | goto out; | |
79787eaa | 1557 | } |
5d4f98a2 YZ |
1558 | |
1559 | leaf = path->nodes[0]; | |
1560 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1561 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1562 | if (item_size < sizeof(*ei)) { | |
1563 | if (!insert) { | |
1564 | err = -ENOENT; | |
1565 | goto out; | |
1566 | } | |
1567 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1568 | extra_size); | |
1569 | if (ret < 0) { | |
1570 | err = ret; | |
1571 | goto out; | |
1572 | } | |
1573 | leaf = path->nodes[0]; | |
1574 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1575 | } | |
1576 | #endif | |
1577 | BUG_ON(item_size < sizeof(*ei)); | |
1578 | ||
5d4f98a2 YZ |
1579 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1580 | flags = btrfs_extent_flags(leaf, ei); | |
1581 | ||
1582 | ptr = (unsigned long)(ei + 1); | |
1583 | end = (unsigned long)ei + item_size; | |
1584 | ||
3173a18f | 1585 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1586 | ptr += sizeof(struct btrfs_tree_block_info); |
1587 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1588 | } |
1589 | ||
1590 | err = -ENOENT; | |
1591 | while (1) { | |
1592 | if (ptr >= end) { | |
1593 | WARN_ON(ptr > end); | |
1594 | break; | |
1595 | } | |
1596 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1597 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1598 | if (want < type) | |
1599 | break; | |
1600 | if (want > type) { | |
1601 | ptr += btrfs_extent_inline_ref_size(type); | |
1602 | continue; | |
1603 | } | |
1604 | ||
1605 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1606 | struct btrfs_extent_data_ref *dref; | |
1607 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1608 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1609 | owner, offset)) { | |
1610 | err = 0; | |
1611 | break; | |
1612 | } | |
1613 | if (hash_extent_data_ref_item(leaf, dref) < | |
1614 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1615 | break; | |
1616 | } else { | |
1617 | u64 ref_offset; | |
1618 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1619 | if (parent > 0) { | |
1620 | if (parent == ref_offset) { | |
1621 | err = 0; | |
1622 | break; | |
1623 | } | |
1624 | if (ref_offset < parent) | |
1625 | break; | |
1626 | } else { | |
1627 | if (root_objectid == ref_offset) { | |
1628 | err = 0; | |
1629 | break; | |
1630 | } | |
1631 | if (ref_offset < root_objectid) | |
1632 | break; | |
1633 | } | |
1634 | } | |
1635 | ptr += btrfs_extent_inline_ref_size(type); | |
1636 | } | |
1637 | if (err == -ENOENT && insert) { | |
1638 | if (item_size + extra_size >= | |
1639 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1640 | err = -EAGAIN; | |
1641 | goto out; | |
1642 | } | |
1643 | /* | |
1644 | * To add new inline back ref, we have to make sure | |
1645 | * there is no corresponding back ref item. | |
1646 | * For simplicity, we just do not add new inline back | |
1647 | * ref if there is any kind of item for this block | |
1648 | */ | |
2c47e605 YZ |
1649 | if (find_next_key(path, 0, &key) == 0 && |
1650 | key.objectid == bytenr && | |
85d4198e | 1651 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1652 | err = -EAGAIN; |
1653 | goto out; | |
1654 | } | |
1655 | } | |
1656 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1657 | out: | |
85d4198e | 1658 | if (insert) { |
5d4f98a2 YZ |
1659 | path->keep_locks = 0; |
1660 | btrfs_unlock_up_safe(path, 1); | |
1661 | } | |
1662 | return err; | |
1663 | } | |
1664 | ||
1665 | /* | |
1666 | * helper to add new inline back ref | |
1667 | */ | |
1668 | static noinline_for_stack | |
fd279fae | 1669 | void setup_inline_extent_backref(struct btrfs_root *root, |
143bede5 JM |
1670 | struct btrfs_path *path, |
1671 | struct btrfs_extent_inline_ref *iref, | |
1672 | u64 parent, u64 root_objectid, | |
1673 | u64 owner, u64 offset, int refs_to_add, | |
1674 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1675 | { |
1676 | struct extent_buffer *leaf; | |
1677 | struct btrfs_extent_item *ei; | |
1678 | unsigned long ptr; | |
1679 | unsigned long end; | |
1680 | unsigned long item_offset; | |
1681 | u64 refs; | |
1682 | int size; | |
1683 | int type; | |
5d4f98a2 YZ |
1684 | |
1685 | leaf = path->nodes[0]; | |
1686 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1687 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1688 | ||
1689 | type = extent_ref_type(parent, owner); | |
1690 | size = btrfs_extent_inline_ref_size(type); | |
1691 | ||
4b90c680 | 1692 | btrfs_extend_item(root, path, size); |
5d4f98a2 YZ |
1693 | |
1694 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1695 | refs = btrfs_extent_refs(leaf, ei); | |
1696 | refs += refs_to_add; | |
1697 | btrfs_set_extent_refs(leaf, ei, refs); | |
1698 | if (extent_op) | |
1699 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1700 | ||
1701 | ptr = (unsigned long)ei + item_offset; | |
1702 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1703 | if (ptr < end - size) | |
1704 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1705 | end - size - ptr); | |
1706 | ||
1707 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1708 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1709 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1710 | struct btrfs_extent_data_ref *dref; | |
1711 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1712 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1713 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1714 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1715 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1716 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1717 | struct btrfs_shared_data_ref *sref; | |
1718 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1719 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1720 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1721 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1722 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1723 | } else { | |
1724 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1725 | } | |
1726 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1727 | } |
1728 | ||
1729 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1730 | struct btrfs_root *root, | |
1731 | struct btrfs_path *path, | |
1732 | struct btrfs_extent_inline_ref **ref_ret, | |
1733 | u64 bytenr, u64 num_bytes, u64 parent, | |
1734 | u64 root_objectid, u64 owner, u64 offset) | |
1735 | { | |
1736 | int ret; | |
1737 | ||
1738 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1739 | bytenr, num_bytes, parent, | |
1740 | root_objectid, owner, offset, 0); | |
1741 | if (ret != -ENOENT) | |
54aa1f4d | 1742 | return ret; |
5d4f98a2 | 1743 | |
b3b4aa74 | 1744 | btrfs_release_path(path); |
5d4f98a2 YZ |
1745 | *ref_ret = NULL; |
1746 | ||
1747 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1748 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1749 | root_objectid); | |
1750 | } else { | |
1751 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1752 | root_objectid, owner, offset); | |
b9473439 | 1753 | } |
5d4f98a2 YZ |
1754 | return ret; |
1755 | } | |
31840ae1 | 1756 | |
5d4f98a2 YZ |
1757 | /* |
1758 | * helper to update/remove inline back ref | |
1759 | */ | |
1760 | static noinline_for_stack | |
afe5fea7 | 1761 | void update_inline_extent_backref(struct btrfs_root *root, |
143bede5 JM |
1762 | struct btrfs_path *path, |
1763 | struct btrfs_extent_inline_ref *iref, | |
1764 | int refs_to_mod, | |
1765 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1766 | { |
1767 | struct extent_buffer *leaf; | |
1768 | struct btrfs_extent_item *ei; | |
1769 | struct btrfs_extent_data_ref *dref = NULL; | |
1770 | struct btrfs_shared_data_ref *sref = NULL; | |
1771 | unsigned long ptr; | |
1772 | unsigned long end; | |
1773 | u32 item_size; | |
1774 | int size; | |
1775 | int type; | |
5d4f98a2 YZ |
1776 | u64 refs; |
1777 | ||
1778 | leaf = path->nodes[0]; | |
1779 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1780 | refs = btrfs_extent_refs(leaf, ei); | |
1781 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1782 | refs += refs_to_mod; | |
1783 | btrfs_set_extent_refs(leaf, ei, refs); | |
1784 | if (extent_op) | |
1785 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1786 | ||
1787 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1788 | ||
1789 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1790 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1791 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1792 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1793 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1794 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1795 | } else { | |
1796 | refs = 1; | |
1797 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1798 | } |
31840ae1 | 1799 | |
5d4f98a2 YZ |
1800 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1801 | refs += refs_to_mod; | |
1802 | ||
1803 | if (refs > 0) { | |
1804 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1805 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1806 | else | |
1807 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1808 | } else { | |
1809 | size = btrfs_extent_inline_ref_size(type); | |
1810 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1811 | ptr = (unsigned long)iref; | |
1812 | end = (unsigned long)ei + item_size; | |
1813 | if (ptr + size < end) | |
1814 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1815 | end - ptr - size); | |
1816 | item_size -= size; | |
afe5fea7 | 1817 | btrfs_truncate_item(root, path, item_size, 1); |
5d4f98a2 YZ |
1818 | } |
1819 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1820 | } |
1821 | ||
1822 | static noinline_for_stack | |
1823 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1824 | struct btrfs_root *root, | |
1825 | struct btrfs_path *path, | |
1826 | u64 bytenr, u64 num_bytes, u64 parent, | |
1827 | u64 root_objectid, u64 owner, | |
1828 | u64 offset, int refs_to_add, | |
1829 | struct btrfs_delayed_extent_op *extent_op) | |
1830 | { | |
1831 | struct btrfs_extent_inline_ref *iref; | |
1832 | int ret; | |
1833 | ||
1834 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1835 | bytenr, num_bytes, parent, | |
1836 | root_objectid, owner, offset, 1); | |
1837 | if (ret == 0) { | |
1838 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
afe5fea7 | 1839 | update_inline_extent_backref(root, path, iref, |
143bede5 | 1840 | refs_to_add, extent_op); |
5d4f98a2 | 1841 | } else if (ret == -ENOENT) { |
fd279fae | 1842 | setup_inline_extent_backref(root, path, iref, parent, |
143bede5 JM |
1843 | root_objectid, owner, offset, |
1844 | refs_to_add, extent_op); | |
1845 | ret = 0; | |
771ed689 | 1846 | } |
5d4f98a2 YZ |
1847 | return ret; |
1848 | } | |
31840ae1 | 1849 | |
5d4f98a2 YZ |
1850 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1851 | struct btrfs_root *root, | |
1852 | struct btrfs_path *path, | |
1853 | u64 bytenr, u64 parent, u64 root_objectid, | |
1854 | u64 owner, u64 offset, int refs_to_add) | |
1855 | { | |
1856 | int ret; | |
1857 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1858 | BUG_ON(refs_to_add != 1); | |
1859 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1860 | parent, root_objectid); | |
1861 | } else { | |
1862 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1863 | parent, root_objectid, | |
1864 | owner, offset, refs_to_add); | |
1865 | } | |
1866 | return ret; | |
1867 | } | |
56bec294 | 1868 | |
5d4f98a2 YZ |
1869 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1870 | struct btrfs_root *root, | |
1871 | struct btrfs_path *path, | |
1872 | struct btrfs_extent_inline_ref *iref, | |
1873 | int refs_to_drop, int is_data) | |
1874 | { | |
143bede5 | 1875 | int ret = 0; |
b9473439 | 1876 | |
5d4f98a2 YZ |
1877 | BUG_ON(!is_data && refs_to_drop != 1); |
1878 | if (iref) { | |
afe5fea7 | 1879 | update_inline_extent_backref(root, path, iref, |
143bede5 | 1880 | -refs_to_drop, NULL); |
5d4f98a2 YZ |
1881 | } else if (is_data) { |
1882 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop); | |
1883 | } else { | |
1884 | ret = btrfs_del_item(trans, root, path); | |
1885 | } | |
1886 | return ret; | |
1887 | } | |
1888 | ||
5378e607 | 1889 | static int btrfs_issue_discard(struct block_device *bdev, |
5d4f98a2 YZ |
1890 | u64 start, u64 len) |
1891 | { | |
5378e607 | 1892 | return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0); |
5d4f98a2 | 1893 | } |
5d4f98a2 YZ |
1894 | |
1895 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 1896 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1897 | { |
5d4f98a2 | 1898 | int ret; |
5378e607 | 1899 | u64 discarded_bytes = 0; |
a1d3c478 | 1900 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1901 | |
e244a0ae | 1902 | |
5d4f98a2 | 1903 | /* Tell the block device(s) that the sectors can be discarded */ |
3ec706c8 | 1904 | ret = btrfs_map_block(root->fs_info, REQ_DISCARD, |
a1d3c478 | 1905 | bytenr, &num_bytes, &bbio, 0); |
79787eaa | 1906 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 1907 | if (!ret) { |
a1d3c478 | 1908 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
1909 | int i; |
1910 | ||
5d4f98a2 | 1911 | |
a1d3c478 | 1912 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d5e2003c JB |
1913 | if (!stripe->dev->can_discard) |
1914 | continue; | |
1915 | ||
5378e607 LD |
1916 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1917 | stripe->physical, | |
1918 | stripe->length); | |
1919 | if (!ret) | |
1920 | discarded_bytes += stripe->length; | |
1921 | else if (ret != -EOPNOTSUPP) | |
79787eaa | 1922 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
1923 | |
1924 | /* | |
1925 | * Just in case we get back EOPNOTSUPP for some reason, | |
1926 | * just ignore the return value so we don't screw up | |
1927 | * people calling discard_extent. | |
1928 | */ | |
1929 | ret = 0; | |
5d4f98a2 | 1930 | } |
a1d3c478 | 1931 | kfree(bbio); |
5d4f98a2 | 1932 | } |
5378e607 LD |
1933 | |
1934 | if (actual_bytes) | |
1935 | *actual_bytes = discarded_bytes; | |
1936 | ||
5d4f98a2 | 1937 | |
53b381b3 DW |
1938 | if (ret == -EOPNOTSUPP) |
1939 | ret = 0; | |
5d4f98a2 | 1940 | return ret; |
5d4f98a2 YZ |
1941 | } |
1942 | ||
79787eaa | 1943 | /* Can return -ENOMEM */ |
5d4f98a2 YZ |
1944 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
1945 | struct btrfs_root *root, | |
1946 | u64 bytenr, u64 num_bytes, u64 parent, | |
66d7e7f0 | 1947 | u64 root_objectid, u64 owner, u64 offset, int for_cow) |
5d4f98a2 YZ |
1948 | { |
1949 | int ret; | |
66d7e7f0 AJ |
1950 | struct btrfs_fs_info *fs_info = root->fs_info; |
1951 | ||
5d4f98a2 YZ |
1952 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
1953 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
1954 | ||
1955 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
66d7e7f0 AJ |
1956 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
1957 | num_bytes, | |
5d4f98a2 | 1958 | parent, root_objectid, (int)owner, |
66d7e7f0 | 1959 | BTRFS_ADD_DELAYED_REF, NULL, for_cow); |
5d4f98a2 | 1960 | } else { |
66d7e7f0 AJ |
1961 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
1962 | num_bytes, | |
5d4f98a2 | 1963 | parent, root_objectid, owner, offset, |
66d7e7f0 | 1964 | BTRFS_ADD_DELAYED_REF, NULL, for_cow); |
5d4f98a2 YZ |
1965 | } |
1966 | return ret; | |
1967 | } | |
1968 | ||
1969 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1970 | struct btrfs_root *root, | |
1971 | u64 bytenr, u64 num_bytes, | |
1972 | u64 parent, u64 root_objectid, | |
1973 | u64 owner, u64 offset, int refs_to_add, | |
1974 | struct btrfs_delayed_extent_op *extent_op) | |
1975 | { | |
1976 | struct btrfs_path *path; | |
1977 | struct extent_buffer *leaf; | |
1978 | struct btrfs_extent_item *item; | |
1979 | u64 refs; | |
1980 | int ret; | |
5d4f98a2 YZ |
1981 | |
1982 | path = btrfs_alloc_path(); | |
1983 | if (!path) | |
1984 | return -ENOMEM; | |
1985 | ||
1986 | path->reada = 1; | |
1987 | path->leave_spinning = 1; | |
1988 | /* this will setup the path even if it fails to insert the back ref */ | |
1989 | ret = insert_inline_extent_backref(trans, root->fs_info->extent_root, | |
1990 | path, bytenr, num_bytes, parent, | |
1991 | root_objectid, owner, offset, | |
1992 | refs_to_add, extent_op); | |
30d133fc | 1993 | if (ret != -EAGAIN) |
5d4f98a2 YZ |
1994 | goto out; |
1995 | ||
5d4f98a2 YZ |
1996 | leaf = path->nodes[0]; |
1997 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1998 | refs = btrfs_extent_refs(leaf, item); | |
1999 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2000 | if (extent_op) | |
2001 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2002 | |
5d4f98a2 | 2003 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2004 | btrfs_release_path(path); |
56bec294 CM |
2005 | |
2006 | path->reada = 1; | |
b9473439 CM |
2007 | path->leave_spinning = 1; |
2008 | ||
56bec294 CM |
2009 | /* now insert the actual backref */ |
2010 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
2011 | path, bytenr, parent, root_objectid, |
2012 | owner, offset, refs_to_add); | |
79787eaa JM |
2013 | if (ret) |
2014 | btrfs_abort_transaction(trans, root, ret); | |
5d4f98a2 | 2015 | out: |
56bec294 | 2016 | btrfs_free_path(path); |
30d133fc | 2017 | return ret; |
56bec294 CM |
2018 | } |
2019 | ||
5d4f98a2 YZ |
2020 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
2021 | struct btrfs_root *root, | |
2022 | struct btrfs_delayed_ref_node *node, | |
2023 | struct btrfs_delayed_extent_op *extent_op, | |
2024 | int insert_reserved) | |
56bec294 | 2025 | { |
5d4f98a2 YZ |
2026 | int ret = 0; |
2027 | struct btrfs_delayed_data_ref *ref; | |
2028 | struct btrfs_key ins; | |
2029 | u64 parent = 0; | |
2030 | u64 ref_root = 0; | |
2031 | u64 flags = 0; | |
2032 | ||
2033 | ins.objectid = node->bytenr; | |
2034 | ins.offset = node->num_bytes; | |
2035 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2036 | ||
2037 | ref = btrfs_delayed_node_to_data_ref(node); | |
599c75ec LB |
2038 | trace_run_delayed_data_ref(node, ref, node->action); |
2039 | ||
5d4f98a2 YZ |
2040 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2041 | parent = ref->parent; | |
2042 | else | |
2043 | ref_root = ref->root; | |
2044 | ||
2045 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2046 | if (extent_op) |
5d4f98a2 | 2047 | flags |= extent_op->flags_to_set; |
5d4f98a2 YZ |
2048 | ret = alloc_reserved_file_extent(trans, root, |
2049 | parent, ref_root, flags, | |
2050 | ref->objectid, ref->offset, | |
2051 | &ins, node->ref_mod); | |
5d4f98a2 YZ |
2052 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2053 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
2054 | node->num_bytes, parent, | |
2055 | ref_root, ref->objectid, | |
2056 | ref->offset, node->ref_mod, | |
2057 | extent_op); | |
2058 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
2059 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
2060 | node->num_bytes, parent, | |
2061 | ref_root, ref->objectid, | |
2062 | ref->offset, node->ref_mod, | |
2063 | extent_op); | |
2064 | } else { | |
2065 | BUG(); | |
2066 | } | |
2067 | return ret; | |
2068 | } | |
2069 | ||
2070 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2071 | struct extent_buffer *leaf, | |
2072 | struct btrfs_extent_item *ei) | |
2073 | { | |
2074 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2075 | if (extent_op->update_flags) { | |
2076 | flags |= extent_op->flags_to_set; | |
2077 | btrfs_set_extent_flags(leaf, ei, flags); | |
2078 | } | |
2079 | ||
2080 | if (extent_op->update_key) { | |
2081 | struct btrfs_tree_block_info *bi; | |
2082 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2083 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2084 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2085 | } | |
2086 | } | |
2087 | ||
2088 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2089 | struct btrfs_root *root, | |
2090 | struct btrfs_delayed_ref_node *node, | |
2091 | struct btrfs_delayed_extent_op *extent_op) | |
2092 | { | |
2093 | struct btrfs_key key; | |
2094 | struct btrfs_path *path; | |
2095 | struct btrfs_extent_item *ei; | |
2096 | struct extent_buffer *leaf; | |
2097 | u32 item_size; | |
56bec294 | 2098 | int ret; |
5d4f98a2 | 2099 | int err = 0; |
b1c79e09 | 2100 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2101 | |
79787eaa JM |
2102 | if (trans->aborted) |
2103 | return 0; | |
2104 | ||
3173a18f JB |
2105 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) |
2106 | metadata = 0; | |
2107 | ||
5d4f98a2 YZ |
2108 | path = btrfs_alloc_path(); |
2109 | if (!path) | |
2110 | return -ENOMEM; | |
2111 | ||
2112 | key.objectid = node->bytenr; | |
5d4f98a2 | 2113 | |
3173a18f | 2114 | if (metadata) { |
3173a18f | 2115 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2116 | key.offset = extent_op->level; |
3173a18f JB |
2117 | } else { |
2118 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2119 | key.offset = node->num_bytes; | |
2120 | } | |
2121 | ||
2122 | again: | |
5d4f98a2 YZ |
2123 | path->reada = 1; |
2124 | path->leave_spinning = 1; | |
2125 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
2126 | path, 0, 1); | |
2127 | if (ret < 0) { | |
2128 | err = ret; | |
2129 | goto out; | |
2130 | } | |
2131 | if (ret > 0) { | |
3173a18f | 2132 | if (metadata) { |
55994887 FDBM |
2133 | if (path->slots[0] > 0) { |
2134 | path->slots[0]--; | |
2135 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2136 | path->slots[0]); | |
2137 | if (key.objectid == node->bytenr && | |
2138 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
2139 | key.offset == node->num_bytes) | |
2140 | ret = 0; | |
2141 | } | |
2142 | if (ret > 0) { | |
2143 | btrfs_release_path(path); | |
2144 | metadata = 0; | |
3173a18f | 2145 | |
55994887 FDBM |
2146 | key.objectid = node->bytenr; |
2147 | key.offset = node->num_bytes; | |
2148 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2149 | goto again; | |
2150 | } | |
2151 | } else { | |
2152 | err = -EIO; | |
2153 | goto out; | |
3173a18f | 2154 | } |
5d4f98a2 YZ |
2155 | } |
2156 | ||
2157 | leaf = path->nodes[0]; | |
2158 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2159 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2160 | if (item_size < sizeof(*ei)) { | |
2161 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
2162 | path, (u64)-1, 0); | |
2163 | if (ret < 0) { | |
2164 | err = ret; | |
2165 | goto out; | |
2166 | } | |
2167 | leaf = path->nodes[0]; | |
2168 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2169 | } | |
2170 | #endif | |
2171 | BUG_ON(item_size < sizeof(*ei)); | |
2172 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2173 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2174 | |
5d4f98a2 YZ |
2175 | btrfs_mark_buffer_dirty(leaf); |
2176 | out: | |
2177 | btrfs_free_path(path); | |
2178 | return err; | |
56bec294 CM |
2179 | } |
2180 | ||
5d4f98a2 YZ |
2181 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2182 | struct btrfs_root *root, | |
2183 | struct btrfs_delayed_ref_node *node, | |
2184 | struct btrfs_delayed_extent_op *extent_op, | |
2185 | int insert_reserved) | |
56bec294 CM |
2186 | { |
2187 | int ret = 0; | |
5d4f98a2 YZ |
2188 | struct btrfs_delayed_tree_ref *ref; |
2189 | struct btrfs_key ins; | |
2190 | u64 parent = 0; | |
2191 | u64 ref_root = 0; | |
3173a18f JB |
2192 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
2193 | SKINNY_METADATA); | |
56bec294 | 2194 | |
5d4f98a2 | 2195 | ref = btrfs_delayed_node_to_tree_ref(node); |
599c75ec LB |
2196 | trace_run_delayed_tree_ref(node, ref, node->action); |
2197 | ||
5d4f98a2 YZ |
2198 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2199 | parent = ref->parent; | |
2200 | else | |
2201 | ref_root = ref->root; | |
2202 | ||
3173a18f JB |
2203 | ins.objectid = node->bytenr; |
2204 | if (skinny_metadata) { | |
2205 | ins.offset = ref->level; | |
2206 | ins.type = BTRFS_METADATA_ITEM_KEY; | |
2207 | } else { | |
2208 | ins.offset = node->num_bytes; | |
2209 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2210 | } | |
2211 | ||
5d4f98a2 YZ |
2212 | BUG_ON(node->ref_mod != 1); |
2213 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2214 | BUG_ON(!extent_op || !extent_op->update_flags); |
5d4f98a2 YZ |
2215 | ret = alloc_reserved_tree_block(trans, root, |
2216 | parent, ref_root, | |
2217 | extent_op->flags_to_set, | |
2218 | &extent_op->key, | |
2219 | ref->level, &ins); | |
5d4f98a2 YZ |
2220 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2221 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
2222 | node->num_bytes, parent, ref_root, | |
2223 | ref->level, 0, 1, extent_op); | |
2224 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
2225 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
2226 | node->num_bytes, parent, ref_root, | |
2227 | ref->level, 0, 1, extent_op); | |
2228 | } else { | |
2229 | BUG(); | |
2230 | } | |
56bec294 CM |
2231 | return ret; |
2232 | } | |
2233 | ||
2234 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 YZ |
2235 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2236 | struct btrfs_root *root, | |
2237 | struct btrfs_delayed_ref_node *node, | |
2238 | struct btrfs_delayed_extent_op *extent_op, | |
2239 | int insert_reserved) | |
56bec294 | 2240 | { |
79787eaa JM |
2241 | int ret = 0; |
2242 | ||
857cc2fc JB |
2243 | if (trans->aborted) { |
2244 | if (insert_reserved) | |
2245 | btrfs_pin_extent(root, node->bytenr, | |
2246 | node->num_bytes, 1); | |
79787eaa | 2247 | return 0; |
857cc2fc | 2248 | } |
79787eaa | 2249 | |
5d4f98a2 | 2250 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2251 | struct btrfs_delayed_ref_head *head; |
2252 | /* | |
2253 | * we've hit the end of the chain and we were supposed | |
2254 | * to insert this extent into the tree. But, it got | |
2255 | * deleted before we ever needed to insert it, so all | |
2256 | * we have to do is clean up the accounting | |
2257 | */ | |
5d4f98a2 YZ |
2258 | BUG_ON(extent_op); |
2259 | head = btrfs_delayed_node_to_head(node); | |
599c75ec LB |
2260 | trace_run_delayed_ref_head(node, head, node->action); |
2261 | ||
56bec294 | 2262 | if (insert_reserved) { |
f0486c68 YZ |
2263 | btrfs_pin_extent(root, node->bytenr, |
2264 | node->num_bytes, 1); | |
5d4f98a2 YZ |
2265 | if (head->is_data) { |
2266 | ret = btrfs_del_csums(trans, root, | |
2267 | node->bytenr, | |
2268 | node->num_bytes); | |
5d4f98a2 | 2269 | } |
56bec294 | 2270 | } |
79787eaa | 2271 | return ret; |
56bec294 CM |
2272 | } |
2273 | ||
5d4f98a2 YZ |
2274 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2275 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2276 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2277 | insert_reserved); | |
2278 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2279 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2280 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2281 | insert_reserved); | |
2282 | else | |
2283 | BUG(); | |
2284 | return ret; | |
56bec294 CM |
2285 | } |
2286 | ||
2287 | static noinline struct btrfs_delayed_ref_node * | |
2288 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
2289 | { | |
2290 | struct rb_node *node; | |
2291 | struct btrfs_delayed_ref_node *ref; | |
2292 | int action = BTRFS_ADD_DELAYED_REF; | |
2293 | again: | |
2294 | /* | |
2295 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2296 | * this prevents ref count from going down to zero when | |
2297 | * there still are pending delayed ref. | |
2298 | */ | |
2299 | node = rb_prev(&head->node.rb_node); | |
2300 | while (1) { | |
2301 | if (!node) | |
2302 | break; | |
2303 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2304 | rb_node); | |
2305 | if (ref->bytenr != head->node.bytenr) | |
2306 | break; | |
5d4f98a2 | 2307 | if (ref->action == action) |
56bec294 CM |
2308 | return ref; |
2309 | node = rb_prev(node); | |
2310 | } | |
2311 | if (action == BTRFS_ADD_DELAYED_REF) { | |
2312 | action = BTRFS_DROP_DELAYED_REF; | |
2313 | goto again; | |
2314 | } | |
2315 | return NULL; | |
2316 | } | |
2317 | ||
79787eaa JM |
2318 | /* |
2319 | * Returns 0 on success or if called with an already aborted transaction. | |
2320 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2321 | */ | |
c3e69d58 CM |
2322 | static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, |
2323 | struct btrfs_root *root, | |
2324 | struct list_head *cluster) | |
56bec294 | 2325 | { |
56bec294 CM |
2326 | struct btrfs_delayed_ref_root *delayed_refs; |
2327 | struct btrfs_delayed_ref_node *ref; | |
2328 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2329 | struct btrfs_delayed_extent_op *extent_op; |
097b8a7c | 2330 | struct btrfs_fs_info *fs_info = root->fs_info; |
56bec294 | 2331 | int ret; |
c3e69d58 | 2332 | int count = 0; |
56bec294 | 2333 | int must_insert_reserved = 0; |
56bec294 CM |
2334 | |
2335 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2336 | while (1) { |
2337 | if (!locked_ref) { | |
c3e69d58 CM |
2338 | /* pick a new head ref from the cluster list */ |
2339 | if (list_empty(cluster)) | |
56bec294 | 2340 | break; |
56bec294 | 2341 | |
c3e69d58 CM |
2342 | locked_ref = list_entry(cluster->next, |
2343 | struct btrfs_delayed_ref_head, cluster); | |
2344 | ||
2345 | /* grab the lock that says we are going to process | |
2346 | * all the refs for this head */ | |
2347 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
2348 | ||
2349 | /* | |
2350 | * we may have dropped the spin lock to get the head | |
2351 | * mutex lock, and that might have given someone else | |
2352 | * time to free the head. If that's true, it has been | |
2353 | * removed from our list and we can move on. | |
2354 | */ | |
2355 | if (ret == -EAGAIN) { | |
2356 | locked_ref = NULL; | |
2357 | count++; | |
2358 | continue; | |
56bec294 CM |
2359 | } |
2360 | } | |
a28ec197 | 2361 | |
ae1e206b JB |
2362 | /* |
2363 | * We need to try and merge add/drops of the same ref since we | |
2364 | * can run into issues with relocate dropping the implicit ref | |
2365 | * and then it being added back again before the drop can | |
2366 | * finish. If we merged anything we need to re-loop so we can | |
2367 | * get a good ref. | |
2368 | */ | |
2369 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, | |
2370 | locked_ref); | |
2371 | ||
d1270cd9 AJ |
2372 | /* |
2373 | * locked_ref is the head node, so we have to go one | |
2374 | * node back for any delayed ref updates | |
2375 | */ | |
2376 | ref = select_delayed_ref(locked_ref); | |
2377 | ||
2378 | if (ref && ref->seq && | |
097b8a7c | 2379 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d1270cd9 AJ |
2380 | /* |
2381 | * there are still refs with lower seq numbers in the | |
2382 | * process of being added. Don't run this ref yet. | |
2383 | */ | |
2384 | list_del_init(&locked_ref->cluster); | |
093486c4 | 2385 | btrfs_delayed_ref_unlock(locked_ref); |
d1270cd9 AJ |
2386 | locked_ref = NULL; |
2387 | delayed_refs->num_heads_ready++; | |
2388 | spin_unlock(&delayed_refs->lock); | |
2389 | cond_resched(); | |
2390 | spin_lock(&delayed_refs->lock); | |
2391 | continue; | |
2392 | } | |
2393 | ||
56bec294 CM |
2394 | /* |
2395 | * record the must insert reserved flag before we | |
2396 | * drop the spin lock. | |
2397 | */ | |
2398 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2399 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2400 | |
5d4f98a2 YZ |
2401 | extent_op = locked_ref->extent_op; |
2402 | locked_ref->extent_op = NULL; | |
2403 | ||
56bec294 CM |
2404 | if (!ref) { |
2405 | /* All delayed refs have been processed, Go ahead | |
2406 | * and send the head node to run_one_delayed_ref, | |
2407 | * so that any accounting fixes can happen | |
2408 | */ | |
2409 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2410 | |
2411 | if (extent_op && must_insert_reserved) { | |
78a6184a | 2412 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2413 | extent_op = NULL; |
2414 | } | |
2415 | ||
2416 | if (extent_op) { | |
2417 | spin_unlock(&delayed_refs->lock); | |
2418 | ||
2419 | ret = run_delayed_extent_op(trans, root, | |
2420 | ref, extent_op); | |
78a6184a | 2421 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 | 2422 | |
79787eaa | 2423 | if (ret) { |
857cc2fc JB |
2424 | /* |
2425 | * Need to reset must_insert_reserved if | |
2426 | * there was an error so the abort stuff | |
2427 | * can cleanup the reserved space | |
2428 | * properly. | |
2429 | */ | |
2430 | if (must_insert_reserved) | |
2431 | locked_ref->must_insert_reserved = 1; | |
c2cf52eb | 2432 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); |
253beebd | 2433 | spin_lock(&delayed_refs->lock); |
093486c4 | 2434 | btrfs_delayed_ref_unlock(locked_ref); |
79787eaa JM |
2435 | return ret; |
2436 | } | |
2437 | ||
203bf287 | 2438 | goto next; |
5d4f98a2 | 2439 | } |
56bec294 | 2440 | } |
02217ed2 | 2441 | |
56bec294 CM |
2442 | ref->in_tree = 0; |
2443 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
2444 | delayed_refs->num_entries--; | |
093486c4 | 2445 | if (!btrfs_delayed_ref_is_head(ref)) { |
22cd2e7d AJ |
2446 | /* |
2447 | * when we play the delayed ref, also correct the | |
2448 | * ref_mod on head | |
2449 | */ | |
2450 | switch (ref->action) { | |
2451 | case BTRFS_ADD_DELAYED_REF: | |
2452 | case BTRFS_ADD_DELAYED_EXTENT: | |
2453 | locked_ref->node.ref_mod -= ref->ref_mod; | |
2454 | break; | |
2455 | case BTRFS_DROP_DELAYED_REF: | |
2456 | locked_ref->node.ref_mod += ref->ref_mod; | |
2457 | break; | |
2458 | default: | |
2459 | WARN_ON(1); | |
2460 | } | |
b8d0c69b JB |
2461 | } else { |
2462 | list_del_init(&locked_ref->cluster); | |
22cd2e7d | 2463 | } |
56bec294 | 2464 | spin_unlock(&delayed_refs->lock); |
925baedd | 2465 | |
5d4f98a2 | 2466 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 | 2467 | must_insert_reserved); |
eb099670 | 2468 | |
78a6184a | 2469 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2470 | if (ret) { |
093486c4 MX |
2471 | btrfs_delayed_ref_unlock(locked_ref); |
2472 | btrfs_put_delayed_ref(ref); | |
c2cf52eb | 2473 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret); |
253beebd | 2474 | spin_lock(&delayed_refs->lock); |
79787eaa JM |
2475 | return ret; |
2476 | } | |
2477 | ||
093486c4 MX |
2478 | /* |
2479 | * If this node is a head, that means all the refs in this head | |
2480 | * have been dealt with, and we will pick the next head to deal | |
2481 | * with, so we must unlock the head and drop it from the cluster | |
2482 | * list before we release it. | |
2483 | */ | |
2484 | if (btrfs_delayed_ref_is_head(ref)) { | |
093486c4 MX |
2485 | btrfs_delayed_ref_unlock(locked_ref); |
2486 | locked_ref = NULL; | |
2487 | } | |
2488 | btrfs_put_delayed_ref(ref); | |
2489 | count++; | |
203bf287 | 2490 | next: |
c3e69d58 CM |
2491 | cond_resched(); |
2492 | spin_lock(&delayed_refs->lock); | |
2493 | } | |
2494 | return count; | |
2495 | } | |
2496 | ||
709c0486 AJ |
2497 | #ifdef SCRAMBLE_DELAYED_REFS |
2498 | /* | |
2499 | * Normally delayed refs get processed in ascending bytenr order. This | |
2500 | * correlates in most cases to the order added. To expose dependencies on this | |
2501 | * order, we start to process the tree in the middle instead of the beginning | |
2502 | */ | |
2503 | static u64 find_middle(struct rb_root *root) | |
2504 | { | |
2505 | struct rb_node *n = root->rb_node; | |
2506 | struct btrfs_delayed_ref_node *entry; | |
2507 | int alt = 1; | |
2508 | u64 middle; | |
2509 | u64 first = 0, last = 0; | |
2510 | ||
2511 | n = rb_first(root); | |
2512 | if (n) { | |
2513 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2514 | first = entry->bytenr; | |
2515 | } | |
2516 | n = rb_last(root); | |
2517 | if (n) { | |
2518 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2519 | last = entry->bytenr; | |
2520 | } | |
2521 | n = root->rb_node; | |
2522 | ||
2523 | while (n) { | |
2524 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2525 | WARN_ON(!entry->in_tree); | |
2526 | ||
2527 | middle = entry->bytenr; | |
2528 | ||
2529 | if (alt) | |
2530 | n = n->rb_left; | |
2531 | else | |
2532 | n = n->rb_right; | |
2533 | ||
2534 | alt = 1 - alt; | |
2535 | } | |
2536 | return middle; | |
2537 | } | |
2538 | #endif | |
2539 | ||
bed92eae AJ |
2540 | int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans, |
2541 | struct btrfs_fs_info *fs_info) | |
2542 | { | |
2543 | struct qgroup_update *qgroup_update; | |
2544 | int ret = 0; | |
2545 | ||
2546 | if (list_empty(&trans->qgroup_ref_list) != | |
2547 | !trans->delayed_ref_elem.seq) { | |
2548 | /* list without seq or seq without list */ | |
c2cf52eb | 2549 | btrfs_err(fs_info, |
fc36ed7e | 2550 | "qgroup accounting update error, list is%s empty, seq is %#x.%x", |
bed92eae | 2551 | list_empty(&trans->qgroup_ref_list) ? "" : " not", |
fc36ed7e JS |
2552 | (u32)(trans->delayed_ref_elem.seq >> 32), |
2553 | (u32)trans->delayed_ref_elem.seq); | |
bed92eae AJ |
2554 | BUG(); |
2555 | } | |
2556 | ||
2557 | if (!trans->delayed_ref_elem.seq) | |
2558 | return 0; | |
2559 | ||
2560 | while (!list_empty(&trans->qgroup_ref_list)) { | |
2561 | qgroup_update = list_first_entry(&trans->qgroup_ref_list, | |
2562 | struct qgroup_update, list); | |
2563 | list_del(&qgroup_update->list); | |
2564 | if (!ret) | |
2565 | ret = btrfs_qgroup_account_ref( | |
2566 | trans, fs_info, qgroup_update->node, | |
2567 | qgroup_update->extent_op); | |
2568 | kfree(qgroup_update); | |
2569 | } | |
2570 | ||
2571 | btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem); | |
2572 | ||
2573 | return ret; | |
2574 | } | |
2575 | ||
bb721703 CM |
2576 | static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq, |
2577 | int count) | |
2578 | { | |
2579 | int val = atomic_read(&delayed_refs->ref_seq); | |
2580 | ||
2581 | if (val < seq || val >= seq + count) | |
2582 | return 1; | |
2583 | return 0; | |
2584 | } | |
2585 | ||
1be41b78 JB |
2586 | static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads) |
2587 | { | |
2588 | u64 num_bytes; | |
2589 | ||
2590 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2591 | sizeof(struct btrfs_extent_inline_ref)); | |
2592 | if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) | |
2593 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); | |
2594 | ||
2595 | /* | |
2596 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
2597 | * closer to what we're really going to want to ouse. | |
2598 | */ | |
2599 | return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root)); | |
2600 | } | |
2601 | ||
2602 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, | |
2603 | struct btrfs_root *root) | |
2604 | { | |
2605 | struct btrfs_block_rsv *global_rsv; | |
2606 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
2607 | u64 num_bytes; | |
2608 | int ret = 0; | |
2609 | ||
2610 | num_bytes = btrfs_calc_trans_metadata_size(root, 1); | |
2611 | num_heads = heads_to_leaves(root, num_heads); | |
2612 | if (num_heads > 1) | |
2613 | num_bytes += (num_heads - 1) * root->leafsize; | |
2614 | num_bytes <<= 1; | |
2615 | global_rsv = &root->fs_info->global_block_rsv; | |
2616 | ||
2617 | /* | |
2618 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2619 | * wiggle room since running delayed refs can create more delayed refs. | |
2620 | */ | |
2621 | if (global_rsv->space_info->full) | |
2622 | num_bytes <<= 1; | |
2623 | ||
2624 | spin_lock(&global_rsv->lock); | |
2625 | if (global_rsv->reserved <= num_bytes) | |
2626 | ret = 1; | |
2627 | spin_unlock(&global_rsv->lock); | |
2628 | return ret; | |
2629 | } | |
2630 | ||
c3e69d58 CM |
2631 | /* |
2632 | * this starts processing the delayed reference count updates and | |
2633 | * extent insertions we have queued up so far. count can be | |
2634 | * 0, which means to process everything in the tree at the start | |
2635 | * of the run (but not newly added entries), or it can be some target | |
2636 | * number you'd like to process. | |
79787eaa JM |
2637 | * |
2638 | * Returns 0 on success or if called with an aborted transaction | |
2639 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2640 | */ |
2641 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2642 | struct btrfs_root *root, unsigned long count) | |
2643 | { | |
2644 | struct rb_node *node; | |
2645 | struct btrfs_delayed_ref_root *delayed_refs; | |
2646 | struct btrfs_delayed_ref_node *ref; | |
2647 | struct list_head cluster; | |
2648 | int ret; | |
a168650c | 2649 | u64 delayed_start; |
c3e69d58 CM |
2650 | int run_all = count == (unsigned long)-1; |
2651 | int run_most = 0; | |
1fa11e26 | 2652 | int loops; |
c3e69d58 | 2653 | |
79787eaa JM |
2654 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2655 | if (trans->aborted) | |
2656 | return 0; | |
2657 | ||
c3e69d58 CM |
2658 | if (root == root->fs_info->extent_root) |
2659 | root = root->fs_info->tree_root; | |
2660 | ||
edf39272 JS |
2661 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); |
2662 | ||
c3e69d58 CM |
2663 | delayed_refs = &trans->transaction->delayed_refs; |
2664 | INIT_LIST_HEAD(&cluster); | |
bb721703 CM |
2665 | if (count == 0) { |
2666 | count = delayed_refs->num_entries * 2; | |
2667 | run_most = 1; | |
2668 | } | |
2669 | ||
2670 | if (!run_all && !run_most) { | |
2671 | int old; | |
2672 | int seq = atomic_read(&delayed_refs->ref_seq); | |
2673 | ||
2674 | progress: | |
2675 | old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1); | |
2676 | if (old) { | |
2677 | DEFINE_WAIT(__wait); | |
1be41b78 JB |
2678 | if (delayed_refs->flushing || |
2679 | !btrfs_should_throttle_delayed_refs(trans, root)) | |
bb721703 CM |
2680 | return 0; |
2681 | ||
2682 | prepare_to_wait(&delayed_refs->wait, &__wait, | |
2683 | TASK_UNINTERRUPTIBLE); | |
2684 | ||
2685 | old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1); | |
2686 | if (old) { | |
2687 | schedule(); | |
2688 | finish_wait(&delayed_refs->wait, &__wait); | |
2689 | ||
2690 | if (!refs_newer(delayed_refs, seq, 256)) | |
2691 | goto progress; | |
2692 | else | |
2693 | return 0; | |
2694 | } else { | |
2695 | finish_wait(&delayed_refs->wait, &__wait); | |
2696 | goto again; | |
2697 | } | |
2698 | } | |
2699 | ||
2700 | } else { | |
2701 | atomic_inc(&delayed_refs->procs_running_refs); | |
2702 | } | |
2703 | ||
c3e69d58 | 2704 | again: |
1fa11e26 | 2705 | loops = 0; |
c3e69d58 | 2706 | spin_lock(&delayed_refs->lock); |
097b8a7c | 2707 | |
709c0486 AJ |
2708 | #ifdef SCRAMBLE_DELAYED_REFS |
2709 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2710 | #endif | |
2711 | ||
c3e69d58 CM |
2712 | while (1) { |
2713 | if (!(run_all || run_most) && | |
1be41b78 | 2714 | !btrfs_should_throttle_delayed_refs(trans, root)) |
c3e69d58 | 2715 | break; |
eb099670 | 2716 | |
56bec294 | 2717 | /* |
c3e69d58 CM |
2718 | * go find something we can process in the rbtree. We start at |
2719 | * the beginning of the tree, and then build a cluster | |
2720 | * of refs to process starting at the first one we are able to | |
2721 | * lock | |
56bec294 | 2722 | */ |
a168650c | 2723 | delayed_start = delayed_refs->run_delayed_start; |
c3e69d58 CM |
2724 | ret = btrfs_find_ref_cluster(trans, &cluster, |
2725 | delayed_refs->run_delayed_start); | |
2726 | if (ret) | |
56bec294 CM |
2727 | break; |
2728 | ||
c3e69d58 | 2729 | ret = run_clustered_refs(trans, root, &cluster); |
79787eaa | 2730 | if (ret < 0) { |
093486c4 | 2731 | btrfs_release_ref_cluster(&cluster); |
79787eaa JM |
2732 | spin_unlock(&delayed_refs->lock); |
2733 | btrfs_abort_transaction(trans, root, ret); | |
bb721703 | 2734 | atomic_dec(&delayed_refs->procs_running_refs); |
f971fe29 | 2735 | wake_up(&delayed_refs->wait); |
79787eaa JM |
2736 | return ret; |
2737 | } | |
c3e69d58 | 2738 | |
bb721703 CM |
2739 | atomic_add(ret, &delayed_refs->ref_seq); |
2740 | ||
c3e69d58 CM |
2741 | count -= min_t(unsigned long, ret, count); |
2742 | ||
2743 | if (count == 0) | |
2744 | break; | |
a168650c | 2745 | |
1fa11e26 AJ |
2746 | if (delayed_start >= delayed_refs->run_delayed_start) { |
2747 | if (loops == 0) { | |
2748 | /* | |
2749 | * btrfs_find_ref_cluster looped. let's do one | |
2750 | * more cycle. if we don't run any delayed ref | |
2751 | * during that cycle (because we can't because | |
2752 | * all of them are blocked), bail out. | |
2753 | */ | |
2754 | loops = 1; | |
2755 | } else { | |
2756 | /* | |
2757 | * no runnable refs left, stop trying | |
2758 | */ | |
2759 | BUG_ON(run_all); | |
2760 | break; | |
2761 | } | |
2762 | } | |
2763 | if (ret) { | |
a168650c | 2764 | /* refs were run, let's reset staleness detection */ |
1fa11e26 | 2765 | loops = 0; |
a168650c | 2766 | } |
eb099670 | 2767 | } |
c3e69d58 | 2768 | |
56bec294 | 2769 | if (run_all) { |
ea658bad JB |
2770 | if (!list_empty(&trans->new_bgs)) { |
2771 | spin_unlock(&delayed_refs->lock); | |
2772 | btrfs_create_pending_block_groups(trans, root); | |
2773 | spin_lock(&delayed_refs->lock); | |
2774 | } | |
2775 | ||
56bec294 | 2776 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 2777 | if (!node) |
56bec294 | 2778 | goto out; |
c3e69d58 | 2779 | count = (unsigned long)-1; |
e9d0b13b | 2780 | |
56bec294 CM |
2781 | while (node) { |
2782 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2783 | rb_node); | |
2784 | if (btrfs_delayed_ref_is_head(ref)) { | |
2785 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 2786 | |
56bec294 CM |
2787 | head = btrfs_delayed_node_to_head(ref); |
2788 | atomic_inc(&ref->refs); | |
2789 | ||
2790 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2791 | /* |
2792 | * Mutex was contended, block until it's | |
2793 | * released and try again | |
2794 | */ | |
56bec294 CM |
2795 | mutex_lock(&head->mutex); |
2796 | mutex_unlock(&head->mutex); | |
2797 | ||
2798 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2799 | cond_resched(); |
56bec294 CM |
2800 | goto again; |
2801 | } | |
2802 | node = rb_next(node); | |
2803 | } | |
2804 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
2805 | schedule_timeout(1); |
2806 | goto again; | |
5f39d397 | 2807 | } |
54aa1f4d | 2808 | out: |
bb721703 CM |
2809 | atomic_dec(&delayed_refs->procs_running_refs); |
2810 | smp_mb(); | |
2811 | if (waitqueue_active(&delayed_refs->wait)) | |
2812 | wake_up(&delayed_refs->wait); | |
2813 | ||
c3e69d58 | 2814 | spin_unlock(&delayed_refs->lock); |
edf39272 | 2815 | assert_qgroups_uptodate(trans); |
a28ec197 CM |
2816 | return 0; |
2817 | } | |
2818 | ||
5d4f98a2 YZ |
2819 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2820 | struct btrfs_root *root, | |
2821 | u64 bytenr, u64 num_bytes, u64 flags, | |
b1c79e09 | 2822 | int level, int is_data) |
5d4f98a2 YZ |
2823 | { |
2824 | struct btrfs_delayed_extent_op *extent_op; | |
2825 | int ret; | |
2826 | ||
78a6184a | 2827 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2828 | if (!extent_op) |
2829 | return -ENOMEM; | |
2830 | ||
2831 | extent_op->flags_to_set = flags; | |
2832 | extent_op->update_flags = 1; | |
2833 | extent_op->update_key = 0; | |
2834 | extent_op->is_data = is_data ? 1 : 0; | |
b1c79e09 | 2835 | extent_op->level = level; |
5d4f98a2 | 2836 | |
66d7e7f0 AJ |
2837 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, |
2838 | num_bytes, extent_op); | |
5d4f98a2 | 2839 | if (ret) |
78a6184a | 2840 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2841 | return ret; |
2842 | } | |
2843 | ||
2844 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2845 | struct btrfs_root *root, | |
2846 | struct btrfs_path *path, | |
2847 | u64 objectid, u64 offset, u64 bytenr) | |
2848 | { | |
2849 | struct btrfs_delayed_ref_head *head; | |
2850 | struct btrfs_delayed_ref_node *ref; | |
2851 | struct btrfs_delayed_data_ref *data_ref; | |
2852 | struct btrfs_delayed_ref_root *delayed_refs; | |
2853 | struct rb_node *node; | |
2854 | int ret = 0; | |
2855 | ||
2856 | ret = -ENOENT; | |
2857 | delayed_refs = &trans->transaction->delayed_refs; | |
2858 | spin_lock(&delayed_refs->lock); | |
2859 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2860 | if (!head) | |
2861 | goto out; | |
2862 | ||
2863 | if (!mutex_trylock(&head->mutex)) { | |
2864 | atomic_inc(&head->node.refs); | |
2865 | spin_unlock(&delayed_refs->lock); | |
2866 | ||
b3b4aa74 | 2867 | btrfs_release_path(path); |
5d4f98a2 | 2868 | |
8cc33e5c DS |
2869 | /* |
2870 | * Mutex was contended, block until it's released and let | |
2871 | * caller try again | |
2872 | */ | |
5d4f98a2 YZ |
2873 | mutex_lock(&head->mutex); |
2874 | mutex_unlock(&head->mutex); | |
2875 | btrfs_put_delayed_ref(&head->node); | |
2876 | return -EAGAIN; | |
2877 | } | |
2878 | ||
2879 | node = rb_prev(&head->node.rb_node); | |
2880 | if (!node) | |
2881 | goto out_unlock; | |
2882 | ||
2883 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2884 | ||
2885 | if (ref->bytenr != bytenr) | |
2886 | goto out_unlock; | |
2887 | ||
2888 | ret = 1; | |
2889 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2890 | goto out_unlock; | |
2891 | ||
2892 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2893 | ||
2894 | node = rb_prev(node); | |
2895 | if (node) { | |
df57dbe6 LB |
2896 | int seq = ref->seq; |
2897 | ||
5d4f98a2 | 2898 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); |
df57dbe6 | 2899 | if (ref->bytenr == bytenr && ref->seq == seq) |
5d4f98a2 YZ |
2900 | goto out_unlock; |
2901 | } | |
2902 | ||
2903 | if (data_ref->root != root->root_key.objectid || | |
2904 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2905 | goto out_unlock; | |
2906 | ||
2907 | ret = 0; | |
2908 | out_unlock: | |
2909 | mutex_unlock(&head->mutex); | |
2910 | out: | |
2911 | spin_unlock(&delayed_refs->lock); | |
2912 | return ret; | |
2913 | } | |
2914 | ||
2915 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2916 | struct btrfs_root *root, | |
2917 | struct btrfs_path *path, | |
2918 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2919 | { |
2920 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2921 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2922 | struct btrfs_extent_data_ref *ref; |
2923 | struct btrfs_extent_inline_ref *iref; | |
2924 | struct btrfs_extent_item *ei; | |
f321e491 | 2925 | struct btrfs_key key; |
5d4f98a2 | 2926 | u32 item_size; |
be20aa9d | 2927 | int ret; |
925baedd | 2928 | |
be20aa9d | 2929 | key.objectid = bytenr; |
31840ae1 | 2930 | key.offset = (u64)-1; |
f321e491 | 2931 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2932 | |
be20aa9d CM |
2933 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2934 | if (ret < 0) | |
2935 | goto out; | |
79787eaa | 2936 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2937 | |
2938 | ret = -ENOENT; | |
2939 | if (path->slots[0] == 0) | |
31840ae1 | 2940 | goto out; |
be20aa9d | 2941 | |
31840ae1 | 2942 | path->slots[0]--; |
f321e491 | 2943 | leaf = path->nodes[0]; |
5d4f98a2 | 2944 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2945 | |
5d4f98a2 | 2946 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2947 | goto out; |
f321e491 | 2948 | |
5d4f98a2 YZ |
2949 | ret = 1; |
2950 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2951 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2952 | if (item_size < sizeof(*ei)) { | |
2953 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2954 | goto out; | |
2955 | } | |
2956 | #endif | |
2957 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2958 | |
5d4f98a2 YZ |
2959 | if (item_size != sizeof(*ei) + |
2960 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2961 | goto out; | |
be20aa9d | 2962 | |
5d4f98a2 YZ |
2963 | if (btrfs_extent_generation(leaf, ei) <= |
2964 | btrfs_root_last_snapshot(&root->root_item)) | |
2965 | goto out; | |
2966 | ||
2967 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2968 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2969 | BTRFS_EXTENT_DATA_REF_KEY) | |
2970 | goto out; | |
2971 | ||
2972 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2973 | if (btrfs_extent_refs(leaf, ei) != | |
2974 | btrfs_extent_data_ref_count(leaf, ref) || | |
2975 | btrfs_extent_data_ref_root(leaf, ref) != | |
2976 | root->root_key.objectid || | |
2977 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2978 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2979 | goto out; | |
2980 | ||
2981 | ret = 0; | |
2982 | out: | |
2983 | return ret; | |
2984 | } | |
2985 | ||
2986 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2987 | struct btrfs_root *root, | |
2988 | u64 objectid, u64 offset, u64 bytenr) | |
2989 | { | |
2990 | struct btrfs_path *path; | |
2991 | int ret; | |
2992 | int ret2; | |
2993 | ||
2994 | path = btrfs_alloc_path(); | |
2995 | if (!path) | |
2996 | return -ENOENT; | |
2997 | ||
2998 | do { | |
2999 | ret = check_committed_ref(trans, root, path, objectid, | |
3000 | offset, bytenr); | |
3001 | if (ret && ret != -ENOENT) | |
f321e491 | 3002 | goto out; |
80ff3856 | 3003 | |
5d4f98a2 YZ |
3004 | ret2 = check_delayed_ref(trans, root, path, objectid, |
3005 | offset, bytenr); | |
3006 | } while (ret2 == -EAGAIN); | |
3007 | ||
3008 | if (ret2 && ret2 != -ENOENT) { | |
3009 | ret = ret2; | |
3010 | goto out; | |
f321e491 | 3011 | } |
5d4f98a2 YZ |
3012 | |
3013 | if (ret != -ENOENT || ret2 != -ENOENT) | |
3014 | ret = 0; | |
be20aa9d | 3015 | out: |
80ff3856 | 3016 | btrfs_free_path(path); |
f0486c68 YZ |
3017 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3018 | WARN_ON(ret > 0); | |
f321e491 | 3019 | return ret; |
be20aa9d | 3020 | } |
c5739bba | 3021 | |
5d4f98a2 | 3022 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3023 | struct btrfs_root *root, |
5d4f98a2 | 3024 | struct extent_buffer *buf, |
66d7e7f0 | 3025 | int full_backref, int inc, int for_cow) |
31840ae1 ZY |
3026 | { |
3027 | u64 bytenr; | |
5d4f98a2 YZ |
3028 | u64 num_bytes; |
3029 | u64 parent; | |
31840ae1 | 3030 | u64 ref_root; |
31840ae1 | 3031 | u32 nritems; |
31840ae1 ZY |
3032 | struct btrfs_key key; |
3033 | struct btrfs_file_extent_item *fi; | |
3034 | int i; | |
3035 | int level; | |
3036 | int ret = 0; | |
31840ae1 | 3037 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
66d7e7f0 | 3038 | u64, u64, u64, u64, u64, u64, int); |
31840ae1 ZY |
3039 | |
3040 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
3041 | nritems = btrfs_header_nritems(buf); |
3042 | level = btrfs_header_level(buf); | |
3043 | ||
5d4f98a2 YZ |
3044 | if (!root->ref_cows && level == 0) |
3045 | return 0; | |
31840ae1 | 3046 | |
5d4f98a2 YZ |
3047 | if (inc) |
3048 | process_func = btrfs_inc_extent_ref; | |
3049 | else | |
3050 | process_func = btrfs_free_extent; | |
31840ae1 | 3051 | |
5d4f98a2 YZ |
3052 | if (full_backref) |
3053 | parent = buf->start; | |
3054 | else | |
3055 | parent = 0; | |
3056 | ||
3057 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3058 | if (level == 0) { |
5d4f98a2 | 3059 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
3060 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
3061 | continue; | |
5d4f98a2 | 3062 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3063 | struct btrfs_file_extent_item); |
3064 | if (btrfs_file_extent_type(buf, fi) == | |
3065 | BTRFS_FILE_EXTENT_INLINE) | |
3066 | continue; | |
3067 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3068 | if (bytenr == 0) | |
3069 | continue; | |
5d4f98a2 YZ |
3070 | |
3071 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3072 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
3073 | ret = process_func(trans, root, bytenr, num_bytes, | |
3074 | parent, ref_root, key.objectid, | |
66d7e7f0 | 3075 | key.offset, for_cow); |
31840ae1 ZY |
3076 | if (ret) |
3077 | goto fail; | |
3078 | } else { | |
5d4f98a2 YZ |
3079 | bytenr = btrfs_node_blockptr(buf, i); |
3080 | num_bytes = btrfs_level_size(root, level - 1); | |
3081 | ret = process_func(trans, root, bytenr, num_bytes, | |
66d7e7f0 AJ |
3082 | parent, ref_root, level - 1, 0, |
3083 | for_cow); | |
31840ae1 ZY |
3084 | if (ret) |
3085 | goto fail; | |
3086 | } | |
3087 | } | |
3088 | return 0; | |
3089 | fail: | |
5d4f98a2 YZ |
3090 | return ret; |
3091 | } | |
3092 | ||
3093 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 3094 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 3095 | { |
66d7e7f0 | 3096 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow); |
5d4f98a2 YZ |
3097 | } |
3098 | ||
3099 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 3100 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 3101 | { |
66d7e7f0 | 3102 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow); |
31840ae1 ZY |
3103 | } |
3104 | ||
9078a3e1 CM |
3105 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
3106 | struct btrfs_root *root, | |
3107 | struct btrfs_path *path, | |
3108 | struct btrfs_block_group_cache *cache) | |
3109 | { | |
3110 | int ret; | |
9078a3e1 | 3111 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
3112 | unsigned long bi; |
3113 | struct extent_buffer *leaf; | |
9078a3e1 | 3114 | |
9078a3e1 | 3115 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
3116 | if (ret < 0) |
3117 | goto fail; | |
79787eaa | 3118 | BUG_ON(ret); /* Corruption */ |
5f39d397 CM |
3119 | |
3120 | leaf = path->nodes[0]; | |
3121 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3122 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3123 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 3124 | btrfs_release_path(path); |
54aa1f4d | 3125 | fail: |
79787eaa JM |
3126 | if (ret) { |
3127 | btrfs_abort_transaction(trans, root, ret); | |
9078a3e1 | 3128 | return ret; |
79787eaa | 3129 | } |
9078a3e1 CM |
3130 | return 0; |
3131 | ||
3132 | } | |
3133 | ||
4a8c9a62 YZ |
3134 | static struct btrfs_block_group_cache * |
3135 | next_block_group(struct btrfs_root *root, | |
3136 | struct btrfs_block_group_cache *cache) | |
3137 | { | |
3138 | struct rb_node *node; | |
3139 | spin_lock(&root->fs_info->block_group_cache_lock); | |
3140 | node = rb_next(&cache->cache_node); | |
3141 | btrfs_put_block_group(cache); | |
3142 | if (node) { | |
3143 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3144 | cache_node); | |
11dfe35a | 3145 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3146 | } else |
3147 | cache = NULL; | |
3148 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
3149 | return cache; | |
3150 | } | |
3151 | ||
0af3d00b JB |
3152 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3153 | struct btrfs_trans_handle *trans, | |
3154 | struct btrfs_path *path) | |
3155 | { | |
3156 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
3157 | struct inode *inode = NULL; | |
3158 | u64 alloc_hint = 0; | |
2b20982e | 3159 | int dcs = BTRFS_DC_ERROR; |
0af3d00b JB |
3160 | int num_pages = 0; |
3161 | int retries = 0; | |
3162 | int ret = 0; | |
3163 | ||
3164 | /* | |
3165 | * If this block group is smaller than 100 megs don't bother caching the | |
3166 | * block group. | |
3167 | */ | |
3168 | if (block_group->key.offset < (100 * 1024 * 1024)) { | |
3169 | spin_lock(&block_group->lock); | |
3170 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3171 | spin_unlock(&block_group->lock); | |
3172 | return 0; | |
3173 | } | |
3174 | ||
3175 | again: | |
3176 | inode = lookup_free_space_inode(root, block_group, path); | |
3177 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
3178 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3179 | btrfs_release_path(path); |
0af3d00b JB |
3180 | goto out; |
3181 | } | |
3182 | ||
3183 | if (IS_ERR(inode)) { | |
3184 | BUG_ON(retries); | |
3185 | retries++; | |
3186 | ||
3187 | if (block_group->ro) | |
3188 | goto out_free; | |
3189 | ||
3190 | ret = create_free_space_inode(root, trans, block_group, path); | |
3191 | if (ret) | |
3192 | goto out_free; | |
3193 | goto again; | |
3194 | } | |
3195 | ||
5b0e95bf JB |
3196 | /* We've already setup this transaction, go ahead and exit */ |
3197 | if (block_group->cache_generation == trans->transid && | |
3198 | i_size_read(inode)) { | |
3199 | dcs = BTRFS_DC_SETUP; | |
3200 | goto out_put; | |
3201 | } | |
3202 | ||
0af3d00b JB |
3203 | /* |
3204 | * We want to set the generation to 0, that way if anything goes wrong | |
3205 | * from here on out we know not to trust this cache when we load up next | |
3206 | * time. | |
3207 | */ | |
3208 | BTRFS_I(inode)->generation = 0; | |
3209 | ret = btrfs_update_inode(trans, root, inode); | |
3210 | WARN_ON(ret); | |
3211 | ||
3212 | if (i_size_read(inode) > 0) { | |
7b61cd92 MX |
3213 | ret = btrfs_check_trunc_cache_free_space(root, |
3214 | &root->fs_info->global_block_rsv); | |
3215 | if (ret) | |
3216 | goto out_put; | |
3217 | ||
74514323 | 3218 | ret = btrfs_truncate_free_space_cache(root, trans, inode); |
0af3d00b JB |
3219 | if (ret) |
3220 | goto out_put; | |
3221 | } | |
3222 | ||
3223 | spin_lock(&block_group->lock); | |
cf7c1ef6 LB |
3224 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
3225 | !btrfs_test_opt(root, SPACE_CACHE)) { | |
3226 | /* | |
3227 | * don't bother trying to write stuff out _if_ | |
3228 | * a) we're not cached, | |
3229 | * b) we're with nospace_cache mount option. | |
3230 | */ | |
2b20982e | 3231 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3232 | spin_unlock(&block_group->lock); |
3233 | goto out_put; | |
3234 | } | |
3235 | spin_unlock(&block_group->lock); | |
3236 | ||
6fc823b1 JB |
3237 | /* |
3238 | * Try to preallocate enough space based on how big the block group is. | |
3239 | * Keep in mind this has to include any pinned space which could end up | |
3240 | * taking up quite a bit since it's not folded into the other space | |
3241 | * cache. | |
3242 | */ | |
3243 | num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024); | |
0af3d00b JB |
3244 | if (!num_pages) |
3245 | num_pages = 1; | |
3246 | ||
0af3d00b JB |
3247 | num_pages *= 16; |
3248 | num_pages *= PAGE_CACHE_SIZE; | |
3249 | ||
3250 | ret = btrfs_check_data_free_space(inode, num_pages); | |
3251 | if (ret) | |
3252 | goto out_put; | |
3253 | ||
3254 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3255 | num_pages, num_pages, | |
3256 | &alloc_hint); | |
2b20982e JB |
3257 | if (!ret) |
3258 | dcs = BTRFS_DC_SETUP; | |
0af3d00b | 3259 | btrfs_free_reserved_data_space(inode, num_pages); |
c09544e0 | 3260 | |
0af3d00b JB |
3261 | out_put: |
3262 | iput(inode); | |
3263 | out_free: | |
b3b4aa74 | 3264 | btrfs_release_path(path); |
0af3d00b JB |
3265 | out: |
3266 | spin_lock(&block_group->lock); | |
e65cbb94 | 3267 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3268 | block_group->cache_generation = trans->transid; |
2b20982e | 3269 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3270 | spin_unlock(&block_group->lock); |
3271 | ||
3272 | return ret; | |
3273 | } | |
3274 | ||
96b5179d CM |
3275 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
3276 | struct btrfs_root *root) | |
9078a3e1 | 3277 | { |
4a8c9a62 | 3278 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 3279 | int err = 0; |
9078a3e1 | 3280 | struct btrfs_path *path; |
96b5179d | 3281 | u64 last = 0; |
9078a3e1 CM |
3282 | |
3283 | path = btrfs_alloc_path(); | |
3284 | if (!path) | |
3285 | return -ENOMEM; | |
3286 | ||
0af3d00b JB |
3287 | again: |
3288 | while (1) { | |
3289 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
3290 | while (cache) { | |
3291 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3292 | break; | |
3293 | cache = next_block_group(root, cache); | |
3294 | } | |
3295 | if (!cache) { | |
3296 | if (last == 0) | |
3297 | break; | |
3298 | last = 0; | |
3299 | continue; | |
3300 | } | |
3301 | err = cache_save_setup(cache, trans, path); | |
3302 | last = cache->key.objectid + cache->key.offset; | |
3303 | btrfs_put_block_group(cache); | |
3304 | } | |
3305 | ||
d397712b | 3306 | while (1) { |
4a8c9a62 YZ |
3307 | if (last == 0) { |
3308 | err = btrfs_run_delayed_refs(trans, root, | |
3309 | (unsigned long)-1); | |
79787eaa JM |
3310 | if (err) /* File system offline */ |
3311 | goto out; | |
0f9dd46c | 3312 | } |
54aa1f4d | 3313 | |
4a8c9a62 YZ |
3314 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
3315 | while (cache) { | |
0af3d00b JB |
3316 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) { |
3317 | btrfs_put_block_group(cache); | |
3318 | goto again; | |
3319 | } | |
3320 | ||
4a8c9a62 YZ |
3321 | if (cache->dirty) |
3322 | break; | |
3323 | cache = next_block_group(root, cache); | |
3324 | } | |
3325 | if (!cache) { | |
3326 | if (last == 0) | |
3327 | break; | |
3328 | last = 0; | |
3329 | continue; | |
3330 | } | |
0f9dd46c | 3331 | |
0cb59c99 JB |
3332 | if (cache->disk_cache_state == BTRFS_DC_SETUP) |
3333 | cache->disk_cache_state = BTRFS_DC_NEED_WRITE; | |
e8569813 | 3334 | cache->dirty = 0; |
4a8c9a62 | 3335 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 3336 | |
4a8c9a62 | 3337 | err = write_one_cache_group(trans, root, path, cache); |
e84cc142 | 3338 | btrfs_put_block_group(cache); |
79787eaa JM |
3339 | if (err) /* File system offline */ |
3340 | goto out; | |
9078a3e1 | 3341 | } |
4a8c9a62 | 3342 | |
0cb59c99 JB |
3343 | while (1) { |
3344 | /* | |
3345 | * I don't think this is needed since we're just marking our | |
3346 | * preallocated extent as written, but just in case it can't | |
3347 | * hurt. | |
3348 | */ | |
3349 | if (last == 0) { | |
3350 | err = btrfs_run_delayed_refs(trans, root, | |
3351 | (unsigned long)-1); | |
79787eaa JM |
3352 | if (err) /* File system offline */ |
3353 | goto out; | |
0cb59c99 JB |
3354 | } |
3355 | ||
3356 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
3357 | while (cache) { | |
3358 | /* | |
3359 | * Really this shouldn't happen, but it could if we | |
3360 | * couldn't write the entire preallocated extent and | |
3361 | * splitting the extent resulted in a new block. | |
3362 | */ | |
3363 | if (cache->dirty) { | |
3364 | btrfs_put_block_group(cache); | |
3365 | goto again; | |
3366 | } | |
3367 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
3368 | break; | |
3369 | cache = next_block_group(root, cache); | |
3370 | } | |
3371 | if (!cache) { | |
3372 | if (last == 0) | |
3373 | break; | |
3374 | last = 0; | |
3375 | continue; | |
3376 | } | |
3377 | ||
79787eaa | 3378 | err = btrfs_write_out_cache(root, trans, cache, path); |
0cb59c99 JB |
3379 | |
3380 | /* | |
3381 | * If we didn't have an error then the cache state is still | |
3382 | * NEED_WRITE, so we can set it to WRITTEN. | |
3383 | */ | |
79787eaa | 3384 | if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE) |
0cb59c99 JB |
3385 | cache->disk_cache_state = BTRFS_DC_WRITTEN; |
3386 | last = cache->key.objectid + cache->key.offset; | |
3387 | btrfs_put_block_group(cache); | |
3388 | } | |
79787eaa | 3389 | out: |
0cb59c99 | 3390 | |
9078a3e1 | 3391 | btrfs_free_path(path); |
79787eaa | 3392 | return err; |
9078a3e1 CM |
3393 | } |
3394 | ||
d2fb3437 YZ |
3395 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
3396 | { | |
3397 | struct btrfs_block_group_cache *block_group; | |
3398 | int readonly = 0; | |
3399 | ||
3400 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
3401 | if (!block_group || block_group->ro) | |
3402 | readonly = 1; | |
3403 | if (block_group) | |
fa9c0d79 | 3404 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3405 | return readonly; |
3406 | } | |
3407 | ||
593060d7 CM |
3408 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3409 | u64 total_bytes, u64 bytes_used, | |
3410 | struct btrfs_space_info **space_info) | |
3411 | { | |
3412 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3413 | int i; |
3414 | int factor; | |
b150a4f1 | 3415 | int ret; |
b742bb82 YZ |
3416 | |
3417 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3418 | BTRFS_BLOCK_GROUP_RAID10)) | |
3419 | factor = 2; | |
3420 | else | |
3421 | factor = 1; | |
593060d7 CM |
3422 | |
3423 | found = __find_space_info(info, flags); | |
3424 | if (found) { | |
25179201 | 3425 | spin_lock(&found->lock); |
593060d7 | 3426 | found->total_bytes += total_bytes; |
89a55897 | 3427 | found->disk_total += total_bytes * factor; |
593060d7 | 3428 | found->bytes_used += bytes_used; |
b742bb82 | 3429 | found->disk_used += bytes_used * factor; |
8f18cf13 | 3430 | found->full = 0; |
25179201 | 3431 | spin_unlock(&found->lock); |
593060d7 CM |
3432 | *space_info = found; |
3433 | return 0; | |
3434 | } | |
c146afad | 3435 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3436 | if (!found) |
3437 | return -ENOMEM; | |
3438 | ||
b150a4f1 JB |
3439 | ret = percpu_counter_init(&found->total_bytes_pinned, 0); |
3440 | if (ret) { | |
3441 | kfree(found); | |
3442 | return ret; | |
3443 | } | |
3444 | ||
b742bb82 YZ |
3445 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
3446 | INIT_LIST_HEAD(&found->block_groups[i]); | |
80eb234a | 3447 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3448 | spin_lock_init(&found->lock); |
52ba6929 | 3449 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
593060d7 | 3450 | found->total_bytes = total_bytes; |
89a55897 | 3451 | found->disk_total = total_bytes * factor; |
593060d7 | 3452 | found->bytes_used = bytes_used; |
b742bb82 | 3453 | found->disk_used = bytes_used * factor; |
593060d7 | 3454 | found->bytes_pinned = 0; |
e8569813 | 3455 | found->bytes_reserved = 0; |
c146afad | 3456 | found->bytes_readonly = 0; |
f0486c68 | 3457 | found->bytes_may_use = 0; |
593060d7 | 3458 | found->full = 0; |
0e4f8f88 | 3459 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3460 | found->chunk_alloc = 0; |
fdb5effd JB |
3461 | found->flush = 0; |
3462 | init_waitqueue_head(&found->wait); | |
593060d7 | 3463 | *space_info = found; |
4184ea7f | 3464 | list_add_rcu(&found->list, &info->space_info); |
b4d7c3c9 LZ |
3465 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3466 | info->data_sinfo = found; | |
593060d7 CM |
3467 | return 0; |
3468 | } | |
3469 | ||
8790d502 CM |
3470 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3471 | { | |
899c81ea ID |
3472 | u64 extra_flags = chunk_to_extended(flags) & |
3473 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 3474 | |
de98ced9 | 3475 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
3476 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3477 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3478 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3479 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3480 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3481 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 3482 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 3483 | } |
593060d7 | 3484 | |
fc67c450 ID |
3485 | /* |
3486 | * returns target flags in extended format or 0 if restripe for this | |
3487 | * chunk_type is not in progress | |
c6664b42 ID |
3488 | * |
3489 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
3490 | */ |
3491 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
3492 | { | |
3493 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3494 | u64 target = 0; | |
3495 | ||
fc67c450 ID |
3496 | if (!bctl) |
3497 | return 0; | |
3498 | ||
3499 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
3500 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3501 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3502 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
3503 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3504 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3505 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
3506 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3507 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3508 | } | |
3509 | ||
3510 | return target; | |
3511 | } | |
3512 | ||
a46d11a8 ID |
3513 | /* |
3514 | * @flags: available profiles in extended format (see ctree.h) | |
3515 | * | |
e4d8ec0f ID |
3516 | * Returns reduced profile in chunk format. If profile changing is in |
3517 | * progress (either running or paused) picks the target profile (if it's | |
3518 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 3519 | */ |
48a3b636 | 3520 | static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 3521 | { |
cd02dca5 CM |
3522 | /* |
3523 | * we add in the count of missing devices because we want | |
3524 | * to make sure that any RAID levels on a degraded FS | |
3525 | * continue to be honored. | |
3526 | */ | |
3527 | u64 num_devices = root->fs_info->fs_devices->rw_devices + | |
3528 | root->fs_info->fs_devices->missing_devices; | |
fc67c450 | 3529 | u64 target; |
53b381b3 | 3530 | u64 tmp; |
a061fc8d | 3531 | |
fc67c450 ID |
3532 | /* |
3533 | * see if restripe for this chunk_type is in progress, if so | |
3534 | * try to reduce to the target profile | |
3535 | */ | |
e4d8ec0f | 3536 | spin_lock(&root->fs_info->balance_lock); |
fc67c450 ID |
3537 | target = get_restripe_target(root->fs_info, flags); |
3538 | if (target) { | |
3539 | /* pick target profile only if it's already available */ | |
3540 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
e4d8ec0f | 3541 | spin_unlock(&root->fs_info->balance_lock); |
fc67c450 | 3542 | return extended_to_chunk(target); |
e4d8ec0f ID |
3543 | } |
3544 | } | |
3545 | spin_unlock(&root->fs_info->balance_lock); | |
3546 | ||
53b381b3 | 3547 | /* First, mask out the RAID levels which aren't possible */ |
a061fc8d | 3548 | if (num_devices == 1) |
53b381b3 DW |
3549 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 | |
3550 | BTRFS_BLOCK_GROUP_RAID5); | |
3551 | if (num_devices < 3) | |
3552 | flags &= ~BTRFS_BLOCK_GROUP_RAID6; | |
a061fc8d CM |
3553 | if (num_devices < 4) |
3554 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
3555 | ||
53b381b3 DW |
3556 | tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 | |
3557 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 | | |
3558 | BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10); | |
3559 | flags &= ~tmp; | |
ec44a35c | 3560 | |
53b381b3 DW |
3561 | if (tmp & BTRFS_BLOCK_GROUP_RAID6) |
3562 | tmp = BTRFS_BLOCK_GROUP_RAID6; | |
3563 | else if (tmp & BTRFS_BLOCK_GROUP_RAID5) | |
3564 | tmp = BTRFS_BLOCK_GROUP_RAID5; | |
3565 | else if (tmp & BTRFS_BLOCK_GROUP_RAID10) | |
3566 | tmp = BTRFS_BLOCK_GROUP_RAID10; | |
3567 | else if (tmp & BTRFS_BLOCK_GROUP_RAID1) | |
3568 | tmp = BTRFS_BLOCK_GROUP_RAID1; | |
3569 | else if (tmp & BTRFS_BLOCK_GROUP_RAID0) | |
3570 | tmp = BTRFS_BLOCK_GROUP_RAID0; | |
a46d11a8 | 3571 | |
53b381b3 | 3572 | return extended_to_chunk(flags | tmp); |
ec44a35c CM |
3573 | } |
3574 | ||
b742bb82 | 3575 | static u64 get_alloc_profile(struct btrfs_root *root, u64 flags) |
6a63209f | 3576 | { |
de98ced9 MX |
3577 | unsigned seq; |
3578 | ||
3579 | do { | |
3580 | seq = read_seqbegin(&root->fs_info->profiles_lock); | |
3581 | ||
3582 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3583 | flags |= root->fs_info->avail_data_alloc_bits; | |
3584 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3585 | flags |= root->fs_info->avail_system_alloc_bits; | |
3586 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3587 | flags |= root->fs_info->avail_metadata_alloc_bits; | |
3588 | } while (read_seqretry(&root->fs_info->profiles_lock, seq)); | |
6fef8df1 | 3589 | |
b742bb82 | 3590 | return btrfs_reduce_alloc_profile(root, flags); |
6a63209f JB |
3591 | } |
3592 | ||
6d07bcec | 3593 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 3594 | { |
b742bb82 | 3595 | u64 flags; |
53b381b3 | 3596 | u64 ret; |
9ed74f2d | 3597 | |
b742bb82 YZ |
3598 | if (data) |
3599 | flags = BTRFS_BLOCK_GROUP_DATA; | |
3600 | else if (root == root->fs_info->chunk_root) | |
3601 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 3602 | else |
b742bb82 | 3603 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3604 | |
53b381b3 DW |
3605 | ret = get_alloc_profile(root, flags); |
3606 | return ret; | |
6a63209f | 3607 | } |
9ed74f2d | 3608 | |
6a63209f | 3609 | /* |
6a63209f JB |
3610 | * This will check the space that the inode allocates from to make sure we have |
3611 | * enough space for bytes. | |
6a63209f | 3612 | */ |
0ca1f7ce | 3613 | int btrfs_check_data_free_space(struct inode *inode, u64 bytes) |
6a63209f | 3614 | { |
6a63209f | 3615 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 3616 | struct btrfs_root *root = BTRFS_I(inode)->root; |
b4d7c3c9 | 3617 | struct btrfs_fs_info *fs_info = root->fs_info; |
ab6e2410 | 3618 | u64 used; |
0af3d00b | 3619 | int ret = 0, committed = 0, alloc_chunk = 1; |
6a63209f | 3620 | |
6a63209f | 3621 | /* make sure bytes are sectorsize aligned */ |
fda2832f | 3622 | bytes = ALIGN(bytes, root->sectorsize); |
6a63209f | 3623 | |
82d5902d LZ |
3624 | if (root == root->fs_info->tree_root || |
3625 | BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) { | |
0af3d00b JB |
3626 | alloc_chunk = 0; |
3627 | committed = 1; | |
3628 | } | |
3629 | ||
b4d7c3c9 | 3630 | data_sinfo = fs_info->data_sinfo; |
33b4d47f CM |
3631 | if (!data_sinfo) |
3632 | goto alloc; | |
9ed74f2d | 3633 | |
6a63209f JB |
3634 | again: |
3635 | /* make sure we have enough space to handle the data first */ | |
3636 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
3637 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
3638 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
3639 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
3640 | |
3641 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 3642 | struct btrfs_trans_handle *trans; |
9ed74f2d | 3643 | |
6a63209f JB |
3644 | /* |
3645 | * if we don't have enough free bytes in this space then we need | |
3646 | * to alloc a new chunk. | |
3647 | */ | |
0af3d00b | 3648 | if (!data_sinfo->full && alloc_chunk) { |
6a63209f | 3649 | u64 alloc_target; |
9ed74f2d | 3650 | |
0e4f8f88 | 3651 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 3652 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 3653 | alloc: |
6a63209f | 3654 | alloc_target = btrfs_get_alloc_profile(root, 1); |
7a7eaa40 | 3655 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3656 | if (IS_ERR(trans)) |
3657 | return PTR_ERR(trans); | |
9ed74f2d | 3658 | |
6a63209f | 3659 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
3660 | alloc_target, |
3661 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 3662 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
3663 | if (ret < 0) { |
3664 | if (ret != -ENOSPC) | |
3665 | return ret; | |
3666 | else | |
3667 | goto commit_trans; | |
3668 | } | |
9ed74f2d | 3669 | |
b4d7c3c9 LZ |
3670 | if (!data_sinfo) |
3671 | data_sinfo = fs_info->data_sinfo; | |
3672 | ||
6a63209f JB |
3673 | goto again; |
3674 | } | |
f2bb8f5c JB |
3675 | |
3676 | /* | |
b150a4f1 JB |
3677 | * If we don't have enough pinned space to deal with this |
3678 | * allocation don't bother committing the transaction. | |
f2bb8f5c | 3679 | */ |
b150a4f1 JB |
3680 | if (percpu_counter_compare(&data_sinfo->total_bytes_pinned, |
3681 | bytes) < 0) | |
f2bb8f5c | 3682 | committed = 1; |
6a63209f | 3683 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3684 | |
4e06bdd6 | 3685 | /* commit the current transaction and try again */ |
d52a5b5f | 3686 | commit_trans: |
a4abeea4 JB |
3687 | if (!committed && |
3688 | !atomic_read(&root->fs_info->open_ioctl_trans)) { | |
4e06bdd6 | 3689 | committed = 1; |
b150a4f1 | 3690 | |
7a7eaa40 | 3691 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3692 | if (IS_ERR(trans)) |
3693 | return PTR_ERR(trans); | |
4e06bdd6 JB |
3694 | ret = btrfs_commit_transaction(trans, root); |
3695 | if (ret) | |
3696 | return ret; | |
3697 | goto again; | |
3698 | } | |
9ed74f2d | 3699 | |
cab45e22 JM |
3700 | trace_btrfs_space_reservation(root->fs_info, |
3701 | "space_info:enospc", | |
3702 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
3703 | return -ENOSPC; |
3704 | } | |
3705 | data_sinfo->bytes_may_use += bytes; | |
8c2a3ca2 | 3706 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3707 | data_sinfo->flags, bytes, 1); |
6a63209f | 3708 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3709 | |
9ed74f2d | 3710 | return 0; |
9ed74f2d | 3711 | } |
6a63209f | 3712 | |
6a63209f | 3713 | /* |
fb25e914 | 3714 | * Called if we need to clear a data reservation for this inode. |
6a63209f | 3715 | */ |
0ca1f7ce | 3716 | void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes) |
e3ccfa98 | 3717 | { |
0ca1f7ce | 3718 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6a63209f | 3719 | struct btrfs_space_info *data_sinfo; |
e3ccfa98 | 3720 | |
6a63209f | 3721 | /* make sure bytes are sectorsize aligned */ |
fda2832f | 3722 | bytes = ALIGN(bytes, root->sectorsize); |
e3ccfa98 | 3723 | |
b4d7c3c9 | 3724 | data_sinfo = root->fs_info->data_sinfo; |
6a63209f | 3725 | spin_lock(&data_sinfo->lock); |
7ee9e440 | 3726 | WARN_ON(data_sinfo->bytes_may_use < bytes); |
6a63209f | 3727 | data_sinfo->bytes_may_use -= bytes; |
8c2a3ca2 | 3728 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3729 | data_sinfo->flags, bytes, 0); |
6a63209f | 3730 | spin_unlock(&data_sinfo->lock); |
e3ccfa98 JB |
3731 | } |
3732 | ||
97e728d4 | 3733 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 3734 | { |
97e728d4 JB |
3735 | struct list_head *head = &info->space_info; |
3736 | struct btrfs_space_info *found; | |
e3ccfa98 | 3737 | |
97e728d4 JB |
3738 | rcu_read_lock(); |
3739 | list_for_each_entry_rcu(found, head, list) { | |
3740 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 3741 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 3742 | } |
97e728d4 | 3743 | rcu_read_unlock(); |
e3ccfa98 JB |
3744 | } |
3745 | ||
3c76cd84 MX |
3746 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
3747 | { | |
3748 | return (global->size << 1); | |
3749 | } | |
3750 | ||
e5bc2458 | 3751 | static int should_alloc_chunk(struct btrfs_root *root, |
698d0082 | 3752 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 3753 | { |
fb25e914 | 3754 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
424499db | 3755 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 3756 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 3757 | u64 thresh; |
e3ccfa98 | 3758 | |
0e4f8f88 CM |
3759 | if (force == CHUNK_ALLOC_FORCE) |
3760 | return 1; | |
3761 | ||
fb25e914 JB |
3762 | /* |
3763 | * We need to take into account the global rsv because for all intents | |
3764 | * and purposes it's used space. Don't worry about locking the | |
3765 | * global_rsv, it doesn't change except when the transaction commits. | |
3766 | */ | |
54338b5c | 3767 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
3c76cd84 | 3768 | num_allocated += calc_global_rsv_need_space(global_rsv); |
fb25e914 | 3769 | |
0e4f8f88 CM |
3770 | /* |
3771 | * in limited mode, we want to have some free space up to | |
3772 | * about 1% of the FS size. | |
3773 | */ | |
3774 | if (force == CHUNK_ALLOC_LIMITED) { | |
6c41761f | 3775 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 CM |
3776 | thresh = max_t(u64, 64 * 1024 * 1024, |
3777 | div_factor_fine(thresh, 1)); | |
3778 | ||
3779 | if (num_bytes - num_allocated < thresh) | |
3780 | return 1; | |
3781 | } | |
0e4f8f88 | 3782 | |
698d0082 | 3783 | if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8)) |
14ed0ca6 | 3784 | return 0; |
424499db | 3785 | return 1; |
32c00aff JB |
3786 | } |
3787 | ||
15d1ff81 LB |
3788 | static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type) |
3789 | { | |
3790 | u64 num_dev; | |
3791 | ||
53b381b3 DW |
3792 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
3793 | BTRFS_BLOCK_GROUP_RAID0 | | |
3794 | BTRFS_BLOCK_GROUP_RAID5 | | |
3795 | BTRFS_BLOCK_GROUP_RAID6)) | |
15d1ff81 LB |
3796 | num_dev = root->fs_info->fs_devices->rw_devices; |
3797 | else if (type & BTRFS_BLOCK_GROUP_RAID1) | |
3798 | num_dev = 2; | |
3799 | else | |
3800 | num_dev = 1; /* DUP or single */ | |
3801 | ||
3802 | /* metadata for updaing devices and chunk tree */ | |
3803 | return btrfs_calc_trans_metadata_size(root, num_dev + 1); | |
3804 | } | |
3805 | ||
3806 | static void check_system_chunk(struct btrfs_trans_handle *trans, | |
3807 | struct btrfs_root *root, u64 type) | |
3808 | { | |
3809 | struct btrfs_space_info *info; | |
3810 | u64 left; | |
3811 | u64 thresh; | |
3812 | ||
3813 | info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
3814 | spin_lock(&info->lock); | |
3815 | left = info->total_bytes - info->bytes_used - info->bytes_pinned - | |
3816 | info->bytes_reserved - info->bytes_readonly; | |
3817 | spin_unlock(&info->lock); | |
3818 | ||
3819 | thresh = get_system_chunk_thresh(root, type); | |
3820 | if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
c2cf52eb SK |
3821 | btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu", |
3822 | left, thresh, type); | |
15d1ff81 LB |
3823 | dump_space_info(info, 0, 0); |
3824 | } | |
3825 | ||
3826 | if (left < thresh) { | |
3827 | u64 flags; | |
3828 | ||
3829 | flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0); | |
3830 | btrfs_alloc_chunk(trans, root, flags); | |
3831 | } | |
3832 | } | |
3833 | ||
6324fbf3 | 3834 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 | 3835 | struct btrfs_root *extent_root, u64 flags, int force) |
9ed74f2d | 3836 | { |
6324fbf3 | 3837 | struct btrfs_space_info *space_info; |
97e728d4 | 3838 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 3839 | int wait_for_alloc = 0; |
9ed74f2d | 3840 | int ret = 0; |
9ed74f2d | 3841 | |
c6b305a8 JB |
3842 | /* Don't re-enter if we're already allocating a chunk */ |
3843 | if (trans->allocating_chunk) | |
3844 | return -ENOSPC; | |
3845 | ||
6324fbf3 | 3846 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
3847 | if (!space_info) { |
3848 | ret = update_space_info(extent_root->fs_info, flags, | |
3849 | 0, 0, &space_info); | |
79787eaa | 3850 | BUG_ON(ret); /* -ENOMEM */ |
9ed74f2d | 3851 | } |
79787eaa | 3852 | BUG_ON(!space_info); /* Logic error */ |
9ed74f2d | 3853 | |
6d74119f | 3854 | again: |
25179201 | 3855 | spin_lock(&space_info->lock); |
9e622d6b | 3856 | if (force < space_info->force_alloc) |
0e4f8f88 | 3857 | force = space_info->force_alloc; |
25179201 | 3858 | if (space_info->full) { |
09fb99a6 FDBM |
3859 | if (should_alloc_chunk(extent_root, space_info, force)) |
3860 | ret = -ENOSPC; | |
3861 | else | |
3862 | ret = 0; | |
25179201 | 3863 | spin_unlock(&space_info->lock); |
09fb99a6 | 3864 | return ret; |
9ed74f2d JB |
3865 | } |
3866 | ||
698d0082 | 3867 | if (!should_alloc_chunk(extent_root, space_info, force)) { |
25179201 | 3868 | spin_unlock(&space_info->lock); |
6d74119f JB |
3869 | return 0; |
3870 | } else if (space_info->chunk_alloc) { | |
3871 | wait_for_alloc = 1; | |
3872 | } else { | |
3873 | space_info->chunk_alloc = 1; | |
9ed74f2d | 3874 | } |
0e4f8f88 | 3875 | |
25179201 | 3876 | spin_unlock(&space_info->lock); |
9ed74f2d | 3877 | |
6d74119f JB |
3878 | mutex_lock(&fs_info->chunk_mutex); |
3879 | ||
3880 | /* | |
3881 | * The chunk_mutex is held throughout the entirety of a chunk | |
3882 | * allocation, so once we've acquired the chunk_mutex we know that the | |
3883 | * other guy is done and we need to recheck and see if we should | |
3884 | * allocate. | |
3885 | */ | |
3886 | if (wait_for_alloc) { | |
3887 | mutex_unlock(&fs_info->chunk_mutex); | |
3888 | wait_for_alloc = 0; | |
3889 | goto again; | |
3890 | } | |
3891 | ||
c6b305a8 JB |
3892 | trans->allocating_chunk = true; |
3893 | ||
67377734 JB |
3894 | /* |
3895 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3896 | * allocating mixed chunks instead of individual chunks. | |
3897 | */ | |
3898 | if (btrfs_mixed_space_info(space_info)) | |
3899 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3900 | ||
97e728d4 JB |
3901 | /* |
3902 | * if we're doing a data chunk, go ahead and make sure that | |
3903 | * we keep a reasonable number of metadata chunks allocated in the | |
3904 | * FS as well. | |
3905 | */ | |
9ed74f2d | 3906 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
3907 | fs_info->data_chunk_allocations++; |
3908 | if (!(fs_info->data_chunk_allocations % | |
3909 | fs_info->metadata_ratio)) | |
3910 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
3911 | } |
3912 | ||
15d1ff81 LB |
3913 | /* |
3914 | * Check if we have enough space in SYSTEM chunk because we may need | |
3915 | * to update devices. | |
3916 | */ | |
3917 | check_system_chunk(trans, extent_root, flags); | |
3918 | ||
2b82032c | 3919 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
c6b305a8 | 3920 | trans->allocating_chunk = false; |
92b8e897 | 3921 | |
9ed74f2d | 3922 | spin_lock(&space_info->lock); |
a81cb9a2 AO |
3923 | if (ret < 0 && ret != -ENOSPC) |
3924 | goto out; | |
9ed74f2d | 3925 | if (ret) |
6324fbf3 | 3926 | space_info->full = 1; |
424499db YZ |
3927 | else |
3928 | ret = 1; | |
6d74119f | 3929 | |
0e4f8f88 | 3930 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 3931 | out: |
6d74119f | 3932 | space_info->chunk_alloc = 0; |
9ed74f2d | 3933 | spin_unlock(&space_info->lock); |
a25c75d5 | 3934 | mutex_unlock(&fs_info->chunk_mutex); |
0f9dd46c | 3935 | return ret; |
6324fbf3 | 3936 | } |
9ed74f2d | 3937 | |
a80c8dcf JB |
3938 | static int can_overcommit(struct btrfs_root *root, |
3939 | struct btrfs_space_info *space_info, u64 bytes, | |
08e007d2 | 3940 | enum btrfs_reserve_flush_enum flush) |
a80c8dcf | 3941 | { |
96f1bb57 | 3942 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
a80c8dcf | 3943 | u64 profile = btrfs_get_alloc_profile(root, 0); |
3c76cd84 | 3944 | u64 space_size; |
a80c8dcf JB |
3945 | u64 avail; |
3946 | u64 used; | |
3947 | ||
3948 | used = space_info->bytes_used + space_info->bytes_reserved + | |
96f1bb57 JB |
3949 | space_info->bytes_pinned + space_info->bytes_readonly; |
3950 | ||
96f1bb57 JB |
3951 | /* |
3952 | * We only want to allow over committing if we have lots of actual space | |
3953 | * free, but if we don't have enough space to handle the global reserve | |
3954 | * space then we could end up having a real enospc problem when trying | |
3955 | * to allocate a chunk or some other such important allocation. | |
3956 | */ | |
3c76cd84 MX |
3957 | spin_lock(&global_rsv->lock); |
3958 | space_size = calc_global_rsv_need_space(global_rsv); | |
3959 | spin_unlock(&global_rsv->lock); | |
3960 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
3961 | return 0; |
3962 | ||
3963 | used += space_info->bytes_may_use; | |
a80c8dcf JB |
3964 | |
3965 | spin_lock(&root->fs_info->free_chunk_lock); | |
3966 | avail = root->fs_info->free_chunk_space; | |
3967 | spin_unlock(&root->fs_info->free_chunk_lock); | |
3968 | ||
3969 | /* | |
3970 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
3971 | * space is actually useable. For raid56, the space info used |
3972 | * doesn't include the parity drive, so we don't have to | |
3973 | * change the math | |
a80c8dcf JB |
3974 | */ |
3975 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
3976 | BTRFS_BLOCK_GROUP_RAID1 | | |
3977 | BTRFS_BLOCK_GROUP_RAID10)) | |
3978 | avail >>= 1; | |
3979 | ||
3980 | /* | |
561c294d MX |
3981 | * If we aren't flushing all things, let us overcommit up to |
3982 | * 1/2th of the space. If we can flush, don't let us overcommit | |
3983 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 3984 | */ |
08e007d2 | 3985 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 3986 | avail >>= 3; |
a80c8dcf | 3987 | else |
14575aef | 3988 | avail >>= 1; |
a80c8dcf | 3989 | |
14575aef | 3990 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
3991 | return 1; |
3992 | return 0; | |
3993 | } | |
3994 | ||
48a3b636 ES |
3995 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root, |
3996 | unsigned long nr_pages) | |
da633a42 MX |
3997 | { |
3998 | struct super_block *sb = root->fs_info->sb; | |
da633a42 | 3999 | |
925a6efb JB |
4000 | if (down_read_trylock(&sb->s_umount)) { |
4001 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4002 | up_read(&sb->s_umount); | |
4003 | } else { | |
da633a42 MX |
4004 | /* |
4005 | * We needn't worry the filesystem going from r/w to r/o though | |
4006 | * we don't acquire ->s_umount mutex, because the filesystem | |
4007 | * should guarantee the delalloc inodes list be empty after | |
4008 | * the filesystem is readonly(all dirty pages are written to | |
4009 | * the disk). | |
4010 | */ | |
eb73c1b7 | 4011 | btrfs_start_all_delalloc_inodes(root->fs_info, 0); |
98ad69cf | 4012 | if (!current->journal_info) |
f0de181c | 4013 | btrfs_wait_all_ordered_extents(root->fs_info); |
da633a42 MX |
4014 | } |
4015 | } | |
4016 | ||
9ed74f2d | 4017 | /* |
5da9d01b | 4018 | * shrink metadata reservation for delalloc |
9ed74f2d | 4019 | */ |
f4c738c2 JB |
4020 | static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig, |
4021 | bool wait_ordered) | |
5da9d01b | 4022 | { |
0ca1f7ce | 4023 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 4024 | struct btrfs_space_info *space_info; |
663350ac | 4025 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4026 | u64 delalloc_bytes; |
5da9d01b | 4027 | u64 max_reclaim; |
b1953bce | 4028 | long time_left; |
877da174 | 4029 | unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT; |
b1953bce | 4030 | int loops = 0; |
08e007d2 | 4031 | enum btrfs_reserve_flush_enum flush; |
5da9d01b | 4032 | |
663350ac | 4033 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0ca1f7ce | 4034 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 4035 | space_info = block_rsv->space_info; |
bf9022e0 CM |
4036 | |
4037 | smp_mb(); | |
963d678b MX |
4038 | delalloc_bytes = percpu_counter_sum_positive( |
4039 | &root->fs_info->delalloc_bytes); | |
f4c738c2 | 4040 | if (delalloc_bytes == 0) { |
fdb5effd | 4041 | if (trans) |
f4c738c2 | 4042 | return; |
f0de181c | 4043 | btrfs_wait_all_ordered_extents(root->fs_info); |
f4c738c2 | 4044 | return; |
fdb5effd JB |
4045 | } |
4046 | ||
f4c738c2 JB |
4047 | while (delalloc_bytes && loops < 3) { |
4048 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
4049 | nr_pages = max_reclaim >> PAGE_CACHE_SHIFT; | |
da633a42 | 4050 | btrfs_writeback_inodes_sb_nr(root, nr_pages); |
dea31f52 JB |
4051 | /* |
4052 | * We need to wait for the async pages to actually start before | |
4053 | * we do anything. | |
4054 | */ | |
4055 | wait_event(root->fs_info->async_submit_wait, | |
4056 | !atomic_read(&root->fs_info->async_delalloc_pages)); | |
4057 | ||
08e007d2 MX |
4058 | if (!trans) |
4059 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
4060 | else | |
4061 | flush = BTRFS_RESERVE_NO_FLUSH; | |
0019f10d | 4062 | spin_lock(&space_info->lock); |
08e007d2 | 4063 | if (can_overcommit(root, space_info, orig, flush)) { |
f4c738c2 JB |
4064 | spin_unlock(&space_info->lock); |
4065 | break; | |
4066 | } | |
0019f10d | 4067 | spin_unlock(&space_info->lock); |
5da9d01b | 4068 | |
36e39c40 | 4069 | loops++; |
f104d044 | 4070 | if (wait_ordered && !trans) { |
f0de181c | 4071 | btrfs_wait_all_ordered_extents(root->fs_info); |
f104d044 | 4072 | } else { |
f4c738c2 | 4073 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4074 | if (time_left) |
4075 | break; | |
4076 | } | |
f4c738c2 | 4077 | smp_mb(); |
963d678b MX |
4078 | delalloc_bytes = percpu_counter_sum_positive( |
4079 | &root->fs_info->delalloc_bytes); | |
5da9d01b | 4080 | } |
5da9d01b YZ |
4081 | } |
4082 | ||
663350ac JB |
4083 | /** |
4084 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4085 | * @root - the root we're allocating for | |
4086 | * @bytes - the number of bytes we want to reserve | |
4087 | * @force - force the commit | |
8bb8ab2e | 4088 | * |
663350ac JB |
4089 | * This will check to make sure that committing the transaction will actually |
4090 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4091 | * will return -ENOSPC. | |
8bb8ab2e | 4092 | */ |
663350ac JB |
4093 | static int may_commit_transaction(struct btrfs_root *root, |
4094 | struct btrfs_space_info *space_info, | |
4095 | u64 bytes, int force) | |
4096 | { | |
4097 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; | |
4098 | struct btrfs_trans_handle *trans; | |
4099 | ||
4100 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4101 | if (trans) | |
4102 | return -EAGAIN; | |
4103 | ||
4104 | if (force) | |
4105 | goto commit; | |
4106 | ||
4107 | /* See if there is enough pinned space to make this reservation */ | |
4108 | spin_lock(&space_info->lock); | |
b150a4f1 JB |
4109 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
4110 | bytes) >= 0) { | |
663350ac JB |
4111 | spin_unlock(&space_info->lock); |
4112 | goto commit; | |
4113 | } | |
4114 | spin_unlock(&space_info->lock); | |
4115 | ||
4116 | /* | |
4117 | * See if there is some space in the delayed insertion reservation for | |
4118 | * this reservation. | |
4119 | */ | |
4120 | if (space_info != delayed_rsv->space_info) | |
4121 | return -ENOSPC; | |
4122 | ||
d9b0218f | 4123 | spin_lock(&space_info->lock); |
663350ac | 4124 | spin_lock(&delayed_rsv->lock); |
b150a4f1 JB |
4125 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
4126 | bytes - delayed_rsv->size) >= 0) { | |
663350ac | 4127 | spin_unlock(&delayed_rsv->lock); |
d9b0218f | 4128 | spin_unlock(&space_info->lock); |
663350ac JB |
4129 | return -ENOSPC; |
4130 | } | |
4131 | spin_unlock(&delayed_rsv->lock); | |
d9b0218f | 4132 | spin_unlock(&space_info->lock); |
663350ac JB |
4133 | |
4134 | commit: | |
4135 | trans = btrfs_join_transaction(root); | |
4136 | if (IS_ERR(trans)) | |
4137 | return -ENOSPC; | |
4138 | ||
4139 | return btrfs_commit_transaction(trans, root); | |
4140 | } | |
4141 | ||
96c3f433 | 4142 | enum flush_state { |
67b0fd63 JB |
4143 | FLUSH_DELAYED_ITEMS_NR = 1, |
4144 | FLUSH_DELAYED_ITEMS = 2, | |
4145 | FLUSH_DELALLOC = 3, | |
4146 | FLUSH_DELALLOC_WAIT = 4, | |
ea658bad JB |
4147 | ALLOC_CHUNK = 5, |
4148 | COMMIT_TRANS = 6, | |
96c3f433 JB |
4149 | }; |
4150 | ||
4151 | static int flush_space(struct btrfs_root *root, | |
4152 | struct btrfs_space_info *space_info, u64 num_bytes, | |
4153 | u64 orig_bytes, int state) | |
4154 | { | |
4155 | struct btrfs_trans_handle *trans; | |
4156 | int nr; | |
f4c738c2 | 4157 | int ret = 0; |
96c3f433 JB |
4158 | |
4159 | switch (state) { | |
96c3f433 JB |
4160 | case FLUSH_DELAYED_ITEMS_NR: |
4161 | case FLUSH_DELAYED_ITEMS: | |
4162 | if (state == FLUSH_DELAYED_ITEMS_NR) { | |
4163 | u64 bytes = btrfs_calc_trans_metadata_size(root, 1); | |
4164 | ||
4165 | nr = (int)div64_u64(num_bytes, bytes); | |
4166 | if (!nr) | |
4167 | nr = 1; | |
4168 | nr *= 2; | |
4169 | } else { | |
4170 | nr = -1; | |
4171 | } | |
4172 | trans = btrfs_join_transaction(root); | |
4173 | if (IS_ERR(trans)) { | |
4174 | ret = PTR_ERR(trans); | |
4175 | break; | |
4176 | } | |
4177 | ret = btrfs_run_delayed_items_nr(trans, root, nr); | |
4178 | btrfs_end_transaction(trans, root); | |
4179 | break; | |
67b0fd63 JB |
4180 | case FLUSH_DELALLOC: |
4181 | case FLUSH_DELALLOC_WAIT: | |
4182 | shrink_delalloc(root, num_bytes, orig_bytes, | |
4183 | state == FLUSH_DELALLOC_WAIT); | |
4184 | break; | |
ea658bad JB |
4185 | case ALLOC_CHUNK: |
4186 | trans = btrfs_join_transaction(root); | |
4187 | if (IS_ERR(trans)) { | |
4188 | ret = PTR_ERR(trans); | |
4189 | break; | |
4190 | } | |
4191 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
ea658bad JB |
4192 | btrfs_get_alloc_profile(root, 0), |
4193 | CHUNK_ALLOC_NO_FORCE); | |
4194 | btrfs_end_transaction(trans, root); | |
4195 | if (ret == -ENOSPC) | |
4196 | ret = 0; | |
4197 | break; | |
96c3f433 JB |
4198 | case COMMIT_TRANS: |
4199 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); | |
4200 | break; | |
4201 | default: | |
4202 | ret = -ENOSPC; | |
4203 | break; | |
4204 | } | |
4205 | ||
4206 | return ret; | |
4207 | } | |
4a92b1b8 JB |
4208 | /** |
4209 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
4210 | * @root - the root we're allocating for | |
4211 | * @block_rsv - the block_rsv we're allocating for | |
4212 | * @orig_bytes - the number of bytes we want | |
48fc7f7e | 4213 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 4214 | * |
4a92b1b8 JB |
4215 | * This will reserve orgi_bytes number of bytes from the space info associated |
4216 | * with the block_rsv. If there is not enough space it will make an attempt to | |
4217 | * flush out space to make room. It will do this by flushing delalloc if | |
4218 | * possible or committing the transaction. If flush is 0 then no attempts to | |
4219 | * regain reservations will be made and this will fail if there is not enough | |
4220 | * space already. | |
8bb8ab2e | 4221 | */ |
4a92b1b8 | 4222 | static int reserve_metadata_bytes(struct btrfs_root *root, |
8bb8ab2e | 4223 | struct btrfs_block_rsv *block_rsv, |
08e007d2 MX |
4224 | u64 orig_bytes, |
4225 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 4226 | { |
f0486c68 | 4227 | struct btrfs_space_info *space_info = block_rsv->space_info; |
2bf64758 | 4228 | u64 used; |
8bb8ab2e | 4229 | u64 num_bytes = orig_bytes; |
67b0fd63 | 4230 | int flush_state = FLUSH_DELAYED_ITEMS_NR; |
8bb8ab2e | 4231 | int ret = 0; |
fdb5effd | 4232 | bool flushing = false; |
9ed74f2d | 4233 | |
8bb8ab2e | 4234 | again: |
fdb5effd | 4235 | ret = 0; |
8bb8ab2e | 4236 | spin_lock(&space_info->lock); |
fdb5effd | 4237 | /* |
08e007d2 MX |
4238 | * We only want to wait if somebody other than us is flushing and we |
4239 | * are actually allowed to flush all things. | |
fdb5effd | 4240 | */ |
08e007d2 MX |
4241 | while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing && |
4242 | space_info->flush) { | |
fdb5effd JB |
4243 | spin_unlock(&space_info->lock); |
4244 | /* | |
4245 | * If we have a trans handle we can't wait because the flusher | |
4246 | * may have to commit the transaction, which would mean we would | |
4247 | * deadlock since we are waiting for the flusher to finish, but | |
4248 | * hold the current transaction open. | |
4249 | */ | |
663350ac | 4250 | if (current->journal_info) |
fdb5effd | 4251 | return -EAGAIN; |
b9688bb8 AJ |
4252 | ret = wait_event_killable(space_info->wait, !space_info->flush); |
4253 | /* Must have been killed, return */ | |
4254 | if (ret) | |
fdb5effd JB |
4255 | return -EINTR; |
4256 | ||
4257 | spin_lock(&space_info->lock); | |
4258 | } | |
4259 | ||
4260 | ret = -ENOSPC; | |
2bf64758 JB |
4261 | used = space_info->bytes_used + space_info->bytes_reserved + |
4262 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4263 | space_info->bytes_may_use; | |
9ed74f2d | 4264 | |
8bb8ab2e JB |
4265 | /* |
4266 | * The idea here is that we've not already over-reserved the block group | |
4267 | * then we can go ahead and save our reservation first and then start | |
4268 | * flushing if we need to. Otherwise if we've already overcommitted | |
4269 | * lets start flushing stuff first and then come back and try to make | |
4270 | * our reservation. | |
4271 | */ | |
2bf64758 JB |
4272 | if (used <= space_info->total_bytes) { |
4273 | if (used + orig_bytes <= space_info->total_bytes) { | |
fb25e914 | 4274 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 4275 | trace_btrfs_space_reservation(root->fs_info, |
2bcc0328 | 4276 | "space_info", space_info->flags, orig_bytes, 1); |
8bb8ab2e JB |
4277 | ret = 0; |
4278 | } else { | |
4279 | /* | |
4280 | * Ok set num_bytes to orig_bytes since we aren't | |
4281 | * overocmmitted, this way we only try and reclaim what | |
4282 | * we need. | |
4283 | */ | |
4284 | num_bytes = orig_bytes; | |
4285 | } | |
4286 | } else { | |
4287 | /* | |
4288 | * Ok we're over committed, set num_bytes to the overcommitted | |
4289 | * amount plus the amount of bytes that we need for this | |
4290 | * reservation. | |
4291 | */ | |
2bf64758 | 4292 | num_bytes = used - space_info->total_bytes + |
96c3f433 | 4293 | (orig_bytes * 2); |
8bb8ab2e | 4294 | } |
9ed74f2d | 4295 | |
44734ed1 JB |
4296 | if (ret && can_overcommit(root, space_info, orig_bytes, flush)) { |
4297 | space_info->bytes_may_use += orig_bytes; | |
4298 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
4299 | space_info->flags, orig_bytes, | |
4300 | 1); | |
4301 | ret = 0; | |
2bf64758 JB |
4302 | } |
4303 | ||
8bb8ab2e JB |
4304 | /* |
4305 | * Couldn't make our reservation, save our place so while we're trying | |
4306 | * to reclaim space we can actually use it instead of somebody else | |
4307 | * stealing it from us. | |
08e007d2 MX |
4308 | * |
4309 | * We make the other tasks wait for the flush only when we can flush | |
4310 | * all things. | |
8bb8ab2e | 4311 | */ |
72bcd99d | 4312 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
fdb5effd JB |
4313 | flushing = true; |
4314 | space_info->flush = 1; | |
8bb8ab2e | 4315 | } |
9ed74f2d | 4316 | |
f0486c68 | 4317 | spin_unlock(&space_info->lock); |
9ed74f2d | 4318 | |
08e007d2 | 4319 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
8bb8ab2e | 4320 | goto out; |
f0486c68 | 4321 | |
96c3f433 JB |
4322 | ret = flush_space(root, space_info, num_bytes, orig_bytes, |
4323 | flush_state); | |
4324 | flush_state++; | |
08e007d2 MX |
4325 | |
4326 | /* | |
4327 | * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock | |
4328 | * would happen. So skip delalloc flush. | |
4329 | */ | |
4330 | if (flush == BTRFS_RESERVE_FLUSH_LIMIT && | |
4331 | (flush_state == FLUSH_DELALLOC || | |
4332 | flush_state == FLUSH_DELALLOC_WAIT)) | |
4333 | flush_state = ALLOC_CHUNK; | |
4334 | ||
96c3f433 | 4335 | if (!ret) |
8bb8ab2e | 4336 | goto again; |
08e007d2 MX |
4337 | else if (flush == BTRFS_RESERVE_FLUSH_LIMIT && |
4338 | flush_state < COMMIT_TRANS) | |
4339 | goto again; | |
4340 | else if (flush == BTRFS_RESERVE_FLUSH_ALL && | |
4341 | flush_state <= COMMIT_TRANS) | |
8bb8ab2e JB |
4342 | goto again; |
4343 | ||
4344 | out: | |
5d80366e JB |
4345 | if (ret == -ENOSPC && |
4346 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
4347 | struct btrfs_block_rsv *global_rsv = | |
4348 | &root->fs_info->global_block_rsv; | |
4349 | ||
4350 | if (block_rsv != global_rsv && | |
4351 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
4352 | ret = 0; | |
4353 | } | |
cab45e22 JM |
4354 | if (ret == -ENOSPC) |
4355 | trace_btrfs_space_reservation(root->fs_info, | |
4356 | "space_info:enospc", | |
4357 | space_info->flags, orig_bytes, 1); | |
fdb5effd | 4358 | if (flushing) { |
8bb8ab2e | 4359 | spin_lock(&space_info->lock); |
fdb5effd JB |
4360 | space_info->flush = 0; |
4361 | wake_up_all(&space_info->wait); | |
8bb8ab2e | 4362 | spin_unlock(&space_info->lock); |
f0486c68 | 4363 | } |
f0486c68 YZ |
4364 | return ret; |
4365 | } | |
4366 | ||
79787eaa JM |
4367 | static struct btrfs_block_rsv *get_block_rsv( |
4368 | const struct btrfs_trans_handle *trans, | |
4369 | const struct btrfs_root *root) | |
f0486c68 | 4370 | { |
4c13d758 JB |
4371 | struct btrfs_block_rsv *block_rsv = NULL; |
4372 | ||
0e721106 JB |
4373 | if (root->ref_cows) |
4374 | block_rsv = trans->block_rsv; | |
4375 | ||
4376 | if (root == root->fs_info->csum_root && trans->adding_csums) | |
f0486c68 | 4377 | block_rsv = trans->block_rsv; |
4c13d758 | 4378 | |
f7a81ea4 SB |
4379 | if (root == root->fs_info->uuid_root) |
4380 | block_rsv = trans->block_rsv; | |
4381 | ||
4c13d758 | 4382 | if (!block_rsv) |
f0486c68 YZ |
4383 | block_rsv = root->block_rsv; |
4384 | ||
4385 | if (!block_rsv) | |
4386 | block_rsv = &root->fs_info->empty_block_rsv; | |
4387 | ||
4388 | return block_rsv; | |
4389 | } | |
4390 | ||
4391 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
4392 | u64 num_bytes) | |
4393 | { | |
4394 | int ret = -ENOSPC; | |
4395 | spin_lock(&block_rsv->lock); | |
4396 | if (block_rsv->reserved >= num_bytes) { | |
4397 | block_rsv->reserved -= num_bytes; | |
4398 | if (block_rsv->reserved < block_rsv->size) | |
4399 | block_rsv->full = 0; | |
4400 | ret = 0; | |
4401 | } | |
4402 | spin_unlock(&block_rsv->lock); | |
4403 | return ret; | |
4404 | } | |
4405 | ||
4406 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
4407 | u64 num_bytes, int update_size) | |
4408 | { | |
4409 | spin_lock(&block_rsv->lock); | |
4410 | block_rsv->reserved += num_bytes; | |
4411 | if (update_size) | |
4412 | block_rsv->size += num_bytes; | |
4413 | else if (block_rsv->reserved >= block_rsv->size) | |
4414 | block_rsv->full = 1; | |
4415 | spin_unlock(&block_rsv->lock); | |
4416 | } | |
4417 | ||
d52be818 JB |
4418 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
4419 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
4420 | int min_factor) | |
4421 | { | |
4422 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
4423 | u64 min_bytes; | |
4424 | ||
4425 | if (global_rsv->space_info != dest->space_info) | |
4426 | return -ENOSPC; | |
4427 | ||
4428 | spin_lock(&global_rsv->lock); | |
4429 | min_bytes = div_factor(global_rsv->size, min_factor); | |
4430 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
4431 | spin_unlock(&global_rsv->lock); | |
4432 | return -ENOSPC; | |
4433 | } | |
4434 | global_rsv->reserved -= num_bytes; | |
4435 | if (global_rsv->reserved < global_rsv->size) | |
4436 | global_rsv->full = 0; | |
4437 | spin_unlock(&global_rsv->lock); | |
4438 | ||
4439 | block_rsv_add_bytes(dest, num_bytes, 1); | |
4440 | return 0; | |
4441 | } | |
4442 | ||
8c2a3ca2 JB |
4443 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
4444 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 4445 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
4446 | { |
4447 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
4448 | ||
4449 | spin_lock(&block_rsv->lock); | |
4450 | if (num_bytes == (u64)-1) | |
4451 | num_bytes = block_rsv->size; | |
4452 | block_rsv->size -= num_bytes; | |
4453 | if (block_rsv->reserved >= block_rsv->size) { | |
4454 | num_bytes = block_rsv->reserved - block_rsv->size; | |
4455 | block_rsv->reserved = block_rsv->size; | |
4456 | block_rsv->full = 1; | |
4457 | } else { | |
4458 | num_bytes = 0; | |
4459 | } | |
4460 | spin_unlock(&block_rsv->lock); | |
4461 | ||
4462 | if (num_bytes > 0) { | |
4463 | if (dest) { | |
e9e22899 JB |
4464 | spin_lock(&dest->lock); |
4465 | if (!dest->full) { | |
4466 | u64 bytes_to_add; | |
4467 | ||
4468 | bytes_to_add = dest->size - dest->reserved; | |
4469 | bytes_to_add = min(num_bytes, bytes_to_add); | |
4470 | dest->reserved += bytes_to_add; | |
4471 | if (dest->reserved >= dest->size) | |
4472 | dest->full = 1; | |
4473 | num_bytes -= bytes_to_add; | |
4474 | } | |
4475 | spin_unlock(&dest->lock); | |
4476 | } | |
4477 | if (num_bytes) { | |
f0486c68 | 4478 | spin_lock(&space_info->lock); |
fb25e914 | 4479 | space_info->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4480 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4481 | space_info->flags, num_bytes, 0); |
f0486c68 | 4482 | spin_unlock(&space_info->lock); |
4e06bdd6 | 4483 | } |
9ed74f2d | 4484 | } |
f0486c68 | 4485 | } |
4e06bdd6 | 4486 | |
f0486c68 YZ |
4487 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
4488 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
4489 | { | |
4490 | int ret; | |
9ed74f2d | 4491 | |
f0486c68 YZ |
4492 | ret = block_rsv_use_bytes(src, num_bytes); |
4493 | if (ret) | |
4494 | return ret; | |
9ed74f2d | 4495 | |
f0486c68 | 4496 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
4497 | return 0; |
4498 | } | |
4499 | ||
66d8f3dd | 4500 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 4501 | { |
f0486c68 YZ |
4502 | memset(rsv, 0, sizeof(*rsv)); |
4503 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 4504 | rsv->type = type; |
f0486c68 YZ |
4505 | } |
4506 | ||
66d8f3dd MX |
4507 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root, |
4508 | unsigned short type) | |
f0486c68 YZ |
4509 | { |
4510 | struct btrfs_block_rsv *block_rsv; | |
4511 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 4512 | |
f0486c68 YZ |
4513 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
4514 | if (!block_rsv) | |
4515 | return NULL; | |
9ed74f2d | 4516 | |
66d8f3dd | 4517 | btrfs_init_block_rsv(block_rsv, type); |
f0486c68 YZ |
4518 | block_rsv->space_info = __find_space_info(fs_info, |
4519 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
4520 | return block_rsv; |
4521 | } | |
9ed74f2d | 4522 | |
f0486c68 YZ |
4523 | void btrfs_free_block_rsv(struct btrfs_root *root, |
4524 | struct btrfs_block_rsv *rsv) | |
4525 | { | |
2aaa6655 JB |
4526 | if (!rsv) |
4527 | return; | |
dabdb640 JB |
4528 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
4529 | kfree(rsv); | |
9ed74f2d JB |
4530 | } |
4531 | ||
08e007d2 MX |
4532 | int btrfs_block_rsv_add(struct btrfs_root *root, |
4533 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
4534 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 4535 | { |
f0486c68 | 4536 | int ret; |
9ed74f2d | 4537 | |
f0486c68 YZ |
4538 | if (num_bytes == 0) |
4539 | return 0; | |
8bb8ab2e | 4540 | |
61b520a9 | 4541 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
4542 | if (!ret) { |
4543 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
4544 | return 0; | |
4545 | } | |
9ed74f2d | 4546 | |
f0486c68 | 4547 | return ret; |
f0486c68 | 4548 | } |
9ed74f2d | 4549 | |
4a92b1b8 | 4550 | int btrfs_block_rsv_check(struct btrfs_root *root, |
36ba022a | 4551 | struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
4552 | { |
4553 | u64 num_bytes = 0; | |
f0486c68 | 4554 | int ret = -ENOSPC; |
9ed74f2d | 4555 | |
f0486c68 YZ |
4556 | if (!block_rsv) |
4557 | return 0; | |
9ed74f2d | 4558 | |
f0486c68 | 4559 | spin_lock(&block_rsv->lock); |
36ba022a JB |
4560 | num_bytes = div_factor(block_rsv->size, min_factor); |
4561 | if (block_rsv->reserved >= num_bytes) | |
4562 | ret = 0; | |
4563 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 4564 | |
36ba022a JB |
4565 | return ret; |
4566 | } | |
4567 | ||
08e007d2 MX |
4568 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
4569 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
4570 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
4571 | { |
4572 | u64 num_bytes = 0; | |
4573 | int ret = -ENOSPC; | |
4574 | ||
4575 | if (!block_rsv) | |
4576 | return 0; | |
4577 | ||
4578 | spin_lock(&block_rsv->lock); | |
4579 | num_bytes = min_reserved; | |
13553e52 | 4580 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 4581 | ret = 0; |
13553e52 | 4582 | else |
f0486c68 | 4583 | num_bytes -= block_rsv->reserved; |
f0486c68 | 4584 | spin_unlock(&block_rsv->lock); |
13553e52 | 4585 | |
f0486c68 YZ |
4586 | if (!ret) |
4587 | return 0; | |
4588 | ||
aa38a711 | 4589 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
4590 | if (!ret) { |
4591 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 4592 | return 0; |
6a63209f | 4593 | } |
9ed74f2d | 4594 | |
13553e52 | 4595 | return ret; |
f0486c68 YZ |
4596 | } |
4597 | ||
4598 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, | |
4599 | struct btrfs_block_rsv *dst_rsv, | |
4600 | u64 num_bytes) | |
4601 | { | |
4602 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
4603 | } | |
4604 | ||
4605 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
4606 | struct btrfs_block_rsv *block_rsv, | |
4607 | u64 num_bytes) | |
4608 | { | |
4609 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
4610 | if (global_rsv->full || global_rsv == block_rsv || | |
4611 | block_rsv->space_info != global_rsv->space_info) | |
4612 | global_rsv = NULL; | |
8c2a3ca2 JB |
4613 | block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv, |
4614 | num_bytes); | |
6a63209f JB |
4615 | } |
4616 | ||
4617 | /* | |
8929ecfa YZ |
4618 | * helper to calculate size of global block reservation. |
4619 | * the desired value is sum of space used by extent tree, | |
4620 | * checksum tree and root tree | |
6a63209f | 4621 | */ |
8929ecfa | 4622 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 4623 | { |
8929ecfa YZ |
4624 | struct btrfs_space_info *sinfo; |
4625 | u64 num_bytes; | |
4626 | u64 meta_used; | |
4627 | u64 data_used; | |
6c41761f | 4628 | int csum_size = btrfs_super_csum_size(fs_info->super_copy); |
6a63209f | 4629 | |
8929ecfa YZ |
4630 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
4631 | spin_lock(&sinfo->lock); | |
4632 | data_used = sinfo->bytes_used; | |
4633 | spin_unlock(&sinfo->lock); | |
33b4d47f | 4634 | |
8929ecfa YZ |
4635 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
4636 | spin_lock(&sinfo->lock); | |
6d48755d JB |
4637 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
4638 | data_used = 0; | |
8929ecfa YZ |
4639 | meta_used = sinfo->bytes_used; |
4640 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 4641 | |
8929ecfa YZ |
4642 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
4643 | csum_size * 2; | |
4644 | num_bytes += div64_u64(data_used + meta_used, 50); | |
4e06bdd6 | 4645 | |
8929ecfa | 4646 | if (num_bytes * 3 > meta_used) |
8e62c2de | 4647 | num_bytes = div64_u64(meta_used, 3); |
ab6e2410 | 4648 | |
8929ecfa YZ |
4649 | return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10); |
4650 | } | |
6a63209f | 4651 | |
8929ecfa YZ |
4652 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
4653 | { | |
4654 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
4655 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
4656 | u64 num_bytes; | |
6a63209f | 4657 | |
8929ecfa | 4658 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 4659 | |
8929ecfa | 4660 | spin_lock(&sinfo->lock); |
1f699d38 | 4661 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 4662 | |
fdf30d1c | 4663 | block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024); |
4e06bdd6 | 4664 | |
8929ecfa | 4665 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
6d48755d JB |
4666 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
4667 | sinfo->bytes_may_use; | |
8929ecfa YZ |
4668 | |
4669 | if (sinfo->total_bytes > num_bytes) { | |
4670 | num_bytes = sinfo->total_bytes - num_bytes; | |
4671 | block_rsv->reserved += num_bytes; | |
fb25e914 | 4672 | sinfo->bytes_may_use += num_bytes; |
8c2a3ca2 | 4673 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4674 | sinfo->flags, num_bytes, 1); |
6a63209f | 4675 | } |
6a63209f | 4676 | |
8929ecfa YZ |
4677 | if (block_rsv->reserved >= block_rsv->size) { |
4678 | num_bytes = block_rsv->reserved - block_rsv->size; | |
fb25e914 | 4679 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4680 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4681 | sinfo->flags, num_bytes, 0); |
8929ecfa YZ |
4682 | block_rsv->reserved = block_rsv->size; |
4683 | block_rsv->full = 1; | |
4684 | } | |
182608c8 | 4685 | |
8929ecfa | 4686 | spin_unlock(&block_rsv->lock); |
1f699d38 | 4687 | spin_unlock(&sinfo->lock); |
6a63209f JB |
4688 | } |
4689 | ||
f0486c68 | 4690 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4691 | { |
f0486c68 | 4692 | struct btrfs_space_info *space_info; |
6a63209f | 4693 | |
f0486c68 YZ |
4694 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
4695 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 4696 | |
f0486c68 | 4697 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 4698 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 4699 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
4700 | fs_info->trans_block_rsv.space_info = space_info; |
4701 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 4702 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 4703 | |
8929ecfa YZ |
4704 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
4705 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
4706 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
4707 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
4708 | if (fs_info->quota_root) |
4709 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 4710 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 4711 | |
8929ecfa | 4712 | update_global_block_rsv(fs_info); |
6a63209f JB |
4713 | } |
4714 | ||
8929ecfa | 4715 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4716 | { |
8c2a3ca2 JB |
4717 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
4718 | (u64)-1); | |
8929ecfa YZ |
4719 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
4720 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
4721 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
4722 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
4723 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
4724 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
4725 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
4726 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
4727 | } |
4728 | ||
a22285a6 YZ |
4729 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
4730 | struct btrfs_root *root) | |
6a63209f | 4731 | { |
0e721106 JB |
4732 | if (!trans->block_rsv) |
4733 | return; | |
4734 | ||
a22285a6 YZ |
4735 | if (!trans->bytes_reserved) |
4736 | return; | |
6a63209f | 4737 | |
e77266e4 | 4738 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 4739 | trans->transid, trans->bytes_reserved, 0); |
b24e03db | 4740 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); |
a22285a6 YZ |
4741 | trans->bytes_reserved = 0; |
4742 | } | |
6a63209f | 4743 | |
79787eaa | 4744 | /* Can only return 0 or -ENOSPC */ |
d68fc57b YZ |
4745 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
4746 | struct inode *inode) | |
4747 | { | |
4748 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4749 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4750 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
4751 | ||
4752 | /* | |
fcb80c2a JB |
4753 | * We need to hold space in order to delete our orphan item once we've |
4754 | * added it, so this takes the reservation so we can release it later | |
4755 | * when we are truly done with the orphan item. | |
d68fc57b | 4756 | */ |
ff5714cc | 4757 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4758 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4759 | btrfs_ino(inode), num_bytes, 1); | |
d68fc57b | 4760 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
6a63209f JB |
4761 | } |
4762 | ||
d68fc57b | 4763 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 4764 | { |
d68fc57b | 4765 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 4766 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4767 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4768 | btrfs_ino(inode), num_bytes, 0); | |
d68fc57b YZ |
4769 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
4770 | } | |
97e728d4 | 4771 | |
d5c12070 MX |
4772 | /* |
4773 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
4774 | * root: the root of the parent directory | |
4775 | * rsv: block reservation | |
4776 | * items: the number of items that we need do reservation | |
4777 | * qgroup_reserved: used to return the reserved size in qgroup | |
4778 | * | |
4779 | * This function is used to reserve the space for snapshot/subvolume | |
4780 | * creation and deletion. Those operations are different with the | |
4781 | * common file/directory operations, they change two fs/file trees | |
4782 | * and root tree, the number of items that the qgroup reserves is | |
4783 | * different with the free space reservation. So we can not use | |
4784 | * the space reseravtion mechanism in start_transaction(). | |
4785 | */ | |
4786 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
4787 | struct btrfs_block_rsv *rsv, | |
4788 | int items, | |
ee3441b4 JM |
4789 | u64 *qgroup_reserved, |
4790 | bool use_global_rsv) | |
a22285a6 | 4791 | { |
d5c12070 MX |
4792 | u64 num_bytes; |
4793 | int ret; | |
ee3441b4 | 4794 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
d5c12070 MX |
4795 | |
4796 | if (root->fs_info->quota_enabled) { | |
4797 | /* One for parent inode, two for dir entries */ | |
4798 | num_bytes = 3 * root->leafsize; | |
4799 | ret = btrfs_qgroup_reserve(root, num_bytes); | |
4800 | if (ret) | |
4801 | return ret; | |
4802 | } else { | |
4803 | num_bytes = 0; | |
4804 | } | |
4805 | ||
4806 | *qgroup_reserved = num_bytes; | |
4807 | ||
4808 | num_bytes = btrfs_calc_trans_metadata_size(root, items); | |
4809 | rsv->space_info = __find_space_info(root->fs_info, | |
4810 | BTRFS_BLOCK_GROUP_METADATA); | |
4811 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
4812 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
4813 | |
4814 | if (ret == -ENOSPC && use_global_rsv) | |
4815 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes); | |
4816 | ||
d5c12070 MX |
4817 | if (ret) { |
4818 | if (*qgroup_reserved) | |
4819 | btrfs_qgroup_free(root, *qgroup_reserved); | |
4820 | } | |
4821 | ||
4822 | return ret; | |
4823 | } | |
4824 | ||
4825 | void btrfs_subvolume_release_metadata(struct btrfs_root *root, | |
4826 | struct btrfs_block_rsv *rsv, | |
4827 | u64 qgroup_reserved) | |
4828 | { | |
4829 | btrfs_block_rsv_release(root, rsv, (u64)-1); | |
4830 | if (qgroup_reserved) | |
4831 | btrfs_qgroup_free(root, qgroup_reserved); | |
97e728d4 JB |
4832 | } |
4833 | ||
7709cde3 JB |
4834 | /** |
4835 | * drop_outstanding_extent - drop an outstanding extent | |
4836 | * @inode: the inode we're dropping the extent for | |
4837 | * | |
4838 | * This is called when we are freeing up an outstanding extent, either called | |
4839 | * after an error or after an extent is written. This will return the number of | |
4840 | * reserved extents that need to be freed. This must be called with | |
4841 | * BTRFS_I(inode)->lock held. | |
4842 | */ | |
9e0baf60 JB |
4843 | static unsigned drop_outstanding_extent(struct inode *inode) |
4844 | { | |
7fd2ae21 | 4845 | unsigned drop_inode_space = 0; |
9e0baf60 JB |
4846 | unsigned dropped_extents = 0; |
4847 | ||
9e0baf60 JB |
4848 | BUG_ON(!BTRFS_I(inode)->outstanding_extents); |
4849 | BTRFS_I(inode)->outstanding_extents--; | |
4850 | ||
7fd2ae21 | 4851 | if (BTRFS_I(inode)->outstanding_extents == 0 && |
72ac3c0d JB |
4852 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4853 | &BTRFS_I(inode)->runtime_flags)) | |
7fd2ae21 | 4854 | drop_inode_space = 1; |
7fd2ae21 | 4855 | |
9e0baf60 JB |
4856 | /* |
4857 | * If we have more or the same amount of outsanding extents than we have | |
4858 | * reserved then we need to leave the reserved extents count alone. | |
4859 | */ | |
4860 | if (BTRFS_I(inode)->outstanding_extents >= | |
4861 | BTRFS_I(inode)->reserved_extents) | |
7fd2ae21 | 4862 | return drop_inode_space; |
9e0baf60 JB |
4863 | |
4864 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
4865 | BTRFS_I(inode)->outstanding_extents; | |
4866 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
7fd2ae21 | 4867 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
4868 | } |
4869 | ||
7709cde3 JB |
4870 | /** |
4871 | * calc_csum_metadata_size - return the amount of metada space that must be | |
4872 | * reserved/free'd for the given bytes. | |
4873 | * @inode: the inode we're manipulating | |
4874 | * @num_bytes: the number of bytes in question | |
4875 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
4876 | * | |
4877 | * This adjusts the number of csum_bytes in the inode and then returns the | |
4878 | * correct amount of metadata that must either be reserved or freed. We | |
4879 | * calculate how many checksums we can fit into one leaf and then divide the | |
4880 | * number of bytes that will need to be checksumed by this value to figure out | |
4881 | * how many checksums will be required. If we are adding bytes then the number | |
4882 | * may go up and we will return the number of additional bytes that must be | |
4883 | * reserved. If it is going down we will return the number of bytes that must | |
4884 | * be freed. | |
4885 | * | |
4886 | * This must be called with BTRFS_I(inode)->lock held. | |
4887 | */ | |
4888 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, | |
4889 | int reserve) | |
6324fbf3 | 4890 | { |
7709cde3 JB |
4891 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4892 | u64 csum_size; | |
4893 | int num_csums_per_leaf; | |
4894 | int num_csums; | |
4895 | int old_csums; | |
4896 | ||
4897 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && | |
4898 | BTRFS_I(inode)->csum_bytes == 0) | |
4899 | return 0; | |
4900 | ||
4901 | old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4902 | if (reserve) | |
4903 | BTRFS_I(inode)->csum_bytes += num_bytes; | |
4904 | else | |
4905 | BTRFS_I(inode)->csum_bytes -= num_bytes; | |
4906 | csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item); | |
4907 | num_csums_per_leaf = (int)div64_u64(csum_size, | |
4908 | sizeof(struct btrfs_csum_item) + | |
4909 | sizeof(struct btrfs_disk_key)); | |
4910 | num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4911 | num_csums = num_csums + num_csums_per_leaf - 1; | |
4912 | num_csums = num_csums / num_csums_per_leaf; | |
4913 | ||
4914 | old_csums = old_csums + num_csums_per_leaf - 1; | |
4915 | old_csums = old_csums / num_csums_per_leaf; | |
4916 | ||
4917 | /* No change, no need to reserve more */ | |
4918 | if (old_csums == num_csums) | |
4919 | return 0; | |
4920 | ||
4921 | if (reserve) | |
4922 | return btrfs_calc_trans_metadata_size(root, | |
4923 | num_csums - old_csums); | |
4924 | ||
4925 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); | |
0ca1f7ce | 4926 | } |
c146afad | 4927 | |
0ca1f7ce YZ |
4928 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
4929 | { | |
4930 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4931 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 | 4932 | u64 to_reserve = 0; |
660d3f6c | 4933 | u64 csum_bytes; |
9e0baf60 | 4934 | unsigned nr_extents = 0; |
660d3f6c | 4935 | int extra_reserve = 0; |
08e007d2 | 4936 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 4937 | int ret = 0; |
c64c2bd8 | 4938 | bool delalloc_lock = true; |
88e081bf WS |
4939 | u64 to_free = 0; |
4940 | unsigned dropped; | |
6324fbf3 | 4941 | |
c64c2bd8 JB |
4942 | /* If we are a free space inode we need to not flush since we will be in |
4943 | * the middle of a transaction commit. We also don't need the delalloc | |
4944 | * mutex since we won't race with anybody. We need this mostly to make | |
4945 | * lockdep shut its filthy mouth. | |
4946 | */ | |
4947 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 4948 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 JB |
4949 | delalloc_lock = false; |
4950 | } | |
c09544e0 | 4951 | |
08e007d2 MX |
4952 | if (flush != BTRFS_RESERVE_NO_FLUSH && |
4953 | btrfs_transaction_in_commit(root->fs_info)) | |
0ca1f7ce | 4954 | schedule_timeout(1); |
ec44a35c | 4955 | |
c64c2bd8 JB |
4956 | if (delalloc_lock) |
4957 | mutex_lock(&BTRFS_I(inode)->delalloc_mutex); | |
4958 | ||
0ca1f7ce | 4959 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 4960 | |
9e0baf60 JB |
4961 | spin_lock(&BTRFS_I(inode)->lock); |
4962 | BTRFS_I(inode)->outstanding_extents++; | |
4963 | ||
4964 | if (BTRFS_I(inode)->outstanding_extents > | |
660d3f6c | 4965 | BTRFS_I(inode)->reserved_extents) |
9e0baf60 JB |
4966 | nr_extents = BTRFS_I(inode)->outstanding_extents - |
4967 | BTRFS_I(inode)->reserved_extents; | |
57a45ced | 4968 | |
7fd2ae21 JB |
4969 | /* |
4970 | * Add an item to reserve for updating the inode when we complete the | |
4971 | * delalloc io. | |
4972 | */ | |
72ac3c0d JB |
4973 | if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4974 | &BTRFS_I(inode)->runtime_flags)) { | |
7fd2ae21 | 4975 | nr_extents++; |
660d3f6c | 4976 | extra_reserve = 1; |
593060d7 | 4977 | } |
7fd2ae21 JB |
4978 | |
4979 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); | |
7709cde3 | 4980 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
660d3f6c | 4981 | csum_bytes = BTRFS_I(inode)->csum_bytes; |
9e0baf60 | 4982 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 4983 | |
88e081bf | 4984 | if (root->fs_info->quota_enabled) { |
c5567237 AJ |
4985 | ret = btrfs_qgroup_reserve(root, num_bytes + |
4986 | nr_extents * root->leafsize); | |
88e081bf WS |
4987 | if (ret) |
4988 | goto out_fail; | |
4989 | } | |
c5567237 | 4990 | |
88e081bf WS |
4991 | ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush); |
4992 | if (unlikely(ret)) { | |
4993 | if (root->fs_info->quota_enabled) | |
4b5829a8 MX |
4994 | btrfs_qgroup_free(root, num_bytes + |
4995 | nr_extents * root->leafsize); | |
88e081bf | 4996 | goto out_fail; |
9e0baf60 | 4997 | } |
25179201 | 4998 | |
660d3f6c JB |
4999 | spin_lock(&BTRFS_I(inode)->lock); |
5000 | if (extra_reserve) { | |
72ac3c0d JB |
5001 | set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
5002 | &BTRFS_I(inode)->runtime_flags); | |
660d3f6c JB |
5003 | nr_extents--; |
5004 | } | |
5005 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
5006 | spin_unlock(&BTRFS_I(inode)->lock); | |
c64c2bd8 JB |
5007 | |
5008 | if (delalloc_lock) | |
5009 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
660d3f6c | 5010 | |
8c2a3ca2 JB |
5011 | if (to_reserve) |
5012 | trace_btrfs_space_reservation(root->fs_info,"delalloc", | |
5013 | btrfs_ino(inode), to_reserve, 1); | |
0ca1f7ce YZ |
5014 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
5015 | ||
0ca1f7ce | 5016 | return 0; |
88e081bf WS |
5017 | |
5018 | out_fail: | |
5019 | spin_lock(&BTRFS_I(inode)->lock); | |
5020 | dropped = drop_outstanding_extent(inode); | |
5021 | /* | |
5022 | * If the inodes csum_bytes is the same as the original | |
5023 | * csum_bytes then we know we haven't raced with any free()ers | |
5024 | * so we can just reduce our inodes csum bytes and carry on. | |
88e081bf | 5025 | */ |
f4881bc7 | 5026 | if (BTRFS_I(inode)->csum_bytes == csum_bytes) { |
88e081bf | 5027 | calc_csum_metadata_size(inode, num_bytes, 0); |
f4881bc7 JB |
5028 | } else { |
5029 | u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes; | |
5030 | u64 bytes; | |
5031 | ||
5032 | /* | |
5033 | * This is tricky, but first we need to figure out how much we | |
5034 | * free'd from any free-ers that occured during this | |
5035 | * reservation, so we reset ->csum_bytes to the csum_bytes | |
5036 | * before we dropped our lock, and then call the free for the | |
5037 | * number of bytes that were freed while we were trying our | |
5038 | * reservation. | |
5039 | */ | |
5040 | bytes = csum_bytes - BTRFS_I(inode)->csum_bytes; | |
5041 | BTRFS_I(inode)->csum_bytes = csum_bytes; | |
5042 | to_free = calc_csum_metadata_size(inode, bytes, 0); | |
5043 | ||
5044 | ||
5045 | /* | |
5046 | * Now we need to see how much we would have freed had we not | |
5047 | * been making this reservation and our ->csum_bytes were not | |
5048 | * artificially inflated. | |
5049 | */ | |
5050 | BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes; | |
5051 | bytes = csum_bytes - orig_csum_bytes; | |
5052 | bytes = calc_csum_metadata_size(inode, bytes, 0); | |
5053 | ||
5054 | /* | |
5055 | * Now reset ->csum_bytes to what it should be. If bytes is | |
5056 | * more than to_free then we would have free'd more space had we | |
5057 | * not had an artificially high ->csum_bytes, so we need to free | |
5058 | * the remainder. If bytes is the same or less then we don't | |
5059 | * need to do anything, the other free-ers did the correct | |
5060 | * thing. | |
5061 | */ | |
5062 | BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes; | |
5063 | if (bytes > to_free) | |
5064 | to_free = bytes - to_free; | |
5065 | else | |
5066 | to_free = 0; | |
5067 | } | |
88e081bf WS |
5068 | spin_unlock(&BTRFS_I(inode)->lock); |
5069 | if (dropped) | |
5070 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
5071 | ||
5072 | if (to_free) { | |
5073 | btrfs_block_rsv_release(root, block_rsv, to_free); | |
5074 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
5075 | btrfs_ino(inode), to_free, 0); | |
5076 | } | |
5077 | if (delalloc_lock) | |
5078 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
5079 | return ret; | |
0ca1f7ce YZ |
5080 | } |
5081 | ||
7709cde3 JB |
5082 | /** |
5083 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
5084 | * @inode: the inode to release the reservation for | |
5085 | * @num_bytes: the number of bytes we're releasing | |
5086 | * | |
5087 | * This will release the metadata reservation for an inode. This can be called | |
5088 | * once we complete IO for a given set of bytes to release their metadata | |
5089 | * reservations. | |
5090 | */ | |
0ca1f7ce YZ |
5091 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) |
5092 | { | |
5093 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
5094 | u64 to_free = 0; |
5095 | unsigned dropped; | |
0ca1f7ce YZ |
5096 | |
5097 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
7709cde3 | 5098 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 5099 | dropped = drop_outstanding_extent(inode); |
97e728d4 | 5100 | |
0934856d MX |
5101 | if (num_bytes) |
5102 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
7709cde3 | 5103 | spin_unlock(&BTRFS_I(inode)->lock); |
9e0baf60 JB |
5104 | if (dropped > 0) |
5105 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce | 5106 | |
8c2a3ca2 JB |
5107 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
5108 | btrfs_ino(inode), to_free, 0); | |
c5567237 AJ |
5109 | if (root->fs_info->quota_enabled) { |
5110 | btrfs_qgroup_free(root, num_bytes + | |
5111 | dropped * root->leafsize); | |
5112 | } | |
5113 | ||
0ca1f7ce YZ |
5114 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, |
5115 | to_free); | |
5116 | } | |
5117 | ||
7709cde3 JB |
5118 | /** |
5119 | * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc | |
5120 | * @inode: inode we're writing to | |
5121 | * @num_bytes: the number of bytes we want to allocate | |
5122 | * | |
5123 | * This will do the following things | |
5124 | * | |
5125 | * o reserve space in the data space info for num_bytes | |
5126 | * o reserve space in the metadata space info based on number of outstanding | |
5127 | * extents and how much csums will be needed | |
5128 | * o add to the inodes ->delalloc_bytes | |
5129 | * o add it to the fs_info's delalloc inodes list. | |
5130 | * | |
5131 | * This will return 0 for success and -ENOSPC if there is no space left. | |
5132 | */ | |
0ca1f7ce YZ |
5133 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes) |
5134 | { | |
5135 | int ret; | |
5136 | ||
5137 | ret = btrfs_check_data_free_space(inode, num_bytes); | |
d397712b | 5138 | if (ret) |
0ca1f7ce YZ |
5139 | return ret; |
5140 | ||
5141 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes); | |
5142 | if (ret) { | |
5143 | btrfs_free_reserved_data_space(inode, num_bytes); | |
5144 | return ret; | |
5145 | } | |
5146 | ||
5147 | return 0; | |
5148 | } | |
5149 | ||
7709cde3 JB |
5150 | /** |
5151 | * btrfs_delalloc_release_space - release data and metadata space for delalloc | |
5152 | * @inode: inode we're releasing space for | |
5153 | * @num_bytes: the number of bytes we want to free up | |
5154 | * | |
5155 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
5156 | * called in the case that we don't need the metadata AND data reservations | |
5157 | * anymore. So if there is an error or we insert an inline extent. | |
5158 | * | |
5159 | * This function will release the metadata space that was not used and will | |
5160 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
5161 | * list if there are no delalloc bytes left. | |
5162 | */ | |
0ca1f7ce YZ |
5163 | void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes) |
5164 | { | |
5165 | btrfs_delalloc_release_metadata(inode, num_bytes); | |
5166 | btrfs_free_reserved_data_space(inode, num_bytes); | |
6324fbf3 CM |
5167 | } |
5168 | ||
c53d613e | 5169 | static int update_block_group(struct btrfs_root *root, |
f0486c68 | 5170 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 5171 | { |
0af3d00b | 5172 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 5173 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 5174 | u64 total = num_bytes; |
9078a3e1 | 5175 | u64 old_val; |
db94535d | 5176 | u64 byte_in_group; |
0af3d00b | 5177 | int factor; |
3e1ad54f | 5178 | |
5d4f98a2 | 5179 | /* block accounting for super block */ |
eb73c1b7 | 5180 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 5181 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
5182 | if (alloc) |
5183 | old_val += num_bytes; | |
5184 | else | |
5185 | old_val -= num_bytes; | |
6c41761f | 5186 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 5187 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 5188 | |
d397712b | 5189 | while (total) { |
db94535d | 5190 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 5191 | if (!cache) |
79787eaa | 5192 | return -ENOENT; |
b742bb82 YZ |
5193 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
5194 | BTRFS_BLOCK_GROUP_RAID1 | | |
5195 | BTRFS_BLOCK_GROUP_RAID10)) | |
5196 | factor = 2; | |
5197 | else | |
5198 | factor = 1; | |
9d66e233 JB |
5199 | /* |
5200 | * If this block group has free space cache written out, we | |
5201 | * need to make sure to load it if we are removing space. This | |
5202 | * is because we need the unpinning stage to actually add the | |
5203 | * space back to the block group, otherwise we will leak space. | |
5204 | */ | |
5205 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 5206 | cache_block_group(cache, 1); |
0af3d00b | 5207 | |
db94535d CM |
5208 | byte_in_group = bytenr - cache->key.objectid; |
5209 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 5210 | |
25179201 | 5211 | spin_lock(&cache->space_info->lock); |
c286ac48 | 5212 | spin_lock(&cache->lock); |
0af3d00b | 5213 | |
73bc1876 | 5214 | if (btrfs_test_opt(root, SPACE_CACHE) && |
0af3d00b JB |
5215 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
5216 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
5217 | ||
0f9dd46c | 5218 | cache->dirty = 1; |
9078a3e1 | 5219 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 5220 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 5221 | if (alloc) { |
db94535d | 5222 | old_val += num_bytes; |
11833d66 YZ |
5223 | btrfs_set_block_group_used(&cache->item, old_val); |
5224 | cache->reserved -= num_bytes; | |
11833d66 | 5225 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
5226 | cache->space_info->bytes_used += num_bytes; |
5227 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 5228 | spin_unlock(&cache->lock); |
25179201 | 5229 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 5230 | } else { |
db94535d | 5231 | old_val -= num_bytes; |
c286ac48 | 5232 | btrfs_set_block_group_used(&cache->item, old_val); |
f0486c68 YZ |
5233 | cache->pinned += num_bytes; |
5234 | cache->space_info->bytes_pinned += num_bytes; | |
6324fbf3 | 5235 | cache->space_info->bytes_used -= num_bytes; |
b742bb82 | 5236 | cache->space_info->disk_used -= num_bytes * factor; |
c286ac48 | 5237 | spin_unlock(&cache->lock); |
25179201 | 5238 | spin_unlock(&cache->space_info->lock); |
1f3c79a2 | 5239 | |
f0486c68 YZ |
5240 | set_extent_dirty(info->pinned_extents, |
5241 | bytenr, bytenr + num_bytes - 1, | |
5242 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 5243 | } |
fa9c0d79 | 5244 | btrfs_put_block_group(cache); |
db94535d CM |
5245 | total -= num_bytes; |
5246 | bytenr += num_bytes; | |
9078a3e1 CM |
5247 | } |
5248 | return 0; | |
5249 | } | |
6324fbf3 | 5250 | |
a061fc8d CM |
5251 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
5252 | { | |
0f9dd46c | 5253 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 5254 | u64 bytenr; |
0f9dd46c | 5255 | |
a1897fdd LB |
5256 | spin_lock(&root->fs_info->block_group_cache_lock); |
5257 | bytenr = root->fs_info->first_logical_byte; | |
5258 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
5259 | ||
5260 | if (bytenr < (u64)-1) | |
5261 | return bytenr; | |
5262 | ||
0f9dd46c JB |
5263 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); |
5264 | if (!cache) | |
a061fc8d | 5265 | return 0; |
0f9dd46c | 5266 | |
d2fb3437 | 5267 | bytenr = cache->key.objectid; |
fa9c0d79 | 5268 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
5269 | |
5270 | return bytenr; | |
a061fc8d CM |
5271 | } |
5272 | ||
f0486c68 YZ |
5273 | static int pin_down_extent(struct btrfs_root *root, |
5274 | struct btrfs_block_group_cache *cache, | |
5275 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 5276 | { |
11833d66 YZ |
5277 | spin_lock(&cache->space_info->lock); |
5278 | spin_lock(&cache->lock); | |
5279 | cache->pinned += num_bytes; | |
5280 | cache->space_info->bytes_pinned += num_bytes; | |
5281 | if (reserved) { | |
5282 | cache->reserved -= num_bytes; | |
5283 | cache->space_info->bytes_reserved -= num_bytes; | |
5284 | } | |
5285 | spin_unlock(&cache->lock); | |
5286 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 5287 | |
f0486c68 YZ |
5288 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
5289 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
0be5dc67 JB |
5290 | if (reserved) |
5291 | trace_btrfs_reserved_extent_free(root, bytenr, num_bytes); | |
f0486c68 YZ |
5292 | return 0; |
5293 | } | |
68b38550 | 5294 | |
f0486c68 YZ |
5295 | /* |
5296 | * this function must be called within transaction | |
5297 | */ | |
5298 | int btrfs_pin_extent(struct btrfs_root *root, | |
5299 | u64 bytenr, u64 num_bytes, int reserved) | |
5300 | { | |
5301 | struct btrfs_block_group_cache *cache; | |
68b38550 | 5302 | |
f0486c68 | 5303 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
79787eaa | 5304 | BUG_ON(!cache); /* Logic error */ |
f0486c68 YZ |
5305 | |
5306 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
5307 | ||
5308 | btrfs_put_block_group(cache); | |
11833d66 YZ |
5309 | return 0; |
5310 | } | |
5311 | ||
f0486c68 | 5312 | /* |
e688b725 CM |
5313 | * this function must be called within transaction |
5314 | */ | |
dcfac415 | 5315 | int btrfs_pin_extent_for_log_replay(struct btrfs_root *root, |
e688b725 CM |
5316 | u64 bytenr, u64 num_bytes) |
5317 | { | |
5318 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 5319 | int ret; |
e688b725 CM |
5320 | |
5321 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
b50c6e25 JB |
5322 | if (!cache) |
5323 | return -EINVAL; | |
e688b725 CM |
5324 | |
5325 | /* | |
5326 | * pull in the free space cache (if any) so that our pin | |
5327 | * removes the free space from the cache. We have load_only set | |
5328 | * to one because the slow code to read in the free extents does check | |
5329 | * the pinned extents. | |
5330 | */ | |
f6373bf3 | 5331 | cache_block_group(cache, 1); |
e688b725 CM |
5332 | |
5333 | pin_down_extent(root, cache, bytenr, num_bytes, 0); | |
5334 | ||
5335 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 5336 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 5337 | btrfs_put_block_group(cache); |
b50c6e25 | 5338 | return ret; |
e688b725 CM |
5339 | } |
5340 | ||
8c2a1a30 JB |
5341 | static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes) |
5342 | { | |
5343 | int ret; | |
5344 | struct btrfs_block_group_cache *block_group; | |
5345 | struct btrfs_caching_control *caching_ctl; | |
5346 | ||
5347 | block_group = btrfs_lookup_block_group(root->fs_info, start); | |
5348 | if (!block_group) | |
5349 | return -EINVAL; | |
5350 | ||
5351 | cache_block_group(block_group, 0); | |
5352 | caching_ctl = get_caching_control(block_group); | |
5353 | ||
5354 | if (!caching_ctl) { | |
5355 | /* Logic error */ | |
5356 | BUG_ON(!block_group_cache_done(block_group)); | |
5357 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
5358 | } else { | |
5359 | mutex_lock(&caching_ctl->mutex); | |
5360 | ||
5361 | if (start >= caching_ctl->progress) { | |
5362 | ret = add_excluded_extent(root, start, num_bytes); | |
5363 | } else if (start + num_bytes <= caching_ctl->progress) { | |
5364 | ret = btrfs_remove_free_space(block_group, | |
5365 | start, num_bytes); | |
5366 | } else { | |
5367 | num_bytes = caching_ctl->progress - start; | |
5368 | ret = btrfs_remove_free_space(block_group, | |
5369 | start, num_bytes); | |
5370 | if (ret) | |
5371 | goto out_lock; | |
5372 | ||
5373 | num_bytes = (start + num_bytes) - | |
5374 | caching_ctl->progress; | |
5375 | start = caching_ctl->progress; | |
5376 | ret = add_excluded_extent(root, start, num_bytes); | |
5377 | } | |
5378 | out_lock: | |
5379 | mutex_unlock(&caching_ctl->mutex); | |
5380 | put_caching_control(caching_ctl); | |
5381 | } | |
5382 | btrfs_put_block_group(block_group); | |
5383 | return ret; | |
5384 | } | |
5385 | ||
5386 | int btrfs_exclude_logged_extents(struct btrfs_root *log, | |
5387 | struct extent_buffer *eb) | |
5388 | { | |
5389 | struct btrfs_file_extent_item *item; | |
5390 | struct btrfs_key key; | |
5391 | int found_type; | |
5392 | int i; | |
5393 | ||
5394 | if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) | |
5395 | return 0; | |
5396 | ||
5397 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
5398 | btrfs_item_key_to_cpu(eb, &key, i); | |
5399 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
5400 | continue; | |
5401 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
5402 | found_type = btrfs_file_extent_type(eb, item); | |
5403 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
5404 | continue; | |
5405 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
5406 | continue; | |
5407 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
5408 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
5409 | __exclude_logged_extent(log, key.objectid, key.offset); | |
5410 | } | |
5411 | ||
5412 | return 0; | |
5413 | } | |
5414 | ||
fb25e914 JB |
5415 | /** |
5416 | * btrfs_update_reserved_bytes - update the block_group and space info counters | |
5417 | * @cache: The cache we are manipulating | |
5418 | * @num_bytes: The number of bytes in question | |
5419 | * @reserve: One of the reservation enums | |
5420 | * | |
5421 | * This is called by the allocator when it reserves space, or by somebody who is | |
5422 | * freeing space that was never actually used on disk. For example if you | |
5423 | * reserve some space for a new leaf in transaction A and before transaction A | |
5424 | * commits you free that leaf, you call this with reserve set to 0 in order to | |
5425 | * clear the reservation. | |
5426 | * | |
5427 | * Metadata reservations should be called with RESERVE_ALLOC so we do the proper | |
5428 | * ENOSPC accounting. For data we handle the reservation through clearing the | |
5429 | * delalloc bits in the io_tree. We have to do this since we could end up | |
5430 | * allocating less disk space for the amount of data we have reserved in the | |
5431 | * case of compression. | |
5432 | * | |
5433 | * If this is a reservation and the block group has become read only we cannot | |
5434 | * make the reservation and return -EAGAIN, otherwise this function always | |
5435 | * succeeds. | |
f0486c68 | 5436 | */ |
fb25e914 JB |
5437 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
5438 | u64 num_bytes, int reserve) | |
11833d66 | 5439 | { |
fb25e914 | 5440 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 5441 | int ret = 0; |
79787eaa | 5442 | |
fb25e914 JB |
5443 | spin_lock(&space_info->lock); |
5444 | spin_lock(&cache->lock); | |
5445 | if (reserve != RESERVE_FREE) { | |
f0486c68 YZ |
5446 | if (cache->ro) { |
5447 | ret = -EAGAIN; | |
5448 | } else { | |
fb25e914 JB |
5449 | cache->reserved += num_bytes; |
5450 | space_info->bytes_reserved += num_bytes; | |
5451 | if (reserve == RESERVE_ALLOC) { | |
8c2a3ca2 | 5452 | trace_btrfs_space_reservation(cache->fs_info, |
2bcc0328 LB |
5453 | "space_info", space_info->flags, |
5454 | num_bytes, 0); | |
fb25e914 JB |
5455 | space_info->bytes_may_use -= num_bytes; |
5456 | } | |
f0486c68 | 5457 | } |
fb25e914 JB |
5458 | } else { |
5459 | if (cache->ro) | |
5460 | space_info->bytes_readonly += num_bytes; | |
5461 | cache->reserved -= num_bytes; | |
5462 | space_info->bytes_reserved -= num_bytes; | |
324ae4df | 5463 | } |
fb25e914 JB |
5464 | spin_unlock(&cache->lock); |
5465 | spin_unlock(&space_info->lock); | |
f0486c68 | 5466 | return ret; |
324ae4df | 5467 | } |
9078a3e1 | 5468 | |
143bede5 | 5469 | void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
11833d66 | 5470 | struct btrfs_root *root) |
e8569813 | 5471 | { |
e8569813 | 5472 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
5473 | struct btrfs_caching_control *next; |
5474 | struct btrfs_caching_control *caching_ctl; | |
5475 | struct btrfs_block_group_cache *cache; | |
b150a4f1 | 5476 | struct btrfs_space_info *space_info; |
e8569813 | 5477 | |
11833d66 | 5478 | down_write(&fs_info->extent_commit_sem); |
25179201 | 5479 | |
11833d66 YZ |
5480 | list_for_each_entry_safe(caching_ctl, next, |
5481 | &fs_info->caching_block_groups, list) { | |
5482 | cache = caching_ctl->block_group; | |
5483 | if (block_group_cache_done(cache)) { | |
5484 | cache->last_byte_to_unpin = (u64)-1; | |
5485 | list_del_init(&caching_ctl->list); | |
5486 | put_caching_control(caching_ctl); | |
e8569813 | 5487 | } else { |
11833d66 | 5488 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 5489 | } |
e8569813 | 5490 | } |
11833d66 YZ |
5491 | |
5492 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
5493 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
5494 | else | |
5495 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
5496 | ||
5497 | up_write(&fs_info->extent_commit_sem); | |
8929ecfa | 5498 | |
b150a4f1 JB |
5499 | list_for_each_entry_rcu(space_info, &fs_info->space_info, list) |
5500 | percpu_counter_set(&space_info->total_bytes_pinned, 0); | |
5501 | ||
8929ecfa | 5502 | update_global_block_rsv(fs_info); |
e8569813 ZY |
5503 | } |
5504 | ||
11833d66 | 5505 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
ccd467d6 | 5506 | { |
11833d66 YZ |
5507 | struct btrfs_fs_info *fs_info = root->fs_info; |
5508 | struct btrfs_block_group_cache *cache = NULL; | |
7b398f8e JB |
5509 | struct btrfs_space_info *space_info; |
5510 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
11833d66 | 5511 | u64 len; |
7b398f8e | 5512 | bool readonly; |
ccd467d6 | 5513 | |
11833d66 | 5514 | while (start <= end) { |
7b398f8e | 5515 | readonly = false; |
11833d66 YZ |
5516 | if (!cache || |
5517 | start >= cache->key.objectid + cache->key.offset) { | |
5518 | if (cache) | |
5519 | btrfs_put_block_group(cache); | |
5520 | cache = btrfs_lookup_block_group(fs_info, start); | |
79787eaa | 5521 | BUG_ON(!cache); /* Logic error */ |
11833d66 YZ |
5522 | } |
5523 | ||
5524 | len = cache->key.objectid + cache->key.offset - start; | |
5525 | len = min(len, end + 1 - start); | |
5526 | ||
5527 | if (start < cache->last_byte_to_unpin) { | |
5528 | len = min(len, cache->last_byte_to_unpin - start); | |
5529 | btrfs_add_free_space(cache, start, len); | |
5530 | } | |
5531 | ||
f0486c68 | 5532 | start += len; |
7b398f8e | 5533 | space_info = cache->space_info; |
f0486c68 | 5534 | |
7b398f8e | 5535 | spin_lock(&space_info->lock); |
11833d66 YZ |
5536 | spin_lock(&cache->lock); |
5537 | cache->pinned -= len; | |
7b398f8e JB |
5538 | space_info->bytes_pinned -= len; |
5539 | if (cache->ro) { | |
5540 | space_info->bytes_readonly += len; | |
5541 | readonly = true; | |
5542 | } | |
11833d66 | 5543 | spin_unlock(&cache->lock); |
7b398f8e JB |
5544 | if (!readonly && global_rsv->space_info == space_info) { |
5545 | spin_lock(&global_rsv->lock); | |
5546 | if (!global_rsv->full) { | |
5547 | len = min(len, global_rsv->size - | |
5548 | global_rsv->reserved); | |
5549 | global_rsv->reserved += len; | |
5550 | space_info->bytes_may_use += len; | |
5551 | if (global_rsv->reserved >= global_rsv->size) | |
5552 | global_rsv->full = 1; | |
5553 | } | |
5554 | spin_unlock(&global_rsv->lock); | |
5555 | } | |
5556 | spin_unlock(&space_info->lock); | |
ccd467d6 | 5557 | } |
11833d66 YZ |
5558 | |
5559 | if (cache) | |
5560 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
5561 | return 0; |
5562 | } | |
5563 | ||
5564 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 5565 | struct btrfs_root *root) |
a28ec197 | 5566 | { |
11833d66 YZ |
5567 | struct btrfs_fs_info *fs_info = root->fs_info; |
5568 | struct extent_io_tree *unpin; | |
1a5bc167 CM |
5569 | u64 start; |
5570 | u64 end; | |
a28ec197 | 5571 | int ret; |
a28ec197 | 5572 | |
79787eaa JM |
5573 | if (trans->aborted) |
5574 | return 0; | |
5575 | ||
11833d66 YZ |
5576 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
5577 | unpin = &fs_info->freed_extents[1]; | |
5578 | else | |
5579 | unpin = &fs_info->freed_extents[0]; | |
5580 | ||
d397712b | 5581 | while (1) { |
1a5bc167 | 5582 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 5583 | EXTENT_DIRTY, NULL); |
1a5bc167 | 5584 | if (ret) |
a28ec197 | 5585 | break; |
1f3c79a2 | 5586 | |
5378e607 LD |
5587 | if (btrfs_test_opt(root, DISCARD)) |
5588 | ret = btrfs_discard_extent(root, start, | |
5589 | end + 1 - start, NULL); | |
1f3c79a2 | 5590 | |
1a5bc167 | 5591 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
11833d66 | 5592 | unpin_extent_range(root, start, end); |
b9473439 | 5593 | cond_resched(); |
a28ec197 | 5594 | } |
817d52f8 | 5595 | |
e20d96d6 CM |
5596 | return 0; |
5597 | } | |
5598 | ||
b150a4f1 JB |
5599 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes, |
5600 | u64 owner, u64 root_objectid) | |
5601 | { | |
5602 | struct btrfs_space_info *space_info; | |
5603 | u64 flags; | |
5604 | ||
5605 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
5606 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) | |
5607 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
5608 | else | |
5609 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
5610 | } else { | |
5611 | flags = BTRFS_BLOCK_GROUP_DATA; | |
5612 | } | |
5613 | ||
5614 | space_info = __find_space_info(fs_info, flags); | |
5615 | BUG_ON(!space_info); /* Logic bug */ | |
5616 | percpu_counter_add(&space_info->total_bytes_pinned, num_bytes); | |
5617 | } | |
5618 | ||
5619 | ||
5d4f98a2 YZ |
5620 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
5621 | struct btrfs_root *root, | |
5622 | u64 bytenr, u64 num_bytes, u64 parent, | |
5623 | u64 root_objectid, u64 owner_objectid, | |
5624 | u64 owner_offset, int refs_to_drop, | |
5625 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 5626 | { |
e2fa7227 | 5627 | struct btrfs_key key; |
5d4f98a2 | 5628 | struct btrfs_path *path; |
1261ec42 CM |
5629 | struct btrfs_fs_info *info = root->fs_info; |
5630 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 5631 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
5632 | struct btrfs_extent_item *ei; |
5633 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 5634 | int ret; |
5d4f98a2 | 5635 | int is_data; |
952fccac CM |
5636 | int extent_slot = 0; |
5637 | int found_extent = 0; | |
5638 | int num_to_del = 1; | |
5d4f98a2 YZ |
5639 | u32 item_size; |
5640 | u64 refs; | |
3173a18f JB |
5641 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
5642 | SKINNY_METADATA); | |
037e6390 | 5643 | |
5caf2a00 | 5644 | path = btrfs_alloc_path(); |
54aa1f4d CM |
5645 | if (!path) |
5646 | return -ENOMEM; | |
5f26f772 | 5647 | |
3c12ac72 | 5648 | path->reada = 1; |
b9473439 | 5649 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5650 | |
5651 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
5652 | BUG_ON(!is_data && refs_to_drop != 1); | |
5653 | ||
3173a18f JB |
5654 | if (is_data) |
5655 | skinny_metadata = 0; | |
5656 | ||
5d4f98a2 YZ |
5657 | ret = lookup_extent_backref(trans, extent_root, path, &iref, |
5658 | bytenr, num_bytes, parent, | |
5659 | root_objectid, owner_objectid, | |
5660 | owner_offset); | |
7bb86316 | 5661 | if (ret == 0) { |
952fccac | 5662 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
5663 | while (extent_slot >= 0) { |
5664 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 5665 | extent_slot); |
5d4f98a2 | 5666 | if (key.objectid != bytenr) |
952fccac | 5667 | break; |
5d4f98a2 YZ |
5668 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
5669 | key.offset == num_bytes) { | |
952fccac CM |
5670 | found_extent = 1; |
5671 | break; | |
5672 | } | |
3173a18f JB |
5673 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
5674 | key.offset == owner_objectid) { | |
5675 | found_extent = 1; | |
5676 | break; | |
5677 | } | |
952fccac CM |
5678 | if (path->slots[0] - extent_slot > 5) |
5679 | break; | |
5d4f98a2 | 5680 | extent_slot--; |
952fccac | 5681 | } |
5d4f98a2 YZ |
5682 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
5683 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
5684 | if (found_extent && item_size < sizeof(*ei)) | |
5685 | found_extent = 0; | |
5686 | #endif | |
31840ae1 | 5687 | if (!found_extent) { |
5d4f98a2 | 5688 | BUG_ON(iref); |
56bec294 | 5689 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
5690 | NULL, refs_to_drop, |
5691 | is_data); | |
005d6427 DS |
5692 | if (ret) { |
5693 | btrfs_abort_transaction(trans, extent_root, ret); | |
5694 | goto out; | |
5695 | } | |
b3b4aa74 | 5696 | btrfs_release_path(path); |
b9473439 | 5697 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5698 | |
5699 | key.objectid = bytenr; | |
5700 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5701 | key.offset = num_bytes; | |
5702 | ||
3173a18f JB |
5703 | if (!is_data && skinny_metadata) { |
5704 | key.type = BTRFS_METADATA_ITEM_KEY; | |
5705 | key.offset = owner_objectid; | |
5706 | } | |
5707 | ||
31840ae1 ZY |
5708 | ret = btrfs_search_slot(trans, extent_root, |
5709 | &key, path, -1, 1); | |
3173a18f JB |
5710 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
5711 | /* | |
5712 | * Couldn't find our skinny metadata item, | |
5713 | * see if we have ye olde extent item. | |
5714 | */ | |
5715 | path->slots[0]--; | |
5716 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
5717 | path->slots[0]); | |
5718 | if (key.objectid == bytenr && | |
5719 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
5720 | key.offset == num_bytes) | |
5721 | ret = 0; | |
5722 | } | |
5723 | ||
5724 | if (ret > 0 && skinny_metadata) { | |
5725 | skinny_metadata = false; | |
5726 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5727 | key.offset = num_bytes; | |
5728 | btrfs_release_path(path); | |
5729 | ret = btrfs_search_slot(trans, extent_root, | |
5730 | &key, path, -1, 1); | |
5731 | } | |
5732 | ||
f3465ca4 | 5733 | if (ret) { |
c2cf52eb | 5734 | btrfs_err(info, "umm, got %d back from search, was looking for %llu", |
c1c9ff7c | 5735 | ret, bytenr); |
b783e62d JB |
5736 | if (ret > 0) |
5737 | btrfs_print_leaf(extent_root, | |
5738 | path->nodes[0]); | |
f3465ca4 | 5739 | } |
005d6427 DS |
5740 | if (ret < 0) { |
5741 | btrfs_abort_transaction(trans, extent_root, ret); | |
5742 | goto out; | |
5743 | } | |
31840ae1 ZY |
5744 | extent_slot = path->slots[0]; |
5745 | } | |
79787eaa | 5746 | } else if (ret == -ENOENT) { |
7bb86316 CM |
5747 | btrfs_print_leaf(extent_root, path->nodes[0]); |
5748 | WARN_ON(1); | |
c2cf52eb SK |
5749 | btrfs_err(info, |
5750 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
5751 | bytenr, parent, root_objectid, owner_objectid, |
5752 | owner_offset); | |
79787eaa | 5753 | } else { |
005d6427 DS |
5754 | btrfs_abort_transaction(trans, extent_root, ret); |
5755 | goto out; | |
7bb86316 | 5756 | } |
5f39d397 CM |
5757 | |
5758 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
5759 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
5760 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
5761 | if (item_size < sizeof(*ei)) { | |
5762 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
5763 | ret = convert_extent_item_v0(trans, extent_root, path, | |
5764 | owner_objectid, 0); | |
005d6427 DS |
5765 | if (ret < 0) { |
5766 | btrfs_abort_transaction(trans, extent_root, ret); | |
5767 | goto out; | |
5768 | } | |
5d4f98a2 | 5769 | |
b3b4aa74 | 5770 | btrfs_release_path(path); |
5d4f98a2 YZ |
5771 | path->leave_spinning = 1; |
5772 | ||
5773 | key.objectid = bytenr; | |
5774 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5775 | key.offset = num_bytes; | |
5776 | ||
5777 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
5778 | -1, 1); | |
5779 | if (ret) { | |
c2cf52eb | 5780 | btrfs_err(info, "umm, got %d back from search, was looking for %llu", |
c1c9ff7c | 5781 | ret, bytenr); |
5d4f98a2 YZ |
5782 | btrfs_print_leaf(extent_root, path->nodes[0]); |
5783 | } | |
005d6427 DS |
5784 | if (ret < 0) { |
5785 | btrfs_abort_transaction(trans, extent_root, ret); | |
5786 | goto out; | |
5787 | } | |
5788 | ||
5d4f98a2 YZ |
5789 | extent_slot = path->slots[0]; |
5790 | leaf = path->nodes[0]; | |
5791 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
5792 | } | |
5793 | #endif | |
5794 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 5795 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 5796 | struct btrfs_extent_item); |
3173a18f JB |
5797 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
5798 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
5799 | struct btrfs_tree_block_info *bi; |
5800 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
5801 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
5802 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
5803 | } | |
56bec294 | 5804 | |
5d4f98a2 | 5805 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 JB |
5806 | if (refs < refs_to_drop) { |
5807 | btrfs_err(info, "trying to drop %d refs but we only have %Lu " | |
5808 | "for bytenr %Lu\n", refs_to_drop, refs, bytenr); | |
5809 | ret = -EINVAL; | |
5810 | btrfs_abort_transaction(trans, extent_root, ret); | |
5811 | goto out; | |
5812 | } | |
56bec294 | 5813 | refs -= refs_to_drop; |
5f39d397 | 5814 | |
5d4f98a2 YZ |
5815 | if (refs > 0) { |
5816 | if (extent_op) | |
5817 | __run_delayed_extent_op(extent_op, leaf, ei); | |
5818 | /* | |
5819 | * In the case of inline back ref, reference count will | |
5820 | * be updated by remove_extent_backref | |
952fccac | 5821 | */ |
5d4f98a2 YZ |
5822 | if (iref) { |
5823 | BUG_ON(!found_extent); | |
5824 | } else { | |
5825 | btrfs_set_extent_refs(leaf, ei, refs); | |
5826 | btrfs_mark_buffer_dirty(leaf); | |
5827 | } | |
5828 | if (found_extent) { | |
5829 | ret = remove_extent_backref(trans, extent_root, path, | |
5830 | iref, refs_to_drop, | |
5831 | is_data); | |
005d6427 DS |
5832 | if (ret) { |
5833 | btrfs_abort_transaction(trans, extent_root, ret); | |
5834 | goto out; | |
5835 | } | |
952fccac | 5836 | } |
b150a4f1 JB |
5837 | add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid, |
5838 | root_objectid); | |
5d4f98a2 | 5839 | } else { |
5d4f98a2 YZ |
5840 | if (found_extent) { |
5841 | BUG_ON(is_data && refs_to_drop != | |
5842 | extent_data_ref_count(root, path, iref)); | |
5843 | if (iref) { | |
5844 | BUG_ON(path->slots[0] != extent_slot); | |
5845 | } else { | |
5846 | BUG_ON(path->slots[0] != extent_slot + 1); | |
5847 | path->slots[0] = extent_slot; | |
5848 | num_to_del = 2; | |
5849 | } | |
78fae27e | 5850 | } |
b9473439 | 5851 | |
952fccac CM |
5852 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
5853 | num_to_del); | |
005d6427 DS |
5854 | if (ret) { |
5855 | btrfs_abort_transaction(trans, extent_root, ret); | |
5856 | goto out; | |
5857 | } | |
b3b4aa74 | 5858 | btrfs_release_path(path); |
21af804c | 5859 | |
5d4f98a2 | 5860 | if (is_data) { |
459931ec | 5861 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
005d6427 DS |
5862 | if (ret) { |
5863 | btrfs_abort_transaction(trans, extent_root, ret); | |
5864 | goto out; | |
5865 | } | |
459931ec CM |
5866 | } |
5867 | ||
c53d613e | 5868 | ret = update_block_group(root, bytenr, num_bytes, 0); |
005d6427 DS |
5869 | if (ret) { |
5870 | btrfs_abort_transaction(trans, extent_root, ret); | |
5871 | goto out; | |
5872 | } | |
a28ec197 | 5873 | } |
79787eaa | 5874 | out: |
5caf2a00 | 5875 | btrfs_free_path(path); |
a28ec197 CM |
5876 | return ret; |
5877 | } | |
5878 | ||
1887be66 | 5879 | /* |
f0486c68 | 5880 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
5881 | * delayed ref for that extent as well. This searches the delayed ref tree for |
5882 | * a given extent, and if there are no other delayed refs to be processed, it | |
5883 | * removes it from the tree. | |
5884 | */ | |
5885 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
5886 | struct btrfs_root *root, u64 bytenr) | |
5887 | { | |
5888 | struct btrfs_delayed_ref_head *head; | |
5889 | struct btrfs_delayed_ref_root *delayed_refs; | |
5890 | struct btrfs_delayed_ref_node *ref; | |
5891 | struct rb_node *node; | |
f0486c68 | 5892 | int ret = 0; |
1887be66 CM |
5893 | |
5894 | delayed_refs = &trans->transaction->delayed_refs; | |
5895 | spin_lock(&delayed_refs->lock); | |
5896 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
5897 | if (!head) | |
5898 | goto out; | |
5899 | ||
5900 | node = rb_prev(&head->node.rb_node); | |
5901 | if (!node) | |
5902 | goto out; | |
5903 | ||
5904 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
5905 | ||
5906 | /* there are still entries for this ref, we can't drop it */ | |
5907 | if (ref->bytenr == bytenr) | |
5908 | goto out; | |
5909 | ||
5d4f98a2 YZ |
5910 | if (head->extent_op) { |
5911 | if (!head->must_insert_reserved) | |
5912 | goto out; | |
78a6184a | 5913 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
5914 | head->extent_op = NULL; |
5915 | } | |
5916 | ||
1887be66 CM |
5917 | /* |
5918 | * waiting for the lock here would deadlock. If someone else has it | |
5919 | * locked they are already in the process of dropping it anyway | |
5920 | */ | |
5921 | if (!mutex_trylock(&head->mutex)) | |
5922 | goto out; | |
5923 | ||
5924 | /* | |
5925 | * at this point we have a head with no other entries. Go | |
5926 | * ahead and process it. | |
5927 | */ | |
5928 | head->node.in_tree = 0; | |
5929 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 5930 | |
1887be66 CM |
5931 | delayed_refs->num_entries--; |
5932 | ||
5933 | /* | |
5934 | * we don't take a ref on the node because we're removing it from the | |
5935 | * tree, so we just steal the ref the tree was holding. | |
5936 | */ | |
c3e69d58 CM |
5937 | delayed_refs->num_heads--; |
5938 | if (list_empty(&head->cluster)) | |
5939 | delayed_refs->num_heads_ready--; | |
5940 | ||
5941 | list_del_init(&head->cluster); | |
1887be66 CM |
5942 | spin_unlock(&delayed_refs->lock); |
5943 | ||
f0486c68 YZ |
5944 | BUG_ON(head->extent_op); |
5945 | if (head->must_insert_reserved) | |
5946 | ret = 1; | |
5947 | ||
5948 | mutex_unlock(&head->mutex); | |
1887be66 | 5949 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 5950 | return ret; |
1887be66 CM |
5951 | out: |
5952 | spin_unlock(&delayed_refs->lock); | |
5953 | return 0; | |
5954 | } | |
5955 | ||
f0486c68 YZ |
5956 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
5957 | struct btrfs_root *root, | |
5958 | struct extent_buffer *buf, | |
5581a51a | 5959 | u64 parent, int last_ref) |
f0486c68 | 5960 | { |
f0486c68 | 5961 | struct btrfs_block_group_cache *cache = NULL; |
b150a4f1 | 5962 | int pin = 1; |
f0486c68 YZ |
5963 | int ret; |
5964 | ||
5965 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
66d7e7f0 AJ |
5966 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
5967 | buf->start, buf->len, | |
5968 | parent, root->root_key.objectid, | |
5969 | btrfs_header_level(buf), | |
5581a51a | 5970 | BTRFS_DROP_DELAYED_REF, NULL, 0); |
79787eaa | 5971 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
5972 | } |
5973 | ||
5974 | if (!last_ref) | |
5975 | return; | |
5976 | ||
f0486c68 | 5977 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); |
f0486c68 YZ |
5978 | |
5979 | if (btrfs_header_generation(buf) == trans->transid) { | |
5980 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
5981 | ret = check_ref_cleanup(trans, root, buf->start); | |
5982 | if (!ret) | |
37be25bc | 5983 | goto out; |
f0486c68 YZ |
5984 | } |
5985 | ||
5986 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
5987 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
37be25bc | 5988 | goto out; |
f0486c68 YZ |
5989 | } |
5990 | ||
5991 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
5992 | ||
5993 | btrfs_add_free_space(cache, buf->start, buf->len); | |
fb25e914 | 5994 | btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE); |
0be5dc67 | 5995 | trace_btrfs_reserved_extent_free(root, buf->start, buf->len); |
b150a4f1 | 5996 | pin = 0; |
f0486c68 YZ |
5997 | } |
5998 | out: | |
b150a4f1 JB |
5999 | if (pin) |
6000 | add_pinned_bytes(root->fs_info, buf->len, | |
6001 | btrfs_header_level(buf), | |
6002 | root->root_key.objectid); | |
6003 | ||
a826d6dc JB |
6004 | /* |
6005 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
6006 | * anymore. | |
6007 | */ | |
6008 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
6009 | btrfs_put_block_group(cache); |
6010 | } | |
6011 | ||
79787eaa | 6012 | /* Can return -ENOMEM */ |
66d7e7f0 AJ |
6013 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
6014 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, | |
6015 | u64 owner, u64 offset, int for_cow) | |
925baedd CM |
6016 | { |
6017 | int ret; | |
66d7e7f0 | 6018 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 6019 | |
b150a4f1 JB |
6020 | add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid); |
6021 | ||
56bec294 CM |
6022 | /* |
6023 | * tree log blocks never actually go into the extent allocation | |
6024 | * tree, just update pinning info and exit early. | |
56bec294 | 6025 | */ |
5d4f98a2 YZ |
6026 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
6027 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 6028 | /* unlocks the pinned mutex */ |
11833d66 | 6029 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 6030 | ret = 0; |
5d4f98a2 | 6031 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
6032 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
6033 | num_bytes, | |
5d4f98a2 | 6034 | parent, root_objectid, (int)owner, |
66d7e7f0 | 6035 | BTRFS_DROP_DELAYED_REF, NULL, for_cow); |
5d4f98a2 | 6036 | } else { |
66d7e7f0 AJ |
6037 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
6038 | num_bytes, | |
6039 | parent, root_objectid, owner, | |
6040 | offset, BTRFS_DROP_DELAYED_REF, | |
6041 | NULL, for_cow); | |
56bec294 | 6042 | } |
925baedd CM |
6043 | return ret; |
6044 | } | |
6045 | ||
53b381b3 DW |
6046 | static u64 stripe_align(struct btrfs_root *root, |
6047 | struct btrfs_block_group_cache *cache, | |
6048 | u64 val, u64 num_bytes) | |
87ee04eb | 6049 | { |
fda2832f | 6050 | u64 ret = ALIGN(val, root->stripesize); |
87ee04eb CM |
6051 | return ret; |
6052 | } | |
6053 | ||
817d52f8 JB |
6054 | /* |
6055 | * when we wait for progress in the block group caching, its because | |
6056 | * our allocation attempt failed at least once. So, we must sleep | |
6057 | * and let some progress happen before we try again. | |
6058 | * | |
6059 | * This function will sleep at least once waiting for new free space to | |
6060 | * show up, and then it will check the block group free space numbers | |
6061 | * for our min num_bytes. Another option is to have it go ahead | |
6062 | * and look in the rbtree for a free extent of a given size, but this | |
6063 | * is a good start. | |
36cce922 JB |
6064 | * |
6065 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
6066 | * any of the information in this block group. | |
817d52f8 | 6067 | */ |
36cce922 | 6068 | static noinline void |
817d52f8 JB |
6069 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
6070 | u64 num_bytes) | |
6071 | { | |
11833d66 | 6072 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 6073 | |
11833d66 YZ |
6074 | caching_ctl = get_caching_control(cache); |
6075 | if (!caching_ctl) | |
36cce922 | 6076 | return; |
817d52f8 | 6077 | |
11833d66 | 6078 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 6079 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
6080 | |
6081 | put_caching_control(caching_ctl); | |
11833d66 YZ |
6082 | } |
6083 | ||
6084 | static noinline int | |
6085 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
6086 | { | |
6087 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 6088 | int ret = 0; |
11833d66 YZ |
6089 | |
6090 | caching_ctl = get_caching_control(cache); | |
6091 | if (!caching_ctl) | |
36cce922 | 6092 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
6093 | |
6094 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
6095 | if (cache->cached == BTRFS_CACHE_ERROR) |
6096 | ret = -EIO; | |
11833d66 | 6097 | put_caching_control(caching_ctl); |
36cce922 | 6098 | return ret; |
817d52f8 JB |
6099 | } |
6100 | ||
31e50229 | 6101 | int __get_raid_index(u64 flags) |
b742bb82 | 6102 | { |
7738a53a | 6103 | if (flags & BTRFS_BLOCK_GROUP_RAID10) |
e6ec716f | 6104 | return BTRFS_RAID_RAID10; |
7738a53a | 6105 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
e6ec716f | 6106 | return BTRFS_RAID_RAID1; |
7738a53a | 6107 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
e6ec716f | 6108 | return BTRFS_RAID_DUP; |
7738a53a | 6109 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
e6ec716f | 6110 | return BTRFS_RAID_RAID0; |
53b381b3 | 6111 | else if (flags & BTRFS_BLOCK_GROUP_RAID5) |
e942f883 | 6112 | return BTRFS_RAID_RAID5; |
53b381b3 | 6113 | else if (flags & BTRFS_BLOCK_GROUP_RAID6) |
e942f883 | 6114 | return BTRFS_RAID_RAID6; |
7738a53a | 6115 | |
e942f883 | 6116 | return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ |
b742bb82 YZ |
6117 | } |
6118 | ||
7738a53a ID |
6119 | static int get_block_group_index(struct btrfs_block_group_cache *cache) |
6120 | { | |
31e50229 | 6121 | return __get_raid_index(cache->flags); |
7738a53a ID |
6122 | } |
6123 | ||
817d52f8 | 6124 | enum btrfs_loop_type { |
285ff5af JB |
6125 | LOOP_CACHING_NOWAIT = 0, |
6126 | LOOP_CACHING_WAIT = 1, | |
6127 | LOOP_ALLOC_CHUNK = 2, | |
6128 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
6129 | }; |
6130 | ||
fec577fb CM |
6131 | /* |
6132 | * walks the btree of allocated extents and find a hole of a given size. | |
6133 | * The key ins is changed to record the hole: | |
a4820398 | 6134 | * ins->objectid == start position |
62e2749e | 6135 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 6136 | * ins->offset == the size of the hole. |
fec577fb | 6137 | * Any available blocks before search_start are skipped. |
a4820398 MX |
6138 | * |
6139 | * If there is no suitable free space, we will record the max size of | |
6140 | * the free space extent currently. | |
fec577fb | 6141 | */ |
00361589 | 6142 | static noinline int find_free_extent(struct btrfs_root *orig_root, |
98ed5174 | 6143 | u64 num_bytes, u64 empty_size, |
98ed5174 | 6144 | u64 hint_byte, struct btrfs_key *ins, |
b6919a58 | 6145 | u64 flags) |
fec577fb | 6146 | { |
80eb234a | 6147 | int ret = 0; |
d397712b | 6148 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 6149 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 6150 | struct btrfs_block_group_cache *block_group = NULL; |
274bd4fb | 6151 | struct btrfs_block_group_cache *used_block_group; |
81c9ad23 | 6152 | u64 search_start = 0; |
a4820398 | 6153 | u64 max_extent_size = 0; |
239b14b3 | 6154 | int empty_cluster = 2 * 1024 * 1024; |
80eb234a | 6155 | struct btrfs_space_info *space_info; |
fa9c0d79 | 6156 | int loop = 0; |
b6919a58 DS |
6157 | int index = __get_raid_index(flags); |
6158 | int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ? | |
fb25e914 | 6159 | RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC; |
817d52f8 | 6160 | bool found_uncached_bg = false; |
0a24325e | 6161 | bool failed_cluster_refill = false; |
1cdda9b8 | 6162 | bool failed_alloc = false; |
67377734 | 6163 | bool use_cluster = true; |
60d2adbb | 6164 | bool have_caching_bg = false; |
fec577fb | 6165 | |
db94535d | 6166 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 6167 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
6168 | ins->objectid = 0; |
6169 | ins->offset = 0; | |
b1a4d965 | 6170 | |
b6919a58 | 6171 | trace_find_free_extent(orig_root, num_bytes, empty_size, flags); |
3f7de037 | 6172 | |
b6919a58 | 6173 | space_info = __find_space_info(root->fs_info, flags); |
1b1d1f66 | 6174 | if (!space_info) { |
b6919a58 | 6175 | btrfs_err(root->fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
6176 | return -ENOSPC; |
6177 | } | |
2552d17e | 6178 | |
67377734 JB |
6179 | /* |
6180 | * If the space info is for both data and metadata it means we have a | |
6181 | * small filesystem and we can't use the clustering stuff. | |
6182 | */ | |
6183 | if (btrfs_mixed_space_info(space_info)) | |
6184 | use_cluster = false; | |
6185 | ||
b6919a58 | 6186 | if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) { |
fa9c0d79 | 6187 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
6188 | if (!btrfs_test_opt(root, SSD)) |
6189 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
6190 | } |
6191 | ||
b6919a58 | 6192 | if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster && |
67377734 | 6193 | btrfs_test_opt(root, SSD)) { |
fa9c0d79 CM |
6194 | last_ptr = &root->fs_info->data_alloc_cluster; |
6195 | } | |
0f9dd46c | 6196 | |
239b14b3 | 6197 | if (last_ptr) { |
fa9c0d79 CM |
6198 | spin_lock(&last_ptr->lock); |
6199 | if (last_ptr->block_group) | |
6200 | hint_byte = last_ptr->window_start; | |
6201 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 6202 | } |
fa9c0d79 | 6203 | |
a061fc8d | 6204 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 6205 | search_start = max(search_start, hint_byte); |
0b86a832 | 6206 | |
817d52f8 | 6207 | if (!last_ptr) |
fa9c0d79 | 6208 | empty_cluster = 0; |
fa9c0d79 | 6209 | |
2552d17e | 6210 | if (search_start == hint_byte) { |
2552d17e JB |
6211 | block_group = btrfs_lookup_block_group(root->fs_info, |
6212 | search_start); | |
274bd4fb | 6213 | used_block_group = block_group; |
817d52f8 JB |
6214 | /* |
6215 | * we don't want to use the block group if it doesn't match our | |
6216 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
6217 | * |
6218 | * However if we are re-searching with an ideal block group | |
6219 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 6220 | */ |
b6919a58 | 6221 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 6222 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 6223 | down_read(&space_info->groups_sem); |
44fb5511 CM |
6224 | if (list_empty(&block_group->list) || |
6225 | block_group->ro) { | |
6226 | /* | |
6227 | * someone is removing this block group, | |
6228 | * we can't jump into the have_block_group | |
6229 | * target because our list pointers are not | |
6230 | * valid | |
6231 | */ | |
6232 | btrfs_put_block_group(block_group); | |
6233 | up_read(&space_info->groups_sem); | |
ccf0e725 | 6234 | } else { |
b742bb82 | 6235 | index = get_block_group_index(block_group); |
44fb5511 | 6236 | goto have_block_group; |
ccf0e725 | 6237 | } |
2552d17e | 6238 | } else if (block_group) { |
fa9c0d79 | 6239 | btrfs_put_block_group(block_group); |
2552d17e | 6240 | } |
42e70e7a | 6241 | } |
2552d17e | 6242 | search: |
60d2adbb | 6243 | have_caching_bg = false; |
80eb234a | 6244 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
6245 | list_for_each_entry(block_group, &space_info->block_groups[index], |
6246 | list) { | |
6226cb0a | 6247 | u64 offset; |
817d52f8 | 6248 | int cached; |
8a1413a2 | 6249 | |
274bd4fb | 6250 | used_block_group = block_group; |
11dfe35a | 6251 | btrfs_get_block_group(block_group); |
2552d17e | 6252 | search_start = block_group->key.objectid; |
42e70e7a | 6253 | |
83a50de9 CM |
6254 | /* |
6255 | * this can happen if we end up cycling through all the | |
6256 | * raid types, but we want to make sure we only allocate | |
6257 | * for the proper type. | |
6258 | */ | |
b6919a58 | 6259 | if (!block_group_bits(block_group, flags)) { |
83a50de9 CM |
6260 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
6261 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
6262 | BTRFS_BLOCK_GROUP_RAID5 | |
6263 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
6264 | BTRFS_BLOCK_GROUP_RAID10; |
6265 | ||
6266 | /* | |
6267 | * if they asked for extra copies and this block group | |
6268 | * doesn't provide them, bail. This does allow us to | |
6269 | * fill raid0 from raid1. | |
6270 | */ | |
b6919a58 | 6271 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
6272 | goto loop; |
6273 | } | |
6274 | ||
2552d17e | 6275 | have_block_group: |
291c7d2f JB |
6276 | cached = block_group_cache_done(block_group); |
6277 | if (unlikely(!cached)) { | |
291c7d2f | 6278 | found_uncached_bg = true; |
f6373bf3 | 6279 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
6280 | BUG_ON(ret < 0); |
6281 | ret = 0; | |
817d52f8 JB |
6282 | } |
6283 | ||
36cce922 JB |
6284 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
6285 | goto loop; | |
ea6a478e | 6286 | if (unlikely(block_group->ro)) |
2552d17e | 6287 | goto loop; |
0f9dd46c | 6288 | |
0a24325e | 6289 | /* |
062c05c4 AO |
6290 | * Ok we want to try and use the cluster allocator, so |
6291 | * lets look there | |
0a24325e | 6292 | */ |
062c05c4 | 6293 | if (last_ptr) { |
8de972b4 | 6294 | unsigned long aligned_cluster; |
fa9c0d79 CM |
6295 | /* |
6296 | * the refill lock keeps out other | |
6297 | * people trying to start a new cluster | |
6298 | */ | |
6299 | spin_lock(&last_ptr->refill_lock); | |
274bd4fb AO |
6300 | used_block_group = last_ptr->block_group; |
6301 | if (used_block_group != block_group && | |
6302 | (!used_block_group || | |
6303 | used_block_group->ro || | |
b6919a58 | 6304 | !block_group_bits(used_block_group, flags))) { |
274bd4fb | 6305 | used_block_group = block_group; |
44fb5511 | 6306 | goto refill_cluster; |
274bd4fb AO |
6307 | } |
6308 | ||
6309 | if (used_block_group != block_group) | |
6310 | btrfs_get_block_group(used_block_group); | |
44fb5511 | 6311 | |
274bd4fb | 6312 | offset = btrfs_alloc_from_cluster(used_block_group, |
a4820398 MX |
6313 | last_ptr, |
6314 | num_bytes, | |
6315 | used_block_group->key.objectid, | |
6316 | &max_extent_size); | |
fa9c0d79 CM |
6317 | if (offset) { |
6318 | /* we have a block, we're done */ | |
6319 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
6320 | trace_btrfs_reserve_extent_cluster(root, |
6321 | block_group, search_start, num_bytes); | |
fa9c0d79 CM |
6322 | goto checks; |
6323 | } | |
6324 | ||
274bd4fb AO |
6325 | WARN_ON(last_ptr->block_group != used_block_group); |
6326 | if (used_block_group != block_group) { | |
6327 | btrfs_put_block_group(used_block_group); | |
6328 | used_block_group = block_group; | |
fa9c0d79 | 6329 | } |
44fb5511 | 6330 | refill_cluster: |
274bd4fb | 6331 | BUG_ON(used_block_group != block_group); |
062c05c4 AO |
6332 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
6333 | * set up a new clusters, so lets just skip it | |
6334 | * and let the allocator find whatever block | |
6335 | * it can find. If we reach this point, we | |
6336 | * will have tried the cluster allocator | |
6337 | * plenty of times and not have found | |
6338 | * anything, so we are likely way too | |
6339 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
6340 | * anything. |
6341 | * | |
6342 | * However, if the cluster is taken from the | |
6343 | * current block group, release the cluster | |
6344 | * first, so that we stand a better chance of | |
6345 | * succeeding in the unclustered | |
6346 | * allocation. */ | |
6347 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
6348 | last_ptr->block_group != block_group) { | |
062c05c4 AO |
6349 | spin_unlock(&last_ptr->refill_lock); |
6350 | goto unclustered_alloc; | |
6351 | } | |
6352 | ||
fa9c0d79 CM |
6353 | /* |
6354 | * this cluster didn't work out, free it and | |
6355 | * start over | |
6356 | */ | |
6357 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
6358 | ||
a5f6f719 AO |
6359 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
6360 | spin_unlock(&last_ptr->refill_lock); | |
6361 | goto unclustered_alloc; | |
6362 | } | |
6363 | ||
8de972b4 CM |
6364 | aligned_cluster = max_t(unsigned long, |
6365 | empty_cluster + empty_size, | |
6366 | block_group->full_stripe_len); | |
6367 | ||
fa9c0d79 | 6368 | /* allocate a cluster in this block group */ |
00361589 JB |
6369 | ret = btrfs_find_space_cluster(root, block_group, |
6370 | last_ptr, search_start, | |
6371 | num_bytes, | |
6372 | aligned_cluster); | |
fa9c0d79 CM |
6373 | if (ret == 0) { |
6374 | /* | |
6375 | * now pull our allocation out of this | |
6376 | * cluster | |
6377 | */ | |
6378 | offset = btrfs_alloc_from_cluster(block_group, | |
a4820398 MX |
6379 | last_ptr, |
6380 | num_bytes, | |
6381 | search_start, | |
6382 | &max_extent_size); | |
fa9c0d79 CM |
6383 | if (offset) { |
6384 | /* we found one, proceed */ | |
6385 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
6386 | trace_btrfs_reserve_extent_cluster(root, |
6387 | block_group, search_start, | |
6388 | num_bytes); | |
fa9c0d79 CM |
6389 | goto checks; |
6390 | } | |
0a24325e JB |
6391 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
6392 | && !failed_cluster_refill) { | |
817d52f8 JB |
6393 | spin_unlock(&last_ptr->refill_lock); |
6394 | ||
0a24325e | 6395 | failed_cluster_refill = true; |
817d52f8 JB |
6396 | wait_block_group_cache_progress(block_group, |
6397 | num_bytes + empty_cluster + empty_size); | |
6398 | goto have_block_group; | |
fa9c0d79 | 6399 | } |
817d52f8 | 6400 | |
fa9c0d79 CM |
6401 | /* |
6402 | * at this point we either didn't find a cluster | |
6403 | * or we weren't able to allocate a block from our | |
6404 | * cluster. Free the cluster we've been trying | |
6405 | * to use, and go to the next block group | |
6406 | */ | |
0a24325e | 6407 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 6408 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 6409 | goto loop; |
fa9c0d79 CM |
6410 | } |
6411 | ||
062c05c4 | 6412 | unclustered_alloc: |
a5f6f719 AO |
6413 | spin_lock(&block_group->free_space_ctl->tree_lock); |
6414 | if (cached && | |
6415 | block_group->free_space_ctl->free_space < | |
6416 | num_bytes + empty_cluster + empty_size) { | |
a4820398 MX |
6417 | if (block_group->free_space_ctl->free_space > |
6418 | max_extent_size) | |
6419 | max_extent_size = | |
6420 | block_group->free_space_ctl->free_space; | |
a5f6f719 AO |
6421 | spin_unlock(&block_group->free_space_ctl->tree_lock); |
6422 | goto loop; | |
6423 | } | |
6424 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
6425 | ||
6226cb0a | 6426 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
a4820398 MX |
6427 | num_bytes, empty_size, |
6428 | &max_extent_size); | |
1cdda9b8 JB |
6429 | /* |
6430 | * If we didn't find a chunk, and we haven't failed on this | |
6431 | * block group before, and this block group is in the middle of | |
6432 | * caching and we are ok with waiting, then go ahead and wait | |
6433 | * for progress to be made, and set failed_alloc to true. | |
6434 | * | |
6435 | * If failed_alloc is true then we've already waited on this | |
6436 | * block group once and should move on to the next block group. | |
6437 | */ | |
6438 | if (!offset && !failed_alloc && !cached && | |
6439 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 6440 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
6441 | num_bytes + empty_size); |
6442 | failed_alloc = true; | |
817d52f8 | 6443 | goto have_block_group; |
1cdda9b8 | 6444 | } else if (!offset) { |
60d2adbb MX |
6445 | if (!cached) |
6446 | have_caching_bg = true; | |
1cdda9b8 | 6447 | goto loop; |
817d52f8 | 6448 | } |
fa9c0d79 | 6449 | checks: |
53b381b3 DW |
6450 | search_start = stripe_align(root, used_block_group, |
6451 | offset, num_bytes); | |
25179201 | 6452 | |
2552d17e JB |
6453 | /* move on to the next group */ |
6454 | if (search_start + num_bytes > | |
274bd4fb AO |
6455 | used_block_group->key.objectid + used_block_group->key.offset) { |
6456 | btrfs_add_free_space(used_block_group, offset, num_bytes); | |
2552d17e | 6457 | goto loop; |
6226cb0a | 6458 | } |
f5a31e16 | 6459 | |
f0486c68 | 6460 | if (offset < search_start) |
274bd4fb | 6461 | btrfs_add_free_space(used_block_group, offset, |
f0486c68 YZ |
6462 | search_start - offset); |
6463 | BUG_ON(offset > search_start); | |
2552d17e | 6464 | |
274bd4fb | 6465 | ret = btrfs_update_reserved_bytes(used_block_group, num_bytes, |
fb25e914 | 6466 | alloc_type); |
f0486c68 | 6467 | if (ret == -EAGAIN) { |
274bd4fb | 6468 | btrfs_add_free_space(used_block_group, offset, num_bytes); |
2552d17e | 6469 | goto loop; |
0f9dd46c | 6470 | } |
0b86a832 | 6471 | |
f0486c68 | 6472 | /* we are all good, lets return */ |
2552d17e JB |
6473 | ins->objectid = search_start; |
6474 | ins->offset = num_bytes; | |
d2fb3437 | 6475 | |
3f7de037 JB |
6476 | trace_btrfs_reserve_extent(orig_root, block_group, |
6477 | search_start, num_bytes); | |
274bd4fb AO |
6478 | if (used_block_group != block_group) |
6479 | btrfs_put_block_group(used_block_group); | |
d82a6f1d | 6480 | btrfs_put_block_group(block_group); |
2552d17e JB |
6481 | break; |
6482 | loop: | |
0a24325e | 6483 | failed_cluster_refill = false; |
1cdda9b8 | 6484 | failed_alloc = false; |
b742bb82 | 6485 | BUG_ON(index != get_block_group_index(block_group)); |
274bd4fb AO |
6486 | if (used_block_group != block_group) |
6487 | btrfs_put_block_group(used_block_group); | |
fa9c0d79 | 6488 | btrfs_put_block_group(block_group); |
2552d17e JB |
6489 | } |
6490 | up_read(&space_info->groups_sem); | |
6491 | ||
60d2adbb MX |
6492 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
6493 | goto search; | |
6494 | ||
b742bb82 YZ |
6495 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
6496 | goto search; | |
6497 | ||
285ff5af | 6498 | /* |
ccf0e725 JB |
6499 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
6500 | * caching kthreads as we move along | |
817d52f8 JB |
6501 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
6502 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
6503 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
6504 | * again | |
fa9c0d79 | 6505 | */ |
723bda20 | 6506 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 6507 | index = 0; |
723bda20 | 6508 | loop++; |
817d52f8 | 6509 | if (loop == LOOP_ALLOC_CHUNK) { |
00361589 JB |
6510 | struct btrfs_trans_handle *trans; |
6511 | ||
6512 | trans = btrfs_join_transaction(root); | |
6513 | if (IS_ERR(trans)) { | |
6514 | ret = PTR_ERR(trans); | |
6515 | goto out; | |
6516 | } | |
6517 | ||
b6919a58 | 6518 | ret = do_chunk_alloc(trans, root, flags, |
ea658bad JB |
6519 | CHUNK_ALLOC_FORCE); |
6520 | /* | |
6521 | * Do not bail out on ENOSPC since we | |
6522 | * can do more things. | |
6523 | */ | |
00361589 | 6524 | if (ret < 0 && ret != -ENOSPC) |
ea658bad JB |
6525 | btrfs_abort_transaction(trans, |
6526 | root, ret); | |
00361589 JB |
6527 | else |
6528 | ret = 0; | |
6529 | btrfs_end_transaction(trans, root); | |
6530 | if (ret) | |
ea658bad | 6531 | goto out; |
2552d17e JB |
6532 | } |
6533 | ||
723bda20 JB |
6534 | if (loop == LOOP_NO_EMPTY_SIZE) { |
6535 | empty_size = 0; | |
6536 | empty_cluster = 0; | |
fa9c0d79 | 6537 | } |
723bda20 JB |
6538 | |
6539 | goto search; | |
2552d17e JB |
6540 | } else if (!ins->objectid) { |
6541 | ret = -ENOSPC; | |
d82a6f1d | 6542 | } else if (ins->objectid) { |
80eb234a | 6543 | ret = 0; |
be744175 | 6544 | } |
79787eaa | 6545 | out: |
a4820398 MX |
6546 | if (ret == -ENOSPC) |
6547 | ins->offset = max_extent_size; | |
0f70abe2 | 6548 | return ret; |
fec577fb | 6549 | } |
ec44a35c | 6550 | |
9ed74f2d JB |
6551 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
6552 | int dump_block_groups) | |
0f9dd46c JB |
6553 | { |
6554 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 6555 | int index = 0; |
0f9dd46c | 6556 | |
9ed74f2d | 6557 | spin_lock(&info->lock); |
fb25e914 | 6558 | printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n", |
c1c9ff7c GU |
6559 | info->flags, |
6560 | info->total_bytes - info->bytes_used - info->bytes_pinned - | |
6561 | info->bytes_reserved - info->bytes_readonly, | |
d397712b | 6562 | (info->full) ? "" : "not "); |
8929ecfa YZ |
6563 | printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, " |
6564 | "reserved=%llu, may_use=%llu, readonly=%llu\n", | |
c1c9ff7c GU |
6565 | info->total_bytes, info->bytes_used, info->bytes_pinned, |
6566 | info->bytes_reserved, info->bytes_may_use, | |
6567 | info->bytes_readonly); | |
9ed74f2d JB |
6568 | spin_unlock(&info->lock); |
6569 | ||
6570 | if (!dump_block_groups) | |
6571 | return; | |
0f9dd46c | 6572 | |
80eb234a | 6573 | down_read(&info->groups_sem); |
b742bb82 YZ |
6574 | again: |
6575 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 6576 | spin_lock(&cache->lock); |
799ffc3c | 6577 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n", |
c1c9ff7c GU |
6578 | cache->key.objectid, cache->key.offset, |
6579 | btrfs_block_group_used(&cache->item), cache->pinned, | |
6580 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
6581 | btrfs_dump_free_space(cache, bytes); |
6582 | spin_unlock(&cache->lock); | |
6583 | } | |
b742bb82 YZ |
6584 | if (++index < BTRFS_NR_RAID_TYPES) |
6585 | goto again; | |
80eb234a | 6586 | up_read(&info->groups_sem); |
0f9dd46c | 6587 | } |
e8569813 | 6588 | |
00361589 | 6589 | int btrfs_reserve_extent(struct btrfs_root *root, |
11833d66 YZ |
6590 | u64 num_bytes, u64 min_alloc_size, |
6591 | u64 empty_size, u64 hint_byte, | |
b6919a58 | 6592 | struct btrfs_key *ins, int is_data) |
fec577fb | 6593 | { |
9e622d6b | 6594 | bool final_tried = false; |
b6919a58 | 6595 | u64 flags; |
fec577fb | 6596 | int ret; |
925baedd | 6597 | |
b6919a58 | 6598 | flags = btrfs_get_alloc_profile(root, is_data); |
98d20f67 | 6599 | again: |
db94535d | 6600 | WARN_ON(num_bytes < root->sectorsize); |
00361589 JB |
6601 | ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins, |
6602 | flags); | |
3b951516 | 6603 | |
9e622d6b | 6604 | if (ret == -ENOSPC) { |
a4820398 MX |
6605 | if (!final_tried && ins->offset) { |
6606 | num_bytes = min(num_bytes >> 1, ins->offset); | |
24542bf7 | 6607 | num_bytes = round_down(num_bytes, root->sectorsize); |
9e622d6b | 6608 | num_bytes = max(num_bytes, min_alloc_size); |
9e622d6b MX |
6609 | if (num_bytes == min_alloc_size) |
6610 | final_tried = true; | |
6611 | goto again; | |
6612 | } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
6613 | struct btrfs_space_info *sinfo; | |
6614 | ||
b6919a58 | 6615 | sinfo = __find_space_info(root->fs_info, flags); |
c2cf52eb | 6616 | btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu", |
c1c9ff7c | 6617 | flags, num_bytes); |
53804280 JM |
6618 | if (sinfo) |
6619 | dump_space_info(sinfo, num_bytes, 1); | |
9e622d6b | 6620 | } |
925baedd | 6621 | } |
0f9dd46c JB |
6622 | |
6623 | return ret; | |
e6dcd2dc CM |
6624 | } |
6625 | ||
e688b725 CM |
6626 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, |
6627 | u64 start, u64 len, int pin) | |
65b51a00 | 6628 | { |
0f9dd46c | 6629 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 6630 | int ret = 0; |
0f9dd46c | 6631 | |
0f9dd46c JB |
6632 | cache = btrfs_lookup_block_group(root->fs_info, start); |
6633 | if (!cache) { | |
c2cf52eb | 6634 | btrfs_err(root->fs_info, "Unable to find block group for %llu", |
c1c9ff7c | 6635 | start); |
0f9dd46c JB |
6636 | return -ENOSPC; |
6637 | } | |
1f3c79a2 | 6638 | |
5378e607 LD |
6639 | if (btrfs_test_opt(root, DISCARD)) |
6640 | ret = btrfs_discard_extent(root, start, len, NULL); | |
1f3c79a2 | 6641 | |
e688b725 CM |
6642 | if (pin) |
6643 | pin_down_extent(root, cache, start, len, 1); | |
6644 | else { | |
6645 | btrfs_add_free_space(cache, start, len); | |
6646 | btrfs_update_reserved_bytes(cache, len, RESERVE_FREE); | |
6647 | } | |
fa9c0d79 | 6648 | btrfs_put_block_group(cache); |
817d52f8 | 6649 | |
1abe9b8a | 6650 | trace_btrfs_reserved_extent_free(root, start, len); |
6651 | ||
e6dcd2dc CM |
6652 | return ret; |
6653 | } | |
6654 | ||
e688b725 CM |
6655 | int btrfs_free_reserved_extent(struct btrfs_root *root, |
6656 | u64 start, u64 len) | |
6657 | { | |
6658 | return __btrfs_free_reserved_extent(root, start, len, 0); | |
6659 | } | |
6660 | ||
6661 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, | |
6662 | u64 start, u64 len) | |
6663 | { | |
6664 | return __btrfs_free_reserved_extent(root, start, len, 1); | |
6665 | } | |
6666 | ||
5d4f98a2 YZ |
6667 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
6668 | struct btrfs_root *root, | |
6669 | u64 parent, u64 root_objectid, | |
6670 | u64 flags, u64 owner, u64 offset, | |
6671 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
6672 | { |
6673 | int ret; | |
5d4f98a2 | 6674 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 6675 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 6676 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 6677 | struct btrfs_path *path; |
5d4f98a2 YZ |
6678 | struct extent_buffer *leaf; |
6679 | int type; | |
6680 | u32 size; | |
26b8003f | 6681 | |
5d4f98a2 YZ |
6682 | if (parent > 0) |
6683 | type = BTRFS_SHARED_DATA_REF_KEY; | |
6684 | else | |
6685 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 6686 | |
5d4f98a2 | 6687 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
6688 | |
6689 | path = btrfs_alloc_path(); | |
db5b493a TI |
6690 | if (!path) |
6691 | return -ENOMEM; | |
47e4bb98 | 6692 | |
b9473439 | 6693 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6694 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
6695 | ins, size); | |
79787eaa JM |
6696 | if (ret) { |
6697 | btrfs_free_path(path); | |
6698 | return ret; | |
6699 | } | |
0f9dd46c | 6700 | |
5d4f98a2 YZ |
6701 | leaf = path->nodes[0]; |
6702 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 6703 | struct btrfs_extent_item); |
5d4f98a2 YZ |
6704 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
6705 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
6706 | btrfs_set_extent_flags(leaf, extent_item, | |
6707 | flags | BTRFS_EXTENT_FLAG_DATA); | |
6708 | ||
6709 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
6710 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
6711 | if (parent > 0) { | |
6712 | struct btrfs_shared_data_ref *ref; | |
6713 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
6714 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
6715 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
6716 | } else { | |
6717 | struct btrfs_extent_data_ref *ref; | |
6718 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
6719 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
6720 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
6721 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
6722 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
6723 | } | |
47e4bb98 CM |
6724 | |
6725 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 6726 | btrfs_free_path(path); |
f510cfec | 6727 | |
c53d613e | 6728 | ret = update_block_group(root, ins->objectid, ins->offset, 1); |
79787eaa | 6729 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 6730 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 6731 | ins->objectid, ins->offset); |
f5947066 CM |
6732 | BUG(); |
6733 | } | |
0be5dc67 | 6734 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
e6dcd2dc CM |
6735 | return ret; |
6736 | } | |
6737 | ||
5d4f98a2 YZ |
6738 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
6739 | struct btrfs_root *root, | |
6740 | u64 parent, u64 root_objectid, | |
6741 | u64 flags, struct btrfs_disk_key *key, | |
6742 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
6743 | { |
6744 | int ret; | |
5d4f98a2 YZ |
6745 | struct btrfs_fs_info *fs_info = root->fs_info; |
6746 | struct btrfs_extent_item *extent_item; | |
6747 | struct btrfs_tree_block_info *block_info; | |
6748 | struct btrfs_extent_inline_ref *iref; | |
6749 | struct btrfs_path *path; | |
6750 | struct extent_buffer *leaf; | |
3173a18f JB |
6751 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
6752 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, | |
6753 | SKINNY_METADATA); | |
6754 | ||
6755 | if (!skinny_metadata) | |
6756 | size += sizeof(*block_info); | |
1c2308f8 | 6757 | |
5d4f98a2 | 6758 | path = btrfs_alloc_path(); |
857cc2fc JB |
6759 | if (!path) { |
6760 | btrfs_free_and_pin_reserved_extent(root, ins->objectid, | |
6761 | root->leafsize); | |
d8926bb3 | 6762 | return -ENOMEM; |
857cc2fc | 6763 | } |
56bec294 | 6764 | |
5d4f98a2 YZ |
6765 | path->leave_spinning = 1; |
6766 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
6767 | ins, size); | |
79787eaa | 6768 | if (ret) { |
857cc2fc JB |
6769 | btrfs_free_and_pin_reserved_extent(root, ins->objectid, |
6770 | root->leafsize); | |
79787eaa JM |
6771 | btrfs_free_path(path); |
6772 | return ret; | |
6773 | } | |
5d4f98a2 YZ |
6774 | |
6775 | leaf = path->nodes[0]; | |
6776 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
6777 | struct btrfs_extent_item); | |
6778 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
6779 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
6780 | btrfs_set_extent_flags(leaf, extent_item, | |
6781 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 6782 | |
3173a18f JB |
6783 | if (skinny_metadata) { |
6784 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
6785 | } else { | |
6786 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
6787 | btrfs_set_tree_block_key(leaf, block_info, key); | |
6788 | btrfs_set_tree_block_level(leaf, block_info, level); | |
6789 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
6790 | } | |
5d4f98a2 | 6791 | |
5d4f98a2 YZ |
6792 | if (parent > 0) { |
6793 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
6794 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6795 | BTRFS_SHARED_BLOCK_REF_KEY); | |
6796 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
6797 | } else { | |
6798 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6799 | BTRFS_TREE_BLOCK_REF_KEY); | |
6800 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
6801 | } | |
6802 | ||
6803 | btrfs_mark_buffer_dirty(leaf); | |
6804 | btrfs_free_path(path); | |
6805 | ||
3173a18f | 6806 | ret = update_block_group(root, ins->objectid, root->leafsize, 1); |
79787eaa | 6807 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 6808 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 6809 | ins->objectid, ins->offset); |
5d4f98a2 YZ |
6810 | BUG(); |
6811 | } | |
0be5dc67 JB |
6812 | |
6813 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->leafsize); | |
5d4f98a2 YZ |
6814 | return ret; |
6815 | } | |
6816 | ||
6817 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
6818 | struct btrfs_root *root, | |
6819 | u64 root_objectid, u64 owner, | |
6820 | u64 offset, struct btrfs_key *ins) | |
6821 | { | |
6822 | int ret; | |
6823 | ||
6824 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
6825 | ||
66d7e7f0 AJ |
6826 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, |
6827 | ins->offset, 0, | |
6828 | root_objectid, owner, offset, | |
6829 | BTRFS_ADD_DELAYED_EXTENT, NULL, 0); | |
e6dcd2dc CM |
6830 | return ret; |
6831 | } | |
e02119d5 CM |
6832 | |
6833 | /* | |
6834 | * this is used by the tree logging recovery code. It records that | |
6835 | * an extent has been allocated and makes sure to clear the free | |
6836 | * space cache bits as well | |
6837 | */ | |
5d4f98a2 YZ |
6838 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
6839 | struct btrfs_root *root, | |
6840 | u64 root_objectid, u64 owner, u64 offset, | |
6841 | struct btrfs_key *ins) | |
e02119d5 CM |
6842 | { |
6843 | int ret; | |
6844 | struct btrfs_block_group_cache *block_group; | |
11833d66 | 6845 | |
8c2a1a30 JB |
6846 | /* |
6847 | * Mixed block groups will exclude before processing the log so we only | |
6848 | * need to do the exlude dance if this fs isn't mixed. | |
6849 | */ | |
6850 | if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) { | |
6851 | ret = __exclude_logged_extent(root, ins->objectid, ins->offset); | |
b50c6e25 | 6852 | if (ret) |
8c2a1a30 | 6853 | return ret; |
11833d66 YZ |
6854 | } |
6855 | ||
8c2a1a30 JB |
6856 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
6857 | if (!block_group) | |
6858 | return -EINVAL; | |
6859 | ||
fb25e914 JB |
6860 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, |
6861 | RESERVE_ALLOC_NO_ACCOUNT); | |
79787eaa | 6862 | BUG_ON(ret); /* logic error */ |
5d4f98a2 YZ |
6863 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
6864 | 0, owner, offset, ins, 1); | |
b50c6e25 | 6865 | btrfs_put_block_group(block_group); |
e02119d5 CM |
6866 | return ret; |
6867 | } | |
6868 | ||
48a3b636 ES |
6869 | static struct extent_buffer * |
6870 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
6871 | u64 bytenr, u32 blocksize, int level) | |
65b51a00 CM |
6872 | { |
6873 | struct extent_buffer *buf; | |
6874 | ||
6875 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
6876 | if (!buf) | |
6877 | return ERR_PTR(-ENOMEM); | |
6878 | btrfs_set_header_generation(buf, trans->transid); | |
85d4e461 | 6879 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 CM |
6880 | btrfs_tree_lock(buf); |
6881 | clean_tree_block(trans, root, buf); | |
3083ee2e | 6882 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
6883 | |
6884 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 6885 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 6886 | |
d0c803c4 | 6887 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
8cef4e16 YZ |
6888 | /* |
6889 | * we allow two log transactions at a time, use different | |
6890 | * EXENT bit to differentiate dirty pages. | |
6891 | */ | |
6892 | if (root->log_transid % 2 == 0) | |
6893 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
6894 | buf->start + buf->len - 1, GFP_NOFS); | |
6895 | else | |
6896 | set_extent_new(&root->dirty_log_pages, buf->start, | |
6897 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 CM |
6898 | } else { |
6899 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 6900 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 6901 | } |
65b51a00 | 6902 | trans->blocks_used++; |
b4ce94de | 6903 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
6904 | return buf; |
6905 | } | |
6906 | ||
f0486c68 YZ |
6907 | static struct btrfs_block_rsv * |
6908 | use_block_rsv(struct btrfs_trans_handle *trans, | |
6909 | struct btrfs_root *root, u32 blocksize) | |
6910 | { | |
6911 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 6912 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 | 6913 | int ret; |
d88033db | 6914 | bool global_updated = false; |
f0486c68 YZ |
6915 | |
6916 | block_rsv = get_block_rsv(trans, root); | |
6917 | ||
b586b323 MX |
6918 | if (unlikely(block_rsv->size == 0)) |
6919 | goto try_reserve; | |
d88033db | 6920 | again: |
f0486c68 YZ |
6921 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
6922 | if (!ret) | |
6923 | return block_rsv; | |
6924 | ||
b586b323 MX |
6925 | if (block_rsv->failfast) |
6926 | return ERR_PTR(ret); | |
6927 | ||
d88033db MX |
6928 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
6929 | global_updated = true; | |
6930 | update_global_block_rsv(root->fs_info); | |
6931 | goto again; | |
6932 | } | |
6933 | ||
b586b323 MX |
6934 | if (btrfs_test_opt(root, ENOSPC_DEBUG)) { |
6935 | static DEFINE_RATELIMIT_STATE(_rs, | |
6936 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
6937 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
6938 | if (__ratelimit(&_rs)) | |
6939 | WARN(1, KERN_DEBUG | |
6940 | "btrfs: block rsv returned %d\n", ret); | |
6941 | } | |
6942 | try_reserve: | |
6943 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
6944 | BTRFS_RESERVE_NO_FLUSH); | |
6945 | if (!ret) | |
6946 | return block_rsv; | |
6947 | /* | |
6948 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
6949 | * the global reserve if its space type is the same as the global |
6950 | * reservation. | |
b586b323 | 6951 | */ |
5881cfc9 MX |
6952 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
6953 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
6954 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
6955 | if (!ret) | |
6956 | return global_rsv; | |
6957 | } | |
6958 | return ERR_PTR(ret); | |
f0486c68 YZ |
6959 | } |
6960 | ||
8c2a3ca2 JB |
6961 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
6962 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
6963 | { |
6964 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 6965 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
6966 | } |
6967 | ||
fec577fb | 6968 | /* |
f0486c68 YZ |
6969 | * finds a free extent and does all the dirty work required for allocation |
6970 | * returns the key for the extent through ins, and a tree buffer for | |
6971 | * the first block of the extent through buf. | |
6972 | * | |
fec577fb CM |
6973 | * returns the tree buffer or NULL. |
6974 | */ | |
5f39d397 | 6975 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
6976 | struct btrfs_root *root, u32 blocksize, |
6977 | u64 parent, u64 root_objectid, | |
6978 | struct btrfs_disk_key *key, int level, | |
5581a51a | 6979 | u64 hint, u64 empty_size) |
fec577fb | 6980 | { |
e2fa7227 | 6981 | struct btrfs_key ins; |
f0486c68 | 6982 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 6983 | struct extent_buffer *buf; |
f0486c68 YZ |
6984 | u64 flags = 0; |
6985 | int ret; | |
3173a18f JB |
6986 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
6987 | SKINNY_METADATA); | |
fec577fb | 6988 | |
f0486c68 YZ |
6989 | block_rsv = use_block_rsv(trans, root, blocksize); |
6990 | if (IS_ERR(block_rsv)) | |
6991 | return ERR_CAST(block_rsv); | |
6992 | ||
00361589 | 6993 | ret = btrfs_reserve_extent(root, blocksize, blocksize, |
81c9ad23 | 6994 | empty_size, hint, &ins, 0); |
fec577fb | 6995 | if (ret) { |
8c2a3ca2 | 6996 | unuse_block_rsv(root->fs_info, block_rsv, blocksize); |
54aa1f4d | 6997 | return ERR_PTR(ret); |
fec577fb | 6998 | } |
55c69072 | 6999 | |
4008c04a CM |
7000 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
7001 | blocksize, level); | |
79787eaa | 7002 | BUG_ON(IS_ERR(buf)); /* -ENOMEM */ |
f0486c68 YZ |
7003 | |
7004 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
7005 | if (parent == 0) | |
7006 | parent = ins.objectid; | |
7007 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
7008 | } else | |
7009 | BUG_ON(parent > 0); | |
7010 | ||
7011 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
7012 | struct btrfs_delayed_extent_op *extent_op; | |
78a6184a | 7013 | extent_op = btrfs_alloc_delayed_extent_op(); |
79787eaa | 7014 | BUG_ON(!extent_op); /* -ENOMEM */ |
f0486c68 YZ |
7015 | if (key) |
7016 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
7017 | else | |
7018 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
7019 | extent_op->flags_to_set = flags; | |
3173a18f JB |
7020 | if (skinny_metadata) |
7021 | extent_op->update_key = 0; | |
7022 | else | |
7023 | extent_op->update_key = 1; | |
f0486c68 YZ |
7024 | extent_op->update_flags = 1; |
7025 | extent_op->is_data = 0; | |
b1c79e09 | 7026 | extent_op->level = level; |
f0486c68 | 7027 | |
66d7e7f0 AJ |
7028 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
7029 | ins.objectid, | |
f0486c68 YZ |
7030 | ins.offset, parent, root_objectid, |
7031 | level, BTRFS_ADD_DELAYED_EXTENT, | |
5581a51a | 7032 | extent_op, 0); |
79787eaa | 7033 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 | 7034 | } |
fec577fb CM |
7035 | return buf; |
7036 | } | |
a28ec197 | 7037 | |
2c47e605 YZ |
7038 | struct walk_control { |
7039 | u64 refs[BTRFS_MAX_LEVEL]; | |
7040 | u64 flags[BTRFS_MAX_LEVEL]; | |
7041 | struct btrfs_key update_progress; | |
7042 | int stage; | |
7043 | int level; | |
7044 | int shared_level; | |
7045 | int update_ref; | |
7046 | int keep_locks; | |
1c4850e2 YZ |
7047 | int reada_slot; |
7048 | int reada_count; | |
66d7e7f0 | 7049 | int for_reloc; |
2c47e605 YZ |
7050 | }; |
7051 | ||
7052 | #define DROP_REFERENCE 1 | |
7053 | #define UPDATE_BACKREF 2 | |
7054 | ||
1c4850e2 YZ |
7055 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
7056 | struct btrfs_root *root, | |
7057 | struct walk_control *wc, | |
7058 | struct btrfs_path *path) | |
6407bf6d | 7059 | { |
1c4850e2 YZ |
7060 | u64 bytenr; |
7061 | u64 generation; | |
7062 | u64 refs; | |
94fcca9f | 7063 | u64 flags; |
5d4f98a2 | 7064 | u32 nritems; |
1c4850e2 YZ |
7065 | u32 blocksize; |
7066 | struct btrfs_key key; | |
7067 | struct extent_buffer *eb; | |
6407bf6d | 7068 | int ret; |
1c4850e2 YZ |
7069 | int slot; |
7070 | int nread = 0; | |
6407bf6d | 7071 | |
1c4850e2 YZ |
7072 | if (path->slots[wc->level] < wc->reada_slot) { |
7073 | wc->reada_count = wc->reada_count * 2 / 3; | |
7074 | wc->reada_count = max(wc->reada_count, 2); | |
7075 | } else { | |
7076 | wc->reada_count = wc->reada_count * 3 / 2; | |
7077 | wc->reada_count = min_t(int, wc->reada_count, | |
7078 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
7079 | } | |
7bb86316 | 7080 | |
1c4850e2 YZ |
7081 | eb = path->nodes[wc->level]; |
7082 | nritems = btrfs_header_nritems(eb); | |
7083 | blocksize = btrfs_level_size(root, wc->level - 1); | |
bd56b302 | 7084 | |
1c4850e2 YZ |
7085 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
7086 | if (nread >= wc->reada_count) | |
7087 | break; | |
bd56b302 | 7088 | |
2dd3e67b | 7089 | cond_resched(); |
1c4850e2 YZ |
7090 | bytenr = btrfs_node_blockptr(eb, slot); |
7091 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 7092 | |
1c4850e2 YZ |
7093 | if (slot == path->slots[wc->level]) |
7094 | goto reada; | |
5d4f98a2 | 7095 | |
1c4850e2 YZ |
7096 | if (wc->stage == UPDATE_BACKREF && |
7097 | generation <= root->root_key.offset) | |
bd56b302 CM |
7098 | continue; |
7099 | ||
94fcca9f | 7100 | /* We don't lock the tree block, it's OK to be racy here */ |
3173a18f JB |
7101 | ret = btrfs_lookup_extent_info(trans, root, bytenr, |
7102 | wc->level - 1, 1, &refs, | |
7103 | &flags); | |
79787eaa JM |
7104 | /* We don't care about errors in readahead. */ |
7105 | if (ret < 0) | |
7106 | continue; | |
94fcca9f YZ |
7107 | BUG_ON(refs == 0); |
7108 | ||
1c4850e2 | 7109 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
7110 | if (refs == 1) |
7111 | goto reada; | |
bd56b302 | 7112 | |
94fcca9f YZ |
7113 | if (wc->level == 1 && |
7114 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
7115 | continue; | |
1c4850e2 YZ |
7116 | if (!wc->update_ref || |
7117 | generation <= root->root_key.offset) | |
7118 | continue; | |
7119 | btrfs_node_key_to_cpu(eb, &key, slot); | |
7120 | ret = btrfs_comp_cpu_keys(&key, | |
7121 | &wc->update_progress); | |
7122 | if (ret < 0) | |
7123 | continue; | |
94fcca9f YZ |
7124 | } else { |
7125 | if (wc->level == 1 && | |
7126 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
7127 | continue; | |
6407bf6d | 7128 | } |
1c4850e2 YZ |
7129 | reada: |
7130 | ret = readahead_tree_block(root, bytenr, blocksize, | |
7131 | generation); | |
7132 | if (ret) | |
bd56b302 | 7133 | break; |
1c4850e2 | 7134 | nread++; |
20524f02 | 7135 | } |
1c4850e2 | 7136 | wc->reada_slot = slot; |
20524f02 | 7137 | } |
2c47e605 | 7138 | |
f82d02d9 | 7139 | /* |
2c016dc2 | 7140 | * helper to process tree block while walking down the tree. |
2c47e605 | 7141 | * |
2c47e605 YZ |
7142 | * when wc->stage == UPDATE_BACKREF, this function updates |
7143 | * back refs for pointers in the block. | |
7144 | * | |
7145 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 7146 | */ |
2c47e605 | 7147 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 7148 | struct btrfs_root *root, |
2c47e605 | 7149 | struct btrfs_path *path, |
94fcca9f | 7150 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 7151 | { |
2c47e605 YZ |
7152 | int level = wc->level; |
7153 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 7154 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
7155 | int ret; |
7156 | ||
2c47e605 YZ |
7157 | if (wc->stage == UPDATE_BACKREF && |
7158 | btrfs_header_owner(eb) != root->root_key.objectid) | |
7159 | return 1; | |
f82d02d9 | 7160 | |
2c47e605 YZ |
7161 | /* |
7162 | * when reference count of tree block is 1, it won't increase | |
7163 | * again. once full backref flag is set, we never clear it. | |
7164 | */ | |
94fcca9f YZ |
7165 | if (lookup_info && |
7166 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
7167 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
7168 | BUG_ON(!path->locks[level]); |
7169 | ret = btrfs_lookup_extent_info(trans, root, | |
3173a18f | 7170 | eb->start, level, 1, |
2c47e605 YZ |
7171 | &wc->refs[level], |
7172 | &wc->flags[level]); | |
79787eaa JM |
7173 | BUG_ON(ret == -ENOMEM); |
7174 | if (ret) | |
7175 | return ret; | |
2c47e605 YZ |
7176 | BUG_ON(wc->refs[level] == 0); |
7177 | } | |
5d4f98a2 | 7178 | |
2c47e605 YZ |
7179 | if (wc->stage == DROP_REFERENCE) { |
7180 | if (wc->refs[level] > 1) | |
7181 | return 1; | |
f82d02d9 | 7182 | |
2c47e605 | 7183 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 7184 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
7185 | path->locks[level] = 0; |
7186 | } | |
7187 | return 0; | |
7188 | } | |
f82d02d9 | 7189 | |
2c47e605 YZ |
7190 | /* wc->stage == UPDATE_BACKREF */ |
7191 | if (!(wc->flags[level] & flag)) { | |
7192 | BUG_ON(!path->locks[level]); | |
66d7e7f0 | 7193 | ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc); |
79787eaa | 7194 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 7195 | ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc); |
79787eaa | 7196 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 | 7197 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, |
b1c79e09 JB |
7198 | eb->len, flag, |
7199 | btrfs_header_level(eb), 0); | |
79787eaa | 7200 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
7201 | wc->flags[level] |= flag; |
7202 | } | |
7203 | ||
7204 | /* | |
7205 | * the block is shared by multiple trees, so it's not good to | |
7206 | * keep the tree lock | |
7207 | */ | |
7208 | if (path->locks[level] && level > 0) { | |
bd681513 | 7209 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
7210 | path->locks[level] = 0; |
7211 | } | |
7212 | return 0; | |
7213 | } | |
7214 | ||
1c4850e2 | 7215 | /* |
2c016dc2 | 7216 | * helper to process tree block pointer. |
1c4850e2 YZ |
7217 | * |
7218 | * when wc->stage == DROP_REFERENCE, this function checks | |
7219 | * reference count of the block pointed to. if the block | |
7220 | * is shared and we need update back refs for the subtree | |
7221 | * rooted at the block, this function changes wc->stage to | |
7222 | * UPDATE_BACKREF. if the block is shared and there is no | |
7223 | * need to update back, this function drops the reference | |
7224 | * to the block. | |
7225 | * | |
7226 | * NOTE: return value 1 means we should stop walking down. | |
7227 | */ | |
7228 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
7229 | struct btrfs_root *root, | |
7230 | struct btrfs_path *path, | |
94fcca9f | 7231 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
7232 | { |
7233 | u64 bytenr; | |
7234 | u64 generation; | |
7235 | u64 parent; | |
7236 | u32 blocksize; | |
7237 | struct btrfs_key key; | |
7238 | struct extent_buffer *next; | |
7239 | int level = wc->level; | |
7240 | int reada = 0; | |
7241 | int ret = 0; | |
7242 | ||
7243 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
7244 | path->slots[level]); | |
7245 | /* | |
7246 | * if the lower level block was created before the snapshot | |
7247 | * was created, we know there is no need to update back refs | |
7248 | * for the subtree | |
7249 | */ | |
7250 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
7251 | generation <= root->root_key.offset) { |
7252 | *lookup_info = 1; | |
1c4850e2 | 7253 | return 1; |
94fcca9f | 7254 | } |
1c4850e2 YZ |
7255 | |
7256 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
7257 | blocksize = btrfs_level_size(root, level - 1); | |
7258 | ||
7259 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
7260 | if (!next) { | |
7261 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
90d2c51d MX |
7262 | if (!next) |
7263 | return -ENOMEM; | |
b2aaaa3b JB |
7264 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
7265 | level - 1); | |
1c4850e2 YZ |
7266 | reada = 1; |
7267 | } | |
7268 | btrfs_tree_lock(next); | |
7269 | btrfs_set_lock_blocking(next); | |
7270 | ||
3173a18f | 7271 | ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1, |
94fcca9f YZ |
7272 | &wc->refs[level - 1], |
7273 | &wc->flags[level - 1]); | |
79787eaa JM |
7274 | if (ret < 0) { |
7275 | btrfs_tree_unlock(next); | |
7276 | return ret; | |
7277 | } | |
7278 | ||
c2cf52eb SK |
7279 | if (unlikely(wc->refs[level - 1] == 0)) { |
7280 | btrfs_err(root->fs_info, "Missing references."); | |
7281 | BUG(); | |
7282 | } | |
94fcca9f | 7283 | *lookup_info = 0; |
1c4850e2 | 7284 | |
94fcca9f | 7285 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 7286 | if (wc->refs[level - 1] > 1) { |
94fcca9f YZ |
7287 | if (level == 1 && |
7288 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
7289 | goto skip; | |
7290 | ||
1c4850e2 YZ |
7291 | if (!wc->update_ref || |
7292 | generation <= root->root_key.offset) | |
7293 | goto skip; | |
7294 | ||
7295 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
7296 | path->slots[level]); | |
7297 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
7298 | if (ret < 0) | |
7299 | goto skip; | |
7300 | ||
7301 | wc->stage = UPDATE_BACKREF; | |
7302 | wc->shared_level = level - 1; | |
7303 | } | |
94fcca9f YZ |
7304 | } else { |
7305 | if (level == 1 && | |
7306 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
7307 | goto skip; | |
1c4850e2 YZ |
7308 | } |
7309 | ||
b9fab919 | 7310 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
7311 | btrfs_tree_unlock(next); |
7312 | free_extent_buffer(next); | |
7313 | next = NULL; | |
94fcca9f | 7314 | *lookup_info = 1; |
1c4850e2 YZ |
7315 | } |
7316 | ||
7317 | if (!next) { | |
7318 | if (reada && level == 1) | |
7319 | reada_walk_down(trans, root, wc, path); | |
7320 | next = read_tree_block(root, bytenr, blocksize, generation); | |
416bc658 JB |
7321 | if (!next || !extent_buffer_uptodate(next)) { |
7322 | free_extent_buffer(next); | |
97d9a8a4 | 7323 | return -EIO; |
416bc658 | 7324 | } |
1c4850e2 YZ |
7325 | btrfs_tree_lock(next); |
7326 | btrfs_set_lock_blocking(next); | |
7327 | } | |
7328 | ||
7329 | level--; | |
7330 | BUG_ON(level != btrfs_header_level(next)); | |
7331 | path->nodes[level] = next; | |
7332 | path->slots[level] = 0; | |
bd681513 | 7333 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
7334 | wc->level = level; |
7335 | if (wc->level == 1) | |
7336 | wc->reada_slot = 0; | |
7337 | return 0; | |
7338 | skip: | |
7339 | wc->refs[level - 1] = 0; | |
7340 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
7341 | if (wc->stage == DROP_REFERENCE) { |
7342 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
7343 | parent = path->nodes[level]->start; | |
7344 | } else { | |
7345 | BUG_ON(root->root_key.objectid != | |
7346 | btrfs_header_owner(path->nodes[level])); | |
7347 | parent = 0; | |
7348 | } | |
1c4850e2 | 7349 | |
94fcca9f | 7350 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
66d7e7f0 | 7351 | root->root_key.objectid, level - 1, 0, 0); |
79787eaa | 7352 | BUG_ON(ret); /* -ENOMEM */ |
1c4850e2 | 7353 | } |
1c4850e2 YZ |
7354 | btrfs_tree_unlock(next); |
7355 | free_extent_buffer(next); | |
94fcca9f | 7356 | *lookup_info = 1; |
1c4850e2 YZ |
7357 | return 1; |
7358 | } | |
7359 | ||
2c47e605 | 7360 | /* |
2c016dc2 | 7361 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
7362 | * |
7363 | * when wc->stage == DROP_REFERENCE, this function drops | |
7364 | * reference count on the block. | |
7365 | * | |
7366 | * when wc->stage == UPDATE_BACKREF, this function changes | |
7367 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
7368 | * to UPDATE_BACKREF previously while processing the block. | |
7369 | * | |
7370 | * NOTE: return value 1 means we should stop walking up. | |
7371 | */ | |
7372 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
7373 | struct btrfs_root *root, | |
7374 | struct btrfs_path *path, | |
7375 | struct walk_control *wc) | |
7376 | { | |
f0486c68 | 7377 | int ret; |
2c47e605 YZ |
7378 | int level = wc->level; |
7379 | struct extent_buffer *eb = path->nodes[level]; | |
7380 | u64 parent = 0; | |
7381 | ||
7382 | if (wc->stage == UPDATE_BACKREF) { | |
7383 | BUG_ON(wc->shared_level < level); | |
7384 | if (level < wc->shared_level) | |
7385 | goto out; | |
7386 | ||
2c47e605 YZ |
7387 | ret = find_next_key(path, level + 1, &wc->update_progress); |
7388 | if (ret > 0) | |
7389 | wc->update_ref = 0; | |
7390 | ||
7391 | wc->stage = DROP_REFERENCE; | |
7392 | wc->shared_level = -1; | |
7393 | path->slots[level] = 0; | |
7394 | ||
7395 | /* | |
7396 | * check reference count again if the block isn't locked. | |
7397 | * we should start walking down the tree again if reference | |
7398 | * count is one. | |
7399 | */ | |
7400 | if (!path->locks[level]) { | |
7401 | BUG_ON(level == 0); | |
7402 | btrfs_tree_lock(eb); | |
7403 | btrfs_set_lock_blocking(eb); | |
bd681513 | 7404 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7405 | |
7406 | ret = btrfs_lookup_extent_info(trans, root, | |
3173a18f | 7407 | eb->start, level, 1, |
2c47e605 YZ |
7408 | &wc->refs[level], |
7409 | &wc->flags[level]); | |
79787eaa JM |
7410 | if (ret < 0) { |
7411 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 7412 | path->locks[level] = 0; |
79787eaa JM |
7413 | return ret; |
7414 | } | |
2c47e605 YZ |
7415 | BUG_ON(wc->refs[level] == 0); |
7416 | if (wc->refs[level] == 1) { | |
bd681513 | 7417 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 7418 | path->locks[level] = 0; |
2c47e605 YZ |
7419 | return 1; |
7420 | } | |
f82d02d9 | 7421 | } |
2c47e605 | 7422 | } |
f82d02d9 | 7423 | |
2c47e605 YZ |
7424 | /* wc->stage == DROP_REFERENCE */ |
7425 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 7426 | |
2c47e605 YZ |
7427 | if (wc->refs[level] == 1) { |
7428 | if (level == 0) { | |
7429 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
66d7e7f0 AJ |
7430 | ret = btrfs_dec_ref(trans, root, eb, 1, |
7431 | wc->for_reloc); | |
2c47e605 | 7432 | else |
66d7e7f0 AJ |
7433 | ret = btrfs_dec_ref(trans, root, eb, 0, |
7434 | wc->for_reloc); | |
79787eaa | 7435 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
7436 | } |
7437 | /* make block locked assertion in clean_tree_block happy */ | |
7438 | if (!path->locks[level] && | |
7439 | btrfs_header_generation(eb) == trans->transid) { | |
7440 | btrfs_tree_lock(eb); | |
7441 | btrfs_set_lock_blocking(eb); | |
bd681513 | 7442 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7443 | } |
7444 | clean_tree_block(trans, root, eb); | |
7445 | } | |
7446 | ||
7447 | if (eb == root->node) { | |
7448 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
7449 | parent = eb->start; | |
7450 | else | |
7451 | BUG_ON(root->root_key.objectid != | |
7452 | btrfs_header_owner(eb)); | |
7453 | } else { | |
7454 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
7455 | parent = path->nodes[level + 1]->start; | |
7456 | else | |
7457 | BUG_ON(root->root_key.objectid != | |
7458 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 7459 | } |
f82d02d9 | 7460 | |
5581a51a | 7461 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
7462 | out: |
7463 | wc->refs[level] = 0; | |
7464 | wc->flags[level] = 0; | |
f0486c68 | 7465 | return 0; |
2c47e605 YZ |
7466 | } |
7467 | ||
7468 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
7469 | struct btrfs_root *root, | |
7470 | struct btrfs_path *path, | |
7471 | struct walk_control *wc) | |
7472 | { | |
2c47e605 | 7473 | int level = wc->level; |
94fcca9f | 7474 | int lookup_info = 1; |
2c47e605 YZ |
7475 | int ret; |
7476 | ||
7477 | while (level >= 0) { | |
94fcca9f | 7478 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
7479 | if (ret > 0) |
7480 | break; | |
7481 | ||
7482 | if (level == 0) | |
7483 | break; | |
7484 | ||
7a7965f8 YZ |
7485 | if (path->slots[level] >= |
7486 | btrfs_header_nritems(path->nodes[level])) | |
7487 | break; | |
7488 | ||
94fcca9f | 7489 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
7490 | if (ret > 0) { |
7491 | path->slots[level]++; | |
7492 | continue; | |
90d2c51d MX |
7493 | } else if (ret < 0) |
7494 | return ret; | |
1c4850e2 | 7495 | level = wc->level; |
f82d02d9 | 7496 | } |
f82d02d9 YZ |
7497 | return 0; |
7498 | } | |
7499 | ||
d397712b | 7500 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 7501 | struct btrfs_root *root, |
f82d02d9 | 7502 | struct btrfs_path *path, |
2c47e605 | 7503 | struct walk_control *wc, int max_level) |
20524f02 | 7504 | { |
2c47e605 | 7505 | int level = wc->level; |
20524f02 | 7506 | int ret; |
9f3a7427 | 7507 | |
2c47e605 YZ |
7508 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
7509 | while (level < max_level && path->nodes[level]) { | |
7510 | wc->level = level; | |
7511 | if (path->slots[level] + 1 < | |
7512 | btrfs_header_nritems(path->nodes[level])) { | |
7513 | path->slots[level]++; | |
20524f02 CM |
7514 | return 0; |
7515 | } else { | |
2c47e605 YZ |
7516 | ret = walk_up_proc(trans, root, path, wc); |
7517 | if (ret > 0) | |
7518 | return 0; | |
bd56b302 | 7519 | |
2c47e605 | 7520 | if (path->locks[level]) { |
bd681513 CM |
7521 | btrfs_tree_unlock_rw(path->nodes[level], |
7522 | path->locks[level]); | |
2c47e605 | 7523 | path->locks[level] = 0; |
f82d02d9 | 7524 | } |
2c47e605 YZ |
7525 | free_extent_buffer(path->nodes[level]); |
7526 | path->nodes[level] = NULL; | |
7527 | level++; | |
20524f02 CM |
7528 | } |
7529 | } | |
7530 | return 1; | |
7531 | } | |
7532 | ||
9aca1d51 | 7533 | /* |
2c47e605 YZ |
7534 | * drop a subvolume tree. |
7535 | * | |
7536 | * this function traverses the tree freeing any blocks that only | |
7537 | * referenced by the tree. | |
7538 | * | |
7539 | * when a shared tree block is found. this function decreases its | |
7540 | * reference count by one. if update_ref is true, this function | |
7541 | * also make sure backrefs for the shared block and all lower level | |
7542 | * blocks are properly updated. | |
9d1a2a3a DS |
7543 | * |
7544 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 7545 | */ |
2c536799 | 7546 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
7547 | struct btrfs_block_rsv *block_rsv, int update_ref, |
7548 | int for_reloc) | |
20524f02 | 7549 | { |
5caf2a00 | 7550 | struct btrfs_path *path; |
2c47e605 YZ |
7551 | struct btrfs_trans_handle *trans; |
7552 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 7553 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
7554 | struct walk_control *wc; |
7555 | struct btrfs_key key; | |
7556 | int err = 0; | |
7557 | int ret; | |
7558 | int level; | |
d29a9f62 | 7559 | bool root_dropped = false; |
20524f02 | 7560 | |
5caf2a00 | 7561 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
7562 | if (!path) { |
7563 | err = -ENOMEM; | |
7564 | goto out; | |
7565 | } | |
20524f02 | 7566 | |
2c47e605 | 7567 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
7568 | if (!wc) { |
7569 | btrfs_free_path(path); | |
cb1b69f4 TI |
7570 | err = -ENOMEM; |
7571 | goto out; | |
38a1a919 | 7572 | } |
2c47e605 | 7573 | |
a22285a6 | 7574 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
7575 | if (IS_ERR(trans)) { |
7576 | err = PTR_ERR(trans); | |
7577 | goto out_free; | |
7578 | } | |
98d5dc13 | 7579 | |
3fd0a558 YZ |
7580 | if (block_rsv) |
7581 | trans->block_rsv = block_rsv; | |
2c47e605 | 7582 | |
9f3a7427 | 7583 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 7584 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
7585 | path->nodes[level] = btrfs_lock_root_node(root); |
7586 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 7587 | path->slots[level] = 0; |
bd681513 | 7588 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7589 | memset(&wc->update_progress, 0, |
7590 | sizeof(wc->update_progress)); | |
9f3a7427 | 7591 | } else { |
9f3a7427 | 7592 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
7593 | memcpy(&wc->update_progress, &key, |
7594 | sizeof(wc->update_progress)); | |
7595 | ||
6702ed49 | 7596 | level = root_item->drop_level; |
2c47e605 | 7597 | BUG_ON(level == 0); |
6702ed49 | 7598 | path->lowest_level = level; |
2c47e605 YZ |
7599 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
7600 | path->lowest_level = 0; | |
7601 | if (ret < 0) { | |
7602 | err = ret; | |
79787eaa | 7603 | goto out_end_trans; |
9f3a7427 | 7604 | } |
1c4850e2 | 7605 | WARN_ON(ret > 0); |
2c47e605 | 7606 | |
7d9eb12c CM |
7607 | /* |
7608 | * unlock our path, this is safe because only this | |
7609 | * function is allowed to delete this snapshot | |
7610 | */ | |
5d4f98a2 | 7611 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
7612 | |
7613 | level = btrfs_header_level(root->node); | |
7614 | while (1) { | |
7615 | btrfs_tree_lock(path->nodes[level]); | |
7616 | btrfs_set_lock_blocking(path->nodes[level]); | |
fec386ac | 7617 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7618 | |
7619 | ret = btrfs_lookup_extent_info(trans, root, | |
7620 | path->nodes[level]->start, | |
3173a18f | 7621 | level, 1, &wc->refs[level], |
2c47e605 | 7622 | &wc->flags[level]); |
79787eaa JM |
7623 | if (ret < 0) { |
7624 | err = ret; | |
7625 | goto out_end_trans; | |
7626 | } | |
2c47e605 YZ |
7627 | BUG_ON(wc->refs[level] == 0); |
7628 | ||
7629 | if (level == root_item->drop_level) | |
7630 | break; | |
7631 | ||
7632 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 7633 | path->locks[level] = 0; |
2c47e605 YZ |
7634 | WARN_ON(wc->refs[level] != 1); |
7635 | level--; | |
7636 | } | |
9f3a7427 | 7637 | } |
2c47e605 YZ |
7638 | |
7639 | wc->level = level; | |
7640 | wc->shared_level = -1; | |
7641 | wc->stage = DROP_REFERENCE; | |
7642 | wc->update_ref = update_ref; | |
7643 | wc->keep_locks = 0; | |
66d7e7f0 | 7644 | wc->for_reloc = for_reloc; |
1c4850e2 | 7645 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 7646 | |
d397712b | 7647 | while (1) { |
9d1a2a3a | 7648 | |
2c47e605 YZ |
7649 | ret = walk_down_tree(trans, root, path, wc); |
7650 | if (ret < 0) { | |
7651 | err = ret; | |
20524f02 | 7652 | break; |
2c47e605 | 7653 | } |
9aca1d51 | 7654 | |
2c47e605 YZ |
7655 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
7656 | if (ret < 0) { | |
7657 | err = ret; | |
20524f02 | 7658 | break; |
2c47e605 YZ |
7659 | } |
7660 | ||
7661 | if (ret > 0) { | |
7662 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
7663 | break; |
7664 | } | |
2c47e605 YZ |
7665 | |
7666 | if (wc->stage == DROP_REFERENCE) { | |
7667 | level = wc->level; | |
7668 | btrfs_node_key(path->nodes[level], | |
7669 | &root_item->drop_progress, | |
7670 | path->slots[level]); | |
7671 | root_item->drop_level = level; | |
7672 | } | |
7673 | ||
7674 | BUG_ON(wc->level == 0); | |
3c8f2422 JB |
7675 | if (btrfs_should_end_transaction(trans, tree_root) || |
7676 | (!for_reloc && btrfs_need_cleaner_sleep(root))) { | |
2c47e605 YZ |
7677 | ret = btrfs_update_root(trans, tree_root, |
7678 | &root->root_key, | |
7679 | root_item); | |
79787eaa JM |
7680 | if (ret) { |
7681 | btrfs_abort_transaction(trans, tree_root, ret); | |
7682 | err = ret; | |
7683 | goto out_end_trans; | |
7684 | } | |
2c47e605 | 7685 | |
3fd0a558 | 7686 | btrfs_end_transaction_throttle(trans, tree_root); |
3c8f2422 JB |
7687 | if (!for_reloc && btrfs_need_cleaner_sleep(root)) { |
7688 | pr_debug("btrfs: drop snapshot early exit\n"); | |
7689 | err = -EAGAIN; | |
7690 | goto out_free; | |
7691 | } | |
7692 | ||
a22285a6 | 7693 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
7694 | if (IS_ERR(trans)) { |
7695 | err = PTR_ERR(trans); | |
7696 | goto out_free; | |
7697 | } | |
3fd0a558 YZ |
7698 | if (block_rsv) |
7699 | trans->block_rsv = block_rsv; | |
c3e69d58 | 7700 | } |
20524f02 | 7701 | } |
b3b4aa74 | 7702 | btrfs_release_path(path); |
79787eaa JM |
7703 | if (err) |
7704 | goto out_end_trans; | |
2c47e605 YZ |
7705 | |
7706 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
79787eaa JM |
7707 | if (ret) { |
7708 | btrfs_abort_transaction(trans, tree_root, ret); | |
7709 | goto out_end_trans; | |
7710 | } | |
2c47e605 | 7711 | |
76dda93c | 7712 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
7713 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
7714 | NULL, NULL); | |
79787eaa JM |
7715 | if (ret < 0) { |
7716 | btrfs_abort_transaction(trans, tree_root, ret); | |
7717 | err = ret; | |
7718 | goto out_end_trans; | |
7719 | } else if (ret > 0) { | |
84cd948c JB |
7720 | /* if we fail to delete the orphan item this time |
7721 | * around, it'll get picked up the next time. | |
7722 | * | |
7723 | * The most common failure here is just -ENOENT. | |
7724 | */ | |
7725 | btrfs_del_orphan_item(trans, tree_root, | |
7726 | root->root_key.objectid); | |
76dda93c YZ |
7727 | } |
7728 | } | |
7729 | ||
7730 | if (root->in_radix) { | |
cb517eab | 7731 | btrfs_drop_and_free_fs_root(tree_root->fs_info, root); |
76dda93c YZ |
7732 | } else { |
7733 | free_extent_buffer(root->node); | |
7734 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 7735 | btrfs_put_fs_root(root); |
76dda93c | 7736 | } |
d29a9f62 | 7737 | root_dropped = true; |
79787eaa | 7738 | out_end_trans: |
3fd0a558 | 7739 | btrfs_end_transaction_throttle(trans, tree_root); |
79787eaa | 7740 | out_free: |
2c47e605 | 7741 | kfree(wc); |
5caf2a00 | 7742 | btrfs_free_path(path); |
cb1b69f4 | 7743 | out: |
d29a9f62 JB |
7744 | /* |
7745 | * So if we need to stop dropping the snapshot for whatever reason we | |
7746 | * need to make sure to add it back to the dead root list so that we | |
7747 | * keep trying to do the work later. This also cleans up roots if we | |
7748 | * don't have it in the radix (like when we recover after a power fail | |
7749 | * or unmount) so we don't leak memory. | |
7750 | */ | |
b37b39cd | 7751 | if (!for_reloc && root_dropped == false) |
d29a9f62 | 7752 | btrfs_add_dead_root(root); |
cb1b69f4 TI |
7753 | if (err) |
7754 | btrfs_std_error(root->fs_info, err); | |
2c536799 | 7755 | return err; |
20524f02 | 7756 | } |
9078a3e1 | 7757 | |
2c47e605 YZ |
7758 | /* |
7759 | * drop subtree rooted at tree block 'node'. | |
7760 | * | |
7761 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 7762 | * only used by relocation code |
2c47e605 | 7763 | */ |
f82d02d9 YZ |
7764 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
7765 | struct btrfs_root *root, | |
7766 | struct extent_buffer *node, | |
7767 | struct extent_buffer *parent) | |
7768 | { | |
7769 | struct btrfs_path *path; | |
2c47e605 | 7770 | struct walk_control *wc; |
f82d02d9 YZ |
7771 | int level; |
7772 | int parent_level; | |
7773 | int ret = 0; | |
7774 | int wret; | |
7775 | ||
2c47e605 YZ |
7776 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
7777 | ||
f82d02d9 | 7778 | path = btrfs_alloc_path(); |
db5b493a TI |
7779 | if (!path) |
7780 | return -ENOMEM; | |
f82d02d9 | 7781 | |
2c47e605 | 7782 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
7783 | if (!wc) { |
7784 | btrfs_free_path(path); | |
7785 | return -ENOMEM; | |
7786 | } | |
2c47e605 | 7787 | |
b9447ef8 | 7788 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
7789 | parent_level = btrfs_header_level(parent); |
7790 | extent_buffer_get(parent); | |
7791 | path->nodes[parent_level] = parent; | |
7792 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
7793 | ||
b9447ef8 | 7794 | btrfs_assert_tree_locked(node); |
f82d02d9 | 7795 | level = btrfs_header_level(node); |
f82d02d9 YZ |
7796 | path->nodes[level] = node; |
7797 | path->slots[level] = 0; | |
bd681513 | 7798 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7799 | |
7800 | wc->refs[parent_level] = 1; | |
7801 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
7802 | wc->level = level; | |
7803 | wc->shared_level = -1; | |
7804 | wc->stage = DROP_REFERENCE; | |
7805 | wc->update_ref = 0; | |
7806 | wc->keep_locks = 1; | |
66d7e7f0 | 7807 | wc->for_reloc = 1; |
1c4850e2 | 7808 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
7809 | |
7810 | while (1) { | |
2c47e605 YZ |
7811 | wret = walk_down_tree(trans, root, path, wc); |
7812 | if (wret < 0) { | |
f82d02d9 | 7813 | ret = wret; |
f82d02d9 | 7814 | break; |
2c47e605 | 7815 | } |
f82d02d9 | 7816 | |
2c47e605 | 7817 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
7818 | if (wret < 0) |
7819 | ret = wret; | |
7820 | if (wret != 0) | |
7821 | break; | |
7822 | } | |
7823 | ||
2c47e605 | 7824 | kfree(wc); |
f82d02d9 YZ |
7825 | btrfs_free_path(path); |
7826 | return ret; | |
7827 | } | |
7828 | ||
ec44a35c CM |
7829 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
7830 | { | |
7831 | u64 num_devices; | |
fc67c450 | 7832 | u64 stripped; |
e4d8ec0f | 7833 | |
fc67c450 ID |
7834 | /* |
7835 | * if restripe for this chunk_type is on pick target profile and | |
7836 | * return, otherwise do the usual balance | |
7837 | */ | |
7838 | stripped = get_restripe_target(root->fs_info, flags); | |
7839 | if (stripped) | |
7840 | return extended_to_chunk(stripped); | |
e4d8ec0f | 7841 | |
cd02dca5 CM |
7842 | /* |
7843 | * we add in the count of missing devices because we want | |
7844 | * to make sure that any RAID levels on a degraded FS | |
7845 | * continue to be honored. | |
7846 | */ | |
7847 | num_devices = root->fs_info->fs_devices->rw_devices + | |
7848 | root->fs_info->fs_devices->missing_devices; | |
7849 | ||
fc67c450 | 7850 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 7851 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
7852 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
7853 | ||
ec44a35c CM |
7854 | if (num_devices == 1) { |
7855 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7856 | stripped = flags & ~stripped; | |
7857 | ||
7858 | /* turn raid0 into single device chunks */ | |
7859 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
7860 | return stripped; | |
7861 | ||
7862 | /* turn mirroring into duplication */ | |
7863 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7864 | BTRFS_BLOCK_GROUP_RAID10)) | |
7865 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
7866 | } else { |
7867 | /* they already had raid on here, just return */ | |
ec44a35c CM |
7868 | if (flags & stripped) |
7869 | return flags; | |
7870 | ||
7871 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7872 | stripped = flags & ~stripped; | |
7873 | ||
7874 | /* switch duplicated blocks with raid1 */ | |
7875 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
7876 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
7877 | ||
e3176ca2 | 7878 | /* this is drive concat, leave it alone */ |
ec44a35c | 7879 | } |
e3176ca2 | 7880 | |
ec44a35c CM |
7881 | return flags; |
7882 | } | |
7883 | ||
199c36ea | 7884 | static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 7885 | { |
f0486c68 YZ |
7886 | struct btrfs_space_info *sinfo = cache->space_info; |
7887 | u64 num_bytes; | |
199c36ea | 7888 | u64 min_allocable_bytes; |
f0486c68 | 7889 | int ret = -ENOSPC; |
0ef3e66b | 7890 | |
c286ac48 | 7891 | |
199c36ea MX |
7892 | /* |
7893 | * We need some metadata space and system metadata space for | |
7894 | * allocating chunks in some corner cases until we force to set | |
7895 | * it to be readonly. | |
7896 | */ | |
7897 | if ((sinfo->flags & | |
7898 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
7899 | !force) | |
7900 | min_allocable_bytes = 1 * 1024 * 1024; | |
7901 | else | |
7902 | min_allocable_bytes = 0; | |
7903 | ||
f0486c68 YZ |
7904 | spin_lock(&sinfo->lock); |
7905 | spin_lock(&cache->lock); | |
61cfea9b W |
7906 | |
7907 | if (cache->ro) { | |
7908 | ret = 0; | |
7909 | goto out; | |
7910 | } | |
7911 | ||
f0486c68 YZ |
7912 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
7913 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7914 | ||
7915 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
37be25bc JB |
7916 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + |
7917 | min_allocable_bytes <= sinfo->total_bytes) { | |
f0486c68 | 7918 | sinfo->bytes_readonly += num_bytes; |
f0486c68 YZ |
7919 | cache->ro = 1; |
7920 | ret = 0; | |
7921 | } | |
61cfea9b | 7922 | out: |
f0486c68 YZ |
7923 | spin_unlock(&cache->lock); |
7924 | spin_unlock(&sinfo->lock); | |
7925 | return ret; | |
7926 | } | |
7d9eb12c | 7927 | |
f0486c68 YZ |
7928 | int btrfs_set_block_group_ro(struct btrfs_root *root, |
7929 | struct btrfs_block_group_cache *cache) | |
c286ac48 | 7930 | |
f0486c68 YZ |
7931 | { |
7932 | struct btrfs_trans_handle *trans; | |
7933 | u64 alloc_flags; | |
7934 | int ret; | |
7d9eb12c | 7935 | |
f0486c68 | 7936 | BUG_ON(cache->ro); |
0ef3e66b | 7937 | |
ff5714cc | 7938 | trans = btrfs_join_transaction(root); |
79787eaa JM |
7939 | if (IS_ERR(trans)) |
7940 | return PTR_ERR(trans); | |
5d4f98a2 | 7941 | |
f0486c68 | 7942 | alloc_flags = update_block_group_flags(root, cache->flags); |
79787eaa | 7943 | if (alloc_flags != cache->flags) { |
698d0082 | 7944 | ret = do_chunk_alloc(trans, root, alloc_flags, |
79787eaa JM |
7945 | CHUNK_ALLOC_FORCE); |
7946 | if (ret < 0) | |
7947 | goto out; | |
7948 | } | |
5d4f98a2 | 7949 | |
199c36ea | 7950 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7951 | if (!ret) |
7952 | goto out; | |
7953 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
698d0082 | 7954 | ret = do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 7955 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
7956 | if (ret < 0) |
7957 | goto out; | |
199c36ea | 7958 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7959 | out: |
7960 | btrfs_end_transaction(trans, root); | |
7961 | return ret; | |
7962 | } | |
5d4f98a2 | 7963 | |
c87f08ca CM |
7964 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
7965 | struct btrfs_root *root, u64 type) | |
7966 | { | |
7967 | u64 alloc_flags = get_alloc_profile(root, type); | |
698d0082 | 7968 | return do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 7969 | CHUNK_ALLOC_FORCE); |
c87f08ca CM |
7970 | } |
7971 | ||
6d07bcec MX |
7972 | /* |
7973 | * helper to account the unused space of all the readonly block group in the | |
7974 | * list. takes mirrors into account. | |
7975 | */ | |
7976 | static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list) | |
7977 | { | |
7978 | struct btrfs_block_group_cache *block_group; | |
7979 | u64 free_bytes = 0; | |
7980 | int factor; | |
7981 | ||
7982 | list_for_each_entry(block_group, groups_list, list) { | |
7983 | spin_lock(&block_group->lock); | |
7984 | ||
7985 | if (!block_group->ro) { | |
7986 | spin_unlock(&block_group->lock); | |
7987 | continue; | |
7988 | } | |
7989 | ||
7990 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7991 | BTRFS_BLOCK_GROUP_RAID10 | | |
7992 | BTRFS_BLOCK_GROUP_DUP)) | |
7993 | factor = 2; | |
7994 | else | |
7995 | factor = 1; | |
7996 | ||
7997 | free_bytes += (block_group->key.offset - | |
7998 | btrfs_block_group_used(&block_group->item)) * | |
7999 | factor; | |
8000 | ||
8001 | spin_unlock(&block_group->lock); | |
8002 | } | |
8003 | ||
8004 | return free_bytes; | |
8005 | } | |
8006 | ||
8007 | /* | |
8008 | * helper to account the unused space of all the readonly block group in the | |
8009 | * space_info. takes mirrors into account. | |
8010 | */ | |
8011 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) | |
8012 | { | |
8013 | int i; | |
8014 | u64 free_bytes = 0; | |
8015 | ||
8016 | spin_lock(&sinfo->lock); | |
8017 | ||
8018 | for(i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
8019 | if (!list_empty(&sinfo->block_groups[i])) | |
8020 | free_bytes += __btrfs_get_ro_block_group_free_space( | |
8021 | &sinfo->block_groups[i]); | |
8022 | ||
8023 | spin_unlock(&sinfo->lock); | |
8024 | ||
8025 | return free_bytes; | |
8026 | } | |
8027 | ||
143bede5 | 8028 | void btrfs_set_block_group_rw(struct btrfs_root *root, |
f0486c68 | 8029 | struct btrfs_block_group_cache *cache) |
5d4f98a2 | 8030 | { |
f0486c68 YZ |
8031 | struct btrfs_space_info *sinfo = cache->space_info; |
8032 | u64 num_bytes; | |
8033 | ||
8034 | BUG_ON(!cache->ro); | |
8035 | ||
8036 | spin_lock(&sinfo->lock); | |
8037 | spin_lock(&cache->lock); | |
8038 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
8039 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
8040 | sinfo->bytes_readonly -= num_bytes; | |
8041 | cache->ro = 0; | |
8042 | spin_unlock(&cache->lock); | |
8043 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
8044 | } |
8045 | ||
ba1bf481 JB |
8046 | /* |
8047 | * checks to see if its even possible to relocate this block group. | |
8048 | * | |
8049 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
8050 | * ok to go ahead and try. | |
8051 | */ | |
8052 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 8053 | { |
ba1bf481 JB |
8054 | struct btrfs_block_group_cache *block_group; |
8055 | struct btrfs_space_info *space_info; | |
8056 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
8057 | struct btrfs_device *device; | |
6df9a95e | 8058 | struct btrfs_trans_handle *trans; |
cdcb725c | 8059 | u64 min_free; |
6719db6a JB |
8060 | u64 dev_min = 1; |
8061 | u64 dev_nr = 0; | |
4a5e98f5 | 8062 | u64 target; |
cdcb725c | 8063 | int index; |
ba1bf481 JB |
8064 | int full = 0; |
8065 | int ret = 0; | |
1a40e23b | 8066 | |
ba1bf481 | 8067 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 8068 | |
ba1bf481 JB |
8069 | /* odd, couldn't find the block group, leave it alone */ |
8070 | if (!block_group) | |
8071 | return -1; | |
1a40e23b | 8072 | |
cdcb725c | 8073 | min_free = btrfs_block_group_used(&block_group->item); |
8074 | ||
ba1bf481 | 8075 | /* no bytes used, we're good */ |
cdcb725c | 8076 | if (!min_free) |
1a40e23b ZY |
8077 | goto out; |
8078 | ||
ba1bf481 JB |
8079 | space_info = block_group->space_info; |
8080 | spin_lock(&space_info->lock); | |
17d217fe | 8081 | |
ba1bf481 | 8082 | full = space_info->full; |
17d217fe | 8083 | |
ba1bf481 JB |
8084 | /* |
8085 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
8086 | * relocate it unless we're able to allocate a new chunk below. |
8087 | * | |
8088 | * Otherwise, we need to make sure we have room in the space to handle | |
8089 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 8090 | */ |
7ce618db | 8091 | if ((space_info->total_bytes != block_group->key.offset) && |
cdcb725c | 8092 | (space_info->bytes_used + space_info->bytes_reserved + |
8093 | space_info->bytes_pinned + space_info->bytes_readonly + | |
8094 | min_free < space_info->total_bytes)) { | |
ba1bf481 JB |
8095 | spin_unlock(&space_info->lock); |
8096 | goto out; | |
17d217fe | 8097 | } |
ba1bf481 | 8098 | spin_unlock(&space_info->lock); |
ea8c2819 | 8099 | |
ba1bf481 JB |
8100 | /* |
8101 | * ok we don't have enough space, but maybe we have free space on our | |
8102 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
8103 | * alloc devices and guess if we have enough space. if this block |
8104 | * group is going to be restriped, run checks against the target | |
8105 | * profile instead of the current one. | |
ba1bf481 JB |
8106 | */ |
8107 | ret = -1; | |
ea8c2819 | 8108 | |
cdcb725c | 8109 | /* |
8110 | * index: | |
8111 | * 0: raid10 | |
8112 | * 1: raid1 | |
8113 | * 2: dup | |
8114 | * 3: raid0 | |
8115 | * 4: single | |
8116 | */ | |
4a5e98f5 ID |
8117 | target = get_restripe_target(root->fs_info, block_group->flags); |
8118 | if (target) { | |
31e50229 | 8119 | index = __get_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
8120 | } else { |
8121 | /* | |
8122 | * this is just a balance, so if we were marked as full | |
8123 | * we know there is no space for a new chunk | |
8124 | */ | |
8125 | if (full) | |
8126 | goto out; | |
8127 | ||
8128 | index = get_block_group_index(block_group); | |
8129 | } | |
8130 | ||
e6ec716f | 8131 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 8132 | dev_min = 4; |
6719db6a JB |
8133 | /* Divide by 2 */ |
8134 | min_free >>= 1; | |
e6ec716f | 8135 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 8136 | dev_min = 2; |
e6ec716f | 8137 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
8138 | /* Multiply by 2 */ |
8139 | min_free <<= 1; | |
e6ec716f | 8140 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 8141 | dev_min = fs_devices->rw_devices; |
6719db6a | 8142 | do_div(min_free, dev_min); |
cdcb725c | 8143 | } |
8144 | ||
6df9a95e JB |
8145 | /* We need to do this so that we can look at pending chunks */ |
8146 | trans = btrfs_join_transaction(root); | |
8147 | if (IS_ERR(trans)) { | |
8148 | ret = PTR_ERR(trans); | |
8149 | goto out; | |
8150 | } | |
8151 | ||
ba1bf481 JB |
8152 | mutex_lock(&root->fs_info->chunk_mutex); |
8153 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7bfc837d | 8154 | u64 dev_offset; |
56bec294 | 8155 | |
ba1bf481 JB |
8156 | /* |
8157 | * check to make sure we can actually find a chunk with enough | |
8158 | * space to fit our block group in. | |
8159 | */ | |
63a212ab SB |
8160 | if (device->total_bytes > device->bytes_used + min_free && |
8161 | !device->is_tgtdev_for_dev_replace) { | |
6df9a95e | 8162 | ret = find_free_dev_extent(trans, device, min_free, |
7bfc837d | 8163 | &dev_offset, NULL); |
ba1bf481 | 8164 | if (!ret) |
cdcb725c | 8165 | dev_nr++; |
8166 | ||
8167 | if (dev_nr >= dev_min) | |
73e48b27 | 8168 | break; |
cdcb725c | 8169 | |
ba1bf481 | 8170 | ret = -1; |
725c8463 | 8171 | } |
edbd8d4e | 8172 | } |
ba1bf481 | 8173 | mutex_unlock(&root->fs_info->chunk_mutex); |
6df9a95e | 8174 | btrfs_end_transaction(trans, root); |
edbd8d4e | 8175 | out: |
ba1bf481 | 8176 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
8177 | return ret; |
8178 | } | |
8179 | ||
b2950863 CH |
8180 | static int find_first_block_group(struct btrfs_root *root, |
8181 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 8182 | { |
925baedd | 8183 | int ret = 0; |
0b86a832 CM |
8184 | struct btrfs_key found_key; |
8185 | struct extent_buffer *leaf; | |
8186 | int slot; | |
edbd8d4e | 8187 | |
0b86a832 CM |
8188 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
8189 | if (ret < 0) | |
925baedd CM |
8190 | goto out; |
8191 | ||
d397712b | 8192 | while (1) { |
0b86a832 | 8193 | slot = path->slots[0]; |
edbd8d4e | 8194 | leaf = path->nodes[0]; |
0b86a832 CM |
8195 | if (slot >= btrfs_header_nritems(leaf)) { |
8196 | ret = btrfs_next_leaf(root, path); | |
8197 | if (ret == 0) | |
8198 | continue; | |
8199 | if (ret < 0) | |
925baedd | 8200 | goto out; |
0b86a832 | 8201 | break; |
edbd8d4e | 8202 | } |
0b86a832 | 8203 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 8204 | |
0b86a832 | 8205 | if (found_key.objectid >= key->objectid && |
925baedd CM |
8206 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
8207 | ret = 0; | |
8208 | goto out; | |
8209 | } | |
0b86a832 | 8210 | path->slots[0]++; |
edbd8d4e | 8211 | } |
925baedd | 8212 | out: |
0b86a832 | 8213 | return ret; |
edbd8d4e CM |
8214 | } |
8215 | ||
0af3d00b JB |
8216 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
8217 | { | |
8218 | struct btrfs_block_group_cache *block_group; | |
8219 | u64 last = 0; | |
8220 | ||
8221 | while (1) { | |
8222 | struct inode *inode; | |
8223 | ||
8224 | block_group = btrfs_lookup_first_block_group(info, last); | |
8225 | while (block_group) { | |
8226 | spin_lock(&block_group->lock); | |
8227 | if (block_group->iref) | |
8228 | break; | |
8229 | spin_unlock(&block_group->lock); | |
8230 | block_group = next_block_group(info->tree_root, | |
8231 | block_group); | |
8232 | } | |
8233 | if (!block_group) { | |
8234 | if (last == 0) | |
8235 | break; | |
8236 | last = 0; | |
8237 | continue; | |
8238 | } | |
8239 | ||
8240 | inode = block_group->inode; | |
8241 | block_group->iref = 0; | |
8242 | block_group->inode = NULL; | |
8243 | spin_unlock(&block_group->lock); | |
8244 | iput(inode); | |
8245 | last = block_group->key.objectid + block_group->key.offset; | |
8246 | btrfs_put_block_group(block_group); | |
8247 | } | |
8248 | } | |
8249 | ||
1a40e23b ZY |
8250 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
8251 | { | |
8252 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 8253 | struct btrfs_space_info *space_info; |
11833d66 | 8254 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
8255 | struct rb_node *n; |
8256 | ||
11833d66 YZ |
8257 | down_write(&info->extent_commit_sem); |
8258 | while (!list_empty(&info->caching_block_groups)) { | |
8259 | caching_ctl = list_entry(info->caching_block_groups.next, | |
8260 | struct btrfs_caching_control, list); | |
8261 | list_del(&caching_ctl->list); | |
8262 | put_caching_control(caching_ctl); | |
8263 | } | |
8264 | up_write(&info->extent_commit_sem); | |
8265 | ||
1a40e23b ZY |
8266 | spin_lock(&info->block_group_cache_lock); |
8267 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
8268 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
8269 | cache_node); | |
1a40e23b ZY |
8270 | rb_erase(&block_group->cache_node, |
8271 | &info->block_group_cache_tree); | |
d899e052 YZ |
8272 | spin_unlock(&info->block_group_cache_lock); |
8273 | ||
80eb234a | 8274 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 8275 | list_del(&block_group->list); |
80eb234a | 8276 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 8277 | |
817d52f8 | 8278 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 8279 | wait_block_group_cache_done(block_group); |
817d52f8 | 8280 | |
3c14874a JB |
8281 | /* |
8282 | * We haven't cached this block group, which means we could | |
8283 | * possibly have excluded extents on this block group. | |
8284 | */ | |
36cce922 JB |
8285 | if (block_group->cached == BTRFS_CACHE_NO || |
8286 | block_group->cached == BTRFS_CACHE_ERROR) | |
3c14874a JB |
8287 | free_excluded_extents(info->extent_root, block_group); |
8288 | ||
817d52f8 | 8289 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 8290 | btrfs_put_block_group(block_group); |
d899e052 YZ |
8291 | |
8292 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
8293 | } |
8294 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
8295 | |
8296 | /* now that all the block groups are freed, go through and | |
8297 | * free all the space_info structs. This is only called during | |
8298 | * the final stages of unmount, and so we know nobody is | |
8299 | * using them. We call synchronize_rcu() once before we start, | |
8300 | * just to be on the safe side. | |
8301 | */ | |
8302 | synchronize_rcu(); | |
8303 | ||
8929ecfa YZ |
8304 | release_global_block_rsv(info); |
8305 | ||
4184ea7f CM |
8306 | while(!list_empty(&info->space_info)) { |
8307 | space_info = list_entry(info->space_info.next, | |
8308 | struct btrfs_space_info, | |
8309 | list); | |
b069e0c3 DS |
8310 | if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) { |
8311 | if (space_info->bytes_pinned > 0 || | |
8312 | space_info->bytes_reserved > 0 || | |
8313 | space_info->bytes_may_use > 0) { | |
8314 | WARN_ON(1); | |
8315 | dump_space_info(space_info, 0, 0); | |
8316 | } | |
f0486c68 | 8317 | } |
b150a4f1 | 8318 | percpu_counter_destroy(&space_info->total_bytes_pinned); |
4184ea7f CM |
8319 | list_del(&space_info->list); |
8320 | kfree(space_info); | |
8321 | } | |
1a40e23b ZY |
8322 | return 0; |
8323 | } | |
8324 | ||
b742bb82 YZ |
8325 | static void __link_block_group(struct btrfs_space_info *space_info, |
8326 | struct btrfs_block_group_cache *cache) | |
8327 | { | |
8328 | int index = get_block_group_index(cache); | |
8329 | ||
8330 | down_write(&space_info->groups_sem); | |
8331 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
8332 | up_write(&space_info->groups_sem); | |
8333 | } | |
8334 | ||
9078a3e1 CM |
8335 | int btrfs_read_block_groups(struct btrfs_root *root) |
8336 | { | |
8337 | struct btrfs_path *path; | |
8338 | int ret; | |
9078a3e1 | 8339 | struct btrfs_block_group_cache *cache; |
be744175 | 8340 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 8341 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
8342 | struct btrfs_key key; |
8343 | struct btrfs_key found_key; | |
5f39d397 | 8344 | struct extent_buffer *leaf; |
0af3d00b JB |
8345 | int need_clear = 0; |
8346 | u64 cache_gen; | |
96b5179d | 8347 | |
be744175 | 8348 | root = info->extent_root; |
9078a3e1 | 8349 | key.objectid = 0; |
0b86a832 | 8350 | key.offset = 0; |
9078a3e1 | 8351 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
8352 | path = btrfs_alloc_path(); |
8353 | if (!path) | |
8354 | return -ENOMEM; | |
026fd317 | 8355 | path->reada = 1; |
9078a3e1 | 8356 | |
6c41761f | 8357 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); |
73bc1876 | 8358 | if (btrfs_test_opt(root, SPACE_CACHE) && |
6c41761f | 8359 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) |
0af3d00b | 8360 | need_clear = 1; |
88c2ba3b JB |
8361 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
8362 | need_clear = 1; | |
0af3d00b | 8363 | |
d397712b | 8364 | while (1) { |
0b86a832 | 8365 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
8366 | if (ret > 0) |
8367 | break; | |
0b86a832 CM |
8368 | if (ret != 0) |
8369 | goto error; | |
5f39d397 CM |
8370 | leaf = path->nodes[0]; |
8371 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 8372 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 8373 | if (!cache) { |
0b86a832 | 8374 | ret = -ENOMEM; |
f0486c68 | 8375 | goto error; |
9078a3e1 | 8376 | } |
34d52cb6 LZ |
8377 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
8378 | GFP_NOFS); | |
8379 | if (!cache->free_space_ctl) { | |
8380 | kfree(cache); | |
8381 | ret = -ENOMEM; | |
8382 | goto error; | |
8383 | } | |
3e1ad54f | 8384 | |
d2fb3437 | 8385 | atomic_set(&cache->count, 1); |
c286ac48 | 8386 | spin_lock_init(&cache->lock); |
817d52f8 | 8387 | cache->fs_info = info; |
0f9dd46c | 8388 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 8389 | INIT_LIST_HEAD(&cache->cluster_list); |
96303081 | 8390 | |
cf7c1ef6 LB |
8391 | if (need_clear) { |
8392 | /* | |
8393 | * When we mount with old space cache, we need to | |
8394 | * set BTRFS_DC_CLEAR and set dirty flag. | |
8395 | * | |
8396 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
8397 | * truncate the old free space cache inode and | |
8398 | * setup a new one. | |
8399 | * b) Setting 'dirty flag' makes sure that we flush | |
8400 | * the new space cache info onto disk. | |
8401 | */ | |
0af3d00b | 8402 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 LB |
8403 | if (btrfs_test_opt(root, SPACE_CACHE)) |
8404 | cache->dirty = 1; | |
8405 | } | |
0af3d00b | 8406 | |
5f39d397 CM |
8407 | read_extent_buffer(leaf, &cache->item, |
8408 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
8409 | sizeof(cache->item)); | |
9078a3e1 | 8410 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 8411 | |
9078a3e1 | 8412 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 8413 | btrfs_release_path(path); |
0b86a832 | 8414 | cache->flags = btrfs_block_group_flags(&cache->item); |
817d52f8 | 8415 | cache->sectorsize = root->sectorsize; |
53b381b3 DW |
8416 | cache->full_stripe_len = btrfs_full_stripe_len(root, |
8417 | &root->fs_info->mapping_tree, | |
8418 | found_key.objectid); | |
34d52cb6 LZ |
8419 | btrfs_init_free_space_ctl(cache); |
8420 | ||
3c14874a JB |
8421 | /* |
8422 | * We need to exclude the super stripes now so that the space | |
8423 | * info has super bytes accounted for, otherwise we'll think | |
8424 | * we have more space than we actually do. | |
8425 | */ | |
835d974f JB |
8426 | ret = exclude_super_stripes(root, cache); |
8427 | if (ret) { | |
8428 | /* | |
8429 | * We may have excluded something, so call this just in | |
8430 | * case. | |
8431 | */ | |
8432 | free_excluded_extents(root, cache); | |
8433 | kfree(cache->free_space_ctl); | |
8434 | kfree(cache); | |
8435 | goto error; | |
8436 | } | |
3c14874a | 8437 | |
817d52f8 JB |
8438 | /* |
8439 | * check for two cases, either we are full, and therefore | |
8440 | * don't need to bother with the caching work since we won't | |
8441 | * find any space, or we are empty, and we can just add all | |
8442 | * the space in and be done with it. This saves us _alot_ of | |
8443 | * time, particularly in the full case. | |
8444 | */ | |
8445 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 8446 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 8447 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 8448 | free_excluded_extents(root, cache); |
817d52f8 | 8449 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 8450 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
8451 | cache->cached = BTRFS_CACHE_FINISHED; |
8452 | add_new_free_space(cache, root->fs_info, | |
8453 | found_key.objectid, | |
8454 | found_key.objectid + | |
8455 | found_key.offset); | |
11833d66 | 8456 | free_excluded_extents(root, cache); |
817d52f8 | 8457 | } |
96b5179d | 8458 | |
8c579fe7 JB |
8459 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
8460 | if (ret) { | |
8461 | btrfs_remove_free_space_cache(cache); | |
8462 | btrfs_put_block_group(cache); | |
8463 | goto error; | |
8464 | } | |
8465 | ||
6324fbf3 CM |
8466 | ret = update_space_info(info, cache->flags, found_key.offset, |
8467 | btrfs_block_group_used(&cache->item), | |
8468 | &space_info); | |
8c579fe7 JB |
8469 | if (ret) { |
8470 | btrfs_remove_free_space_cache(cache); | |
8471 | spin_lock(&info->block_group_cache_lock); | |
8472 | rb_erase(&cache->cache_node, | |
8473 | &info->block_group_cache_tree); | |
8474 | spin_unlock(&info->block_group_cache_lock); | |
8475 | btrfs_put_block_group(cache); | |
8476 | goto error; | |
8477 | } | |
8478 | ||
6324fbf3 | 8479 | cache->space_info = space_info; |
1b2da372 | 8480 | spin_lock(&cache->space_info->lock); |
f0486c68 | 8481 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
8482 | spin_unlock(&cache->space_info->lock); |
8483 | ||
b742bb82 | 8484 | __link_block_group(space_info, cache); |
0f9dd46c | 8485 | |
75ccf47d | 8486 | set_avail_alloc_bits(root->fs_info, cache->flags); |
2b82032c | 8487 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
199c36ea | 8488 | set_block_group_ro(cache, 1); |
9078a3e1 | 8489 | } |
b742bb82 YZ |
8490 | |
8491 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
8492 | if (!(get_alloc_profile(root, space_info->flags) & | |
8493 | (BTRFS_BLOCK_GROUP_RAID10 | | |
8494 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
8495 | BTRFS_BLOCK_GROUP_RAID5 | |
8496 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
8497 | BTRFS_BLOCK_GROUP_DUP))) |
8498 | continue; | |
8499 | /* | |
8500 | * avoid allocating from un-mirrored block group if there are | |
8501 | * mirrored block groups. | |
8502 | */ | |
1095cc0d | 8503 | list_for_each_entry(cache, |
8504 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
8505 | list) | |
199c36ea | 8506 | set_block_group_ro(cache, 1); |
1095cc0d | 8507 | list_for_each_entry(cache, |
8508 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
8509 | list) | |
199c36ea | 8510 | set_block_group_ro(cache, 1); |
9078a3e1 | 8511 | } |
f0486c68 YZ |
8512 | |
8513 | init_global_block_rsv(info); | |
0b86a832 CM |
8514 | ret = 0; |
8515 | error: | |
9078a3e1 | 8516 | btrfs_free_path(path); |
0b86a832 | 8517 | return ret; |
9078a3e1 | 8518 | } |
6324fbf3 | 8519 | |
ea658bad JB |
8520 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans, |
8521 | struct btrfs_root *root) | |
8522 | { | |
8523 | struct btrfs_block_group_cache *block_group, *tmp; | |
8524 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
8525 | struct btrfs_block_group_item item; | |
8526 | struct btrfs_key key; | |
8527 | int ret = 0; | |
8528 | ||
8529 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, | |
8530 | new_bg_list) { | |
8531 | list_del_init(&block_group->new_bg_list); | |
8532 | ||
8533 | if (ret) | |
8534 | continue; | |
8535 | ||
8536 | spin_lock(&block_group->lock); | |
8537 | memcpy(&item, &block_group->item, sizeof(item)); | |
8538 | memcpy(&key, &block_group->key, sizeof(key)); | |
8539 | spin_unlock(&block_group->lock); | |
8540 | ||
8541 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
8542 | sizeof(item)); | |
8543 | if (ret) | |
8544 | btrfs_abort_transaction(trans, extent_root, ret); | |
6df9a95e JB |
8545 | ret = btrfs_finish_chunk_alloc(trans, extent_root, |
8546 | key.objectid, key.offset); | |
8547 | if (ret) | |
8548 | btrfs_abort_transaction(trans, extent_root, ret); | |
ea658bad JB |
8549 | } |
8550 | } | |
8551 | ||
6324fbf3 CM |
8552 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
8553 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 8554 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
8555 | u64 size) |
8556 | { | |
8557 | int ret; | |
6324fbf3 CM |
8558 | struct btrfs_root *extent_root; |
8559 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
8560 | |
8561 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 8562 | |
12fcfd22 | 8563 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 8564 | |
8f18cf13 | 8565 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
8566 | if (!cache) |
8567 | return -ENOMEM; | |
34d52cb6 LZ |
8568 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
8569 | GFP_NOFS); | |
8570 | if (!cache->free_space_ctl) { | |
8571 | kfree(cache); | |
8572 | return -ENOMEM; | |
8573 | } | |
0f9dd46c | 8574 | |
e17cade2 | 8575 | cache->key.objectid = chunk_offset; |
6324fbf3 | 8576 | cache->key.offset = size; |
d2fb3437 | 8577 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
96303081 | 8578 | cache->sectorsize = root->sectorsize; |
0af3d00b | 8579 | cache->fs_info = root->fs_info; |
53b381b3 DW |
8580 | cache->full_stripe_len = btrfs_full_stripe_len(root, |
8581 | &root->fs_info->mapping_tree, | |
8582 | chunk_offset); | |
96303081 | 8583 | |
d2fb3437 | 8584 | atomic_set(&cache->count, 1); |
c286ac48 | 8585 | spin_lock_init(&cache->lock); |
0f9dd46c | 8586 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 8587 | INIT_LIST_HEAD(&cache->cluster_list); |
ea658bad | 8588 | INIT_LIST_HEAD(&cache->new_bg_list); |
0ef3e66b | 8589 | |
34d52cb6 LZ |
8590 | btrfs_init_free_space_ctl(cache); |
8591 | ||
6324fbf3 | 8592 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
8593 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
8594 | cache->flags = type; | |
8595 | btrfs_set_block_group_flags(&cache->item, type); | |
8596 | ||
11833d66 | 8597 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 8598 | cache->cached = BTRFS_CACHE_FINISHED; |
835d974f JB |
8599 | ret = exclude_super_stripes(root, cache); |
8600 | if (ret) { | |
8601 | /* | |
8602 | * We may have excluded something, so call this just in | |
8603 | * case. | |
8604 | */ | |
8605 | free_excluded_extents(root, cache); | |
8606 | kfree(cache->free_space_ctl); | |
8607 | kfree(cache); | |
8608 | return ret; | |
8609 | } | |
96303081 | 8610 | |
817d52f8 JB |
8611 | add_new_free_space(cache, root->fs_info, chunk_offset, |
8612 | chunk_offset + size); | |
8613 | ||
11833d66 YZ |
8614 | free_excluded_extents(root, cache); |
8615 | ||
8c579fe7 JB |
8616 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
8617 | if (ret) { | |
8618 | btrfs_remove_free_space_cache(cache); | |
8619 | btrfs_put_block_group(cache); | |
8620 | return ret; | |
8621 | } | |
8622 | ||
6324fbf3 CM |
8623 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
8624 | &cache->space_info); | |
8c579fe7 JB |
8625 | if (ret) { |
8626 | btrfs_remove_free_space_cache(cache); | |
8627 | spin_lock(&root->fs_info->block_group_cache_lock); | |
8628 | rb_erase(&cache->cache_node, | |
8629 | &root->fs_info->block_group_cache_tree); | |
8630 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
8631 | btrfs_put_block_group(cache); | |
8632 | return ret; | |
8633 | } | |
c7c144db | 8634 | update_global_block_rsv(root->fs_info); |
1b2da372 JB |
8635 | |
8636 | spin_lock(&cache->space_info->lock); | |
f0486c68 | 8637 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
8638 | spin_unlock(&cache->space_info->lock); |
8639 | ||
b742bb82 | 8640 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 8641 | |
ea658bad | 8642 | list_add_tail(&cache->new_bg_list, &trans->new_bgs); |
6324fbf3 | 8643 | |
d18a2c44 | 8644 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 8645 | |
6324fbf3 CM |
8646 | return 0; |
8647 | } | |
1a40e23b | 8648 | |
10ea00f5 ID |
8649 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
8650 | { | |
899c81ea ID |
8651 | u64 extra_flags = chunk_to_extended(flags) & |
8652 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 8653 | |
de98ced9 | 8654 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
8655 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
8656 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
8657 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
8658 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
8659 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
8660 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 8661 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
8662 | } |
8663 | ||
1a40e23b ZY |
8664 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
8665 | struct btrfs_root *root, u64 group_start) | |
8666 | { | |
8667 | struct btrfs_path *path; | |
8668 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 8669 | struct btrfs_free_cluster *cluster; |
0af3d00b | 8670 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 8671 | struct btrfs_key key; |
0af3d00b | 8672 | struct inode *inode; |
1a40e23b | 8673 | int ret; |
10ea00f5 | 8674 | int index; |
89a55897 | 8675 | int factor; |
1a40e23b | 8676 | |
1a40e23b ZY |
8677 | root = root->fs_info->extent_root; |
8678 | ||
8679 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
8680 | BUG_ON(!block_group); | |
c146afad | 8681 | BUG_ON(!block_group->ro); |
1a40e23b | 8682 | |
9f7c43c9 | 8683 | /* |
8684 | * Free the reserved super bytes from this block group before | |
8685 | * remove it. | |
8686 | */ | |
8687 | free_excluded_extents(root, block_group); | |
8688 | ||
1a40e23b | 8689 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 8690 | index = get_block_group_index(block_group); |
89a55897 JB |
8691 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
8692 | BTRFS_BLOCK_GROUP_RAID1 | | |
8693 | BTRFS_BLOCK_GROUP_RAID10)) | |
8694 | factor = 2; | |
8695 | else | |
8696 | factor = 1; | |
1a40e23b | 8697 | |
44fb5511 CM |
8698 | /* make sure this block group isn't part of an allocation cluster */ |
8699 | cluster = &root->fs_info->data_alloc_cluster; | |
8700 | spin_lock(&cluster->refill_lock); | |
8701 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
8702 | spin_unlock(&cluster->refill_lock); | |
8703 | ||
8704 | /* | |
8705 | * make sure this block group isn't part of a metadata | |
8706 | * allocation cluster | |
8707 | */ | |
8708 | cluster = &root->fs_info->meta_alloc_cluster; | |
8709 | spin_lock(&cluster->refill_lock); | |
8710 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
8711 | spin_unlock(&cluster->refill_lock); | |
8712 | ||
1a40e23b | 8713 | path = btrfs_alloc_path(); |
d8926bb3 MF |
8714 | if (!path) { |
8715 | ret = -ENOMEM; | |
8716 | goto out; | |
8717 | } | |
1a40e23b | 8718 | |
10b2f34d | 8719 | inode = lookup_free_space_inode(tree_root, block_group, path); |
0af3d00b | 8720 | if (!IS_ERR(inode)) { |
b532402e | 8721 | ret = btrfs_orphan_add(trans, inode); |
79787eaa JM |
8722 | if (ret) { |
8723 | btrfs_add_delayed_iput(inode); | |
8724 | goto out; | |
8725 | } | |
0af3d00b JB |
8726 | clear_nlink(inode); |
8727 | /* One for the block groups ref */ | |
8728 | spin_lock(&block_group->lock); | |
8729 | if (block_group->iref) { | |
8730 | block_group->iref = 0; | |
8731 | block_group->inode = NULL; | |
8732 | spin_unlock(&block_group->lock); | |
8733 | iput(inode); | |
8734 | } else { | |
8735 | spin_unlock(&block_group->lock); | |
8736 | } | |
8737 | /* One for our lookup ref */ | |
455757c3 | 8738 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
8739 | } |
8740 | ||
8741 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
8742 | key.offset = block_group->key.objectid; | |
8743 | key.type = 0; | |
8744 | ||
8745 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
8746 | if (ret < 0) | |
8747 | goto out; | |
8748 | if (ret > 0) | |
b3b4aa74 | 8749 | btrfs_release_path(path); |
0af3d00b JB |
8750 | if (ret == 0) { |
8751 | ret = btrfs_del_item(trans, tree_root, path); | |
8752 | if (ret) | |
8753 | goto out; | |
b3b4aa74 | 8754 | btrfs_release_path(path); |
0af3d00b JB |
8755 | } |
8756 | ||
3dfdb934 | 8757 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
8758 | rb_erase(&block_group->cache_node, |
8759 | &root->fs_info->block_group_cache_tree); | |
a1897fdd LB |
8760 | |
8761 | if (root->fs_info->first_logical_byte == block_group->key.objectid) | |
8762 | root->fs_info->first_logical_byte = (u64)-1; | |
3dfdb934 | 8763 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 8764 | |
80eb234a | 8765 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
8766 | /* |
8767 | * we must use list_del_init so people can check to see if they | |
8768 | * are still on the list after taking the semaphore | |
8769 | */ | |
8770 | list_del_init(&block_group->list); | |
10ea00f5 ID |
8771 | if (list_empty(&block_group->space_info->block_groups[index])) |
8772 | clear_avail_alloc_bits(root->fs_info, block_group->flags); | |
80eb234a | 8773 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 8774 | |
817d52f8 | 8775 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 8776 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
8777 | |
8778 | btrfs_remove_free_space_cache(block_group); | |
8779 | ||
c146afad YZ |
8780 | spin_lock(&block_group->space_info->lock); |
8781 | block_group->space_info->total_bytes -= block_group->key.offset; | |
8782 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 8783 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
c146afad | 8784 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 8785 | |
0af3d00b JB |
8786 | memcpy(&key, &block_group->key, sizeof(key)); |
8787 | ||
283bb197 | 8788 | btrfs_clear_space_info_full(root->fs_info); |
c146afad | 8789 | |
fa9c0d79 CM |
8790 | btrfs_put_block_group(block_group); |
8791 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
8792 | |
8793 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
8794 | if (ret > 0) | |
8795 | ret = -EIO; | |
8796 | if (ret < 0) | |
8797 | goto out; | |
8798 | ||
8799 | ret = btrfs_del_item(trans, root, path); | |
8800 | out: | |
8801 | btrfs_free_path(path); | |
8802 | return ret; | |
8803 | } | |
acce952b | 8804 | |
c59021f8 | 8805 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
8806 | { | |
8807 | struct btrfs_space_info *space_info; | |
1aba86d6 | 8808 | struct btrfs_super_block *disk_super; |
8809 | u64 features; | |
8810 | u64 flags; | |
8811 | int mixed = 0; | |
c59021f8 | 8812 | int ret; |
8813 | ||
6c41761f | 8814 | disk_super = fs_info->super_copy; |
1aba86d6 | 8815 | if (!btrfs_super_root(disk_super)) |
8816 | return 1; | |
c59021f8 | 8817 | |
1aba86d6 | 8818 | features = btrfs_super_incompat_flags(disk_super); |
8819 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
8820 | mixed = 1; | |
c59021f8 | 8821 | |
1aba86d6 | 8822 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
8823 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
c59021f8 | 8824 | if (ret) |
1aba86d6 | 8825 | goto out; |
c59021f8 | 8826 | |
1aba86d6 | 8827 | if (mixed) { |
8828 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
8829 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8830 | } else { | |
8831 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
8832 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8833 | if (ret) | |
8834 | goto out; | |
8835 | ||
8836 | flags = BTRFS_BLOCK_GROUP_DATA; | |
8837 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8838 | } | |
8839 | out: | |
c59021f8 | 8840 | return ret; |
8841 | } | |
8842 | ||
acce952b | 8843 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
8844 | { | |
8845 | return unpin_extent_range(root, start, end); | |
8846 | } | |
8847 | ||
8848 | int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 8849 | u64 num_bytes, u64 *actual_bytes) |
acce952b | 8850 | { |
5378e607 | 8851 | return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes); |
acce952b | 8852 | } |
f7039b1d LD |
8853 | |
8854 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) | |
8855 | { | |
8856 | struct btrfs_fs_info *fs_info = root->fs_info; | |
8857 | struct btrfs_block_group_cache *cache = NULL; | |
8858 | u64 group_trimmed; | |
8859 | u64 start; | |
8860 | u64 end; | |
8861 | u64 trimmed = 0; | |
2cac13e4 | 8862 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
8863 | int ret = 0; |
8864 | ||
2cac13e4 LB |
8865 | /* |
8866 | * try to trim all FS space, our block group may start from non-zero. | |
8867 | */ | |
8868 | if (range->len == total_bytes) | |
8869 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
8870 | else | |
8871 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
8872 | |
8873 | while (cache) { | |
8874 | if (cache->key.objectid >= (range->start + range->len)) { | |
8875 | btrfs_put_block_group(cache); | |
8876 | break; | |
8877 | } | |
8878 | ||
8879 | start = max(range->start, cache->key.objectid); | |
8880 | end = min(range->start + range->len, | |
8881 | cache->key.objectid + cache->key.offset); | |
8882 | ||
8883 | if (end - start >= range->minlen) { | |
8884 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 8885 | ret = cache_block_group(cache, 0); |
1be41b78 JB |
8886 | if (ret) { |
8887 | btrfs_put_block_group(cache); | |
8888 | break; | |
8889 | } | |
8890 | ret = wait_block_group_cache_done(cache); | |
8891 | if (ret) { | |
8892 | btrfs_put_block_group(cache); | |
8893 | break; | |
8894 | } | |
f7039b1d LD |
8895 | } |
8896 | ret = btrfs_trim_block_group(cache, | |
8897 | &group_trimmed, | |
8898 | start, | |
8899 | end, | |
8900 | range->minlen); | |
8901 | ||
8902 | trimmed += group_trimmed; | |
8903 | if (ret) { | |
8904 | btrfs_put_block_group(cache); | |
8905 | break; | |
8906 | } | |
8907 | } | |
8908 | ||
8909 | cache = next_block_group(fs_info->tree_root, cache); | |
8910 | } | |
8911 | ||
8912 | range->len = trimmed; | |
8913 | return ret; | |
8914 | } |