]>
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> |
4b4e25f2 | 26 | #include "compat.h" |
74493f7a | 27 | #include "hash.h" |
fec577fb CM |
28 | #include "ctree.h" |
29 | #include "disk-io.h" | |
30 | #include "print-tree.h" | |
e089f05c | 31 | #include "transaction.h" |
0b86a832 | 32 | #include "volumes.h" |
925baedd | 33 | #include "locking.h" |
fa9c0d79 | 34 | #include "free-space-cache.h" |
fec577fb | 35 | |
0e4f8f88 CM |
36 | /* control flags for do_chunk_alloc's force field |
37 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk | |
38 | * if we really need one. | |
39 | * | |
40 | * CHUNK_ALLOC_FORCE means it must try to allocate one | |
41 | * | |
42 | * CHUNK_ALLOC_LIMITED means to only try and allocate one | |
43 | * if we have very few chunks already allocated. This is | |
44 | * used as part of the clustering code to help make sure | |
45 | * we have a good pool of storage to cluster in, without | |
46 | * filling the FS with empty chunks | |
47 | * | |
48 | */ | |
49 | enum { | |
50 | CHUNK_ALLOC_NO_FORCE = 0, | |
51 | CHUNK_ALLOC_FORCE = 1, | |
52 | CHUNK_ALLOC_LIMITED = 2, | |
53 | }; | |
54 | ||
f3465ca4 JB |
55 | static int update_block_group(struct btrfs_trans_handle *trans, |
56 | struct btrfs_root *root, | |
f0486c68 | 57 | u64 bytenr, u64 num_bytes, int alloc); |
5d4f98a2 YZ |
58 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
59 | struct btrfs_root *root, | |
60 | u64 bytenr, u64 num_bytes, u64 parent, | |
61 | u64 root_objectid, u64 owner_objectid, | |
62 | u64 owner_offset, int refs_to_drop, | |
63 | struct btrfs_delayed_extent_op *extra_op); | |
64 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
65 | struct extent_buffer *leaf, | |
66 | struct btrfs_extent_item *ei); | |
67 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
68 | struct btrfs_root *root, | |
69 | u64 parent, u64 root_objectid, | |
70 | u64 flags, u64 owner, u64 offset, | |
71 | struct btrfs_key *ins, int ref_mod); | |
72 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
73 | struct btrfs_root *root, | |
74 | u64 parent, u64 root_objectid, | |
75 | u64 flags, struct btrfs_disk_key *key, | |
76 | int level, struct btrfs_key *ins); | |
6a63209f JB |
77 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
78 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
79 | u64 flags, int force); | |
11833d66 YZ |
80 | static int find_next_key(struct btrfs_path *path, int level, |
81 | struct btrfs_key *key); | |
9ed74f2d JB |
82 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
83 | int dump_block_groups); | |
6a63209f | 84 | |
817d52f8 JB |
85 | static noinline int |
86 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
87 | { | |
88 | smp_mb(); | |
89 | return cache->cached == BTRFS_CACHE_FINISHED; | |
90 | } | |
91 | ||
0f9dd46c JB |
92 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
93 | { | |
94 | return (cache->flags & bits) == bits; | |
95 | } | |
96 | ||
62a45b60 | 97 | static void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
98 | { |
99 | atomic_inc(&cache->count); | |
100 | } | |
101 | ||
102 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
103 | { | |
f0486c68 YZ |
104 | if (atomic_dec_and_test(&cache->count)) { |
105 | WARN_ON(cache->pinned > 0); | |
106 | WARN_ON(cache->reserved > 0); | |
107 | WARN_ON(cache->reserved_pinned > 0); | |
34d52cb6 | 108 | kfree(cache->free_space_ctl); |
11dfe35a | 109 | kfree(cache); |
f0486c68 | 110 | } |
11dfe35a JB |
111 | } |
112 | ||
0f9dd46c JB |
113 | /* |
114 | * this adds the block group to the fs_info rb tree for the block group | |
115 | * cache | |
116 | */ | |
b2950863 | 117 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
118 | struct btrfs_block_group_cache *block_group) |
119 | { | |
120 | struct rb_node **p; | |
121 | struct rb_node *parent = NULL; | |
122 | struct btrfs_block_group_cache *cache; | |
123 | ||
124 | spin_lock(&info->block_group_cache_lock); | |
125 | p = &info->block_group_cache_tree.rb_node; | |
126 | ||
127 | while (*p) { | |
128 | parent = *p; | |
129 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
130 | cache_node); | |
131 | if (block_group->key.objectid < cache->key.objectid) { | |
132 | p = &(*p)->rb_left; | |
133 | } else if (block_group->key.objectid > cache->key.objectid) { | |
134 | p = &(*p)->rb_right; | |
135 | } else { | |
136 | spin_unlock(&info->block_group_cache_lock); | |
137 | return -EEXIST; | |
138 | } | |
139 | } | |
140 | ||
141 | rb_link_node(&block_group->cache_node, parent, p); | |
142 | rb_insert_color(&block_group->cache_node, | |
143 | &info->block_group_cache_tree); | |
144 | spin_unlock(&info->block_group_cache_lock); | |
145 | ||
146 | return 0; | |
147 | } | |
148 | ||
149 | /* | |
150 | * This will return the block group at or after bytenr if contains is 0, else | |
151 | * it will return the block group that contains the bytenr | |
152 | */ | |
153 | static struct btrfs_block_group_cache * | |
154 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
155 | int contains) | |
156 | { | |
157 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
158 | struct rb_node *n; | |
159 | u64 end, start; | |
160 | ||
161 | spin_lock(&info->block_group_cache_lock); | |
162 | n = info->block_group_cache_tree.rb_node; | |
163 | ||
164 | while (n) { | |
165 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
166 | cache_node); | |
167 | end = cache->key.objectid + cache->key.offset - 1; | |
168 | start = cache->key.objectid; | |
169 | ||
170 | if (bytenr < start) { | |
171 | if (!contains && (!ret || start < ret->key.objectid)) | |
172 | ret = cache; | |
173 | n = n->rb_left; | |
174 | } else if (bytenr > start) { | |
175 | if (contains && bytenr <= end) { | |
176 | ret = cache; | |
177 | break; | |
178 | } | |
179 | n = n->rb_right; | |
180 | } else { | |
181 | ret = cache; | |
182 | break; | |
183 | } | |
184 | } | |
d2fb3437 | 185 | if (ret) |
11dfe35a | 186 | btrfs_get_block_group(ret); |
0f9dd46c JB |
187 | spin_unlock(&info->block_group_cache_lock); |
188 | ||
189 | return ret; | |
190 | } | |
191 | ||
11833d66 YZ |
192 | static int add_excluded_extent(struct btrfs_root *root, |
193 | u64 start, u64 num_bytes) | |
817d52f8 | 194 | { |
11833d66 YZ |
195 | u64 end = start + num_bytes - 1; |
196 | set_extent_bits(&root->fs_info->freed_extents[0], | |
197 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
198 | set_extent_bits(&root->fs_info->freed_extents[1], | |
199 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
200 | return 0; | |
201 | } | |
817d52f8 | 202 | |
11833d66 YZ |
203 | static void free_excluded_extents(struct btrfs_root *root, |
204 | struct btrfs_block_group_cache *cache) | |
205 | { | |
206 | u64 start, end; | |
817d52f8 | 207 | |
11833d66 YZ |
208 | start = cache->key.objectid; |
209 | end = start + cache->key.offset - 1; | |
210 | ||
211 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
212 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
213 | clear_extent_bits(&root->fs_info->freed_extents[1], | |
214 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
817d52f8 JB |
215 | } |
216 | ||
11833d66 YZ |
217 | static int exclude_super_stripes(struct btrfs_root *root, |
218 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 219 | { |
817d52f8 JB |
220 | u64 bytenr; |
221 | u64 *logical; | |
222 | int stripe_len; | |
223 | int i, nr, ret; | |
224 | ||
06b2331f YZ |
225 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
226 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
227 | cache->bytes_super += stripe_len; | |
228 | ret = add_excluded_extent(root, cache->key.objectid, | |
229 | stripe_len); | |
230 | BUG_ON(ret); | |
231 | } | |
232 | ||
817d52f8 JB |
233 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
234 | bytenr = btrfs_sb_offset(i); | |
235 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
236 | cache->key.objectid, bytenr, | |
237 | 0, &logical, &nr, &stripe_len); | |
238 | BUG_ON(ret); | |
11833d66 | 239 | |
817d52f8 | 240 | while (nr--) { |
1b2da372 | 241 | cache->bytes_super += stripe_len; |
11833d66 YZ |
242 | ret = add_excluded_extent(root, logical[nr], |
243 | stripe_len); | |
244 | BUG_ON(ret); | |
817d52f8 | 245 | } |
11833d66 | 246 | |
817d52f8 JB |
247 | kfree(logical); |
248 | } | |
817d52f8 JB |
249 | return 0; |
250 | } | |
251 | ||
11833d66 YZ |
252 | static struct btrfs_caching_control * |
253 | get_caching_control(struct btrfs_block_group_cache *cache) | |
254 | { | |
255 | struct btrfs_caching_control *ctl; | |
256 | ||
257 | spin_lock(&cache->lock); | |
258 | if (cache->cached != BTRFS_CACHE_STARTED) { | |
259 | spin_unlock(&cache->lock); | |
260 | return NULL; | |
261 | } | |
262 | ||
dde5abee JB |
263 | /* We're loading it the fast way, so we don't have a caching_ctl. */ |
264 | if (!cache->caching_ctl) { | |
265 | spin_unlock(&cache->lock); | |
11833d66 YZ |
266 | return NULL; |
267 | } | |
268 | ||
269 | ctl = cache->caching_ctl; | |
270 | atomic_inc(&ctl->count); | |
271 | spin_unlock(&cache->lock); | |
272 | return ctl; | |
273 | } | |
274 | ||
275 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
276 | { | |
277 | if (atomic_dec_and_test(&ctl->count)) | |
278 | kfree(ctl); | |
279 | } | |
280 | ||
0f9dd46c JB |
281 | /* |
282 | * this is only called by cache_block_group, since we could have freed extents | |
283 | * we need to check the pinned_extents for any extents that can't be used yet | |
284 | * since their free space will be released as soon as the transaction commits. | |
285 | */ | |
817d52f8 | 286 | static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
0f9dd46c JB |
287 | struct btrfs_fs_info *info, u64 start, u64 end) |
288 | { | |
817d52f8 | 289 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
290 | int ret; |
291 | ||
292 | while (start < end) { | |
11833d66 | 293 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 294 | &extent_start, &extent_end, |
11833d66 | 295 | EXTENT_DIRTY | EXTENT_UPTODATE); |
0f9dd46c JB |
296 | if (ret) |
297 | break; | |
298 | ||
06b2331f | 299 | if (extent_start <= start) { |
0f9dd46c JB |
300 | start = extent_end + 1; |
301 | } else if (extent_start > start && extent_start < end) { | |
302 | size = extent_start - start; | |
817d52f8 | 303 | total_added += size; |
ea6a478e JB |
304 | ret = btrfs_add_free_space(block_group, start, |
305 | size); | |
0f9dd46c JB |
306 | BUG_ON(ret); |
307 | start = extent_end + 1; | |
308 | } else { | |
309 | break; | |
310 | } | |
311 | } | |
312 | ||
313 | if (start < end) { | |
314 | size = end - start; | |
817d52f8 | 315 | total_added += size; |
ea6a478e | 316 | ret = btrfs_add_free_space(block_group, start, size); |
0f9dd46c JB |
317 | BUG_ON(ret); |
318 | } | |
319 | ||
817d52f8 | 320 | return total_added; |
0f9dd46c JB |
321 | } |
322 | ||
bab39bf9 | 323 | static noinline void caching_thread(struct btrfs_work *work) |
e37c9e69 | 324 | { |
bab39bf9 JB |
325 | struct btrfs_block_group_cache *block_group; |
326 | struct btrfs_fs_info *fs_info; | |
327 | struct btrfs_caching_control *caching_ctl; | |
328 | struct btrfs_root *extent_root; | |
e37c9e69 | 329 | struct btrfs_path *path; |
5f39d397 | 330 | struct extent_buffer *leaf; |
11833d66 | 331 | struct btrfs_key key; |
817d52f8 | 332 | u64 total_found = 0; |
11833d66 YZ |
333 | u64 last = 0; |
334 | u32 nritems; | |
335 | int ret = 0; | |
f510cfec | 336 | |
bab39bf9 JB |
337 | caching_ctl = container_of(work, struct btrfs_caching_control, work); |
338 | block_group = caching_ctl->block_group; | |
339 | fs_info = block_group->fs_info; | |
340 | extent_root = fs_info->extent_root; | |
341 | ||
e37c9e69 CM |
342 | path = btrfs_alloc_path(); |
343 | if (!path) | |
bab39bf9 | 344 | goto out; |
7d7d6068 | 345 | |
817d52f8 | 346 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 347 | |
5cd57b2c | 348 | /* |
817d52f8 JB |
349 | * We don't want to deadlock with somebody trying to allocate a new |
350 | * extent for the extent root while also trying to search the extent | |
351 | * root to add free space. So we skip locking and search the commit | |
352 | * root, since its read-only | |
5cd57b2c CM |
353 | */ |
354 | path->skip_locking = 1; | |
817d52f8 | 355 | path->search_commit_root = 1; |
026fd317 | 356 | path->reada = 1; |
817d52f8 | 357 | |
e4404d6e | 358 | key.objectid = last; |
e37c9e69 | 359 | key.offset = 0; |
11833d66 | 360 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 361 | again: |
11833d66 | 362 | mutex_lock(&caching_ctl->mutex); |
013f1b12 CM |
363 | /* need to make sure the commit_root doesn't disappear */ |
364 | down_read(&fs_info->extent_commit_sem); | |
365 | ||
11833d66 | 366 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 367 | if (ret < 0) |
ef8bbdfe | 368 | goto err; |
a512bbf8 | 369 | |
11833d66 YZ |
370 | leaf = path->nodes[0]; |
371 | nritems = btrfs_header_nritems(leaf); | |
372 | ||
d397712b | 373 | while (1) { |
7841cb28 | 374 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 375 | last = (u64)-1; |
817d52f8 | 376 | break; |
f25784b3 | 377 | } |
817d52f8 | 378 | |
11833d66 YZ |
379 | if (path->slots[0] < nritems) { |
380 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
381 | } else { | |
382 | ret = find_next_key(path, 0, &key); | |
383 | if (ret) | |
e37c9e69 | 384 | break; |
817d52f8 | 385 | |
589d8ade JB |
386 | if (need_resched() || |
387 | btrfs_next_leaf(extent_root, path)) { | |
388 | caching_ctl->progress = last; | |
ff5714cc | 389 | btrfs_release_path(path); |
589d8ade JB |
390 | up_read(&fs_info->extent_commit_sem); |
391 | mutex_unlock(&caching_ctl->mutex); | |
11833d66 | 392 | cond_resched(); |
589d8ade JB |
393 | goto again; |
394 | } | |
395 | leaf = path->nodes[0]; | |
396 | nritems = btrfs_header_nritems(leaf); | |
397 | continue; | |
11833d66 | 398 | } |
817d52f8 | 399 | |
11833d66 YZ |
400 | if (key.objectid < block_group->key.objectid) { |
401 | path->slots[0]++; | |
817d52f8 | 402 | continue; |
e37c9e69 | 403 | } |
0f9dd46c | 404 | |
e37c9e69 | 405 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 406 | block_group->key.offset) |
e37c9e69 | 407 | break; |
7d7d6068 | 408 | |
11833d66 | 409 | if (key.type == BTRFS_EXTENT_ITEM_KEY) { |
817d52f8 JB |
410 | total_found += add_new_free_space(block_group, |
411 | fs_info, last, | |
412 | key.objectid); | |
7d7d6068 | 413 | last = key.objectid + key.offset; |
817d52f8 | 414 | |
11833d66 YZ |
415 | if (total_found > (1024 * 1024 * 2)) { |
416 | total_found = 0; | |
417 | wake_up(&caching_ctl->wait); | |
418 | } | |
817d52f8 | 419 | } |
e37c9e69 CM |
420 | path->slots[0]++; |
421 | } | |
817d52f8 | 422 | ret = 0; |
e37c9e69 | 423 | |
817d52f8 JB |
424 | total_found += add_new_free_space(block_group, fs_info, last, |
425 | block_group->key.objectid + | |
426 | block_group->key.offset); | |
11833d66 | 427 | caching_ctl->progress = (u64)-1; |
817d52f8 JB |
428 | |
429 | spin_lock(&block_group->lock); | |
11833d66 | 430 | block_group->caching_ctl = NULL; |
817d52f8 JB |
431 | block_group->cached = BTRFS_CACHE_FINISHED; |
432 | spin_unlock(&block_group->lock); | |
0f9dd46c | 433 | |
54aa1f4d | 434 | err: |
e37c9e69 | 435 | btrfs_free_path(path); |
276e680d | 436 | up_read(&fs_info->extent_commit_sem); |
817d52f8 | 437 | |
11833d66 YZ |
438 | free_excluded_extents(extent_root, block_group); |
439 | ||
440 | mutex_unlock(&caching_ctl->mutex); | |
bab39bf9 | 441 | out: |
11833d66 YZ |
442 | wake_up(&caching_ctl->wait); |
443 | ||
444 | put_caching_control(caching_ctl); | |
11dfe35a | 445 | btrfs_put_block_group(block_group); |
817d52f8 JB |
446 | } |
447 | ||
9d66e233 JB |
448 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
449 | struct btrfs_trans_handle *trans, | |
b8399dee | 450 | struct btrfs_root *root, |
9d66e233 | 451 | int load_cache_only) |
817d52f8 | 452 | { |
11833d66 YZ |
453 | struct btrfs_fs_info *fs_info = cache->fs_info; |
454 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
455 | int ret = 0; |
456 | ||
11833d66 YZ |
457 | smp_mb(); |
458 | if (cache->cached != BTRFS_CACHE_NO) | |
459 | return 0; | |
460 | ||
9d66e233 JB |
461 | /* |
462 | * We can't do the read from on-disk cache during a commit since we need | |
b8399dee JB |
463 | * to have the normal tree locking. Also if we are currently trying to |
464 | * allocate blocks for the tree root we can't do the fast caching since | |
465 | * we likely hold important locks. | |
9d66e233 | 466 | */ |
f7039b1d | 467 | if (trans && (!trans->transaction->in_commit) && |
b8399dee | 468 | (root && root != root->fs_info->tree_root)) { |
9d66e233 JB |
469 | spin_lock(&cache->lock); |
470 | if (cache->cached != BTRFS_CACHE_NO) { | |
471 | spin_unlock(&cache->lock); | |
472 | return 0; | |
473 | } | |
474 | cache->cached = BTRFS_CACHE_STARTED; | |
475 | spin_unlock(&cache->lock); | |
476 | ||
477 | ret = load_free_space_cache(fs_info, cache); | |
478 | ||
479 | spin_lock(&cache->lock); | |
480 | if (ret == 1) { | |
481 | cache->cached = BTRFS_CACHE_FINISHED; | |
482 | cache->last_byte_to_unpin = (u64)-1; | |
483 | } else { | |
484 | cache->cached = BTRFS_CACHE_NO; | |
485 | } | |
486 | spin_unlock(&cache->lock); | |
3c14874a JB |
487 | if (ret == 1) { |
488 | free_excluded_extents(fs_info->extent_root, cache); | |
9d66e233 | 489 | return 0; |
3c14874a | 490 | } |
9d66e233 JB |
491 | } |
492 | ||
493 | if (load_cache_only) | |
494 | return 0; | |
495 | ||
fc0e4a31 | 496 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
11833d66 YZ |
497 | BUG_ON(!caching_ctl); |
498 | ||
499 | INIT_LIST_HEAD(&caching_ctl->list); | |
500 | mutex_init(&caching_ctl->mutex); | |
501 | init_waitqueue_head(&caching_ctl->wait); | |
502 | caching_ctl->block_group = cache; | |
503 | caching_ctl->progress = cache->key.objectid; | |
504 | /* one for caching kthread, one for caching block group list */ | |
505 | atomic_set(&caching_ctl->count, 2); | |
bab39bf9 | 506 | caching_ctl->work.func = caching_thread; |
11833d66 | 507 | |
817d52f8 JB |
508 | spin_lock(&cache->lock); |
509 | if (cache->cached != BTRFS_CACHE_NO) { | |
510 | spin_unlock(&cache->lock); | |
11833d66 YZ |
511 | kfree(caching_ctl); |
512 | return 0; | |
817d52f8 | 513 | } |
11833d66 | 514 | cache->caching_ctl = caching_ctl; |
817d52f8 JB |
515 | cache->cached = BTRFS_CACHE_STARTED; |
516 | spin_unlock(&cache->lock); | |
517 | ||
11833d66 YZ |
518 | down_write(&fs_info->extent_commit_sem); |
519 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); | |
520 | up_write(&fs_info->extent_commit_sem); | |
521 | ||
11dfe35a | 522 | btrfs_get_block_group(cache); |
11833d66 | 523 | |
bab39bf9 | 524 | btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 525 | |
ef8bbdfe | 526 | return ret; |
e37c9e69 CM |
527 | } |
528 | ||
0f9dd46c JB |
529 | /* |
530 | * return the block group that starts at or after bytenr | |
531 | */ | |
d397712b CM |
532 | static struct btrfs_block_group_cache * |
533 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 534 | { |
0f9dd46c | 535 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 536 | |
0f9dd46c | 537 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 538 | |
0f9dd46c | 539 | return cache; |
0ef3e66b CM |
540 | } |
541 | ||
0f9dd46c | 542 | /* |
9f55684c | 543 | * return the block group that contains the given bytenr |
0f9dd46c | 544 | */ |
d397712b CM |
545 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
546 | struct btrfs_fs_info *info, | |
547 | u64 bytenr) | |
be744175 | 548 | { |
0f9dd46c | 549 | struct btrfs_block_group_cache *cache; |
be744175 | 550 | |
0f9dd46c | 551 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 552 | |
0f9dd46c | 553 | return cache; |
be744175 | 554 | } |
0b86a832 | 555 | |
0f9dd46c JB |
556 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
557 | u64 flags) | |
6324fbf3 | 558 | { |
0f9dd46c | 559 | struct list_head *head = &info->space_info; |
0f9dd46c | 560 | struct btrfs_space_info *found; |
4184ea7f | 561 | |
b742bb82 YZ |
562 | flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM | |
563 | BTRFS_BLOCK_GROUP_METADATA; | |
564 | ||
4184ea7f CM |
565 | rcu_read_lock(); |
566 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 567 | if (found->flags & flags) { |
4184ea7f | 568 | rcu_read_unlock(); |
0f9dd46c | 569 | return found; |
4184ea7f | 570 | } |
0f9dd46c | 571 | } |
4184ea7f | 572 | rcu_read_unlock(); |
0f9dd46c | 573 | return NULL; |
6324fbf3 CM |
574 | } |
575 | ||
4184ea7f CM |
576 | /* |
577 | * after adding space to the filesystem, we need to clear the full flags | |
578 | * on all the space infos. | |
579 | */ | |
580 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
581 | { | |
582 | struct list_head *head = &info->space_info; | |
583 | struct btrfs_space_info *found; | |
584 | ||
585 | rcu_read_lock(); | |
586 | list_for_each_entry_rcu(found, head, list) | |
587 | found->full = 0; | |
588 | rcu_read_unlock(); | |
589 | } | |
590 | ||
80eb234a JB |
591 | static u64 div_factor(u64 num, int factor) |
592 | { | |
593 | if (factor == 10) | |
594 | return num; | |
595 | num *= factor; | |
596 | do_div(num, 10); | |
597 | return num; | |
598 | } | |
599 | ||
e5bc2458 CM |
600 | static u64 div_factor_fine(u64 num, int factor) |
601 | { | |
602 | if (factor == 100) | |
603 | return num; | |
604 | num *= factor; | |
605 | do_div(num, 100); | |
606 | return num; | |
607 | } | |
608 | ||
d2fb3437 YZ |
609 | u64 btrfs_find_block_group(struct btrfs_root *root, |
610 | u64 search_start, u64 search_hint, int owner) | |
cd1bc465 | 611 | { |
96b5179d | 612 | struct btrfs_block_group_cache *cache; |
cd1bc465 | 613 | u64 used; |
d2fb3437 YZ |
614 | u64 last = max(search_hint, search_start); |
615 | u64 group_start = 0; | |
31f3c99b | 616 | int full_search = 0; |
d2fb3437 | 617 | int factor = 9; |
0ef3e66b | 618 | int wrapped = 0; |
31f3c99b | 619 | again: |
e8569813 ZY |
620 | while (1) { |
621 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
0f9dd46c JB |
622 | if (!cache) |
623 | break; | |
96b5179d | 624 | |
c286ac48 | 625 | spin_lock(&cache->lock); |
96b5179d CM |
626 | last = cache->key.objectid + cache->key.offset; |
627 | used = btrfs_block_group_used(&cache->item); | |
628 | ||
d2fb3437 YZ |
629 | if ((full_search || !cache->ro) && |
630 | block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) { | |
e8569813 | 631 | if (used + cache->pinned + cache->reserved < |
d2fb3437 YZ |
632 | div_factor(cache->key.offset, factor)) { |
633 | group_start = cache->key.objectid; | |
c286ac48 | 634 | spin_unlock(&cache->lock); |
fa9c0d79 | 635 | btrfs_put_block_group(cache); |
8790d502 CM |
636 | goto found; |
637 | } | |
6324fbf3 | 638 | } |
c286ac48 | 639 | spin_unlock(&cache->lock); |
fa9c0d79 | 640 | btrfs_put_block_group(cache); |
de428b63 | 641 | cond_resched(); |
cd1bc465 | 642 | } |
0ef3e66b CM |
643 | if (!wrapped) { |
644 | last = search_start; | |
645 | wrapped = 1; | |
646 | goto again; | |
647 | } | |
648 | if (!full_search && factor < 10) { | |
be744175 | 649 | last = search_start; |
31f3c99b | 650 | full_search = 1; |
0ef3e66b | 651 | factor = 10; |
31f3c99b CM |
652 | goto again; |
653 | } | |
be744175 | 654 | found: |
d2fb3437 | 655 | return group_start; |
925baedd | 656 | } |
0f9dd46c | 657 | |
e02119d5 | 658 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 659 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
660 | { |
661 | int ret; | |
662 | struct btrfs_key key; | |
31840ae1 | 663 | struct btrfs_path *path; |
e02119d5 | 664 | |
31840ae1 | 665 | path = btrfs_alloc_path(); |
d8926bb3 MF |
666 | if (!path) |
667 | return -ENOMEM; | |
668 | ||
e02119d5 CM |
669 | key.objectid = start; |
670 | key.offset = len; | |
671 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
672 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, | |
673 | 0, 0); | |
31840ae1 | 674 | btrfs_free_path(path); |
7bb86316 CM |
675 | return ret; |
676 | } | |
677 | ||
a22285a6 YZ |
678 | /* |
679 | * helper function to lookup reference count and flags of extent. | |
680 | * | |
681 | * the head node for delayed ref is used to store the sum of all the | |
682 | * reference count modifications queued up in the rbtree. the head | |
683 | * node may also store the extent flags to set. This way you can check | |
684 | * to see what the reference count and extent flags would be if all of | |
685 | * the delayed refs are not processed. | |
686 | */ | |
687 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
688 | struct btrfs_root *root, u64 bytenr, | |
689 | u64 num_bytes, u64 *refs, u64 *flags) | |
690 | { | |
691 | struct btrfs_delayed_ref_head *head; | |
692 | struct btrfs_delayed_ref_root *delayed_refs; | |
693 | struct btrfs_path *path; | |
694 | struct btrfs_extent_item *ei; | |
695 | struct extent_buffer *leaf; | |
696 | struct btrfs_key key; | |
697 | u32 item_size; | |
698 | u64 num_refs; | |
699 | u64 extent_flags; | |
700 | int ret; | |
701 | ||
702 | path = btrfs_alloc_path(); | |
703 | if (!path) | |
704 | return -ENOMEM; | |
705 | ||
706 | key.objectid = bytenr; | |
707 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
708 | key.offset = num_bytes; | |
709 | if (!trans) { | |
710 | path->skip_locking = 1; | |
711 | path->search_commit_root = 1; | |
712 | } | |
713 | again: | |
714 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, | |
715 | &key, path, 0, 0); | |
716 | if (ret < 0) | |
717 | goto out_free; | |
718 | ||
719 | if (ret == 0) { | |
720 | leaf = path->nodes[0]; | |
721 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
722 | if (item_size >= sizeof(*ei)) { | |
723 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
724 | struct btrfs_extent_item); | |
725 | num_refs = btrfs_extent_refs(leaf, ei); | |
726 | extent_flags = btrfs_extent_flags(leaf, ei); | |
727 | } else { | |
728 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
729 | struct btrfs_extent_item_v0 *ei0; | |
730 | BUG_ON(item_size != sizeof(*ei0)); | |
731 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
732 | struct btrfs_extent_item_v0); | |
733 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
734 | /* FIXME: this isn't correct for data */ | |
735 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
736 | #else | |
737 | BUG(); | |
738 | #endif | |
739 | } | |
740 | BUG_ON(num_refs == 0); | |
741 | } else { | |
742 | num_refs = 0; | |
743 | extent_flags = 0; | |
744 | ret = 0; | |
745 | } | |
746 | ||
747 | if (!trans) | |
748 | goto out; | |
749 | ||
750 | delayed_refs = &trans->transaction->delayed_refs; | |
751 | spin_lock(&delayed_refs->lock); | |
752 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
753 | if (head) { | |
754 | if (!mutex_trylock(&head->mutex)) { | |
755 | atomic_inc(&head->node.refs); | |
756 | spin_unlock(&delayed_refs->lock); | |
757 | ||
b3b4aa74 | 758 | btrfs_release_path(path); |
a22285a6 | 759 | |
8cc33e5c DS |
760 | /* |
761 | * Mutex was contended, block until it's released and try | |
762 | * again | |
763 | */ | |
a22285a6 YZ |
764 | mutex_lock(&head->mutex); |
765 | mutex_unlock(&head->mutex); | |
766 | btrfs_put_delayed_ref(&head->node); | |
767 | goto again; | |
768 | } | |
769 | if (head->extent_op && head->extent_op->update_flags) | |
770 | extent_flags |= head->extent_op->flags_to_set; | |
771 | else | |
772 | BUG_ON(num_refs == 0); | |
773 | ||
774 | num_refs += head->node.ref_mod; | |
775 | mutex_unlock(&head->mutex); | |
776 | } | |
777 | spin_unlock(&delayed_refs->lock); | |
778 | out: | |
779 | WARN_ON(num_refs == 0); | |
780 | if (refs) | |
781 | *refs = num_refs; | |
782 | if (flags) | |
783 | *flags = extent_flags; | |
784 | out_free: | |
785 | btrfs_free_path(path); | |
786 | return ret; | |
787 | } | |
788 | ||
d8d5f3e1 CM |
789 | /* |
790 | * Back reference rules. Back refs have three main goals: | |
791 | * | |
792 | * 1) differentiate between all holders of references to an extent so that | |
793 | * when a reference is dropped we can make sure it was a valid reference | |
794 | * before freeing the extent. | |
795 | * | |
796 | * 2) Provide enough information to quickly find the holders of an extent | |
797 | * if we notice a given block is corrupted or bad. | |
798 | * | |
799 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
800 | * maintenance. This is actually the same as #2, but with a slightly | |
801 | * different use case. | |
802 | * | |
5d4f98a2 YZ |
803 | * There are two kinds of back refs. The implicit back refs is optimized |
804 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
805 | * back refs of this kind provide information about the block's owner tree | |
806 | * and the pointer's key. These information allow us to find the block by | |
807 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
808 | * referenced by their owner trees. The location of tree block is recorded | |
809 | * in the back refs. Actually the full back refs is generic, and can be | |
810 | * used in all cases the implicit back refs is used. The major shortcoming | |
811 | * of the full back refs is its overhead. Every time a tree block gets | |
812 | * COWed, we have to update back refs entry for all pointers in it. | |
813 | * | |
814 | * For a newly allocated tree block, we use implicit back refs for | |
815 | * pointers in it. This means most tree related operations only involve | |
816 | * implicit back refs. For a tree block created in old transaction, the | |
817 | * only way to drop a reference to it is COW it. So we can detect the | |
818 | * event that tree block loses its owner tree's reference and do the | |
819 | * back refs conversion. | |
820 | * | |
821 | * When a tree block is COW'd through a tree, there are four cases: | |
822 | * | |
823 | * The reference count of the block is one and the tree is the block's | |
824 | * owner tree. Nothing to do in this case. | |
825 | * | |
826 | * The reference count of the block is one and the tree is not the | |
827 | * block's owner tree. In this case, full back refs is used for pointers | |
828 | * in the block. Remove these full back refs, add implicit back refs for | |
829 | * every pointers in the new block. | |
830 | * | |
831 | * The reference count of the block is greater than one and the tree is | |
832 | * the block's owner tree. In this case, implicit back refs is used for | |
833 | * pointers in the block. Add full back refs for every pointers in the | |
834 | * block, increase lower level extents' reference counts. The original | |
835 | * implicit back refs are entailed to the new block. | |
836 | * | |
837 | * The reference count of the block is greater than one and the tree is | |
838 | * not the block's owner tree. Add implicit back refs for every pointer in | |
839 | * the new block, increase lower level extents' reference count. | |
840 | * | |
841 | * Back Reference Key composing: | |
842 | * | |
843 | * The key objectid corresponds to the first byte in the extent, | |
844 | * The key type is used to differentiate between types of back refs. | |
845 | * There are different meanings of the key offset for different types | |
846 | * of back refs. | |
847 | * | |
d8d5f3e1 CM |
848 | * File extents can be referenced by: |
849 | * | |
850 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 851 | * - different files inside a single subvolume |
d8d5f3e1 CM |
852 | * - different offsets inside a file (bookend extents in file.c) |
853 | * | |
5d4f98a2 | 854 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
855 | * |
856 | * - Objectid of the subvolume root | |
d8d5f3e1 | 857 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
858 | * - original offset in the file |
859 | * - how many bookend extents | |
d8d5f3e1 | 860 | * |
5d4f98a2 YZ |
861 | * The key offset for the implicit back refs is hash of the first |
862 | * three fields. | |
d8d5f3e1 | 863 | * |
5d4f98a2 | 864 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 865 | * |
5d4f98a2 | 866 | * - number of pointers in the tree leaf |
d8d5f3e1 | 867 | * |
5d4f98a2 YZ |
868 | * The key offset for the implicit back refs is the first byte of |
869 | * the tree leaf | |
d8d5f3e1 | 870 | * |
5d4f98a2 YZ |
871 | * When a file extent is allocated, The implicit back refs is used. |
872 | * the fields are filled in: | |
d8d5f3e1 | 873 | * |
5d4f98a2 | 874 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 875 | * |
5d4f98a2 YZ |
876 | * When a file extent is removed file truncation, we find the |
877 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 878 | * |
5d4f98a2 | 879 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 880 | * |
5d4f98a2 | 881 | * Btree extents can be referenced by: |
d8d5f3e1 | 882 | * |
5d4f98a2 | 883 | * - Different subvolumes |
d8d5f3e1 | 884 | * |
5d4f98a2 YZ |
885 | * Both the implicit back refs and the full back refs for tree blocks |
886 | * only consist of key. The key offset for the implicit back refs is | |
887 | * objectid of block's owner tree. The key offset for the full back refs | |
888 | * is the first byte of parent block. | |
d8d5f3e1 | 889 | * |
5d4f98a2 YZ |
890 | * When implicit back refs is used, information about the lowest key and |
891 | * level of the tree block are required. These information are stored in | |
892 | * tree block info structure. | |
d8d5f3e1 | 893 | */ |
31840ae1 | 894 | |
5d4f98a2 YZ |
895 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
896 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
897 | struct btrfs_root *root, | |
898 | struct btrfs_path *path, | |
899 | u64 owner, u32 extra_size) | |
7bb86316 | 900 | { |
5d4f98a2 YZ |
901 | struct btrfs_extent_item *item; |
902 | struct btrfs_extent_item_v0 *ei0; | |
903 | struct btrfs_extent_ref_v0 *ref0; | |
904 | struct btrfs_tree_block_info *bi; | |
905 | struct extent_buffer *leaf; | |
7bb86316 | 906 | struct btrfs_key key; |
5d4f98a2 YZ |
907 | struct btrfs_key found_key; |
908 | u32 new_size = sizeof(*item); | |
909 | u64 refs; | |
910 | int ret; | |
911 | ||
912 | leaf = path->nodes[0]; | |
913 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
914 | ||
915 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
916 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
917 | struct btrfs_extent_item_v0); | |
918 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
919 | ||
920 | if (owner == (u64)-1) { | |
921 | while (1) { | |
922 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
923 | ret = btrfs_next_leaf(root, path); | |
924 | if (ret < 0) | |
925 | return ret; | |
926 | BUG_ON(ret > 0); | |
927 | leaf = path->nodes[0]; | |
928 | } | |
929 | btrfs_item_key_to_cpu(leaf, &found_key, | |
930 | path->slots[0]); | |
931 | BUG_ON(key.objectid != found_key.objectid); | |
932 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
933 | path->slots[0]++; | |
934 | continue; | |
935 | } | |
936 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
937 | struct btrfs_extent_ref_v0); | |
938 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
939 | break; | |
940 | } | |
941 | } | |
b3b4aa74 | 942 | btrfs_release_path(path); |
5d4f98a2 YZ |
943 | |
944 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
945 | new_size += sizeof(*bi); | |
946 | ||
947 | new_size -= sizeof(*ei0); | |
948 | ret = btrfs_search_slot(trans, root, &key, path, | |
949 | new_size + extra_size, 1); | |
950 | if (ret < 0) | |
951 | return ret; | |
952 | BUG_ON(ret); | |
953 | ||
954 | ret = btrfs_extend_item(trans, root, path, new_size); | |
5d4f98a2 YZ |
955 | |
956 | leaf = path->nodes[0]; | |
957 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
958 | btrfs_set_extent_refs(leaf, item, refs); | |
959 | /* FIXME: get real generation */ | |
960 | btrfs_set_extent_generation(leaf, item, 0); | |
961 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
962 | btrfs_set_extent_flags(leaf, item, | |
963 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
964 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
965 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
966 | /* FIXME: get first key of the block */ | |
967 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
968 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
969 | } else { | |
970 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
971 | } | |
972 | btrfs_mark_buffer_dirty(leaf); | |
973 | return 0; | |
974 | } | |
975 | #endif | |
976 | ||
977 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
978 | { | |
979 | u32 high_crc = ~(u32)0; | |
980 | u32 low_crc = ~(u32)0; | |
981 | __le64 lenum; | |
982 | ||
983 | lenum = cpu_to_le64(root_objectid); | |
163e783e | 984 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 985 | lenum = cpu_to_le64(owner); |
163e783e | 986 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 987 | lenum = cpu_to_le64(offset); |
163e783e | 988 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
989 | |
990 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
991 | } | |
992 | ||
993 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
994 | struct btrfs_extent_data_ref *ref) | |
995 | { | |
996 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
997 | btrfs_extent_data_ref_objectid(leaf, ref), | |
998 | btrfs_extent_data_ref_offset(leaf, ref)); | |
999 | } | |
1000 | ||
1001 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1002 | struct btrfs_extent_data_ref *ref, | |
1003 | u64 root_objectid, u64 owner, u64 offset) | |
1004 | { | |
1005 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1006 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1007 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1008 | return 0; | |
1009 | return 1; | |
1010 | } | |
1011 | ||
1012 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
1013 | struct btrfs_root *root, | |
1014 | struct btrfs_path *path, | |
1015 | u64 bytenr, u64 parent, | |
1016 | u64 root_objectid, | |
1017 | u64 owner, u64 offset) | |
1018 | { | |
1019 | struct btrfs_key key; | |
1020 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1021 | struct extent_buffer *leaf; |
5d4f98a2 | 1022 | u32 nritems; |
74493f7a | 1023 | int ret; |
5d4f98a2 YZ |
1024 | int recow; |
1025 | int err = -ENOENT; | |
74493f7a | 1026 | |
31840ae1 | 1027 | key.objectid = bytenr; |
5d4f98a2 YZ |
1028 | if (parent) { |
1029 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1030 | key.offset = parent; | |
1031 | } else { | |
1032 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1033 | key.offset = hash_extent_data_ref(root_objectid, | |
1034 | owner, offset); | |
1035 | } | |
1036 | again: | |
1037 | recow = 0; | |
1038 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1039 | if (ret < 0) { | |
1040 | err = ret; | |
1041 | goto fail; | |
1042 | } | |
31840ae1 | 1043 | |
5d4f98a2 YZ |
1044 | if (parent) { |
1045 | if (!ret) | |
1046 | return 0; | |
1047 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1048 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1049 | btrfs_release_path(path); |
5d4f98a2 YZ |
1050 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1051 | if (ret < 0) { | |
1052 | err = ret; | |
1053 | goto fail; | |
1054 | } | |
1055 | if (!ret) | |
1056 | return 0; | |
1057 | #endif | |
1058 | goto fail; | |
31840ae1 ZY |
1059 | } |
1060 | ||
1061 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1062 | nritems = btrfs_header_nritems(leaf); |
1063 | while (1) { | |
1064 | if (path->slots[0] >= nritems) { | |
1065 | ret = btrfs_next_leaf(root, path); | |
1066 | if (ret < 0) | |
1067 | err = ret; | |
1068 | if (ret) | |
1069 | goto fail; | |
1070 | ||
1071 | leaf = path->nodes[0]; | |
1072 | nritems = btrfs_header_nritems(leaf); | |
1073 | recow = 1; | |
1074 | } | |
1075 | ||
1076 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1077 | if (key.objectid != bytenr || | |
1078 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1079 | goto fail; | |
1080 | ||
1081 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1082 | struct btrfs_extent_data_ref); | |
1083 | ||
1084 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1085 | owner, offset)) { | |
1086 | if (recow) { | |
b3b4aa74 | 1087 | btrfs_release_path(path); |
5d4f98a2 YZ |
1088 | goto again; |
1089 | } | |
1090 | err = 0; | |
1091 | break; | |
1092 | } | |
1093 | path->slots[0]++; | |
31840ae1 | 1094 | } |
5d4f98a2 YZ |
1095 | fail: |
1096 | return err; | |
31840ae1 ZY |
1097 | } |
1098 | ||
5d4f98a2 YZ |
1099 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
1100 | struct btrfs_root *root, | |
1101 | struct btrfs_path *path, | |
1102 | u64 bytenr, u64 parent, | |
1103 | u64 root_objectid, u64 owner, | |
1104 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
1105 | { |
1106 | struct btrfs_key key; | |
1107 | struct extent_buffer *leaf; | |
5d4f98a2 | 1108 | u32 size; |
31840ae1 ZY |
1109 | u32 num_refs; |
1110 | int ret; | |
74493f7a | 1111 | |
74493f7a | 1112 | key.objectid = bytenr; |
5d4f98a2 YZ |
1113 | if (parent) { |
1114 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1115 | key.offset = parent; | |
1116 | size = sizeof(struct btrfs_shared_data_ref); | |
1117 | } else { | |
1118 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1119 | key.offset = hash_extent_data_ref(root_objectid, | |
1120 | owner, offset); | |
1121 | size = sizeof(struct btrfs_extent_data_ref); | |
1122 | } | |
74493f7a | 1123 | |
5d4f98a2 YZ |
1124 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1125 | if (ret && ret != -EEXIST) | |
1126 | goto fail; | |
1127 | ||
1128 | leaf = path->nodes[0]; | |
1129 | if (parent) { | |
1130 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1131 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1132 | struct btrfs_shared_data_ref); |
1133 | if (ret == 0) { | |
1134 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1135 | } else { | |
1136 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1137 | num_refs += refs_to_add; | |
1138 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1139 | } |
5d4f98a2 YZ |
1140 | } else { |
1141 | struct btrfs_extent_data_ref *ref; | |
1142 | while (ret == -EEXIST) { | |
1143 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1144 | struct btrfs_extent_data_ref); | |
1145 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1146 | owner, offset)) | |
1147 | break; | |
b3b4aa74 | 1148 | btrfs_release_path(path); |
5d4f98a2 YZ |
1149 | key.offset++; |
1150 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1151 | size); | |
1152 | if (ret && ret != -EEXIST) | |
1153 | goto fail; | |
31840ae1 | 1154 | |
5d4f98a2 YZ |
1155 | leaf = path->nodes[0]; |
1156 | } | |
1157 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1158 | struct btrfs_extent_data_ref); | |
1159 | if (ret == 0) { | |
1160 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1161 | root_objectid); | |
1162 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1163 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1164 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1165 | } else { | |
1166 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1167 | num_refs += refs_to_add; | |
1168 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1169 | } |
31840ae1 | 1170 | } |
5d4f98a2 YZ |
1171 | btrfs_mark_buffer_dirty(leaf); |
1172 | ret = 0; | |
1173 | fail: | |
b3b4aa74 | 1174 | btrfs_release_path(path); |
7bb86316 | 1175 | return ret; |
74493f7a CM |
1176 | } |
1177 | ||
5d4f98a2 YZ |
1178 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
1179 | struct btrfs_root *root, | |
1180 | struct btrfs_path *path, | |
1181 | int refs_to_drop) | |
31840ae1 | 1182 | { |
5d4f98a2 YZ |
1183 | struct btrfs_key key; |
1184 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1185 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1186 | struct extent_buffer *leaf; |
5d4f98a2 | 1187 | u32 num_refs = 0; |
31840ae1 ZY |
1188 | int ret = 0; |
1189 | ||
1190 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1191 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1192 | ||
1193 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1194 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1195 | struct btrfs_extent_data_ref); | |
1196 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1197 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1198 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1199 | struct btrfs_shared_data_ref); | |
1200 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1201 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1202 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1203 | struct btrfs_extent_ref_v0 *ref0; | |
1204 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1205 | struct btrfs_extent_ref_v0); | |
1206 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1207 | #endif | |
1208 | } else { | |
1209 | BUG(); | |
1210 | } | |
1211 | ||
56bec294 CM |
1212 | BUG_ON(num_refs < refs_to_drop); |
1213 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1214 | |
31840ae1 ZY |
1215 | if (num_refs == 0) { |
1216 | ret = btrfs_del_item(trans, root, path); | |
1217 | } else { | |
5d4f98a2 YZ |
1218 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1219 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1220 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1221 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1222 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1223 | else { | |
1224 | struct btrfs_extent_ref_v0 *ref0; | |
1225 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1226 | struct btrfs_extent_ref_v0); | |
1227 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1228 | } | |
1229 | #endif | |
31840ae1 ZY |
1230 | btrfs_mark_buffer_dirty(leaf); |
1231 | } | |
31840ae1 ZY |
1232 | return ret; |
1233 | } | |
1234 | ||
5d4f98a2 YZ |
1235 | static noinline u32 extent_data_ref_count(struct btrfs_root *root, |
1236 | struct btrfs_path *path, | |
1237 | struct btrfs_extent_inline_ref *iref) | |
15916de8 | 1238 | { |
5d4f98a2 YZ |
1239 | struct btrfs_key key; |
1240 | struct extent_buffer *leaf; | |
1241 | struct btrfs_extent_data_ref *ref1; | |
1242 | struct btrfs_shared_data_ref *ref2; | |
1243 | u32 num_refs = 0; | |
1244 | ||
1245 | leaf = path->nodes[0]; | |
1246 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1247 | if (iref) { | |
1248 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1249 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1250 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1251 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1252 | } else { | |
1253 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1254 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1255 | } | |
1256 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1257 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1258 | struct btrfs_extent_data_ref); | |
1259 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1260 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1261 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1262 | struct btrfs_shared_data_ref); | |
1263 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1264 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1265 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1266 | struct btrfs_extent_ref_v0 *ref0; | |
1267 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1268 | struct btrfs_extent_ref_v0); | |
1269 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1270 | #endif |
5d4f98a2 YZ |
1271 | } else { |
1272 | WARN_ON(1); | |
1273 | } | |
1274 | return num_refs; | |
1275 | } | |
15916de8 | 1276 | |
5d4f98a2 YZ |
1277 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1278 | struct btrfs_root *root, | |
1279 | struct btrfs_path *path, | |
1280 | u64 bytenr, u64 parent, | |
1281 | u64 root_objectid) | |
1f3c79a2 | 1282 | { |
5d4f98a2 | 1283 | struct btrfs_key key; |
1f3c79a2 | 1284 | int ret; |
1f3c79a2 | 1285 | |
5d4f98a2 YZ |
1286 | key.objectid = bytenr; |
1287 | if (parent) { | |
1288 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1289 | key.offset = parent; | |
1290 | } else { | |
1291 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1292 | key.offset = root_objectid; | |
1f3c79a2 LH |
1293 | } |
1294 | ||
5d4f98a2 YZ |
1295 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1296 | if (ret > 0) | |
1297 | ret = -ENOENT; | |
1298 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1299 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1300 | btrfs_release_path(path); |
5d4f98a2 YZ |
1301 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1302 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1303 | if (ret > 0) | |
1304 | ret = -ENOENT; | |
1305 | } | |
1f3c79a2 | 1306 | #endif |
5d4f98a2 | 1307 | return ret; |
1f3c79a2 LH |
1308 | } |
1309 | ||
5d4f98a2 YZ |
1310 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1311 | struct btrfs_root *root, | |
1312 | struct btrfs_path *path, | |
1313 | u64 bytenr, u64 parent, | |
1314 | u64 root_objectid) | |
31840ae1 | 1315 | { |
5d4f98a2 | 1316 | struct btrfs_key key; |
31840ae1 | 1317 | int ret; |
31840ae1 | 1318 | |
5d4f98a2 YZ |
1319 | key.objectid = bytenr; |
1320 | if (parent) { | |
1321 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1322 | key.offset = parent; | |
1323 | } else { | |
1324 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1325 | key.offset = root_objectid; | |
1326 | } | |
1327 | ||
1328 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
b3b4aa74 | 1329 | btrfs_release_path(path); |
31840ae1 ZY |
1330 | return ret; |
1331 | } | |
1332 | ||
5d4f98a2 | 1333 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1334 | { |
5d4f98a2 YZ |
1335 | int type; |
1336 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1337 | if (parent > 0) | |
1338 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1339 | else | |
1340 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1341 | } else { | |
1342 | if (parent > 0) | |
1343 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1344 | else | |
1345 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1346 | } | |
1347 | return type; | |
31840ae1 | 1348 | } |
56bec294 | 1349 | |
2c47e605 YZ |
1350 | static int find_next_key(struct btrfs_path *path, int level, |
1351 | struct btrfs_key *key) | |
56bec294 | 1352 | |
02217ed2 | 1353 | { |
2c47e605 | 1354 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1355 | if (!path->nodes[level]) |
1356 | break; | |
5d4f98a2 YZ |
1357 | if (path->slots[level] + 1 >= |
1358 | btrfs_header_nritems(path->nodes[level])) | |
1359 | continue; | |
1360 | if (level == 0) | |
1361 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1362 | path->slots[level] + 1); | |
1363 | else | |
1364 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1365 | path->slots[level] + 1); | |
1366 | return 0; | |
1367 | } | |
1368 | return 1; | |
1369 | } | |
037e6390 | 1370 | |
5d4f98a2 YZ |
1371 | /* |
1372 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1373 | * to the address of inline back ref, and 0 is returned. | |
1374 | * | |
1375 | * if back ref isn't found, *ref_ret is set to the address where it | |
1376 | * should be inserted, and -ENOENT is returned. | |
1377 | * | |
1378 | * if insert is true and there are too many inline back refs, the path | |
1379 | * points to the extent item, and -EAGAIN is returned. | |
1380 | * | |
1381 | * NOTE: inline back refs are ordered in the same way that back ref | |
1382 | * items in the tree are ordered. | |
1383 | */ | |
1384 | static noinline_for_stack | |
1385 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1386 | struct btrfs_root *root, | |
1387 | struct btrfs_path *path, | |
1388 | struct btrfs_extent_inline_ref **ref_ret, | |
1389 | u64 bytenr, u64 num_bytes, | |
1390 | u64 parent, u64 root_objectid, | |
1391 | u64 owner, u64 offset, int insert) | |
1392 | { | |
1393 | struct btrfs_key key; | |
1394 | struct extent_buffer *leaf; | |
1395 | struct btrfs_extent_item *ei; | |
1396 | struct btrfs_extent_inline_ref *iref; | |
1397 | u64 flags; | |
1398 | u64 item_size; | |
1399 | unsigned long ptr; | |
1400 | unsigned long end; | |
1401 | int extra_size; | |
1402 | int type; | |
1403 | int want; | |
1404 | int ret; | |
1405 | int err = 0; | |
26b8003f | 1406 | |
db94535d | 1407 | key.objectid = bytenr; |
31840ae1 | 1408 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1409 | key.offset = num_bytes; |
31840ae1 | 1410 | |
5d4f98a2 YZ |
1411 | want = extent_ref_type(parent, owner); |
1412 | if (insert) { | |
1413 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1414 | path->keep_locks = 1; |
5d4f98a2 YZ |
1415 | } else |
1416 | extra_size = -1; | |
1417 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); | |
b9473439 | 1418 | if (ret < 0) { |
5d4f98a2 YZ |
1419 | err = ret; |
1420 | goto out; | |
1421 | } | |
1422 | BUG_ON(ret); | |
1423 | ||
1424 | leaf = path->nodes[0]; | |
1425 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1426 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1427 | if (item_size < sizeof(*ei)) { | |
1428 | if (!insert) { | |
1429 | err = -ENOENT; | |
1430 | goto out; | |
1431 | } | |
1432 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1433 | extra_size); | |
1434 | if (ret < 0) { | |
1435 | err = ret; | |
1436 | goto out; | |
1437 | } | |
1438 | leaf = path->nodes[0]; | |
1439 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1440 | } | |
1441 | #endif | |
1442 | BUG_ON(item_size < sizeof(*ei)); | |
1443 | ||
5d4f98a2 YZ |
1444 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1445 | flags = btrfs_extent_flags(leaf, ei); | |
1446 | ||
1447 | ptr = (unsigned long)(ei + 1); | |
1448 | end = (unsigned long)ei + item_size; | |
1449 | ||
1450 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
1451 | ptr += sizeof(struct btrfs_tree_block_info); | |
1452 | BUG_ON(ptr > end); | |
1453 | } else { | |
1454 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); | |
1455 | } | |
1456 | ||
1457 | err = -ENOENT; | |
1458 | while (1) { | |
1459 | if (ptr >= end) { | |
1460 | WARN_ON(ptr > end); | |
1461 | break; | |
1462 | } | |
1463 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1464 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1465 | if (want < type) | |
1466 | break; | |
1467 | if (want > type) { | |
1468 | ptr += btrfs_extent_inline_ref_size(type); | |
1469 | continue; | |
1470 | } | |
1471 | ||
1472 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1473 | struct btrfs_extent_data_ref *dref; | |
1474 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1475 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1476 | owner, offset)) { | |
1477 | err = 0; | |
1478 | break; | |
1479 | } | |
1480 | if (hash_extent_data_ref_item(leaf, dref) < | |
1481 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1482 | break; | |
1483 | } else { | |
1484 | u64 ref_offset; | |
1485 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1486 | if (parent > 0) { | |
1487 | if (parent == ref_offset) { | |
1488 | err = 0; | |
1489 | break; | |
1490 | } | |
1491 | if (ref_offset < parent) | |
1492 | break; | |
1493 | } else { | |
1494 | if (root_objectid == ref_offset) { | |
1495 | err = 0; | |
1496 | break; | |
1497 | } | |
1498 | if (ref_offset < root_objectid) | |
1499 | break; | |
1500 | } | |
1501 | } | |
1502 | ptr += btrfs_extent_inline_ref_size(type); | |
1503 | } | |
1504 | if (err == -ENOENT && insert) { | |
1505 | if (item_size + extra_size >= | |
1506 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1507 | err = -EAGAIN; | |
1508 | goto out; | |
1509 | } | |
1510 | /* | |
1511 | * To add new inline back ref, we have to make sure | |
1512 | * there is no corresponding back ref item. | |
1513 | * For simplicity, we just do not add new inline back | |
1514 | * ref if there is any kind of item for this block | |
1515 | */ | |
2c47e605 YZ |
1516 | if (find_next_key(path, 0, &key) == 0 && |
1517 | key.objectid == bytenr && | |
85d4198e | 1518 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1519 | err = -EAGAIN; |
1520 | goto out; | |
1521 | } | |
1522 | } | |
1523 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1524 | out: | |
85d4198e | 1525 | if (insert) { |
5d4f98a2 YZ |
1526 | path->keep_locks = 0; |
1527 | btrfs_unlock_up_safe(path, 1); | |
1528 | } | |
1529 | return err; | |
1530 | } | |
1531 | ||
1532 | /* | |
1533 | * helper to add new inline back ref | |
1534 | */ | |
1535 | static noinline_for_stack | |
1536 | int setup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1537 | struct btrfs_root *root, | |
1538 | struct btrfs_path *path, | |
1539 | struct btrfs_extent_inline_ref *iref, | |
1540 | u64 parent, u64 root_objectid, | |
1541 | u64 owner, u64 offset, int refs_to_add, | |
1542 | struct btrfs_delayed_extent_op *extent_op) | |
1543 | { | |
1544 | struct extent_buffer *leaf; | |
1545 | struct btrfs_extent_item *ei; | |
1546 | unsigned long ptr; | |
1547 | unsigned long end; | |
1548 | unsigned long item_offset; | |
1549 | u64 refs; | |
1550 | int size; | |
1551 | int type; | |
1552 | int ret; | |
1553 | ||
1554 | leaf = path->nodes[0]; | |
1555 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1556 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1557 | ||
1558 | type = extent_ref_type(parent, owner); | |
1559 | size = btrfs_extent_inline_ref_size(type); | |
1560 | ||
1561 | ret = btrfs_extend_item(trans, root, path, size); | |
5d4f98a2 YZ |
1562 | |
1563 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1564 | refs = btrfs_extent_refs(leaf, ei); | |
1565 | refs += refs_to_add; | |
1566 | btrfs_set_extent_refs(leaf, ei, refs); | |
1567 | if (extent_op) | |
1568 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1569 | ||
1570 | ptr = (unsigned long)ei + item_offset; | |
1571 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1572 | if (ptr < end - size) | |
1573 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1574 | end - size - ptr); | |
1575 | ||
1576 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1577 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1578 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1579 | struct btrfs_extent_data_ref *dref; | |
1580 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1581 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1582 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1583 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1584 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1585 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1586 | struct btrfs_shared_data_ref *sref; | |
1587 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1588 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1589 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1590 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1591 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1592 | } else { | |
1593 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1594 | } | |
1595 | btrfs_mark_buffer_dirty(leaf); | |
1596 | return 0; | |
1597 | } | |
1598 | ||
1599 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1600 | struct btrfs_root *root, | |
1601 | struct btrfs_path *path, | |
1602 | struct btrfs_extent_inline_ref **ref_ret, | |
1603 | u64 bytenr, u64 num_bytes, u64 parent, | |
1604 | u64 root_objectid, u64 owner, u64 offset) | |
1605 | { | |
1606 | int ret; | |
1607 | ||
1608 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1609 | bytenr, num_bytes, parent, | |
1610 | root_objectid, owner, offset, 0); | |
1611 | if (ret != -ENOENT) | |
54aa1f4d | 1612 | return ret; |
5d4f98a2 | 1613 | |
b3b4aa74 | 1614 | btrfs_release_path(path); |
5d4f98a2 YZ |
1615 | *ref_ret = NULL; |
1616 | ||
1617 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1618 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1619 | root_objectid); | |
1620 | } else { | |
1621 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1622 | root_objectid, owner, offset); | |
b9473439 | 1623 | } |
5d4f98a2 YZ |
1624 | return ret; |
1625 | } | |
31840ae1 | 1626 | |
5d4f98a2 YZ |
1627 | /* |
1628 | * helper to update/remove inline back ref | |
1629 | */ | |
1630 | static noinline_for_stack | |
1631 | int update_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1632 | struct btrfs_root *root, | |
1633 | struct btrfs_path *path, | |
1634 | struct btrfs_extent_inline_ref *iref, | |
1635 | int refs_to_mod, | |
1636 | struct btrfs_delayed_extent_op *extent_op) | |
1637 | { | |
1638 | struct extent_buffer *leaf; | |
1639 | struct btrfs_extent_item *ei; | |
1640 | struct btrfs_extent_data_ref *dref = NULL; | |
1641 | struct btrfs_shared_data_ref *sref = NULL; | |
1642 | unsigned long ptr; | |
1643 | unsigned long end; | |
1644 | u32 item_size; | |
1645 | int size; | |
1646 | int type; | |
1647 | int ret; | |
1648 | u64 refs; | |
1649 | ||
1650 | leaf = path->nodes[0]; | |
1651 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1652 | refs = btrfs_extent_refs(leaf, ei); | |
1653 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1654 | refs += refs_to_mod; | |
1655 | btrfs_set_extent_refs(leaf, ei, refs); | |
1656 | if (extent_op) | |
1657 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1658 | ||
1659 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1660 | ||
1661 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1662 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1663 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1664 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1665 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1666 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1667 | } else { | |
1668 | refs = 1; | |
1669 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1670 | } |
31840ae1 | 1671 | |
5d4f98a2 YZ |
1672 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1673 | refs += refs_to_mod; | |
1674 | ||
1675 | if (refs > 0) { | |
1676 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1677 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1678 | else | |
1679 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1680 | } else { | |
1681 | size = btrfs_extent_inline_ref_size(type); | |
1682 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1683 | ptr = (unsigned long)iref; | |
1684 | end = (unsigned long)ei + item_size; | |
1685 | if (ptr + size < end) | |
1686 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1687 | end - ptr - size); | |
1688 | item_size -= size; | |
1689 | ret = btrfs_truncate_item(trans, root, path, item_size, 1); | |
5d4f98a2 YZ |
1690 | } |
1691 | btrfs_mark_buffer_dirty(leaf); | |
1692 | return 0; | |
1693 | } | |
1694 | ||
1695 | static noinline_for_stack | |
1696 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1697 | struct btrfs_root *root, | |
1698 | struct btrfs_path *path, | |
1699 | u64 bytenr, u64 num_bytes, u64 parent, | |
1700 | u64 root_objectid, u64 owner, | |
1701 | u64 offset, int refs_to_add, | |
1702 | struct btrfs_delayed_extent_op *extent_op) | |
1703 | { | |
1704 | struct btrfs_extent_inline_ref *iref; | |
1705 | int ret; | |
1706 | ||
1707 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1708 | bytenr, num_bytes, parent, | |
1709 | root_objectid, owner, offset, 1); | |
1710 | if (ret == 0) { | |
1711 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
1712 | ret = update_inline_extent_backref(trans, root, path, iref, | |
1713 | refs_to_add, extent_op); | |
1714 | } else if (ret == -ENOENT) { | |
1715 | ret = setup_inline_extent_backref(trans, root, path, iref, | |
1716 | parent, root_objectid, | |
1717 | owner, offset, refs_to_add, | |
1718 | extent_op); | |
771ed689 | 1719 | } |
5d4f98a2 YZ |
1720 | return ret; |
1721 | } | |
31840ae1 | 1722 | |
5d4f98a2 YZ |
1723 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1724 | struct btrfs_root *root, | |
1725 | struct btrfs_path *path, | |
1726 | u64 bytenr, u64 parent, u64 root_objectid, | |
1727 | u64 owner, u64 offset, int refs_to_add) | |
1728 | { | |
1729 | int ret; | |
1730 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1731 | BUG_ON(refs_to_add != 1); | |
1732 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1733 | parent, root_objectid); | |
1734 | } else { | |
1735 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1736 | parent, root_objectid, | |
1737 | owner, offset, refs_to_add); | |
1738 | } | |
1739 | return ret; | |
1740 | } | |
56bec294 | 1741 | |
5d4f98a2 YZ |
1742 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1743 | struct btrfs_root *root, | |
1744 | struct btrfs_path *path, | |
1745 | struct btrfs_extent_inline_ref *iref, | |
1746 | int refs_to_drop, int is_data) | |
1747 | { | |
1748 | int ret; | |
b9473439 | 1749 | |
5d4f98a2 YZ |
1750 | BUG_ON(!is_data && refs_to_drop != 1); |
1751 | if (iref) { | |
1752 | ret = update_inline_extent_backref(trans, root, path, iref, | |
1753 | -refs_to_drop, NULL); | |
1754 | } else if (is_data) { | |
1755 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop); | |
1756 | } else { | |
1757 | ret = btrfs_del_item(trans, root, path); | |
1758 | } | |
1759 | return ret; | |
1760 | } | |
1761 | ||
5378e607 | 1762 | static int btrfs_issue_discard(struct block_device *bdev, |
5d4f98a2 YZ |
1763 | u64 start, u64 len) |
1764 | { | |
5378e607 | 1765 | return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0); |
5d4f98a2 | 1766 | } |
5d4f98a2 YZ |
1767 | |
1768 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 1769 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1770 | { |
5d4f98a2 | 1771 | int ret; |
5378e607 | 1772 | u64 discarded_bytes = 0; |
5d4f98a2 YZ |
1773 | struct btrfs_multi_bio *multi = NULL; |
1774 | ||
e244a0ae | 1775 | |
5d4f98a2 | 1776 | /* Tell the block device(s) that the sectors can be discarded */ |
5378e607 LD |
1777 | ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD, |
1778 | bytenr, &num_bytes, &multi, 0); | |
5d4f98a2 YZ |
1779 | if (!ret) { |
1780 | struct btrfs_bio_stripe *stripe = multi->stripes; | |
1781 | int i; | |
1782 | ||
5d4f98a2 YZ |
1783 | |
1784 | for (i = 0; i < multi->num_stripes; i++, stripe++) { | |
5378e607 LD |
1785 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1786 | stripe->physical, | |
1787 | stripe->length); | |
1788 | if (!ret) | |
1789 | discarded_bytes += stripe->length; | |
1790 | else if (ret != -EOPNOTSUPP) | |
1791 | break; | |
5d4f98a2 YZ |
1792 | } |
1793 | kfree(multi); | |
1794 | } | |
5378e607 LD |
1795 | if (discarded_bytes && ret == -EOPNOTSUPP) |
1796 | ret = 0; | |
1797 | ||
1798 | if (actual_bytes) | |
1799 | *actual_bytes = discarded_bytes; | |
1800 | ||
5d4f98a2 YZ |
1801 | |
1802 | return ret; | |
5d4f98a2 YZ |
1803 | } |
1804 | ||
1805 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1806 | struct btrfs_root *root, | |
1807 | u64 bytenr, u64 num_bytes, u64 parent, | |
1808 | u64 root_objectid, u64 owner, u64 offset) | |
1809 | { | |
1810 | int ret; | |
1811 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && | |
1812 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
1813 | ||
1814 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1815 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes, | |
1816 | parent, root_objectid, (int)owner, | |
1817 | BTRFS_ADD_DELAYED_REF, NULL); | |
1818 | } else { | |
1819 | ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes, | |
1820 | parent, root_objectid, owner, offset, | |
1821 | BTRFS_ADD_DELAYED_REF, NULL); | |
1822 | } | |
1823 | return ret; | |
1824 | } | |
1825 | ||
1826 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1827 | struct btrfs_root *root, | |
1828 | u64 bytenr, u64 num_bytes, | |
1829 | u64 parent, u64 root_objectid, | |
1830 | u64 owner, u64 offset, int refs_to_add, | |
1831 | struct btrfs_delayed_extent_op *extent_op) | |
1832 | { | |
1833 | struct btrfs_path *path; | |
1834 | struct extent_buffer *leaf; | |
1835 | struct btrfs_extent_item *item; | |
1836 | u64 refs; | |
1837 | int ret; | |
1838 | int err = 0; | |
1839 | ||
1840 | path = btrfs_alloc_path(); | |
1841 | if (!path) | |
1842 | return -ENOMEM; | |
1843 | ||
1844 | path->reada = 1; | |
1845 | path->leave_spinning = 1; | |
1846 | /* this will setup the path even if it fails to insert the back ref */ | |
1847 | ret = insert_inline_extent_backref(trans, root->fs_info->extent_root, | |
1848 | path, bytenr, num_bytes, parent, | |
1849 | root_objectid, owner, offset, | |
1850 | refs_to_add, extent_op); | |
1851 | if (ret == 0) | |
1852 | goto out; | |
1853 | ||
1854 | if (ret != -EAGAIN) { | |
1855 | err = ret; | |
1856 | goto out; | |
1857 | } | |
1858 | ||
1859 | leaf = path->nodes[0]; | |
1860 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1861 | refs = btrfs_extent_refs(leaf, item); | |
1862 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1863 | if (extent_op) | |
1864 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1865 | |
5d4f98a2 | 1866 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 1867 | btrfs_release_path(path); |
56bec294 CM |
1868 | |
1869 | path->reada = 1; | |
b9473439 CM |
1870 | path->leave_spinning = 1; |
1871 | ||
56bec294 CM |
1872 | /* now insert the actual backref */ |
1873 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
1874 | path, bytenr, parent, root_objectid, |
1875 | owner, offset, refs_to_add); | |
56bec294 | 1876 | BUG_ON(ret); |
5d4f98a2 | 1877 | out: |
56bec294 | 1878 | btrfs_free_path(path); |
5d4f98a2 | 1879 | return err; |
56bec294 CM |
1880 | } |
1881 | ||
5d4f98a2 YZ |
1882 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
1883 | struct btrfs_root *root, | |
1884 | struct btrfs_delayed_ref_node *node, | |
1885 | struct btrfs_delayed_extent_op *extent_op, | |
1886 | int insert_reserved) | |
56bec294 | 1887 | { |
5d4f98a2 YZ |
1888 | int ret = 0; |
1889 | struct btrfs_delayed_data_ref *ref; | |
1890 | struct btrfs_key ins; | |
1891 | u64 parent = 0; | |
1892 | u64 ref_root = 0; | |
1893 | u64 flags = 0; | |
1894 | ||
1895 | ins.objectid = node->bytenr; | |
1896 | ins.offset = node->num_bytes; | |
1897 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1898 | ||
1899 | ref = btrfs_delayed_node_to_data_ref(node); | |
1900 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) | |
1901 | parent = ref->parent; | |
1902 | else | |
1903 | ref_root = ref->root; | |
1904 | ||
1905 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
1906 | if (extent_op) { | |
1907 | BUG_ON(extent_op->update_key); | |
1908 | flags |= extent_op->flags_to_set; | |
1909 | } | |
1910 | ret = alloc_reserved_file_extent(trans, root, | |
1911 | parent, ref_root, flags, | |
1912 | ref->objectid, ref->offset, | |
1913 | &ins, node->ref_mod); | |
5d4f98a2 YZ |
1914 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
1915 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
1916 | node->num_bytes, parent, | |
1917 | ref_root, ref->objectid, | |
1918 | ref->offset, node->ref_mod, | |
1919 | extent_op); | |
1920 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
1921 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
1922 | node->num_bytes, parent, | |
1923 | ref_root, ref->objectid, | |
1924 | ref->offset, node->ref_mod, | |
1925 | extent_op); | |
1926 | } else { | |
1927 | BUG(); | |
1928 | } | |
1929 | return ret; | |
1930 | } | |
1931 | ||
1932 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1933 | struct extent_buffer *leaf, | |
1934 | struct btrfs_extent_item *ei) | |
1935 | { | |
1936 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1937 | if (extent_op->update_flags) { | |
1938 | flags |= extent_op->flags_to_set; | |
1939 | btrfs_set_extent_flags(leaf, ei, flags); | |
1940 | } | |
1941 | ||
1942 | if (extent_op->update_key) { | |
1943 | struct btrfs_tree_block_info *bi; | |
1944 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
1945 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
1946 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
1947 | } | |
1948 | } | |
1949 | ||
1950 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
1951 | struct btrfs_root *root, | |
1952 | struct btrfs_delayed_ref_node *node, | |
1953 | struct btrfs_delayed_extent_op *extent_op) | |
1954 | { | |
1955 | struct btrfs_key key; | |
1956 | struct btrfs_path *path; | |
1957 | struct btrfs_extent_item *ei; | |
1958 | struct extent_buffer *leaf; | |
1959 | u32 item_size; | |
56bec294 | 1960 | int ret; |
5d4f98a2 YZ |
1961 | int err = 0; |
1962 | ||
1963 | path = btrfs_alloc_path(); | |
1964 | if (!path) | |
1965 | return -ENOMEM; | |
1966 | ||
1967 | key.objectid = node->bytenr; | |
1968 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1969 | key.offset = node->num_bytes; | |
1970 | ||
1971 | path->reada = 1; | |
1972 | path->leave_spinning = 1; | |
1973 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
1974 | path, 0, 1); | |
1975 | if (ret < 0) { | |
1976 | err = ret; | |
1977 | goto out; | |
1978 | } | |
1979 | if (ret > 0) { | |
1980 | err = -EIO; | |
1981 | goto out; | |
1982 | } | |
1983 | ||
1984 | leaf = path->nodes[0]; | |
1985 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1986 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1987 | if (item_size < sizeof(*ei)) { | |
1988 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
1989 | path, (u64)-1, 0); | |
1990 | if (ret < 0) { | |
1991 | err = ret; | |
1992 | goto out; | |
1993 | } | |
1994 | leaf = path->nodes[0]; | |
1995 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1996 | } | |
1997 | #endif | |
1998 | BUG_ON(item_size < sizeof(*ei)); | |
1999 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2000 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2001 | |
5d4f98a2 YZ |
2002 | btrfs_mark_buffer_dirty(leaf); |
2003 | out: | |
2004 | btrfs_free_path(path); | |
2005 | return err; | |
56bec294 CM |
2006 | } |
2007 | ||
5d4f98a2 YZ |
2008 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2009 | struct btrfs_root *root, | |
2010 | struct btrfs_delayed_ref_node *node, | |
2011 | struct btrfs_delayed_extent_op *extent_op, | |
2012 | int insert_reserved) | |
56bec294 CM |
2013 | { |
2014 | int ret = 0; | |
5d4f98a2 YZ |
2015 | struct btrfs_delayed_tree_ref *ref; |
2016 | struct btrfs_key ins; | |
2017 | u64 parent = 0; | |
2018 | u64 ref_root = 0; | |
56bec294 | 2019 | |
5d4f98a2 YZ |
2020 | ins.objectid = node->bytenr; |
2021 | ins.offset = node->num_bytes; | |
2022 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
56bec294 | 2023 | |
5d4f98a2 YZ |
2024 | ref = btrfs_delayed_node_to_tree_ref(node); |
2025 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2026 | parent = ref->parent; | |
2027 | else | |
2028 | ref_root = ref->root; | |
2029 | ||
2030 | BUG_ON(node->ref_mod != 1); | |
2031 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
2032 | BUG_ON(!extent_op || !extent_op->update_flags || | |
2033 | !extent_op->update_key); | |
2034 | ret = alloc_reserved_tree_block(trans, root, | |
2035 | parent, ref_root, | |
2036 | extent_op->flags_to_set, | |
2037 | &extent_op->key, | |
2038 | ref->level, &ins); | |
5d4f98a2 YZ |
2039 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2040 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
2041 | node->num_bytes, parent, ref_root, | |
2042 | ref->level, 0, 1, extent_op); | |
2043 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
2044 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
2045 | node->num_bytes, parent, ref_root, | |
2046 | ref->level, 0, 1, extent_op); | |
2047 | } else { | |
2048 | BUG(); | |
2049 | } | |
56bec294 CM |
2050 | return ret; |
2051 | } | |
2052 | ||
2053 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 YZ |
2054 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2055 | struct btrfs_root *root, | |
2056 | struct btrfs_delayed_ref_node *node, | |
2057 | struct btrfs_delayed_extent_op *extent_op, | |
2058 | int insert_reserved) | |
56bec294 CM |
2059 | { |
2060 | int ret; | |
5d4f98a2 | 2061 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2062 | struct btrfs_delayed_ref_head *head; |
2063 | /* | |
2064 | * we've hit the end of the chain and we were supposed | |
2065 | * to insert this extent into the tree. But, it got | |
2066 | * deleted before we ever needed to insert it, so all | |
2067 | * we have to do is clean up the accounting | |
2068 | */ | |
5d4f98a2 YZ |
2069 | BUG_ON(extent_op); |
2070 | head = btrfs_delayed_node_to_head(node); | |
56bec294 | 2071 | if (insert_reserved) { |
f0486c68 YZ |
2072 | btrfs_pin_extent(root, node->bytenr, |
2073 | node->num_bytes, 1); | |
5d4f98a2 YZ |
2074 | if (head->is_data) { |
2075 | ret = btrfs_del_csums(trans, root, | |
2076 | node->bytenr, | |
2077 | node->num_bytes); | |
2078 | BUG_ON(ret); | |
2079 | } | |
56bec294 | 2080 | } |
56bec294 CM |
2081 | mutex_unlock(&head->mutex); |
2082 | return 0; | |
2083 | } | |
2084 | ||
5d4f98a2 YZ |
2085 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2086 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2087 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2088 | insert_reserved); | |
2089 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2090 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2091 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2092 | insert_reserved); | |
2093 | else | |
2094 | BUG(); | |
2095 | return ret; | |
56bec294 CM |
2096 | } |
2097 | ||
2098 | static noinline struct btrfs_delayed_ref_node * | |
2099 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
2100 | { | |
2101 | struct rb_node *node; | |
2102 | struct btrfs_delayed_ref_node *ref; | |
2103 | int action = BTRFS_ADD_DELAYED_REF; | |
2104 | again: | |
2105 | /* | |
2106 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2107 | * this prevents ref count from going down to zero when | |
2108 | * there still are pending delayed ref. | |
2109 | */ | |
2110 | node = rb_prev(&head->node.rb_node); | |
2111 | while (1) { | |
2112 | if (!node) | |
2113 | break; | |
2114 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2115 | rb_node); | |
2116 | if (ref->bytenr != head->node.bytenr) | |
2117 | break; | |
5d4f98a2 | 2118 | if (ref->action == action) |
56bec294 CM |
2119 | return ref; |
2120 | node = rb_prev(node); | |
2121 | } | |
2122 | if (action == BTRFS_ADD_DELAYED_REF) { | |
2123 | action = BTRFS_DROP_DELAYED_REF; | |
2124 | goto again; | |
2125 | } | |
2126 | return NULL; | |
2127 | } | |
2128 | ||
c3e69d58 CM |
2129 | static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, |
2130 | struct btrfs_root *root, | |
2131 | struct list_head *cluster) | |
56bec294 | 2132 | { |
56bec294 CM |
2133 | struct btrfs_delayed_ref_root *delayed_refs; |
2134 | struct btrfs_delayed_ref_node *ref; | |
2135 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2136 | struct btrfs_delayed_extent_op *extent_op; |
56bec294 | 2137 | int ret; |
c3e69d58 | 2138 | int count = 0; |
56bec294 | 2139 | int must_insert_reserved = 0; |
56bec294 CM |
2140 | |
2141 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2142 | while (1) { |
2143 | if (!locked_ref) { | |
c3e69d58 CM |
2144 | /* pick a new head ref from the cluster list */ |
2145 | if (list_empty(cluster)) | |
56bec294 | 2146 | break; |
56bec294 | 2147 | |
c3e69d58 CM |
2148 | locked_ref = list_entry(cluster->next, |
2149 | struct btrfs_delayed_ref_head, cluster); | |
2150 | ||
2151 | /* grab the lock that says we are going to process | |
2152 | * all the refs for this head */ | |
2153 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
2154 | ||
2155 | /* | |
2156 | * we may have dropped the spin lock to get the head | |
2157 | * mutex lock, and that might have given someone else | |
2158 | * time to free the head. If that's true, it has been | |
2159 | * removed from our list and we can move on. | |
2160 | */ | |
2161 | if (ret == -EAGAIN) { | |
2162 | locked_ref = NULL; | |
2163 | count++; | |
2164 | continue; | |
56bec294 CM |
2165 | } |
2166 | } | |
a28ec197 | 2167 | |
56bec294 CM |
2168 | /* |
2169 | * record the must insert reserved flag before we | |
2170 | * drop the spin lock. | |
2171 | */ | |
2172 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2173 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2174 | |
5d4f98a2 YZ |
2175 | extent_op = locked_ref->extent_op; |
2176 | locked_ref->extent_op = NULL; | |
2177 | ||
56bec294 CM |
2178 | /* |
2179 | * locked_ref is the head node, so we have to go one | |
2180 | * node back for any delayed ref updates | |
2181 | */ | |
56bec294 CM |
2182 | ref = select_delayed_ref(locked_ref); |
2183 | if (!ref) { | |
2184 | /* All delayed refs have been processed, Go ahead | |
2185 | * and send the head node to run_one_delayed_ref, | |
2186 | * so that any accounting fixes can happen | |
2187 | */ | |
2188 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2189 | |
2190 | if (extent_op && must_insert_reserved) { | |
2191 | kfree(extent_op); | |
2192 | extent_op = NULL; | |
2193 | } | |
2194 | ||
2195 | if (extent_op) { | |
2196 | spin_unlock(&delayed_refs->lock); | |
2197 | ||
2198 | ret = run_delayed_extent_op(trans, root, | |
2199 | ref, extent_op); | |
2200 | BUG_ON(ret); | |
2201 | kfree(extent_op); | |
2202 | ||
2203 | cond_resched(); | |
2204 | spin_lock(&delayed_refs->lock); | |
2205 | continue; | |
2206 | } | |
2207 | ||
c3e69d58 | 2208 | list_del_init(&locked_ref->cluster); |
56bec294 CM |
2209 | locked_ref = NULL; |
2210 | } | |
02217ed2 | 2211 | |
56bec294 CM |
2212 | ref->in_tree = 0; |
2213 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
2214 | delayed_refs->num_entries--; | |
5d4f98a2 | 2215 | |
56bec294 | 2216 | spin_unlock(&delayed_refs->lock); |
925baedd | 2217 | |
5d4f98a2 | 2218 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 CM |
2219 | must_insert_reserved); |
2220 | BUG_ON(ret); | |
eb099670 | 2221 | |
5d4f98a2 YZ |
2222 | btrfs_put_delayed_ref(ref); |
2223 | kfree(extent_op); | |
c3e69d58 | 2224 | count++; |
5d4f98a2 | 2225 | |
c3e69d58 CM |
2226 | cond_resched(); |
2227 | spin_lock(&delayed_refs->lock); | |
2228 | } | |
2229 | return count; | |
2230 | } | |
2231 | ||
2232 | /* | |
2233 | * this starts processing the delayed reference count updates and | |
2234 | * extent insertions we have queued up so far. count can be | |
2235 | * 0, which means to process everything in the tree at the start | |
2236 | * of the run (but not newly added entries), or it can be some target | |
2237 | * number you'd like to process. | |
2238 | */ | |
2239 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2240 | struct btrfs_root *root, unsigned long count) | |
2241 | { | |
2242 | struct rb_node *node; | |
2243 | struct btrfs_delayed_ref_root *delayed_refs; | |
2244 | struct btrfs_delayed_ref_node *ref; | |
2245 | struct list_head cluster; | |
2246 | int ret; | |
2247 | int run_all = count == (unsigned long)-1; | |
2248 | int run_most = 0; | |
2249 | ||
2250 | if (root == root->fs_info->extent_root) | |
2251 | root = root->fs_info->tree_root; | |
2252 | ||
2253 | delayed_refs = &trans->transaction->delayed_refs; | |
2254 | INIT_LIST_HEAD(&cluster); | |
2255 | again: | |
2256 | spin_lock(&delayed_refs->lock); | |
2257 | if (count == 0) { | |
2258 | count = delayed_refs->num_entries * 2; | |
2259 | run_most = 1; | |
2260 | } | |
2261 | while (1) { | |
2262 | if (!(run_all || run_most) && | |
2263 | delayed_refs->num_heads_ready < 64) | |
2264 | break; | |
eb099670 | 2265 | |
56bec294 | 2266 | /* |
c3e69d58 CM |
2267 | * go find something we can process in the rbtree. We start at |
2268 | * the beginning of the tree, and then build a cluster | |
2269 | * of refs to process starting at the first one we are able to | |
2270 | * lock | |
56bec294 | 2271 | */ |
c3e69d58 CM |
2272 | ret = btrfs_find_ref_cluster(trans, &cluster, |
2273 | delayed_refs->run_delayed_start); | |
2274 | if (ret) | |
56bec294 CM |
2275 | break; |
2276 | ||
c3e69d58 CM |
2277 | ret = run_clustered_refs(trans, root, &cluster); |
2278 | BUG_ON(ret < 0); | |
2279 | ||
2280 | count -= min_t(unsigned long, ret, count); | |
2281 | ||
2282 | if (count == 0) | |
2283 | break; | |
eb099670 | 2284 | } |
c3e69d58 | 2285 | |
56bec294 | 2286 | if (run_all) { |
56bec294 | 2287 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 2288 | if (!node) |
56bec294 | 2289 | goto out; |
c3e69d58 | 2290 | count = (unsigned long)-1; |
e9d0b13b | 2291 | |
56bec294 CM |
2292 | while (node) { |
2293 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2294 | rb_node); | |
2295 | if (btrfs_delayed_ref_is_head(ref)) { | |
2296 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 2297 | |
56bec294 CM |
2298 | head = btrfs_delayed_node_to_head(ref); |
2299 | atomic_inc(&ref->refs); | |
2300 | ||
2301 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2302 | /* |
2303 | * Mutex was contended, block until it's | |
2304 | * released and try again | |
2305 | */ | |
56bec294 CM |
2306 | mutex_lock(&head->mutex); |
2307 | mutex_unlock(&head->mutex); | |
2308 | ||
2309 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2310 | cond_resched(); |
56bec294 CM |
2311 | goto again; |
2312 | } | |
2313 | node = rb_next(node); | |
2314 | } | |
2315 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
2316 | schedule_timeout(1); |
2317 | goto again; | |
5f39d397 | 2318 | } |
54aa1f4d | 2319 | out: |
c3e69d58 | 2320 | spin_unlock(&delayed_refs->lock); |
a28ec197 CM |
2321 | return 0; |
2322 | } | |
2323 | ||
5d4f98a2 YZ |
2324 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2325 | struct btrfs_root *root, | |
2326 | u64 bytenr, u64 num_bytes, u64 flags, | |
2327 | int is_data) | |
2328 | { | |
2329 | struct btrfs_delayed_extent_op *extent_op; | |
2330 | int ret; | |
2331 | ||
2332 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
2333 | if (!extent_op) | |
2334 | return -ENOMEM; | |
2335 | ||
2336 | extent_op->flags_to_set = flags; | |
2337 | extent_op->update_flags = 1; | |
2338 | extent_op->update_key = 0; | |
2339 | extent_op->is_data = is_data ? 1 : 0; | |
2340 | ||
2341 | ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op); | |
2342 | if (ret) | |
2343 | kfree(extent_op); | |
2344 | return ret; | |
2345 | } | |
2346 | ||
2347 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2348 | struct btrfs_root *root, | |
2349 | struct btrfs_path *path, | |
2350 | u64 objectid, u64 offset, u64 bytenr) | |
2351 | { | |
2352 | struct btrfs_delayed_ref_head *head; | |
2353 | struct btrfs_delayed_ref_node *ref; | |
2354 | struct btrfs_delayed_data_ref *data_ref; | |
2355 | struct btrfs_delayed_ref_root *delayed_refs; | |
2356 | struct rb_node *node; | |
2357 | int ret = 0; | |
2358 | ||
2359 | ret = -ENOENT; | |
2360 | delayed_refs = &trans->transaction->delayed_refs; | |
2361 | spin_lock(&delayed_refs->lock); | |
2362 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2363 | if (!head) | |
2364 | goto out; | |
2365 | ||
2366 | if (!mutex_trylock(&head->mutex)) { | |
2367 | atomic_inc(&head->node.refs); | |
2368 | spin_unlock(&delayed_refs->lock); | |
2369 | ||
b3b4aa74 | 2370 | btrfs_release_path(path); |
5d4f98a2 | 2371 | |
8cc33e5c DS |
2372 | /* |
2373 | * Mutex was contended, block until it's released and let | |
2374 | * caller try again | |
2375 | */ | |
5d4f98a2 YZ |
2376 | mutex_lock(&head->mutex); |
2377 | mutex_unlock(&head->mutex); | |
2378 | btrfs_put_delayed_ref(&head->node); | |
2379 | return -EAGAIN; | |
2380 | } | |
2381 | ||
2382 | node = rb_prev(&head->node.rb_node); | |
2383 | if (!node) | |
2384 | goto out_unlock; | |
2385 | ||
2386 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2387 | ||
2388 | if (ref->bytenr != bytenr) | |
2389 | goto out_unlock; | |
2390 | ||
2391 | ret = 1; | |
2392 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2393 | goto out_unlock; | |
2394 | ||
2395 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2396 | ||
2397 | node = rb_prev(node); | |
2398 | if (node) { | |
2399 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2400 | if (ref->bytenr == bytenr) | |
2401 | goto out_unlock; | |
2402 | } | |
2403 | ||
2404 | if (data_ref->root != root->root_key.objectid || | |
2405 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2406 | goto out_unlock; | |
2407 | ||
2408 | ret = 0; | |
2409 | out_unlock: | |
2410 | mutex_unlock(&head->mutex); | |
2411 | out: | |
2412 | spin_unlock(&delayed_refs->lock); | |
2413 | return ret; | |
2414 | } | |
2415 | ||
2416 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2417 | struct btrfs_root *root, | |
2418 | struct btrfs_path *path, | |
2419 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2420 | { |
2421 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2422 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2423 | struct btrfs_extent_data_ref *ref; |
2424 | struct btrfs_extent_inline_ref *iref; | |
2425 | struct btrfs_extent_item *ei; | |
f321e491 | 2426 | struct btrfs_key key; |
5d4f98a2 | 2427 | u32 item_size; |
be20aa9d | 2428 | int ret; |
925baedd | 2429 | |
be20aa9d | 2430 | key.objectid = bytenr; |
31840ae1 | 2431 | key.offset = (u64)-1; |
f321e491 | 2432 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2433 | |
be20aa9d CM |
2434 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2435 | if (ret < 0) | |
2436 | goto out; | |
2437 | BUG_ON(ret == 0); | |
80ff3856 YZ |
2438 | |
2439 | ret = -ENOENT; | |
2440 | if (path->slots[0] == 0) | |
31840ae1 | 2441 | goto out; |
be20aa9d | 2442 | |
31840ae1 | 2443 | path->slots[0]--; |
f321e491 | 2444 | leaf = path->nodes[0]; |
5d4f98a2 | 2445 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2446 | |
5d4f98a2 | 2447 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2448 | goto out; |
f321e491 | 2449 | |
5d4f98a2 YZ |
2450 | ret = 1; |
2451 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2452 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2453 | if (item_size < sizeof(*ei)) { | |
2454 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2455 | goto out; | |
2456 | } | |
2457 | #endif | |
2458 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2459 | |
5d4f98a2 YZ |
2460 | if (item_size != sizeof(*ei) + |
2461 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2462 | goto out; | |
be20aa9d | 2463 | |
5d4f98a2 YZ |
2464 | if (btrfs_extent_generation(leaf, ei) <= |
2465 | btrfs_root_last_snapshot(&root->root_item)) | |
2466 | goto out; | |
2467 | ||
2468 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2469 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2470 | BTRFS_EXTENT_DATA_REF_KEY) | |
2471 | goto out; | |
2472 | ||
2473 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2474 | if (btrfs_extent_refs(leaf, ei) != | |
2475 | btrfs_extent_data_ref_count(leaf, ref) || | |
2476 | btrfs_extent_data_ref_root(leaf, ref) != | |
2477 | root->root_key.objectid || | |
2478 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2479 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2480 | goto out; | |
2481 | ||
2482 | ret = 0; | |
2483 | out: | |
2484 | return ret; | |
2485 | } | |
2486 | ||
2487 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2488 | struct btrfs_root *root, | |
2489 | u64 objectid, u64 offset, u64 bytenr) | |
2490 | { | |
2491 | struct btrfs_path *path; | |
2492 | int ret; | |
2493 | int ret2; | |
2494 | ||
2495 | path = btrfs_alloc_path(); | |
2496 | if (!path) | |
2497 | return -ENOENT; | |
2498 | ||
2499 | do { | |
2500 | ret = check_committed_ref(trans, root, path, objectid, | |
2501 | offset, bytenr); | |
2502 | if (ret && ret != -ENOENT) | |
f321e491 | 2503 | goto out; |
80ff3856 | 2504 | |
5d4f98a2 YZ |
2505 | ret2 = check_delayed_ref(trans, root, path, objectid, |
2506 | offset, bytenr); | |
2507 | } while (ret2 == -EAGAIN); | |
2508 | ||
2509 | if (ret2 && ret2 != -ENOENT) { | |
2510 | ret = ret2; | |
2511 | goto out; | |
f321e491 | 2512 | } |
5d4f98a2 YZ |
2513 | |
2514 | if (ret != -ENOENT || ret2 != -ENOENT) | |
2515 | ret = 0; | |
be20aa9d | 2516 | out: |
80ff3856 | 2517 | btrfs_free_path(path); |
f0486c68 YZ |
2518 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2519 | WARN_ON(ret > 0); | |
f321e491 | 2520 | return ret; |
be20aa9d | 2521 | } |
c5739bba | 2522 | |
5d4f98a2 | 2523 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2524 | struct btrfs_root *root, |
5d4f98a2 YZ |
2525 | struct extent_buffer *buf, |
2526 | int full_backref, int inc) | |
31840ae1 ZY |
2527 | { |
2528 | u64 bytenr; | |
5d4f98a2 YZ |
2529 | u64 num_bytes; |
2530 | u64 parent; | |
31840ae1 | 2531 | u64 ref_root; |
31840ae1 | 2532 | u32 nritems; |
31840ae1 ZY |
2533 | struct btrfs_key key; |
2534 | struct btrfs_file_extent_item *fi; | |
2535 | int i; | |
2536 | int level; | |
2537 | int ret = 0; | |
31840ae1 | 2538 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
5d4f98a2 | 2539 | u64, u64, u64, u64, u64, u64); |
31840ae1 ZY |
2540 | |
2541 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
2542 | nritems = btrfs_header_nritems(buf); |
2543 | level = btrfs_header_level(buf); | |
2544 | ||
5d4f98a2 YZ |
2545 | if (!root->ref_cows && level == 0) |
2546 | return 0; | |
31840ae1 | 2547 | |
5d4f98a2 YZ |
2548 | if (inc) |
2549 | process_func = btrfs_inc_extent_ref; | |
2550 | else | |
2551 | process_func = btrfs_free_extent; | |
31840ae1 | 2552 | |
5d4f98a2 YZ |
2553 | if (full_backref) |
2554 | parent = buf->start; | |
2555 | else | |
2556 | parent = 0; | |
2557 | ||
2558 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2559 | if (level == 0) { |
5d4f98a2 | 2560 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
2561 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
2562 | continue; | |
5d4f98a2 | 2563 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2564 | struct btrfs_file_extent_item); |
2565 | if (btrfs_file_extent_type(buf, fi) == | |
2566 | BTRFS_FILE_EXTENT_INLINE) | |
2567 | continue; | |
2568 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2569 | if (bytenr == 0) | |
2570 | continue; | |
5d4f98a2 YZ |
2571 | |
2572 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2573 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2574 | ret = process_func(trans, root, bytenr, num_bytes, | |
2575 | parent, ref_root, key.objectid, | |
2576 | key.offset); | |
31840ae1 ZY |
2577 | if (ret) |
2578 | goto fail; | |
2579 | } else { | |
5d4f98a2 YZ |
2580 | bytenr = btrfs_node_blockptr(buf, i); |
2581 | num_bytes = btrfs_level_size(root, level - 1); | |
2582 | ret = process_func(trans, root, bytenr, num_bytes, | |
2583 | parent, ref_root, level - 1, 0); | |
31840ae1 ZY |
2584 | if (ret) |
2585 | goto fail; | |
2586 | } | |
2587 | } | |
2588 | return 0; | |
2589 | fail: | |
5d4f98a2 YZ |
2590 | BUG(); |
2591 | return ret; | |
2592 | } | |
2593 | ||
2594 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
2595 | struct extent_buffer *buf, int full_backref) | |
2596 | { | |
2597 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); | |
2598 | } | |
2599 | ||
2600 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
2601 | struct extent_buffer *buf, int full_backref) | |
2602 | { | |
2603 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); | |
31840ae1 ZY |
2604 | } |
2605 | ||
9078a3e1 CM |
2606 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2607 | struct btrfs_root *root, | |
2608 | struct btrfs_path *path, | |
2609 | struct btrfs_block_group_cache *cache) | |
2610 | { | |
2611 | int ret; | |
9078a3e1 | 2612 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
2613 | unsigned long bi; |
2614 | struct extent_buffer *leaf; | |
9078a3e1 | 2615 | |
9078a3e1 | 2616 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
2617 | if (ret < 0) |
2618 | goto fail; | |
9078a3e1 | 2619 | BUG_ON(ret); |
5f39d397 CM |
2620 | |
2621 | leaf = path->nodes[0]; | |
2622 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2623 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
2624 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 2625 | btrfs_release_path(path); |
54aa1f4d | 2626 | fail: |
9078a3e1 CM |
2627 | if (ret) |
2628 | return ret; | |
9078a3e1 CM |
2629 | return 0; |
2630 | ||
2631 | } | |
2632 | ||
4a8c9a62 YZ |
2633 | static struct btrfs_block_group_cache * |
2634 | next_block_group(struct btrfs_root *root, | |
2635 | struct btrfs_block_group_cache *cache) | |
2636 | { | |
2637 | struct rb_node *node; | |
2638 | spin_lock(&root->fs_info->block_group_cache_lock); | |
2639 | node = rb_next(&cache->cache_node); | |
2640 | btrfs_put_block_group(cache); | |
2641 | if (node) { | |
2642 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
2643 | cache_node); | |
11dfe35a | 2644 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
2645 | } else |
2646 | cache = NULL; | |
2647 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
2648 | return cache; | |
2649 | } | |
2650 | ||
0af3d00b JB |
2651 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
2652 | struct btrfs_trans_handle *trans, | |
2653 | struct btrfs_path *path) | |
2654 | { | |
2655 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
2656 | struct inode *inode = NULL; | |
2657 | u64 alloc_hint = 0; | |
2b20982e | 2658 | int dcs = BTRFS_DC_ERROR; |
0af3d00b JB |
2659 | int num_pages = 0; |
2660 | int retries = 0; | |
2661 | int ret = 0; | |
2662 | ||
2663 | /* | |
2664 | * If this block group is smaller than 100 megs don't bother caching the | |
2665 | * block group. | |
2666 | */ | |
2667 | if (block_group->key.offset < (100 * 1024 * 1024)) { | |
2668 | spin_lock(&block_group->lock); | |
2669 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
2670 | spin_unlock(&block_group->lock); | |
2671 | return 0; | |
2672 | } | |
2673 | ||
2674 | again: | |
2675 | inode = lookup_free_space_inode(root, block_group, path); | |
2676 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
2677 | ret = PTR_ERR(inode); | |
b3b4aa74 | 2678 | btrfs_release_path(path); |
0af3d00b JB |
2679 | goto out; |
2680 | } | |
2681 | ||
2682 | if (IS_ERR(inode)) { | |
2683 | BUG_ON(retries); | |
2684 | retries++; | |
2685 | ||
2686 | if (block_group->ro) | |
2687 | goto out_free; | |
2688 | ||
2689 | ret = create_free_space_inode(root, trans, block_group, path); | |
2690 | if (ret) | |
2691 | goto out_free; | |
2692 | goto again; | |
2693 | } | |
2694 | ||
2695 | /* | |
2696 | * We want to set the generation to 0, that way if anything goes wrong | |
2697 | * from here on out we know not to trust this cache when we load up next | |
2698 | * time. | |
2699 | */ | |
2700 | BTRFS_I(inode)->generation = 0; | |
2701 | ret = btrfs_update_inode(trans, root, inode); | |
2702 | WARN_ON(ret); | |
2703 | ||
2704 | if (i_size_read(inode) > 0) { | |
2705 | ret = btrfs_truncate_free_space_cache(root, trans, path, | |
2706 | inode); | |
2707 | if (ret) | |
2708 | goto out_put; | |
2709 | } | |
2710 | ||
2711 | spin_lock(&block_group->lock); | |
2712 | if (block_group->cached != BTRFS_CACHE_FINISHED) { | |
2b20982e JB |
2713 | /* We're not cached, don't bother trying to write stuff out */ |
2714 | dcs = BTRFS_DC_WRITTEN; | |
0af3d00b JB |
2715 | spin_unlock(&block_group->lock); |
2716 | goto out_put; | |
2717 | } | |
2718 | spin_unlock(&block_group->lock); | |
2719 | ||
2720 | num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024); | |
2721 | if (!num_pages) | |
2722 | num_pages = 1; | |
2723 | ||
2724 | /* | |
2725 | * Just to make absolutely sure we have enough space, we're going to | |
2726 | * preallocate 12 pages worth of space for each block group. In | |
2727 | * practice we ought to use at most 8, but we need extra space so we can | |
2728 | * add our header and have a terminator between the extents and the | |
2729 | * bitmaps. | |
2730 | */ | |
2731 | num_pages *= 16; | |
2732 | num_pages *= PAGE_CACHE_SIZE; | |
2733 | ||
2734 | ret = btrfs_check_data_free_space(inode, num_pages); | |
2735 | if (ret) | |
2736 | goto out_put; | |
2737 | ||
2738 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
2739 | num_pages, num_pages, | |
2740 | &alloc_hint); | |
2b20982e JB |
2741 | if (!ret) |
2742 | dcs = BTRFS_DC_SETUP; | |
0af3d00b JB |
2743 | btrfs_free_reserved_data_space(inode, num_pages); |
2744 | out_put: | |
2745 | iput(inode); | |
2746 | out_free: | |
b3b4aa74 | 2747 | btrfs_release_path(path); |
0af3d00b JB |
2748 | out: |
2749 | spin_lock(&block_group->lock); | |
2b20982e | 2750 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
2751 | spin_unlock(&block_group->lock); |
2752 | ||
2753 | return ret; | |
2754 | } | |
2755 | ||
96b5179d CM |
2756 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
2757 | struct btrfs_root *root) | |
9078a3e1 | 2758 | { |
4a8c9a62 | 2759 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 2760 | int err = 0; |
9078a3e1 | 2761 | struct btrfs_path *path; |
96b5179d | 2762 | u64 last = 0; |
9078a3e1 CM |
2763 | |
2764 | path = btrfs_alloc_path(); | |
2765 | if (!path) | |
2766 | return -ENOMEM; | |
2767 | ||
0af3d00b JB |
2768 | again: |
2769 | while (1) { | |
2770 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
2771 | while (cache) { | |
2772 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
2773 | break; | |
2774 | cache = next_block_group(root, cache); | |
2775 | } | |
2776 | if (!cache) { | |
2777 | if (last == 0) | |
2778 | break; | |
2779 | last = 0; | |
2780 | continue; | |
2781 | } | |
2782 | err = cache_save_setup(cache, trans, path); | |
2783 | last = cache->key.objectid + cache->key.offset; | |
2784 | btrfs_put_block_group(cache); | |
2785 | } | |
2786 | ||
d397712b | 2787 | while (1) { |
4a8c9a62 YZ |
2788 | if (last == 0) { |
2789 | err = btrfs_run_delayed_refs(trans, root, | |
2790 | (unsigned long)-1); | |
2791 | BUG_ON(err); | |
0f9dd46c | 2792 | } |
54aa1f4d | 2793 | |
4a8c9a62 YZ |
2794 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
2795 | while (cache) { | |
0af3d00b JB |
2796 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) { |
2797 | btrfs_put_block_group(cache); | |
2798 | goto again; | |
2799 | } | |
2800 | ||
4a8c9a62 YZ |
2801 | if (cache->dirty) |
2802 | break; | |
2803 | cache = next_block_group(root, cache); | |
2804 | } | |
2805 | if (!cache) { | |
2806 | if (last == 0) | |
2807 | break; | |
2808 | last = 0; | |
2809 | continue; | |
2810 | } | |
0f9dd46c | 2811 | |
0cb59c99 JB |
2812 | if (cache->disk_cache_state == BTRFS_DC_SETUP) |
2813 | cache->disk_cache_state = BTRFS_DC_NEED_WRITE; | |
e8569813 | 2814 | cache->dirty = 0; |
4a8c9a62 | 2815 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 2816 | |
4a8c9a62 YZ |
2817 | err = write_one_cache_group(trans, root, path, cache); |
2818 | BUG_ON(err); | |
2819 | btrfs_put_block_group(cache); | |
9078a3e1 | 2820 | } |
4a8c9a62 | 2821 | |
0cb59c99 JB |
2822 | while (1) { |
2823 | /* | |
2824 | * I don't think this is needed since we're just marking our | |
2825 | * preallocated extent as written, but just in case it can't | |
2826 | * hurt. | |
2827 | */ | |
2828 | if (last == 0) { | |
2829 | err = btrfs_run_delayed_refs(trans, root, | |
2830 | (unsigned long)-1); | |
2831 | BUG_ON(err); | |
2832 | } | |
2833 | ||
2834 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
2835 | while (cache) { | |
2836 | /* | |
2837 | * Really this shouldn't happen, but it could if we | |
2838 | * couldn't write the entire preallocated extent and | |
2839 | * splitting the extent resulted in a new block. | |
2840 | */ | |
2841 | if (cache->dirty) { | |
2842 | btrfs_put_block_group(cache); | |
2843 | goto again; | |
2844 | } | |
2845 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
2846 | break; | |
2847 | cache = next_block_group(root, cache); | |
2848 | } | |
2849 | if (!cache) { | |
2850 | if (last == 0) | |
2851 | break; | |
2852 | last = 0; | |
2853 | continue; | |
2854 | } | |
2855 | ||
2856 | btrfs_write_out_cache(root, trans, cache, path); | |
2857 | ||
2858 | /* | |
2859 | * If we didn't have an error then the cache state is still | |
2860 | * NEED_WRITE, so we can set it to WRITTEN. | |
2861 | */ | |
2862 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
2863 | cache->disk_cache_state = BTRFS_DC_WRITTEN; | |
2864 | last = cache->key.objectid + cache->key.offset; | |
2865 | btrfs_put_block_group(cache); | |
2866 | } | |
2867 | ||
9078a3e1 | 2868 | btrfs_free_path(path); |
4a8c9a62 | 2869 | return 0; |
9078a3e1 CM |
2870 | } |
2871 | ||
d2fb3437 YZ |
2872 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
2873 | { | |
2874 | struct btrfs_block_group_cache *block_group; | |
2875 | int readonly = 0; | |
2876 | ||
2877 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
2878 | if (!block_group || block_group->ro) | |
2879 | readonly = 1; | |
2880 | if (block_group) | |
fa9c0d79 | 2881 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
2882 | return readonly; |
2883 | } | |
2884 | ||
593060d7 CM |
2885 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
2886 | u64 total_bytes, u64 bytes_used, | |
2887 | struct btrfs_space_info **space_info) | |
2888 | { | |
2889 | struct btrfs_space_info *found; | |
b742bb82 YZ |
2890 | int i; |
2891 | int factor; | |
2892 | ||
2893 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
2894 | BTRFS_BLOCK_GROUP_RAID10)) | |
2895 | factor = 2; | |
2896 | else | |
2897 | factor = 1; | |
593060d7 CM |
2898 | |
2899 | found = __find_space_info(info, flags); | |
2900 | if (found) { | |
25179201 | 2901 | spin_lock(&found->lock); |
593060d7 | 2902 | found->total_bytes += total_bytes; |
89a55897 | 2903 | found->disk_total += total_bytes * factor; |
593060d7 | 2904 | found->bytes_used += bytes_used; |
b742bb82 | 2905 | found->disk_used += bytes_used * factor; |
8f18cf13 | 2906 | found->full = 0; |
25179201 | 2907 | spin_unlock(&found->lock); |
593060d7 CM |
2908 | *space_info = found; |
2909 | return 0; | |
2910 | } | |
c146afad | 2911 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
2912 | if (!found) |
2913 | return -ENOMEM; | |
2914 | ||
b742bb82 YZ |
2915 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
2916 | INIT_LIST_HEAD(&found->block_groups[i]); | |
80eb234a | 2917 | init_rwsem(&found->groups_sem); |
0f9dd46c | 2918 | spin_lock_init(&found->lock); |
b742bb82 YZ |
2919 | found->flags = flags & (BTRFS_BLOCK_GROUP_DATA | |
2920 | BTRFS_BLOCK_GROUP_SYSTEM | | |
2921 | BTRFS_BLOCK_GROUP_METADATA); | |
593060d7 | 2922 | found->total_bytes = total_bytes; |
89a55897 | 2923 | found->disk_total = total_bytes * factor; |
593060d7 | 2924 | found->bytes_used = bytes_used; |
b742bb82 | 2925 | found->disk_used = bytes_used * factor; |
593060d7 | 2926 | found->bytes_pinned = 0; |
e8569813 | 2927 | found->bytes_reserved = 0; |
c146afad | 2928 | found->bytes_readonly = 0; |
f0486c68 | 2929 | found->bytes_may_use = 0; |
593060d7 | 2930 | found->full = 0; |
0e4f8f88 | 2931 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 2932 | found->chunk_alloc = 0; |
fdb5effd JB |
2933 | found->flush = 0; |
2934 | init_waitqueue_head(&found->wait); | |
593060d7 | 2935 | *space_info = found; |
4184ea7f | 2936 | list_add_rcu(&found->list, &info->space_info); |
593060d7 CM |
2937 | return 0; |
2938 | } | |
2939 | ||
8790d502 CM |
2940 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
2941 | { | |
2942 | u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 | | |
611f0e00 | 2943 | BTRFS_BLOCK_GROUP_RAID1 | |
321aecc6 | 2944 | BTRFS_BLOCK_GROUP_RAID10 | |
611f0e00 | 2945 | BTRFS_BLOCK_GROUP_DUP); |
8790d502 CM |
2946 | if (extra_flags) { |
2947 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
2948 | fs_info->avail_data_alloc_bits |= extra_flags; | |
2949 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
2950 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
2951 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
2952 | fs_info->avail_system_alloc_bits |= extra_flags; | |
2953 | } | |
2954 | } | |
593060d7 | 2955 | |
2b82032c | 2956 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 2957 | { |
cd02dca5 CM |
2958 | /* |
2959 | * we add in the count of missing devices because we want | |
2960 | * to make sure that any RAID levels on a degraded FS | |
2961 | * continue to be honored. | |
2962 | */ | |
2963 | u64 num_devices = root->fs_info->fs_devices->rw_devices + | |
2964 | root->fs_info->fs_devices->missing_devices; | |
a061fc8d CM |
2965 | |
2966 | if (num_devices == 1) | |
2967 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
2968 | if (num_devices < 4) | |
2969 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
2970 | ||
ec44a35c CM |
2971 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
2972 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 2973 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 2974 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 2975 | } |
ec44a35c CM |
2976 | |
2977 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 2978 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 2979 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 2980 | } |
ec44a35c CM |
2981 | |
2982 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
2983 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
2984 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
2985 | (flags & BTRFS_BLOCK_GROUP_DUP))) | |
2986 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; | |
2987 | return flags; | |
2988 | } | |
2989 | ||
b742bb82 | 2990 | static u64 get_alloc_profile(struct btrfs_root *root, u64 flags) |
6a63209f | 2991 | { |
b742bb82 YZ |
2992 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
2993 | flags |= root->fs_info->avail_data_alloc_bits & | |
2994 | root->fs_info->data_alloc_profile; | |
2995 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
2996 | flags |= root->fs_info->avail_system_alloc_bits & | |
2997 | root->fs_info->system_alloc_profile; | |
2998 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
2999 | flags |= root->fs_info->avail_metadata_alloc_bits & | |
3000 | root->fs_info->metadata_alloc_profile; | |
3001 | return btrfs_reduce_alloc_profile(root, flags); | |
6a63209f JB |
3002 | } |
3003 | ||
6d07bcec | 3004 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 3005 | { |
b742bb82 | 3006 | u64 flags; |
9ed74f2d | 3007 | |
b742bb82 YZ |
3008 | if (data) |
3009 | flags = BTRFS_BLOCK_GROUP_DATA; | |
3010 | else if (root == root->fs_info->chunk_root) | |
3011 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 3012 | else |
b742bb82 | 3013 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3014 | |
b742bb82 | 3015 | return get_alloc_profile(root, flags); |
6a63209f | 3016 | } |
9ed74f2d | 3017 | |
6a63209f JB |
3018 | void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode) |
3019 | { | |
6a63209f | 3020 | BTRFS_I(inode)->space_info = __find_space_info(root->fs_info, |
f0486c68 | 3021 | BTRFS_BLOCK_GROUP_DATA); |
9ed74f2d JB |
3022 | } |
3023 | ||
6a63209f | 3024 | /* |
6a63209f JB |
3025 | * This will check the space that the inode allocates from to make sure we have |
3026 | * enough space for bytes. | |
6a63209f | 3027 | */ |
0ca1f7ce | 3028 | int btrfs_check_data_free_space(struct inode *inode, u64 bytes) |
6a63209f | 3029 | { |
6a63209f | 3030 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 3031 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ab6e2410 | 3032 | u64 used; |
0af3d00b | 3033 | int ret = 0, committed = 0, alloc_chunk = 1; |
6a63209f | 3034 | |
6a63209f JB |
3035 | /* make sure bytes are sectorsize aligned */ |
3036 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
6a63209f | 3037 | |
82d5902d LZ |
3038 | if (root == root->fs_info->tree_root || |
3039 | BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) { | |
0af3d00b JB |
3040 | alloc_chunk = 0; |
3041 | committed = 1; | |
3042 | } | |
3043 | ||
6a63209f | 3044 | data_sinfo = BTRFS_I(inode)->space_info; |
33b4d47f CM |
3045 | if (!data_sinfo) |
3046 | goto alloc; | |
9ed74f2d | 3047 | |
6a63209f JB |
3048 | again: |
3049 | /* make sure we have enough space to handle the data first */ | |
3050 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
3051 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
3052 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
3053 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
3054 | |
3055 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 3056 | struct btrfs_trans_handle *trans; |
9ed74f2d | 3057 | |
6a63209f JB |
3058 | /* |
3059 | * if we don't have enough free bytes in this space then we need | |
3060 | * to alloc a new chunk. | |
3061 | */ | |
0af3d00b | 3062 | if (!data_sinfo->full && alloc_chunk) { |
6a63209f | 3063 | u64 alloc_target; |
9ed74f2d | 3064 | |
0e4f8f88 | 3065 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 3066 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 3067 | alloc: |
6a63209f | 3068 | alloc_target = btrfs_get_alloc_profile(root, 1); |
7a7eaa40 | 3069 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3070 | if (IS_ERR(trans)) |
3071 | return PTR_ERR(trans); | |
9ed74f2d | 3072 | |
6a63209f JB |
3073 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
3074 | bytes + 2 * 1024 * 1024, | |
0e4f8f88 CM |
3075 | alloc_target, |
3076 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 3077 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
3078 | if (ret < 0) { |
3079 | if (ret != -ENOSPC) | |
3080 | return ret; | |
3081 | else | |
3082 | goto commit_trans; | |
3083 | } | |
9ed74f2d | 3084 | |
33b4d47f CM |
3085 | if (!data_sinfo) { |
3086 | btrfs_set_inode_space_info(root, inode); | |
3087 | data_sinfo = BTRFS_I(inode)->space_info; | |
3088 | } | |
6a63209f JB |
3089 | goto again; |
3090 | } | |
f2bb8f5c JB |
3091 | |
3092 | /* | |
3093 | * If we have less pinned bytes than we want to allocate then | |
3094 | * don't bother committing the transaction, it won't help us. | |
3095 | */ | |
3096 | if (data_sinfo->bytes_pinned < bytes) | |
3097 | committed = 1; | |
6a63209f | 3098 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3099 | |
4e06bdd6 | 3100 | /* commit the current transaction and try again */ |
d52a5b5f | 3101 | commit_trans: |
a4abeea4 JB |
3102 | if (!committed && |
3103 | !atomic_read(&root->fs_info->open_ioctl_trans)) { | |
4e06bdd6 | 3104 | committed = 1; |
7a7eaa40 | 3105 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3106 | if (IS_ERR(trans)) |
3107 | return PTR_ERR(trans); | |
4e06bdd6 JB |
3108 | ret = btrfs_commit_transaction(trans, root); |
3109 | if (ret) | |
3110 | return ret; | |
3111 | goto again; | |
3112 | } | |
9ed74f2d | 3113 | |
6a63209f JB |
3114 | return -ENOSPC; |
3115 | } | |
3116 | data_sinfo->bytes_may_use += bytes; | |
3117 | BTRFS_I(inode)->reserved_bytes += bytes; | |
3118 | spin_unlock(&data_sinfo->lock); | |
6a63209f | 3119 | |
9ed74f2d | 3120 | return 0; |
9ed74f2d | 3121 | } |
6a63209f | 3122 | |
6a63209f | 3123 | /* |
0ca1f7ce YZ |
3124 | * called when we are clearing an delalloc extent from the |
3125 | * inode's io_tree or there was an error for whatever reason | |
3126 | * after calling btrfs_check_data_free_space | |
6a63209f | 3127 | */ |
0ca1f7ce | 3128 | void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes) |
e3ccfa98 | 3129 | { |
0ca1f7ce | 3130 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6a63209f | 3131 | struct btrfs_space_info *data_sinfo; |
e3ccfa98 | 3132 | |
6a63209f JB |
3133 | /* make sure bytes are sectorsize aligned */ |
3134 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
e3ccfa98 | 3135 | |
6a63209f JB |
3136 | data_sinfo = BTRFS_I(inode)->space_info; |
3137 | spin_lock(&data_sinfo->lock); | |
3138 | data_sinfo->bytes_may_use -= bytes; | |
3139 | BTRFS_I(inode)->reserved_bytes -= bytes; | |
3140 | spin_unlock(&data_sinfo->lock); | |
e3ccfa98 JB |
3141 | } |
3142 | ||
97e728d4 | 3143 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 3144 | { |
97e728d4 JB |
3145 | struct list_head *head = &info->space_info; |
3146 | struct btrfs_space_info *found; | |
e3ccfa98 | 3147 | |
97e728d4 JB |
3148 | rcu_read_lock(); |
3149 | list_for_each_entry_rcu(found, head, list) { | |
3150 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 3151 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 3152 | } |
97e728d4 | 3153 | rcu_read_unlock(); |
e3ccfa98 JB |
3154 | } |
3155 | ||
e5bc2458 | 3156 | static int should_alloc_chunk(struct btrfs_root *root, |
0e4f8f88 CM |
3157 | struct btrfs_space_info *sinfo, u64 alloc_bytes, |
3158 | int force) | |
32c00aff | 3159 | { |
424499db | 3160 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 3161 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 3162 | u64 thresh; |
e3ccfa98 | 3163 | |
0e4f8f88 CM |
3164 | if (force == CHUNK_ALLOC_FORCE) |
3165 | return 1; | |
3166 | ||
3167 | /* | |
3168 | * in limited mode, we want to have some free space up to | |
3169 | * about 1% of the FS size. | |
3170 | */ | |
3171 | if (force == CHUNK_ALLOC_LIMITED) { | |
3172 | thresh = btrfs_super_total_bytes(&root->fs_info->super_copy); | |
3173 | thresh = max_t(u64, 64 * 1024 * 1024, | |
3174 | div_factor_fine(thresh, 1)); | |
3175 | ||
3176 | if (num_bytes - num_allocated < thresh) | |
3177 | return 1; | |
3178 | } | |
3179 | ||
3180 | /* | |
3181 | * we have two similar checks here, one based on percentage | |
3182 | * and once based on a hard number of 256MB. The idea | |
3183 | * is that if we have a good amount of free | |
3184 | * room, don't allocate a chunk. A good mount is | |
3185 | * less than 80% utilized of the chunks we have allocated, | |
3186 | * or more than 256MB free | |
3187 | */ | |
3188 | if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes) | |
424499db | 3189 | return 0; |
e3ccfa98 | 3190 | |
0e4f8f88 | 3191 | if (num_allocated + alloc_bytes < div_factor(num_bytes, 8)) |
424499db | 3192 | return 0; |
32c00aff | 3193 | |
e5bc2458 | 3194 | thresh = btrfs_super_total_bytes(&root->fs_info->super_copy); |
0e4f8f88 CM |
3195 | |
3196 | /* 256MB or 5% of the FS */ | |
e5bc2458 CM |
3197 | thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5)); |
3198 | ||
3199 | if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3)) | |
14ed0ca6 | 3200 | return 0; |
424499db | 3201 | return 1; |
32c00aff JB |
3202 | } |
3203 | ||
6324fbf3 CM |
3204 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
3205 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 3206 | u64 flags, int force) |
9ed74f2d | 3207 | { |
6324fbf3 | 3208 | struct btrfs_space_info *space_info; |
97e728d4 | 3209 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 3210 | int wait_for_alloc = 0; |
9ed74f2d | 3211 | int ret = 0; |
9ed74f2d | 3212 | |
2b82032c | 3213 | flags = btrfs_reduce_alloc_profile(extent_root, flags); |
ec44a35c | 3214 | |
6324fbf3 | 3215 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
3216 | if (!space_info) { |
3217 | ret = update_space_info(extent_root->fs_info, flags, | |
3218 | 0, 0, &space_info); | |
3219 | BUG_ON(ret); | |
9ed74f2d | 3220 | } |
6324fbf3 | 3221 | BUG_ON(!space_info); |
9ed74f2d | 3222 | |
6d74119f | 3223 | again: |
25179201 | 3224 | spin_lock(&space_info->lock); |
9ed74f2d | 3225 | if (space_info->force_alloc) |
0e4f8f88 | 3226 | force = space_info->force_alloc; |
25179201 JB |
3227 | if (space_info->full) { |
3228 | spin_unlock(&space_info->lock); | |
6d74119f | 3229 | return 0; |
9ed74f2d JB |
3230 | } |
3231 | ||
0e4f8f88 | 3232 | if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) { |
25179201 | 3233 | spin_unlock(&space_info->lock); |
6d74119f JB |
3234 | return 0; |
3235 | } else if (space_info->chunk_alloc) { | |
3236 | wait_for_alloc = 1; | |
3237 | } else { | |
3238 | space_info->chunk_alloc = 1; | |
9ed74f2d | 3239 | } |
0e4f8f88 | 3240 | |
25179201 | 3241 | spin_unlock(&space_info->lock); |
9ed74f2d | 3242 | |
6d74119f JB |
3243 | mutex_lock(&fs_info->chunk_mutex); |
3244 | ||
3245 | /* | |
3246 | * The chunk_mutex is held throughout the entirety of a chunk | |
3247 | * allocation, so once we've acquired the chunk_mutex we know that the | |
3248 | * other guy is done and we need to recheck and see if we should | |
3249 | * allocate. | |
3250 | */ | |
3251 | if (wait_for_alloc) { | |
3252 | mutex_unlock(&fs_info->chunk_mutex); | |
3253 | wait_for_alloc = 0; | |
3254 | goto again; | |
3255 | } | |
3256 | ||
67377734 JB |
3257 | /* |
3258 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3259 | * allocating mixed chunks instead of individual chunks. | |
3260 | */ | |
3261 | if (btrfs_mixed_space_info(space_info)) | |
3262 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3263 | ||
97e728d4 JB |
3264 | /* |
3265 | * if we're doing a data chunk, go ahead and make sure that | |
3266 | * we keep a reasonable number of metadata chunks allocated in the | |
3267 | * FS as well. | |
3268 | */ | |
9ed74f2d | 3269 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
3270 | fs_info->data_chunk_allocations++; |
3271 | if (!(fs_info->data_chunk_allocations % | |
3272 | fs_info->metadata_ratio)) | |
3273 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
3274 | } |
3275 | ||
2b82032c | 3276 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
92b8e897 MF |
3277 | if (ret < 0 && ret != -ENOSPC) |
3278 | goto out; | |
3279 | ||
9ed74f2d | 3280 | spin_lock(&space_info->lock); |
9ed74f2d | 3281 | if (ret) |
6324fbf3 | 3282 | space_info->full = 1; |
424499db YZ |
3283 | else |
3284 | ret = 1; | |
6d74119f | 3285 | |
0e4f8f88 | 3286 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3287 | space_info->chunk_alloc = 0; |
9ed74f2d | 3288 | spin_unlock(&space_info->lock); |
92b8e897 | 3289 | out: |
c146afad | 3290 | mutex_unlock(&extent_root->fs_info->chunk_mutex); |
0f9dd46c | 3291 | return ret; |
6324fbf3 | 3292 | } |
9ed74f2d | 3293 | |
9ed74f2d | 3294 | /* |
5da9d01b | 3295 | * shrink metadata reservation for delalloc |
9ed74f2d | 3296 | */ |
5da9d01b | 3297 | static int shrink_delalloc(struct btrfs_trans_handle *trans, |
0019f10d | 3298 | struct btrfs_root *root, u64 to_reclaim, int sync) |
5da9d01b | 3299 | { |
0ca1f7ce | 3300 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 3301 | struct btrfs_space_info *space_info; |
5da9d01b YZ |
3302 | u64 reserved; |
3303 | u64 max_reclaim; | |
3304 | u64 reclaimed = 0; | |
b1953bce | 3305 | long time_left; |
bf9022e0 | 3306 | int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT; |
b1953bce | 3307 | int loops = 0; |
36e39c40 | 3308 | unsigned long progress; |
5da9d01b | 3309 | |
0ca1f7ce | 3310 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 3311 | space_info = block_rsv->space_info; |
bf9022e0 CM |
3312 | |
3313 | smp_mb(); | |
0019f10d | 3314 | reserved = space_info->bytes_reserved; |
36e39c40 | 3315 | progress = space_info->reservation_progress; |
5da9d01b YZ |
3316 | |
3317 | if (reserved == 0) | |
3318 | return 0; | |
c4f675cd | 3319 | |
fdb5effd JB |
3320 | smp_mb(); |
3321 | if (root->fs_info->delalloc_bytes == 0) { | |
3322 | if (trans) | |
3323 | return 0; | |
3324 | btrfs_wait_ordered_extents(root, 0, 0); | |
3325 | return 0; | |
3326 | } | |
3327 | ||
5da9d01b YZ |
3328 | max_reclaim = min(reserved, to_reclaim); |
3329 | ||
b1953bce | 3330 | while (loops < 1024) { |
bf9022e0 CM |
3331 | /* have the flusher threads jump in and do some IO */ |
3332 | smp_mb(); | |
3333 | nr_pages = min_t(unsigned long, nr_pages, | |
3334 | root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT); | |
3335 | writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages); | |
5da9d01b | 3336 | |
0019f10d | 3337 | spin_lock(&space_info->lock); |
36e39c40 | 3338 | if (reserved > space_info->bytes_reserved) |
0019f10d JB |
3339 | reclaimed += reserved - space_info->bytes_reserved; |
3340 | reserved = space_info->bytes_reserved; | |
3341 | spin_unlock(&space_info->lock); | |
5da9d01b | 3342 | |
36e39c40 CM |
3343 | loops++; |
3344 | ||
5da9d01b YZ |
3345 | if (reserved == 0 || reclaimed >= max_reclaim) |
3346 | break; | |
3347 | ||
3348 | if (trans && trans->transaction->blocked) | |
3349 | return -EAGAIN; | |
bf9022e0 | 3350 | |
36e39c40 | 3351 | time_left = schedule_timeout_interruptible(1); |
b1953bce JB |
3352 | |
3353 | /* We were interrupted, exit */ | |
3354 | if (time_left) | |
3355 | break; | |
3356 | ||
36e39c40 CM |
3357 | /* we've kicked the IO a few times, if anything has been freed, |
3358 | * exit. There is no sense in looping here for a long time | |
3359 | * when we really need to commit the transaction, or there are | |
3360 | * just too many writers without enough free space | |
3361 | */ | |
3362 | ||
3363 | if (loops > 3) { | |
3364 | smp_mb(); | |
3365 | if (progress != space_info->reservation_progress) | |
3366 | break; | |
3367 | } | |
bf9022e0 | 3368 | |
5da9d01b | 3369 | } |
fdb5effd JB |
3370 | if (reclaimed >= to_reclaim && !trans) |
3371 | btrfs_wait_ordered_extents(root, 0, 0); | |
5da9d01b YZ |
3372 | return reclaimed >= to_reclaim; |
3373 | } | |
3374 | ||
8bb8ab2e JB |
3375 | /* |
3376 | * Retries tells us how many times we've called reserve_metadata_bytes. The | |
3377 | * idea is if this is the first call (retries == 0) then we will add to our | |
3378 | * reserved count if we can't make the allocation in order to hold our place | |
3379 | * while we go and try and free up space. That way for retries > 1 we don't try | |
3380 | * and add space, we just check to see if the amount of unused space is >= the | |
3381 | * total space, meaning that our reservation is valid. | |
3382 | * | |
3383 | * However if we don't intend to retry this reservation, pass -1 as retries so | |
3384 | * that it short circuits this logic. | |
3385 | */ | |
3386 | static int reserve_metadata_bytes(struct btrfs_trans_handle *trans, | |
3387 | struct btrfs_root *root, | |
3388 | struct btrfs_block_rsv *block_rsv, | |
3389 | u64 orig_bytes, int flush) | |
9ed74f2d | 3390 | { |
f0486c68 | 3391 | struct btrfs_space_info *space_info = block_rsv->space_info; |
8bb8ab2e JB |
3392 | u64 unused; |
3393 | u64 num_bytes = orig_bytes; | |
3394 | int retries = 0; | |
3395 | int ret = 0; | |
38227933 | 3396 | bool committed = false; |
fdb5effd | 3397 | bool flushing = false; |
9ed74f2d | 3398 | |
8bb8ab2e | 3399 | again: |
fdb5effd | 3400 | ret = 0; |
8bb8ab2e | 3401 | spin_lock(&space_info->lock); |
fdb5effd JB |
3402 | /* |
3403 | * We only want to wait if somebody other than us is flushing and we are | |
3404 | * actually alloed to flush. | |
3405 | */ | |
3406 | while (flush && !flushing && space_info->flush) { | |
3407 | spin_unlock(&space_info->lock); | |
3408 | /* | |
3409 | * If we have a trans handle we can't wait because the flusher | |
3410 | * may have to commit the transaction, which would mean we would | |
3411 | * deadlock since we are waiting for the flusher to finish, but | |
3412 | * hold the current transaction open. | |
3413 | */ | |
3414 | if (trans) | |
3415 | return -EAGAIN; | |
3416 | ret = wait_event_interruptible(space_info->wait, | |
3417 | !space_info->flush); | |
3418 | /* Must have been interrupted, return */ | |
3419 | if (ret) | |
3420 | return -EINTR; | |
3421 | ||
3422 | spin_lock(&space_info->lock); | |
3423 | } | |
3424 | ||
3425 | ret = -ENOSPC; | |
8bb8ab2e JB |
3426 | unused = space_info->bytes_used + space_info->bytes_reserved + |
3427 | space_info->bytes_pinned + space_info->bytes_readonly + | |
3428 | space_info->bytes_may_use; | |
9ed74f2d | 3429 | |
8bb8ab2e JB |
3430 | /* |
3431 | * The idea here is that we've not already over-reserved the block group | |
3432 | * then we can go ahead and save our reservation first and then start | |
3433 | * flushing if we need to. Otherwise if we've already overcommitted | |
3434 | * lets start flushing stuff first and then come back and try to make | |
3435 | * our reservation. | |
3436 | */ | |
3437 | if (unused <= space_info->total_bytes) { | |
6f334348 | 3438 | unused = space_info->total_bytes - unused; |
8bb8ab2e | 3439 | if (unused >= num_bytes) { |
fdb5effd | 3440 | space_info->bytes_reserved += orig_bytes; |
8bb8ab2e JB |
3441 | ret = 0; |
3442 | } else { | |
3443 | /* | |
3444 | * Ok set num_bytes to orig_bytes since we aren't | |
3445 | * overocmmitted, this way we only try and reclaim what | |
3446 | * we need. | |
3447 | */ | |
3448 | num_bytes = orig_bytes; | |
3449 | } | |
3450 | } else { | |
3451 | /* | |
3452 | * Ok we're over committed, set num_bytes to the overcommitted | |
3453 | * amount plus the amount of bytes that we need for this | |
3454 | * reservation. | |
3455 | */ | |
3456 | num_bytes = unused - space_info->total_bytes + | |
3457 | (orig_bytes * (retries + 1)); | |
3458 | } | |
9ed74f2d | 3459 | |
8bb8ab2e JB |
3460 | /* |
3461 | * Couldn't make our reservation, save our place so while we're trying | |
3462 | * to reclaim space we can actually use it instead of somebody else | |
3463 | * stealing it from us. | |
3464 | */ | |
fdb5effd JB |
3465 | if (ret && flush) { |
3466 | flushing = true; | |
3467 | space_info->flush = 1; | |
8bb8ab2e | 3468 | } |
9ed74f2d | 3469 | |
f0486c68 | 3470 | spin_unlock(&space_info->lock); |
9ed74f2d | 3471 | |
fdb5effd | 3472 | if (!ret || !flush) |
8bb8ab2e | 3473 | goto out; |
f0486c68 | 3474 | |
8bb8ab2e JB |
3475 | /* |
3476 | * We do synchronous shrinking since we don't actually unreserve | |
3477 | * metadata until after the IO is completed. | |
3478 | */ | |
3479 | ret = shrink_delalloc(trans, root, num_bytes, 1); | |
fdb5effd | 3480 | if (ret < 0) |
8bb8ab2e | 3481 | goto out; |
f0486c68 | 3482 | |
75c195a2 CM |
3483 | ret = 0; |
3484 | ||
8bb8ab2e JB |
3485 | /* |
3486 | * So if we were overcommitted it's possible that somebody else flushed | |
3487 | * out enough space and we simply didn't have enough space to reclaim, | |
3488 | * so go back around and try again. | |
3489 | */ | |
3490 | if (retries < 2) { | |
3491 | retries++; | |
3492 | goto again; | |
3493 | } | |
f0486c68 | 3494 | |
8bb8ab2e JB |
3495 | /* |
3496 | * Not enough space to be reclaimed, don't bother committing the | |
3497 | * transaction. | |
3498 | */ | |
fdb5effd | 3499 | spin_lock(&space_info->lock); |
8bb8ab2e JB |
3500 | if (space_info->bytes_pinned < orig_bytes) |
3501 | ret = -ENOSPC; | |
3502 | spin_unlock(&space_info->lock); | |
3503 | if (ret) | |
3504 | goto out; | |
f0486c68 | 3505 | |
8bb8ab2e | 3506 | ret = -EAGAIN; |
75c195a2 | 3507 | if (trans) |
8bb8ab2e | 3508 | goto out; |
f0486c68 | 3509 | |
8bb8ab2e | 3510 | ret = -ENOSPC; |
75c195a2 CM |
3511 | if (committed) |
3512 | goto out; | |
3513 | ||
7a7eaa40 | 3514 | trans = btrfs_join_transaction(root); |
8bb8ab2e JB |
3515 | if (IS_ERR(trans)) |
3516 | goto out; | |
3517 | ret = btrfs_commit_transaction(trans, root); | |
38227933 JB |
3518 | if (!ret) { |
3519 | trans = NULL; | |
3520 | committed = true; | |
8bb8ab2e | 3521 | goto again; |
38227933 | 3522 | } |
8bb8ab2e JB |
3523 | |
3524 | out: | |
fdb5effd | 3525 | if (flushing) { |
8bb8ab2e | 3526 | spin_lock(&space_info->lock); |
fdb5effd JB |
3527 | space_info->flush = 0; |
3528 | wake_up_all(&space_info->wait); | |
8bb8ab2e | 3529 | spin_unlock(&space_info->lock); |
f0486c68 | 3530 | } |
f0486c68 YZ |
3531 | return ret; |
3532 | } | |
3533 | ||
3534 | static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans, | |
3535 | struct btrfs_root *root) | |
3536 | { | |
3537 | struct btrfs_block_rsv *block_rsv; | |
3538 | if (root->ref_cows) | |
3539 | block_rsv = trans->block_rsv; | |
3540 | else | |
3541 | block_rsv = root->block_rsv; | |
3542 | ||
3543 | if (!block_rsv) | |
3544 | block_rsv = &root->fs_info->empty_block_rsv; | |
3545 | ||
3546 | return block_rsv; | |
3547 | } | |
3548 | ||
3549 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
3550 | u64 num_bytes) | |
3551 | { | |
3552 | int ret = -ENOSPC; | |
3553 | spin_lock(&block_rsv->lock); | |
3554 | if (block_rsv->reserved >= num_bytes) { | |
3555 | block_rsv->reserved -= num_bytes; | |
3556 | if (block_rsv->reserved < block_rsv->size) | |
3557 | block_rsv->full = 0; | |
3558 | ret = 0; | |
3559 | } | |
3560 | spin_unlock(&block_rsv->lock); | |
3561 | return ret; | |
3562 | } | |
3563 | ||
3564 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
3565 | u64 num_bytes, int update_size) | |
3566 | { | |
3567 | spin_lock(&block_rsv->lock); | |
3568 | block_rsv->reserved += num_bytes; | |
3569 | if (update_size) | |
3570 | block_rsv->size += num_bytes; | |
3571 | else if (block_rsv->reserved >= block_rsv->size) | |
3572 | block_rsv->full = 1; | |
3573 | spin_unlock(&block_rsv->lock); | |
3574 | } | |
3575 | ||
62a45b60 DS |
3576 | static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv, |
3577 | struct btrfs_block_rsv *dest, u64 num_bytes) | |
f0486c68 YZ |
3578 | { |
3579 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
3580 | ||
3581 | spin_lock(&block_rsv->lock); | |
3582 | if (num_bytes == (u64)-1) | |
3583 | num_bytes = block_rsv->size; | |
3584 | block_rsv->size -= num_bytes; | |
3585 | if (block_rsv->reserved >= block_rsv->size) { | |
3586 | num_bytes = block_rsv->reserved - block_rsv->size; | |
3587 | block_rsv->reserved = block_rsv->size; | |
3588 | block_rsv->full = 1; | |
3589 | } else { | |
3590 | num_bytes = 0; | |
3591 | } | |
3592 | spin_unlock(&block_rsv->lock); | |
3593 | ||
3594 | if (num_bytes > 0) { | |
3595 | if (dest) { | |
e9e22899 JB |
3596 | spin_lock(&dest->lock); |
3597 | if (!dest->full) { | |
3598 | u64 bytes_to_add; | |
3599 | ||
3600 | bytes_to_add = dest->size - dest->reserved; | |
3601 | bytes_to_add = min(num_bytes, bytes_to_add); | |
3602 | dest->reserved += bytes_to_add; | |
3603 | if (dest->reserved >= dest->size) | |
3604 | dest->full = 1; | |
3605 | num_bytes -= bytes_to_add; | |
3606 | } | |
3607 | spin_unlock(&dest->lock); | |
3608 | } | |
3609 | if (num_bytes) { | |
f0486c68 YZ |
3610 | spin_lock(&space_info->lock); |
3611 | space_info->bytes_reserved -= num_bytes; | |
36e39c40 | 3612 | space_info->reservation_progress++; |
f0486c68 | 3613 | spin_unlock(&space_info->lock); |
4e06bdd6 | 3614 | } |
9ed74f2d | 3615 | } |
f0486c68 | 3616 | } |
4e06bdd6 | 3617 | |
f0486c68 YZ |
3618 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
3619 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
3620 | { | |
3621 | int ret; | |
9ed74f2d | 3622 | |
f0486c68 YZ |
3623 | ret = block_rsv_use_bytes(src, num_bytes); |
3624 | if (ret) | |
3625 | return ret; | |
9ed74f2d | 3626 | |
f0486c68 | 3627 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
3628 | return 0; |
3629 | } | |
3630 | ||
f0486c68 | 3631 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv) |
9ed74f2d | 3632 | { |
f0486c68 YZ |
3633 | memset(rsv, 0, sizeof(*rsv)); |
3634 | spin_lock_init(&rsv->lock); | |
3635 | atomic_set(&rsv->usage, 1); | |
3636 | rsv->priority = 6; | |
3637 | INIT_LIST_HEAD(&rsv->list); | |
3638 | } | |
3639 | ||
3640 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root) | |
3641 | { | |
3642 | struct btrfs_block_rsv *block_rsv; | |
3643 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 3644 | |
f0486c68 YZ |
3645 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
3646 | if (!block_rsv) | |
3647 | return NULL; | |
9ed74f2d | 3648 | |
f0486c68 | 3649 | btrfs_init_block_rsv(block_rsv); |
f0486c68 YZ |
3650 | block_rsv->space_info = __find_space_info(fs_info, |
3651 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
3652 | return block_rsv; |
3653 | } | |
9ed74f2d | 3654 | |
f0486c68 YZ |
3655 | void btrfs_free_block_rsv(struct btrfs_root *root, |
3656 | struct btrfs_block_rsv *rsv) | |
3657 | { | |
3658 | if (rsv && atomic_dec_and_test(&rsv->usage)) { | |
3659 | btrfs_block_rsv_release(root, rsv, (u64)-1); | |
3660 | if (!rsv->durable) | |
3661 | kfree(rsv); | |
3662 | } | |
9ed74f2d JB |
3663 | } |
3664 | ||
3665 | /* | |
f0486c68 YZ |
3666 | * make the block_rsv struct be able to capture freed space. |
3667 | * the captured space will re-add to the the block_rsv struct | |
3668 | * after transaction commit | |
9ed74f2d | 3669 | */ |
f0486c68 YZ |
3670 | void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info, |
3671 | struct btrfs_block_rsv *block_rsv) | |
9ed74f2d | 3672 | { |
f0486c68 YZ |
3673 | block_rsv->durable = 1; |
3674 | mutex_lock(&fs_info->durable_block_rsv_mutex); | |
3675 | list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list); | |
3676 | mutex_unlock(&fs_info->durable_block_rsv_mutex); | |
3677 | } | |
9ed74f2d | 3678 | |
f0486c68 YZ |
3679 | int btrfs_block_rsv_add(struct btrfs_trans_handle *trans, |
3680 | struct btrfs_root *root, | |
3681 | struct btrfs_block_rsv *block_rsv, | |
8bb8ab2e | 3682 | u64 num_bytes) |
f0486c68 YZ |
3683 | { |
3684 | int ret; | |
9ed74f2d | 3685 | |
f0486c68 YZ |
3686 | if (num_bytes == 0) |
3687 | return 0; | |
8bb8ab2e JB |
3688 | |
3689 | ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1); | |
f0486c68 YZ |
3690 | if (!ret) { |
3691 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
3692 | return 0; | |
3693 | } | |
9ed74f2d | 3694 | |
f0486c68 YZ |
3695 | return ret; |
3696 | } | |
9ed74f2d | 3697 | |
f0486c68 YZ |
3698 | int btrfs_block_rsv_check(struct btrfs_trans_handle *trans, |
3699 | struct btrfs_root *root, | |
3700 | struct btrfs_block_rsv *block_rsv, | |
3701 | u64 min_reserved, int min_factor) | |
3702 | { | |
3703 | u64 num_bytes = 0; | |
3704 | int commit_trans = 0; | |
3705 | int ret = -ENOSPC; | |
9ed74f2d | 3706 | |
f0486c68 YZ |
3707 | if (!block_rsv) |
3708 | return 0; | |
9ed74f2d | 3709 | |
f0486c68 YZ |
3710 | spin_lock(&block_rsv->lock); |
3711 | if (min_factor > 0) | |
3712 | num_bytes = div_factor(block_rsv->size, min_factor); | |
3713 | if (min_reserved > num_bytes) | |
3714 | num_bytes = min_reserved; | |
9ed74f2d | 3715 | |
f0486c68 YZ |
3716 | if (block_rsv->reserved >= num_bytes) { |
3717 | ret = 0; | |
3718 | } else { | |
3719 | num_bytes -= block_rsv->reserved; | |
3720 | if (block_rsv->durable && | |
3721 | block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes) | |
3722 | commit_trans = 1; | |
3723 | } | |
3724 | spin_unlock(&block_rsv->lock); | |
3725 | if (!ret) | |
3726 | return 0; | |
3727 | ||
3728 | if (block_rsv->refill_used) { | |
8bb8ab2e JB |
3729 | ret = reserve_metadata_bytes(trans, root, block_rsv, |
3730 | num_bytes, 0); | |
f0486c68 YZ |
3731 | if (!ret) { |
3732 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
3733 | return 0; | |
4e06bdd6 | 3734 | } |
f0486c68 | 3735 | } |
9ed74f2d | 3736 | |
f0486c68 YZ |
3737 | if (commit_trans) { |
3738 | if (trans) | |
3739 | return -EAGAIN; | |
7a7eaa40 | 3740 | trans = btrfs_join_transaction(root); |
f0486c68 YZ |
3741 | BUG_ON(IS_ERR(trans)); |
3742 | ret = btrfs_commit_transaction(trans, root); | |
3743 | return 0; | |
6a63209f | 3744 | } |
9ed74f2d | 3745 | |
f0486c68 YZ |
3746 | return -ENOSPC; |
3747 | } | |
3748 | ||
3749 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, | |
3750 | struct btrfs_block_rsv *dst_rsv, | |
3751 | u64 num_bytes) | |
3752 | { | |
3753 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
3754 | } | |
3755 | ||
3756 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
3757 | struct btrfs_block_rsv *block_rsv, | |
3758 | u64 num_bytes) | |
3759 | { | |
3760 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
3761 | if (global_rsv->full || global_rsv == block_rsv || | |
3762 | block_rsv->space_info != global_rsv->space_info) | |
3763 | global_rsv = NULL; | |
3764 | block_rsv_release_bytes(block_rsv, global_rsv, num_bytes); | |
6a63209f JB |
3765 | } |
3766 | ||
3767 | /* | |
8929ecfa YZ |
3768 | * helper to calculate size of global block reservation. |
3769 | * the desired value is sum of space used by extent tree, | |
3770 | * checksum tree and root tree | |
6a63209f | 3771 | */ |
8929ecfa | 3772 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 3773 | { |
8929ecfa YZ |
3774 | struct btrfs_space_info *sinfo; |
3775 | u64 num_bytes; | |
3776 | u64 meta_used; | |
3777 | u64 data_used; | |
3778 | int csum_size = btrfs_super_csum_size(&fs_info->super_copy); | |
6a63209f | 3779 | |
8929ecfa YZ |
3780 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
3781 | spin_lock(&sinfo->lock); | |
3782 | data_used = sinfo->bytes_used; | |
3783 | spin_unlock(&sinfo->lock); | |
33b4d47f | 3784 | |
8929ecfa YZ |
3785 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
3786 | spin_lock(&sinfo->lock); | |
6d48755d JB |
3787 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
3788 | data_used = 0; | |
8929ecfa YZ |
3789 | meta_used = sinfo->bytes_used; |
3790 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 3791 | |
8929ecfa YZ |
3792 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
3793 | csum_size * 2; | |
3794 | num_bytes += div64_u64(data_used + meta_used, 50); | |
4e06bdd6 | 3795 | |
8929ecfa YZ |
3796 | if (num_bytes * 3 > meta_used) |
3797 | num_bytes = div64_u64(meta_used, 3); | |
ab6e2410 | 3798 | |
8929ecfa YZ |
3799 | return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10); |
3800 | } | |
6a63209f | 3801 | |
8929ecfa YZ |
3802 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
3803 | { | |
3804 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
3805 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
3806 | u64 num_bytes; | |
6a63209f | 3807 | |
8929ecfa | 3808 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 3809 | |
8929ecfa YZ |
3810 | spin_lock(&block_rsv->lock); |
3811 | spin_lock(&sinfo->lock); | |
4e06bdd6 | 3812 | |
8929ecfa | 3813 | block_rsv->size = num_bytes; |
4e06bdd6 | 3814 | |
8929ecfa | 3815 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
6d48755d JB |
3816 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
3817 | sinfo->bytes_may_use; | |
8929ecfa YZ |
3818 | |
3819 | if (sinfo->total_bytes > num_bytes) { | |
3820 | num_bytes = sinfo->total_bytes - num_bytes; | |
3821 | block_rsv->reserved += num_bytes; | |
3822 | sinfo->bytes_reserved += num_bytes; | |
6a63209f | 3823 | } |
6a63209f | 3824 | |
8929ecfa YZ |
3825 | if (block_rsv->reserved >= block_rsv->size) { |
3826 | num_bytes = block_rsv->reserved - block_rsv->size; | |
3827 | sinfo->bytes_reserved -= num_bytes; | |
36e39c40 | 3828 | sinfo->reservation_progress++; |
8929ecfa YZ |
3829 | block_rsv->reserved = block_rsv->size; |
3830 | block_rsv->full = 1; | |
3831 | } | |
182608c8 | 3832 | |
8929ecfa YZ |
3833 | spin_unlock(&sinfo->lock); |
3834 | spin_unlock(&block_rsv->lock); | |
6a63209f JB |
3835 | } |
3836 | ||
f0486c68 | 3837 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 3838 | { |
f0486c68 | 3839 | struct btrfs_space_info *space_info; |
6a63209f | 3840 | |
f0486c68 YZ |
3841 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
3842 | fs_info->chunk_block_rsv.space_info = space_info; | |
3843 | fs_info->chunk_block_rsv.priority = 10; | |
6a63209f | 3844 | |
f0486c68 | 3845 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa YZ |
3846 | fs_info->global_block_rsv.space_info = space_info; |
3847 | fs_info->global_block_rsv.priority = 10; | |
3848 | fs_info->global_block_rsv.refill_used = 1; | |
3849 | fs_info->delalloc_block_rsv.space_info = space_info; | |
f0486c68 YZ |
3850 | fs_info->trans_block_rsv.space_info = space_info; |
3851 | fs_info->empty_block_rsv.space_info = space_info; | |
3852 | fs_info->empty_block_rsv.priority = 10; | |
3853 | ||
8929ecfa YZ |
3854 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
3855 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
3856 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
3857 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 3858 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa YZ |
3859 | |
3860 | btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv); | |
3861 | ||
3862 | btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv); | |
3863 | ||
3864 | update_global_block_rsv(fs_info); | |
6a63209f JB |
3865 | } |
3866 | ||
8929ecfa | 3867 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 3868 | { |
8929ecfa YZ |
3869 | block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1); |
3870 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); | |
3871 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
3872 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
3873 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
3874 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
3875 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
f0486c68 | 3876 | } |
6a63209f | 3877 | |
fcb80c2a JB |
3878 | int btrfs_truncate_reserve_metadata(struct btrfs_trans_handle *trans, |
3879 | struct btrfs_root *root, | |
3880 | struct btrfs_block_rsv *rsv) | |
3881 | { | |
3882 | struct btrfs_block_rsv *trans_rsv = &root->fs_info->trans_block_rsv; | |
3883 | u64 num_bytes; | |
3884 | int ret; | |
3885 | ||
3886 | /* | |
3887 | * Truncate should be freeing data, but give us 2 items just in case it | |
3888 | * needs to use some space. We may want to be smarter about this in the | |
3889 | * future. | |
3890 | */ | |
ff5714cc | 3891 | num_bytes = btrfs_calc_trans_metadata_size(root, 2); |
fcb80c2a JB |
3892 | |
3893 | /* We already have enough bytes, just return */ | |
3894 | if (rsv->reserved >= num_bytes) | |
3895 | return 0; | |
3896 | ||
3897 | num_bytes -= rsv->reserved; | |
3898 | ||
3899 | /* | |
3900 | * You should have reserved enough space before hand to do this, so this | |
3901 | * should not fail. | |
3902 | */ | |
3903 | ret = block_rsv_migrate_bytes(trans_rsv, rsv, num_bytes); | |
3904 | BUG_ON(ret); | |
3905 | ||
3906 | return 0; | |
3907 | } | |
3908 | ||
a22285a6 YZ |
3909 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
3910 | struct btrfs_root *root) | |
6a63209f | 3911 | { |
a22285a6 YZ |
3912 | if (!trans->bytes_reserved) |
3913 | return; | |
6a63209f | 3914 | |
a22285a6 YZ |
3915 | BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv); |
3916 | btrfs_block_rsv_release(root, trans->block_rsv, | |
3917 | trans->bytes_reserved); | |
3918 | trans->bytes_reserved = 0; | |
3919 | } | |
6a63209f | 3920 | |
d68fc57b YZ |
3921 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
3922 | struct inode *inode) | |
3923 | { | |
3924 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3925 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
3926 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
3927 | ||
3928 | /* | |
fcb80c2a JB |
3929 | * We need to hold space in order to delete our orphan item once we've |
3930 | * added it, so this takes the reservation so we can release it later | |
3931 | * when we are truly done with the orphan item. | |
d68fc57b | 3932 | */ |
ff5714cc | 3933 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
d68fc57b | 3934 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
6a63209f JB |
3935 | } |
3936 | ||
d68fc57b | 3937 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 3938 | { |
d68fc57b | 3939 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 3940 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
d68fc57b YZ |
3941 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
3942 | } | |
97e728d4 | 3943 | |
a22285a6 YZ |
3944 | int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans, |
3945 | struct btrfs_pending_snapshot *pending) | |
3946 | { | |
3947 | struct btrfs_root *root = pending->root; | |
3948 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
3949 | struct btrfs_block_rsv *dst_rsv = &pending->block_rsv; | |
3950 | /* | |
3951 | * two for root back/forward refs, two for directory entries | |
3952 | * and one for root of the snapshot. | |
3953 | */ | |
16cdcec7 | 3954 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5); |
a22285a6 YZ |
3955 | dst_rsv->space_info = src_rsv->space_info; |
3956 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
97e728d4 JB |
3957 | } |
3958 | ||
9e0baf60 JB |
3959 | static unsigned drop_outstanding_extent(struct inode *inode) |
3960 | { | |
3961 | unsigned dropped_extents = 0; | |
3962 | ||
3963 | spin_lock(&BTRFS_I(inode)->lock); | |
3964 | BUG_ON(!BTRFS_I(inode)->outstanding_extents); | |
3965 | BTRFS_I(inode)->outstanding_extents--; | |
3966 | ||
3967 | /* | |
3968 | * If we have more or the same amount of outsanding extents than we have | |
3969 | * reserved then we need to leave the reserved extents count alone. | |
3970 | */ | |
3971 | if (BTRFS_I(inode)->outstanding_extents >= | |
3972 | BTRFS_I(inode)->reserved_extents) | |
3973 | goto out; | |
3974 | ||
3975 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
3976 | BTRFS_I(inode)->outstanding_extents; | |
3977 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
3978 | out: | |
3979 | spin_unlock(&BTRFS_I(inode)->lock); | |
3980 | return dropped_extents; | |
3981 | } | |
3982 | ||
0ca1f7ce | 3983 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes) |
6324fbf3 | 3984 | { |
0ca1f7ce YZ |
3985 | return num_bytes >>= 3; |
3986 | } | |
c146afad | 3987 | |
0ca1f7ce YZ |
3988 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
3989 | { | |
3990 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3991 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 JB |
3992 | u64 to_reserve = 0; |
3993 | unsigned nr_extents = 0; | |
0ca1f7ce | 3994 | int ret; |
6324fbf3 | 3995 | |
0ca1f7ce YZ |
3996 | if (btrfs_transaction_in_commit(root->fs_info)) |
3997 | schedule_timeout(1); | |
ec44a35c | 3998 | |
0ca1f7ce | 3999 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 4000 | |
9e0baf60 JB |
4001 | spin_lock(&BTRFS_I(inode)->lock); |
4002 | BTRFS_I(inode)->outstanding_extents++; | |
4003 | ||
4004 | if (BTRFS_I(inode)->outstanding_extents > | |
4005 | BTRFS_I(inode)->reserved_extents) { | |
4006 | nr_extents = BTRFS_I(inode)->outstanding_extents - | |
4007 | BTRFS_I(inode)->reserved_extents; | |
4008 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
57a45ced | 4009 | |
16cdcec7 | 4010 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); |
593060d7 | 4011 | } |
9e0baf60 | 4012 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 4013 | |
0ca1f7ce | 4014 | to_reserve += calc_csum_metadata_size(inode, num_bytes); |
8bb8ab2e | 4015 | ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1); |
9e0baf60 JB |
4016 | if (ret) { |
4017 | unsigned dropped; | |
4018 | /* | |
4019 | * We don't need the return value since our reservation failed, | |
4020 | * we just need to clean up our counter. | |
4021 | */ | |
4022 | dropped = drop_outstanding_extent(inode); | |
4023 | WARN_ON(dropped > 1); | |
0ca1f7ce | 4024 | return ret; |
9e0baf60 | 4025 | } |
25179201 | 4026 | |
0ca1f7ce YZ |
4027 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
4028 | ||
0ca1f7ce YZ |
4029 | return 0; |
4030 | } | |
4031 | ||
4032 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) | |
4033 | { | |
4034 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
4035 | u64 to_free = 0; |
4036 | unsigned dropped; | |
0ca1f7ce YZ |
4037 | |
4038 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
9e0baf60 | 4039 | dropped = drop_outstanding_extent(inode); |
97e728d4 | 4040 | |
0ca1f7ce | 4041 | to_free = calc_csum_metadata_size(inode, num_bytes); |
9e0baf60 JB |
4042 | if (dropped > 0) |
4043 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce YZ |
4044 | |
4045 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, | |
4046 | to_free); | |
4047 | } | |
4048 | ||
4049 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes) | |
4050 | { | |
4051 | int ret; | |
4052 | ||
4053 | ret = btrfs_check_data_free_space(inode, num_bytes); | |
d397712b | 4054 | if (ret) |
0ca1f7ce YZ |
4055 | return ret; |
4056 | ||
4057 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes); | |
4058 | if (ret) { | |
4059 | btrfs_free_reserved_data_space(inode, num_bytes); | |
4060 | return ret; | |
4061 | } | |
4062 | ||
4063 | return 0; | |
4064 | } | |
4065 | ||
4066 | void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes) | |
4067 | { | |
4068 | btrfs_delalloc_release_metadata(inode, num_bytes); | |
4069 | btrfs_free_reserved_data_space(inode, num_bytes); | |
6324fbf3 CM |
4070 | } |
4071 | ||
9078a3e1 CM |
4072 | static int update_block_group(struct btrfs_trans_handle *trans, |
4073 | struct btrfs_root *root, | |
f0486c68 | 4074 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 4075 | { |
0af3d00b | 4076 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 4077 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 4078 | u64 total = num_bytes; |
9078a3e1 | 4079 | u64 old_val; |
db94535d | 4080 | u64 byte_in_group; |
0af3d00b | 4081 | int factor; |
3e1ad54f | 4082 | |
5d4f98a2 YZ |
4083 | /* block accounting for super block */ |
4084 | spin_lock(&info->delalloc_lock); | |
4085 | old_val = btrfs_super_bytes_used(&info->super_copy); | |
4086 | if (alloc) | |
4087 | old_val += num_bytes; | |
4088 | else | |
4089 | old_val -= num_bytes; | |
4090 | btrfs_set_super_bytes_used(&info->super_copy, old_val); | |
5d4f98a2 YZ |
4091 | spin_unlock(&info->delalloc_lock); |
4092 | ||
d397712b | 4093 | while (total) { |
db94535d | 4094 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 4095 | if (!cache) |
9078a3e1 | 4096 | return -1; |
b742bb82 YZ |
4097 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
4098 | BTRFS_BLOCK_GROUP_RAID1 | | |
4099 | BTRFS_BLOCK_GROUP_RAID10)) | |
4100 | factor = 2; | |
4101 | else | |
4102 | factor = 1; | |
9d66e233 JB |
4103 | /* |
4104 | * If this block group has free space cache written out, we | |
4105 | * need to make sure to load it if we are removing space. This | |
4106 | * is because we need the unpinning stage to actually add the | |
4107 | * space back to the block group, otherwise we will leak space. | |
4108 | */ | |
4109 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
b8399dee | 4110 | cache_block_group(cache, trans, NULL, 1); |
0af3d00b | 4111 | |
db94535d CM |
4112 | byte_in_group = bytenr - cache->key.objectid; |
4113 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 4114 | |
25179201 | 4115 | spin_lock(&cache->space_info->lock); |
c286ac48 | 4116 | spin_lock(&cache->lock); |
0af3d00b JB |
4117 | |
4118 | if (btrfs_super_cache_generation(&info->super_copy) != 0 && | |
4119 | cache->disk_cache_state < BTRFS_DC_CLEAR) | |
4120 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
4121 | ||
0f9dd46c | 4122 | cache->dirty = 1; |
9078a3e1 | 4123 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 4124 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 4125 | if (alloc) { |
db94535d | 4126 | old_val += num_bytes; |
11833d66 YZ |
4127 | btrfs_set_block_group_used(&cache->item, old_val); |
4128 | cache->reserved -= num_bytes; | |
11833d66 | 4129 | cache->space_info->bytes_reserved -= num_bytes; |
36e39c40 | 4130 | cache->space_info->reservation_progress++; |
b742bb82 YZ |
4131 | cache->space_info->bytes_used += num_bytes; |
4132 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 4133 | spin_unlock(&cache->lock); |
25179201 | 4134 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 4135 | } else { |
db94535d | 4136 | old_val -= num_bytes; |
c286ac48 | 4137 | btrfs_set_block_group_used(&cache->item, old_val); |
f0486c68 YZ |
4138 | cache->pinned += num_bytes; |
4139 | cache->space_info->bytes_pinned += num_bytes; | |
6324fbf3 | 4140 | cache->space_info->bytes_used -= num_bytes; |
b742bb82 | 4141 | cache->space_info->disk_used -= num_bytes * factor; |
c286ac48 | 4142 | spin_unlock(&cache->lock); |
25179201 | 4143 | spin_unlock(&cache->space_info->lock); |
1f3c79a2 | 4144 | |
f0486c68 YZ |
4145 | set_extent_dirty(info->pinned_extents, |
4146 | bytenr, bytenr + num_bytes - 1, | |
4147 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 4148 | } |
fa9c0d79 | 4149 | btrfs_put_block_group(cache); |
db94535d CM |
4150 | total -= num_bytes; |
4151 | bytenr += num_bytes; | |
9078a3e1 CM |
4152 | } |
4153 | return 0; | |
4154 | } | |
6324fbf3 | 4155 | |
a061fc8d CM |
4156 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
4157 | { | |
0f9dd46c | 4158 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 4159 | u64 bytenr; |
0f9dd46c JB |
4160 | |
4161 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); | |
4162 | if (!cache) | |
a061fc8d | 4163 | return 0; |
0f9dd46c | 4164 | |
d2fb3437 | 4165 | bytenr = cache->key.objectid; |
fa9c0d79 | 4166 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
4167 | |
4168 | return bytenr; | |
a061fc8d CM |
4169 | } |
4170 | ||
f0486c68 YZ |
4171 | static int pin_down_extent(struct btrfs_root *root, |
4172 | struct btrfs_block_group_cache *cache, | |
4173 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 4174 | { |
11833d66 YZ |
4175 | spin_lock(&cache->space_info->lock); |
4176 | spin_lock(&cache->lock); | |
4177 | cache->pinned += num_bytes; | |
4178 | cache->space_info->bytes_pinned += num_bytes; | |
4179 | if (reserved) { | |
4180 | cache->reserved -= num_bytes; | |
4181 | cache->space_info->bytes_reserved -= num_bytes; | |
36e39c40 | 4182 | cache->space_info->reservation_progress++; |
11833d66 YZ |
4183 | } |
4184 | spin_unlock(&cache->lock); | |
4185 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 4186 | |
f0486c68 YZ |
4187 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
4188 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
4189 | return 0; | |
4190 | } | |
68b38550 | 4191 | |
f0486c68 YZ |
4192 | /* |
4193 | * this function must be called within transaction | |
4194 | */ | |
4195 | int btrfs_pin_extent(struct btrfs_root *root, | |
4196 | u64 bytenr, u64 num_bytes, int reserved) | |
4197 | { | |
4198 | struct btrfs_block_group_cache *cache; | |
68b38550 | 4199 | |
f0486c68 YZ |
4200 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
4201 | BUG_ON(!cache); | |
4202 | ||
4203 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
4204 | ||
4205 | btrfs_put_block_group(cache); | |
11833d66 YZ |
4206 | return 0; |
4207 | } | |
4208 | ||
f0486c68 YZ |
4209 | /* |
4210 | * update size of reserved extents. this function may return -EAGAIN | |
4211 | * if 'reserve' is true or 'sinfo' is false. | |
4212 | */ | |
b4d00d56 LD |
4213 | int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
4214 | u64 num_bytes, int reserve, int sinfo) | |
11833d66 | 4215 | { |
f0486c68 YZ |
4216 | int ret = 0; |
4217 | if (sinfo) { | |
4218 | struct btrfs_space_info *space_info = cache->space_info; | |
4219 | spin_lock(&space_info->lock); | |
4220 | spin_lock(&cache->lock); | |
4221 | if (reserve) { | |
4222 | if (cache->ro) { | |
4223 | ret = -EAGAIN; | |
4224 | } else { | |
4225 | cache->reserved += num_bytes; | |
4226 | space_info->bytes_reserved += num_bytes; | |
4227 | } | |
4228 | } else { | |
4229 | if (cache->ro) | |
4230 | space_info->bytes_readonly += num_bytes; | |
4231 | cache->reserved -= num_bytes; | |
4232 | space_info->bytes_reserved -= num_bytes; | |
36e39c40 | 4233 | space_info->reservation_progress++; |
f0486c68 YZ |
4234 | } |
4235 | spin_unlock(&cache->lock); | |
4236 | spin_unlock(&space_info->lock); | |
11833d66 | 4237 | } else { |
f0486c68 YZ |
4238 | spin_lock(&cache->lock); |
4239 | if (cache->ro) { | |
4240 | ret = -EAGAIN; | |
4241 | } else { | |
4242 | if (reserve) | |
4243 | cache->reserved += num_bytes; | |
4244 | else | |
4245 | cache->reserved -= num_bytes; | |
4246 | } | |
4247 | spin_unlock(&cache->lock); | |
324ae4df | 4248 | } |
f0486c68 | 4249 | return ret; |
324ae4df | 4250 | } |
9078a3e1 | 4251 | |
11833d66 YZ |
4252 | int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
4253 | struct btrfs_root *root) | |
e8569813 | 4254 | { |
e8569813 | 4255 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
4256 | struct btrfs_caching_control *next; |
4257 | struct btrfs_caching_control *caching_ctl; | |
4258 | struct btrfs_block_group_cache *cache; | |
e8569813 | 4259 | |
11833d66 | 4260 | down_write(&fs_info->extent_commit_sem); |
25179201 | 4261 | |
11833d66 YZ |
4262 | list_for_each_entry_safe(caching_ctl, next, |
4263 | &fs_info->caching_block_groups, list) { | |
4264 | cache = caching_ctl->block_group; | |
4265 | if (block_group_cache_done(cache)) { | |
4266 | cache->last_byte_to_unpin = (u64)-1; | |
4267 | list_del_init(&caching_ctl->list); | |
4268 | put_caching_control(caching_ctl); | |
e8569813 | 4269 | } else { |
11833d66 | 4270 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 4271 | } |
e8569813 | 4272 | } |
11833d66 YZ |
4273 | |
4274 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
4275 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
4276 | else | |
4277 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
4278 | ||
4279 | up_write(&fs_info->extent_commit_sem); | |
8929ecfa YZ |
4280 | |
4281 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
4282 | return 0; |
4283 | } | |
4284 | ||
11833d66 | 4285 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
ccd467d6 | 4286 | { |
11833d66 YZ |
4287 | struct btrfs_fs_info *fs_info = root->fs_info; |
4288 | struct btrfs_block_group_cache *cache = NULL; | |
4289 | u64 len; | |
ccd467d6 | 4290 | |
11833d66 YZ |
4291 | while (start <= end) { |
4292 | if (!cache || | |
4293 | start >= cache->key.objectid + cache->key.offset) { | |
4294 | if (cache) | |
4295 | btrfs_put_block_group(cache); | |
4296 | cache = btrfs_lookup_block_group(fs_info, start); | |
4297 | BUG_ON(!cache); | |
4298 | } | |
4299 | ||
4300 | len = cache->key.objectid + cache->key.offset - start; | |
4301 | len = min(len, end + 1 - start); | |
4302 | ||
4303 | if (start < cache->last_byte_to_unpin) { | |
4304 | len = min(len, cache->last_byte_to_unpin - start); | |
4305 | btrfs_add_free_space(cache, start, len); | |
4306 | } | |
4307 | ||
f0486c68 YZ |
4308 | start += len; |
4309 | ||
11833d66 YZ |
4310 | spin_lock(&cache->space_info->lock); |
4311 | spin_lock(&cache->lock); | |
4312 | cache->pinned -= len; | |
4313 | cache->space_info->bytes_pinned -= len; | |
f0486c68 YZ |
4314 | if (cache->ro) { |
4315 | cache->space_info->bytes_readonly += len; | |
4316 | } else if (cache->reserved_pinned > 0) { | |
4317 | len = min(len, cache->reserved_pinned); | |
4318 | cache->reserved_pinned -= len; | |
4319 | cache->space_info->bytes_reserved += len; | |
4320 | } | |
11833d66 YZ |
4321 | spin_unlock(&cache->lock); |
4322 | spin_unlock(&cache->space_info->lock); | |
ccd467d6 | 4323 | } |
11833d66 YZ |
4324 | |
4325 | if (cache) | |
4326 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
4327 | return 0; |
4328 | } | |
4329 | ||
4330 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 4331 | struct btrfs_root *root) |
a28ec197 | 4332 | { |
11833d66 YZ |
4333 | struct btrfs_fs_info *fs_info = root->fs_info; |
4334 | struct extent_io_tree *unpin; | |
f0486c68 YZ |
4335 | struct btrfs_block_rsv *block_rsv; |
4336 | struct btrfs_block_rsv *next_rsv; | |
1a5bc167 CM |
4337 | u64 start; |
4338 | u64 end; | |
f0486c68 | 4339 | int idx; |
a28ec197 | 4340 | int ret; |
a28ec197 | 4341 | |
11833d66 YZ |
4342 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
4343 | unpin = &fs_info->freed_extents[1]; | |
4344 | else | |
4345 | unpin = &fs_info->freed_extents[0]; | |
4346 | ||
d397712b | 4347 | while (1) { |
1a5bc167 CM |
4348 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
4349 | EXTENT_DIRTY); | |
4350 | if (ret) | |
a28ec197 | 4351 | break; |
1f3c79a2 | 4352 | |
5378e607 LD |
4353 | if (btrfs_test_opt(root, DISCARD)) |
4354 | ret = btrfs_discard_extent(root, start, | |
4355 | end + 1 - start, NULL); | |
1f3c79a2 | 4356 | |
1a5bc167 | 4357 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
11833d66 | 4358 | unpin_extent_range(root, start, end); |
b9473439 | 4359 | cond_resched(); |
a28ec197 | 4360 | } |
817d52f8 | 4361 | |
f0486c68 YZ |
4362 | mutex_lock(&fs_info->durable_block_rsv_mutex); |
4363 | list_for_each_entry_safe(block_rsv, next_rsv, | |
4364 | &fs_info->durable_block_rsv_list, list) { | |
444528b3 | 4365 | |
f0486c68 YZ |
4366 | idx = trans->transid & 0x1; |
4367 | if (block_rsv->freed[idx] > 0) { | |
4368 | block_rsv_add_bytes(block_rsv, | |
4369 | block_rsv->freed[idx], 0); | |
4370 | block_rsv->freed[idx] = 0; | |
4371 | } | |
4372 | if (atomic_read(&block_rsv->usage) == 0) { | |
4373 | btrfs_block_rsv_release(root, block_rsv, (u64)-1); | |
31840ae1 | 4374 | |
f0486c68 YZ |
4375 | if (block_rsv->freed[0] == 0 && |
4376 | block_rsv->freed[1] == 0) { | |
4377 | list_del_init(&block_rsv->list); | |
4378 | kfree(block_rsv); | |
4379 | } | |
4380 | } else { | |
4381 | btrfs_block_rsv_release(root, block_rsv, 0); | |
8ef97622 | 4382 | } |
f4b9aa8d | 4383 | } |
f0486c68 | 4384 | mutex_unlock(&fs_info->durable_block_rsv_mutex); |
31840ae1 | 4385 | |
e20d96d6 CM |
4386 | return 0; |
4387 | } | |
4388 | ||
5d4f98a2 YZ |
4389 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
4390 | struct btrfs_root *root, | |
4391 | u64 bytenr, u64 num_bytes, u64 parent, | |
4392 | u64 root_objectid, u64 owner_objectid, | |
4393 | u64 owner_offset, int refs_to_drop, | |
4394 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 4395 | { |
e2fa7227 | 4396 | struct btrfs_key key; |
5d4f98a2 | 4397 | struct btrfs_path *path; |
1261ec42 CM |
4398 | struct btrfs_fs_info *info = root->fs_info; |
4399 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 4400 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
4401 | struct btrfs_extent_item *ei; |
4402 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 4403 | int ret; |
5d4f98a2 | 4404 | int is_data; |
952fccac CM |
4405 | int extent_slot = 0; |
4406 | int found_extent = 0; | |
4407 | int num_to_del = 1; | |
5d4f98a2 YZ |
4408 | u32 item_size; |
4409 | u64 refs; | |
037e6390 | 4410 | |
5caf2a00 | 4411 | path = btrfs_alloc_path(); |
54aa1f4d CM |
4412 | if (!path) |
4413 | return -ENOMEM; | |
5f26f772 | 4414 | |
3c12ac72 | 4415 | path->reada = 1; |
b9473439 | 4416 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4417 | |
4418 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
4419 | BUG_ON(!is_data && refs_to_drop != 1); | |
4420 | ||
4421 | ret = lookup_extent_backref(trans, extent_root, path, &iref, | |
4422 | bytenr, num_bytes, parent, | |
4423 | root_objectid, owner_objectid, | |
4424 | owner_offset); | |
7bb86316 | 4425 | if (ret == 0) { |
952fccac | 4426 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
4427 | while (extent_slot >= 0) { |
4428 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 4429 | extent_slot); |
5d4f98a2 | 4430 | if (key.objectid != bytenr) |
952fccac | 4431 | break; |
5d4f98a2 YZ |
4432 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
4433 | key.offset == num_bytes) { | |
952fccac CM |
4434 | found_extent = 1; |
4435 | break; | |
4436 | } | |
4437 | if (path->slots[0] - extent_slot > 5) | |
4438 | break; | |
5d4f98a2 | 4439 | extent_slot--; |
952fccac | 4440 | } |
5d4f98a2 YZ |
4441 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
4442 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
4443 | if (found_extent && item_size < sizeof(*ei)) | |
4444 | found_extent = 0; | |
4445 | #endif | |
31840ae1 | 4446 | if (!found_extent) { |
5d4f98a2 | 4447 | BUG_ON(iref); |
56bec294 | 4448 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
4449 | NULL, refs_to_drop, |
4450 | is_data); | |
31840ae1 | 4451 | BUG_ON(ret); |
b3b4aa74 | 4452 | btrfs_release_path(path); |
b9473439 | 4453 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4454 | |
4455 | key.objectid = bytenr; | |
4456 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
4457 | key.offset = num_bytes; | |
4458 | ||
31840ae1 ZY |
4459 | ret = btrfs_search_slot(trans, extent_root, |
4460 | &key, path, -1, 1); | |
f3465ca4 JB |
4461 | if (ret) { |
4462 | printk(KERN_ERR "umm, got %d back from search" | |
d397712b CM |
4463 | ", was looking for %llu\n", ret, |
4464 | (unsigned long long)bytenr); | |
f3465ca4 JB |
4465 | btrfs_print_leaf(extent_root, path->nodes[0]); |
4466 | } | |
31840ae1 ZY |
4467 | BUG_ON(ret); |
4468 | extent_slot = path->slots[0]; | |
4469 | } | |
7bb86316 CM |
4470 | } else { |
4471 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
4472 | WARN_ON(1); | |
d397712b | 4473 | printk(KERN_ERR "btrfs unable to find ref byte nr %llu " |
5d4f98a2 | 4474 | "parent %llu root %llu owner %llu offset %llu\n", |
d397712b | 4475 | (unsigned long long)bytenr, |
56bec294 | 4476 | (unsigned long long)parent, |
d397712b | 4477 | (unsigned long long)root_objectid, |
5d4f98a2 YZ |
4478 | (unsigned long long)owner_objectid, |
4479 | (unsigned long long)owner_offset); | |
7bb86316 | 4480 | } |
5f39d397 CM |
4481 | |
4482 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
4483 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
4484 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
4485 | if (item_size < sizeof(*ei)) { | |
4486 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
4487 | ret = convert_extent_item_v0(trans, extent_root, path, | |
4488 | owner_objectid, 0); | |
4489 | BUG_ON(ret < 0); | |
4490 | ||
b3b4aa74 | 4491 | btrfs_release_path(path); |
5d4f98a2 YZ |
4492 | path->leave_spinning = 1; |
4493 | ||
4494 | key.objectid = bytenr; | |
4495 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
4496 | key.offset = num_bytes; | |
4497 | ||
4498 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
4499 | -1, 1); | |
4500 | if (ret) { | |
4501 | printk(KERN_ERR "umm, got %d back from search" | |
4502 | ", was looking for %llu\n", ret, | |
4503 | (unsigned long long)bytenr); | |
4504 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
4505 | } | |
4506 | BUG_ON(ret); | |
4507 | extent_slot = path->slots[0]; | |
4508 | leaf = path->nodes[0]; | |
4509 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
4510 | } | |
4511 | #endif | |
4512 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 4513 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 4514 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4515 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { |
4516 | struct btrfs_tree_block_info *bi; | |
4517 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
4518 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
4519 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
4520 | } | |
56bec294 | 4521 | |
5d4f98a2 | 4522 | refs = btrfs_extent_refs(leaf, ei); |
56bec294 CM |
4523 | BUG_ON(refs < refs_to_drop); |
4524 | refs -= refs_to_drop; | |
5f39d397 | 4525 | |
5d4f98a2 YZ |
4526 | if (refs > 0) { |
4527 | if (extent_op) | |
4528 | __run_delayed_extent_op(extent_op, leaf, ei); | |
4529 | /* | |
4530 | * In the case of inline back ref, reference count will | |
4531 | * be updated by remove_extent_backref | |
952fccac | 4532 | */ |
5d4f98a2 YZ |
4533 | if (iref) { |
4534 | BUG_ON(!found_extent); | |
4535 | } else { | |
4536 | btrfs_set_extent_refs(leaf, ei, refs); | |
4537 | btrfs_mark_buffer_dirty(leaf); | |
4538 | } | |
4539 | if (found_extent) { | |
4540 | ret = remove_extent_backref(trans, extent_root, path, | |
4541 | iref, refs_to_drop, | |
4542 | is_data); | |
952fccac CM |
4543 | BUG_ON(ret); |
4544 | } | |
5d4f98a2 | 4545 | } else { |
5d4f98a2 YZ |
4546 | if (found_extent) { |
4547 | BUG_ON(is_data && refs_to_drop != | |
4548 | extent_data_ref_count(root, path, iref)); | |
4549 | if (iref) { | |
4550 | BUG_ON(path->slots[0] != extent_slot); | |
4551 | } else { | |
4552 | BUG_ON(path->slots[0] != extent_slot + 1); | |
4553 | path->slots[0] = extent_slot; | |
4554 | num_to_del = 2; | |
4555 | } | |
78fae27e | 4556 | } |
b9473439 | 4557 | |
952fccac CM |
4558 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
4559 | num_to_del); | |
31840ae1 | 4560 | BUG_ON(ret); |
b3b4aa74 | 4561 | btrfs_release_path(path); |
21af804c | 4562 | |
5d4f98a2 | 4563 | if (is_data) { |
459931ec CM |
4564 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
4565 | BUG_ON(ret); | |
d57e62b8 CM |
4566 | } else { |
4567 | invalidate_mapping_pages(info->btree_inode->i_mapping, | |
4568 | bytenr >> PAGE_CACHE_SHIFT, | |
4569 | (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT); | |
459931ec CM |
4570 | } |
4571 | ||
f0486c68 | 4572 | ret = update_block_group(trans, root, bytenr, num_bytes, 0); |
dcbdd4dc | 4573 | BUG_ON(ret); |
a28ec197 | 4574 | } |
5caf2a00 | 4575 | btrfs_free_path(path); |
a28ec197 CM |
4576 | return ret; |
4577 | } | |
4578 | ||
1887be66 | 4579 | /* |
f0486c68 | 4580 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
4581 | * delayed ref for that extent as well. This searches the delayed ref tree for |
4582 | * a given extent, and if there are no other delayed refs to be processed, it | |
4583 | * removes it from the tree. | |
4584 | */ | |
4585 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
4586 | struct btrfs_root *root, u64 bytenr) | |
4587 | { | |
4588 | struct btrfs_delayed_ref_head *head; | |
4589 | struct btrfs_delayed_ref_root *delayed_refs; | |
4590 | struct btrfs_delayed_ref_node *ref; | |
4591 | struct rb_node *node; | |
f0486c68 | 4592 | int ret = 0; |
1887be66 CM |
4593 | |
4594 | delayed_refs = &trans->transaction->delayed_refs; | |
4595 | spin_lock(&delayed_refs->lock); | |
4596 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
4597 | if (!head) | |
4598 | goto out; | |
4599 | ||
4600 | node = rb_prev(&head->node.rb_node); | |
4601 | if (!node) | |
4602 | goto out; | |
4603 | ||
4604 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
4605 | ||
4606 | /* there are still entries for this ref, we can't drop it */ | |
4607 | if (ref->bytenr == bytenr) | |
4608 | goto out; | |
4609 | ||
5d4f98a2 YZ |
4610 | if (head->extent_op) { |
4611 | if (!head->must_insert_reserved) | |
4612 | goto out; | |
4613 | kfree(head->extent_op); | |
4614 | head->extent_op = NULL; | |
4615 | } | |
4616 | ||
1887be66 CM |
4617 | /* |
4618 | * waiting for the lock here would deadlock. If someone else has it | |
4619 | * locked they are already in the process of dropping it anyway | |
4620 | */ | |
4621 | if (!mutex_trylock(&head->mutex)) | |
4622 | goto out; | |
4623 | ||
4624 | /* | |
4625 | * at this point we have a head with no other entries. Go | |
4626 | * ahead and process it. | |
4627 | */ | |
4628 | head->node.in_tree = 0; | |
4629 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 4630 | |
1887be66 CM |
4631 | delayed_refs->num_entries--; |
4632 | ||
4633 | /* | |
4634 | * we don't take a ref on the node because we're removing it from the | |
4635 | * tree, so we just steal the ref the tree was holding. | |
4636 | */ | |
c3e69d58 CM |
4637 | delayed_refs->num_heads--; |
4638 | if (list_empty(&head->cluster)) | |
4639 | delayed_refs->num_heads_ready--; | |
4640 | ||
4641 | list_del_init(&head->cluster); | |
1887be66 CM |
4642 | spin_unlock(&delayed_refs->lock); |
4643 | ||
f0486c68 YZ |
4644 | BUG_ON(head->extent_op); |
4645 | if (head->must_insert_reserved) | |
4646 | ret = 1; | |
4647 | ||
4648 | mutex_unlock(&head->mutex); | |
1887be66 | 4649 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 4650 | return ret; |
1887be66 CM |
4651 | out: |
4652 | spin_unlock(&delayed_refs->lock); | |
4653 | return 0; | |
4654 | } | |
4655 | ||
f0486c68 YZ |
4656 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
4657 | struct btrfs_root *root, | |
4658 | struct extent_buffer *buf, | |
4659 | u64 parent, int last_ref) | |
4660 | { | |
4661 | struct btrfs_block_rsv *block_rsv; | |
4662 | struct btrfs_block_group_cache *cache = NULL; | |
4663 | int ret; | |
4664 | ||
4665 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
4666 | ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len, | |
4667 | parent, root->root_key.objectid, | |
4668 | btrfs_header_level(buf), | |
4669 | BTRFS_DROP_DELAYED_REF, NULL); | |
4670 | BUG_ON(ret); | |
4671 | } | |
4672 | ||
4673 | if (!last_ref) | |
4674 | return; | |
4675 | ||
4676 | block_rsv = get_block_rsv(trans, root); | |
4677 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); | |
3bf84a5a YZ |
4678 | if (block_rsv->space_info != cache->space_info) |
4679 | goto out; | |
f0486c68 YZ |
4680 | |
4681 | if (btrfs_header_generation(buf) == trans->transid) { | |
4682 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
4683 | ret = check_ref_cleanup(trans, root, buf->start); | |
4684 | if (!ret) | |
4685 | goto pin; | |
4686 | } | |
4687 | ||
4688 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
4689 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
4690 | goto pin; | |
4691 | } | |
4692 | ||
4693 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
4694 | ||
4695 | btrfs_add_free_space(cache, buf->start, buf->len); | |
b4d00d56 | 4696 | ret = btrfs_update_reserved_bytes(cache, buf->len, 0, 0); |
f0486c68 YZ |
4697 | if (ret == -EAGAIN) { |
4698 | /* block group became read-only */ | |
b4d00d56 | 4699 | btrfs_update_reserved_bytes(cache, buf->len, 0, 1); |
f0486c68 YZ |
4700 | goto out; |
4701 | } | |
4702 | ||
4703 | ret = 1; | |
4704 | spin_lock(&block_rsv->lock); | |
4705 | if (block_rsv->reserved < block_rsv->size) { | |
4706 | block_rsv->reserved += buf->len; | |
4707 | ret = 0; | |
4708 | } | |
4709 | spin_unlock(&block_rsv->lock); | |
4710 | ||
4711 | if (ret) { | |
4712 | spin_lock(&cache->space_info->lock); | |
4713 | cache->space_info->bytes_reserved -= buf->len; | |
36e39c40 | 4714 | cache->space_info->reservation_progress++; |
f0486c68 YZ |
4715 | spin_unlock(&cache->space_info->lock); |
4716 | } | |
4717 | goto out; | |
4718 | } | |
4719 | pin: | |
4720 | if (block_rsv->durable && !cache->ro) { | |
4721 | ret = 0; | |
4722 | spin_lock(&cache->lock); | |
4723 | if (!cache->ro) { | |
4724 | cache->reserved_pinned += buf->len; | |
4725 | ret = 1; | |
4726 | } | |
4727 | spin_unlock(&cache->lock); | |
4728 | ||
4729 | if (ret) { | |
4730 | spin_lock(&block_rsv->lock); | |
4731 | block_rsv->freed[trans->transid & 0x1] += buf->len; | |
4732 | spin_unlock(&block_rsv->lock); | |
4733 | } | |
4734 | } | |
4735 | out: | |
a826d6dc JB |
4736 | /* |
4737 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
4738 | * anymore. | |
4739 | */ | |
4740 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
4741 | btrfs_put_block_group(cache); |
4742 | } | |
4743 | ||
925baedd | 4744 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
31840ae1 ZY |
4745 | struct btrfs_root *root, |
4746 | u64 bytenr, u64 num_bytes, u64 parent, | |
5d4f98a2 | 4747 | u64 root_objectid, u64 owner, u64 offset) |
925baedd CM |
4748 | { |
4749 | int ret; | |
4750 | ||
56bec294 CM |
4751 | /* |
4752 | * tree log blocks never actually go into the extent allocation | |
4753 | * tree, just update pinning info and exit early. | |
56bec294 | 4754 | */ |
5d4f98a2 YZ |
4755 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
4756 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 4757 | /* unlocks the pinned mutex */ |
11833d66 | 4758 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 4759 | ret = 0; |
5d4f98a2 YZ |
4760 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
4761 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes, | |
4762 | parent, root_objectid, (int)owner, | |
4763 | BTRFS_DROP_DELAYED_REF, NULL); | |
1887be66 | 4764 | BUG_ON(ret); |
5d4f98a2 YZ |
4765 | } else { |
4766 | ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes, | |
4767 | parent, root_objectid, owner, | |
4768 | offset, BTRFS_DROP_DELAYED_REF, NULL); | |
4769 | BUG_ON(ret); | |
56bec294 | 4770 | } |
925baedd CM |
4771 | return ret; |
4772 | } | |
4773 | ||
87ee04eb CM |
4774 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
4775 | { | |
4776 | u64 mask = ((u64)root->stripesize - 1); | |
4777 | u64 ret = (val + mask) & ~mask; | |
4778 | return ret; | |
4779 | } | |
4780 | ||
817d52f8 JB |
4781 | /* |
4782 | * when we wait for progress in the block group caching, its because | |
4783 | * our allocation attempt failed at least once. So, we must sleep | |
4784 | * and let some progress happen before we try again. | |
4785 | * | |
4786 | * This function will sleep at least once waiting for new free space to | |
4787 | * show up, and then it will check the block group free space numbers | |
4788 | * for our min num_bytes. Another option is to have it go ahead | |
4789 | * and look in the rbtree for a free extent of a given size, but this | |
4790 | * is a good start. | |
4791 | */ | |
4792 | static noinline int | |
4793 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
4794 | u64 num_bytes) | |
4795 | { | |
11833d66 | 4796 | struct btrfs_caching_control *caching_ctl; |
817d52f8 JB |
4797 | DEFINE_WAIT(wait); |
4798 | ||
11833d66 YZ |
4799 | caching_ctl = get_caching_control(cache); |
4800 | if (!caching_ctl) | |
817d52f8 | 4801 | return 0; |
817d52f8 | 4802 | |
11833d66 | 4803 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 4804 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
4805 | |
4806 | put_caching_control(caching_ctl); | |
4807 | return 0; | |
4808 | } | |
4809 | ||
4810 | static noinline int | |
4811 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
4812 | { | |
4813 | struct btrfs_caching_control *caching_ctl; | |
4814 | DEFINE_WAIT(wait); | |
4815 | ||
4816 | caching_ctl = get_caching_control(cache); | |
4817 | if (!caching_ctl) | |
4818 | return 0; | |
4819 | ||
4820 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
4821 | ||
4822 | put_caching_control(caching_ctl); | |
817d52f8 JB |
4823 | return 0; |
4824 | } | |
4825 | ||
b742bb82 YZ |
4826 | static int get_block_group_index(struct btrfs_block_group_cache *cache) |
4827 | { | |
4828 | int index; | |
4829 | if (cache->flags & BTRFS_BLOCK_GROUP_RAID10) | |
4830 | index = 0; | |
4831 | else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1) | |
4832 | index = 1; | |
4833 | else if (cache->flags & BTRFS_BLOCK_GROUP_DUP) | |
4834 | index = 2; | |
4835 | else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0) | |
4836 | index = 3; | |
4837 | else | |
4838 | index = 4; | |
4839 | return index; | |
4840 | } | |
4841 | ||
817d52f8 | 4842 | enum btrfs_loop_type { |
ccf0e725 | 4843 | LOOP_FIND_IDEAL = 0, |
817d52f8 JB |
4844 | LOOP_CACHING_NOWAIT = 1, |
4845 | LOOP_CACHING_WAIT = 2, | |
4846 | LOOP_ALLOC_CHUNK = 3, | |
4847 | LOOP_NO_EMPTY_SIZE = 4, | |
4848 | }; | |
4849 | ||
fec577fb CM |
4850 | /* |
4851 | * walks the btree of allocated extents and find a hole of a given size. | |
4852 | * The key ins is changed to record the hole: | |
4853 | * ins->objectid == block start | |
62e2749e | 4854 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
4855 | * ins->offset == number of blocks |
4856 | * Any available blocks before search_start are skipped. | |
4857 | */ | |
d397712b | 4858 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
4859 | struct btrfs_root *orig_root, |
4860 | u64 num_bytes, u64 empty_size, | |
4861 | u64 search_start, u64 search_end, | |
4862 | u64 hint_byte, struct btrfs_key *ins, | |
e0f54067 | 4863 | u64 data) |
fec577fb | 4864 | { |
80eb234a | 4865 | int ret = 0; |
d397712b | 4866 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 4867 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 4868 | struct btrfs_block_group_cache *block_group = NULL; |
239b14b3 | 4869 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 4870 | int allowed_chunk_alloc = 0; |
ccf0e725 | 4871 | int done_chunk_alloc = 0; |
80eb234a | 4872 | struct btrfs_space_info *space_info; |
fa9c0d79 CM |
4873 | int last_ptr_loop = 0; |
4874 | int loop = 0; | |
f0486c68 | 4875 | int index = 0; |
817d52f8 | 4876 | bool found_uncached_bg = false; |
0a24325e | 4877 | bool failed_cluster_refill = false; |
1cdda9b8 | 4878 | bool failed_alloc = false; |
67377734 | 4879 | bool use_cluster = true; |
ccf0e725 JB |
4880 | u64 ideal_cache_percent = 0; |
4881 | u64 ideal_cache_offset = 0; | |
fec577fb | 4882 | |
db94535d | 4883 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 4884 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
4885 | ins->objectid = 0; |
4886 | ins->offset = 0; | |
b1a4d965 | 4887 | |
2552d17e | 4888 | space_info = __find_space_info(root->fs_info, data); |
1b1d1f66 | 4889 | if (!space_info) { |
e0f54067 | 4890 | printk(KERN_ERR "No space info for %llu\n", data); |
1b1d1f66 JB |
4891 | return -ENOSPC; |
4892 | } | |
2552d17e | 4893 | |
67377734 JB |
4894 | /* |
4895 | * If the space info is for both data and metadata it means we have a | |
4896 | * small filesystem and we can't use the clustering stuff. | |
4897 | */ | |
4898 | if (btrfs_mixed_space_info(space_info)) | |
4899 | use_cluster = false; | |
4900 | ||
0ef3e66b CM |
4901 | if (orig_root->ref_cows || empty_size) |
4902 | allowed_chunk_alloc = 1; | |
4903 | ||
67377734 | 4904 | if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) { |
fa9c0d79 | 4905 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
4906 | if (!btrfs_test_opt(root, SSD)) |
4907 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
4908 | } |
4909 | ||
67377734 JB |
4910 | if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster && |
4911 | btrfs_test_opt(root, SSD)) { | |
fa9c0d79 CM |
4912 | last_ptr = &root->fs_info->data_alloc_cluster; |
4913 | } | |
0f9dd46c | 4914 | |
239b14b3 | 4915 | if (last_ptr) { |
fa9c0d79 CM |
4916 | spin_lock(&last_ptr->lock); |
4917 | if (last_ptr->block_group) | |
4918 | hint_byte = last_ptr->window_start; | |
4919 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 4920 | } |
fa9c0d79 | 4921 | |
a061fc8d | 4922 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 4923 | search_start = max(search_start, hint_byte); |
0b86a832 | 4924 | |
817d52f8 | 4925 | if (!last_ptr) |
fa9c0d79 | 4926 | empty_cluster = 0; |
fa9c0d79 | 4927 | |
2552d17e | 4928 | if (search_start == hint_byte) { |
ccf0e725 | 4929 | ideal_cache: |
2552d17e JB |
4930 | block_group = btrfs_lookup_block_group(root->fs_info, |
4931 | search_start); | |
817d52f8 JB |
4932 | /* |
4933 | * we don't want to use the block group if it doesn't match our | |
4934 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
4935 | * |
4936 | * However if we are re-searching with an ideal block group | |
4937 | * picked out then we don't care that the block group is cached. | |
817d52f8 JB |
4938 | */ |
4939 | if (block_group && block_group_bits(block_group, data) && | |
ccf0e725 JB |
4940 | (block_group->cached != BTRFS_CACHE_NO || |
4941 | search_start == ideal_cache_offset)) { | |
2552d17e | 4942 | down_read(&space_info->groups_sem); |
44fb5511 CM |
4943 | if (list_empty(&block_group->list) || |
4944 | block_group->ro) { | |
4945 | /* | |
4946 | * someone is removing this block group, | |
4947 | * we can't jump into the have_block_group | |
4948 | * target because our list pointers are not | |
4949 | * valid | |
4950 | */ | |
4951 | btrfs_put_block_group(block_group); | |
4952 | up_read(&space_info->groups_sem); | |
ccf0e725 | 4953 | } else { |
b742bb82 | 4954 | index = get_block_group_index(block_group); |
44fb5511 | 4955 | goto have_block_group; |
ccf0e725 | 4956 | } |
2552d17e | 4957 | } else if (block_group) { |
fa9c0d79 | 4958 | btrfs_put_block_group(block_group); |
2552d17e | 4959 | } |
42e70e7a | 4960 | } |
2552d17e | 4961 | search: |
80eb234a | 4962 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
4963 | list_for_each_entry(block_group, &space_info->block_groups[index], |
4964 | list) { | |
6226cb0a | 4965 | u64 offset; |
817d52f8 | 4966 | int cached; |
8a1413a2 | 4967 | |
11dfe35a | 4968 | btrfs_get_block_group(block_group); |
2552d17e | 4969 | search_start = block_group->key.objectid; |
42e70e7a | 4970 | |
83a50de9 CM |
4971 | /* |
4972 | * this can happen if we end up cycling through all the | |
4973 | * raid types, but we want to make sure we only allocate | |
4974 | * for the proper type. | |
4975 | */ | |
4976 | if (!block_group_bits(block_group, data)) { | |
4977 | u64 extra = BTRFS_BLOCK_GROUP_DUP | | |
4978 | BTRFS_BLOCK_GROUP_RAID1 | | |
4979 | BTRFS_BLOCK_GROUP_RAID10; | |
4980 | ||
4981 | /* | |
4982 | * if they asked for extra copies and this block group | |
4983 | * doesn't provide them, bail. This does allow us to | |
4984 | * fill raid0 from raid1. | |
4985 | */ | |
4986 | if ((data & extra) && !(block_group->flags & extra)) | |
4987 | goto loop; | |
4988 | } | |
4989 | ||
2552d17e | 4990 | have_block_group: |
817d52f8 | 4991 | if (unlikely(block_group->cached == BTRFS_CACHE_NO)) { |
ccf0e725 JB |
4992 | u64 free_percent; |
4993 | ||
b8399dee JB |
4994 | ret = cache_block_group(block_group, trans, |
4995 | orig_root, 1); | |
9d66e233 JB |
4996 | if (block_group->cached == BTRFS_CACHE_FINISHED) |
4997 | goto have_block_group; | |
4998 | ||
ccf0e725 JB |
4999 | free_percent = btrfs_block_group_used(&block_group->item); |
5000 | free_percent *= 100; | |
5001 | free_percent = div64_u64(free_percent, | |
5002 | block_group->key.offset); | |
5003 | free_percent = 100 - free_percent; | |
5004 | if (free_percent > ideal_cache_percent && | |
5005 | likely(!block_group->ro)) { | |
5006 | ideal_cache_offset = block_group->key.objectid; | |
5007 | ideal_cache_percent = free_percent; | |
5008 | } | |
5009 | ||
817d52f8 | 5010 | /* |
bab39bf9 JB |
5011 | * The caching workers are limited to 2 threads, so we |
5012 | * can queue as much work as we care to. | |
817d52f8 | 5013 | */ |
bab39bf9 | 5014 | if (loop > LOOP_FIND_IDEAL) { |
b8399dee JB |
5015 | ret = cache_block_group(block_group, trans, |
5016 | orig_root, 0); | |
817d52f8 | 5017 | BUG_ON(ret); |
2552d17e | 5018 | } |
817d52f8 JB |
5019 | found_uncached_bg = true; |
5020 | ||
ccf0e725 JB |
5021 | /* |
5022 | * If loop is set for cached only, try the next block | |
5023 | * group. | |
5024 | */ | |
5025 | if (loop == LOOP_FIND_IDEAL) | |
817d52f8 JB |
5026 | goto loop; |
5027 | } | |
5028 | ||
ccf0e725 JB |
5029 | cached = block_group_cache_done(block_group); |
5030 | if (unlikely(!cached)) | |
5031 | found_uncached_bg = true; | |
5032 | ||
ea6a478e | 5033 | if (unlikely(block_group->ro)) |
2552d17e | 5034 | goto loop; |
0f9dd46c | 5035 | |
ff5714cc | 5036 | spin_lock(&block_group->free_space_ctl->tree_lock); |
cca1c81f | 5037 | if (cached && |
ff5714cc CM |
5038 | block_group->free_space_ctl->free_space < |
5039 | num_bytes + empty_size) { | |
5040 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
cca1c81f JB |
5041 | goto loop; |
5042 | } | |
ff5714cc | 5043 | spin_unlock(&block_group->free_space_ctl->tree_lock); |
cca1c81f | 5044 | |
0a24325e JB |
5045 | /* |
5046 | * Ok we want to try and use the cluster allocator, so lets look | |
5047 | * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will | |
5048 | * have tried the cluster allocator plenty of times at this | |
5049 | * point and not have found anything, so we are likely way too | |
5050 | * fragmented for the clustering stuff to find anything, so lets | |
5051 | * just skip it and let the allocator find whatever block it can | |
5052 | * find | |
5053 | */ | |
5054 | if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) { | |
fa9c0d79 CM |
5055 | /* |
5056 | * the refill lock keeps out other | |
5057 | * people trying to start a new cluster | |
5058 | */ | |
5059 | spin_lock(&last_ptr->refill_lock); | |
44fb5511 CM |
5060 | if (last_ptr->block_group && |
5061 | (last_ptr->block_group->ro || | |
5062 | !block_group_bits(last_ptr->block_group, data))) { | |
5063 | offset = 0; | |
5064 | goto refill_cluster; | |
5065 | } | |
5066 | ||
fa9c0d79 CM |
5067 | offset = btrfs_alloc_from_cluster(block_group, last_ptr, |
5068 | num_bytes, search_start); | |
5069 | if (offset) { | |
5070 | /* we have a block, we're done */ | |
5071 | spin_unlock(&last_ptr->refill_lock); | |
5072 | goto checks; | |
5073 | } | |
5074 | ||
5075 | spin_lock(&last_ptr->lock); | |
5076 | /* | |
5077 | * whoops, this cluster doesn't actually point to | |
5078 | * this block group. Get a ref on the block | |
5079 | * group is does point to and try again | |
5080 | */ | |
5081 | if (!last_ptr_loop && last_ptr->block_group && | |
5082 | last_ptr->block_group != block_group) { | |
5083 | ||
5084 | btrfs_put_block_group(block_group); | |
5085 | block_group = last_ptr->block_group; | |
11dfe35a | 5086 | btrfs_get_block_group(block_group); |
fa9c0d79 CM |
5087 | spin_unlock(&last_ptr->lock); |
5088 | spin_unlock(&last_ptr->refill_lock); | |
5089 | ||
5090 | last_ptr_loop = 1; | |
5091 | search_start = block_group->key.objectid; | |
44fb5511 CM |
5092 | /* |
5093 | * we know this block group is properly | |
5094 | * in the list because | |
5095 | * btrfs_remove_block_group, drops the | |
5096 | * cluster before it removes the block | |
5097 | * group from the list | |
5098 | */ | |
fa9c0d79 CM |
5099 | goto have_block_group; |
5100 | } | |
5101 | spin_unlock(&last_ptr->lock); | |
44fb5511 | 5102 | refill_cluster: |
fa9c0d79 CM |
5103 | /* |
5104 | * this cluster didn't work out, free it and | |
5105 | * start over | |
5106 | */ | |
5107 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
5108 | ||
5109 | last_ptr_loop = 0; | |
5110 | ||
5111 | /* allocate a cluster in this block group */ | |
451d7585 | 5112 | ret = btrfs_find_space_cluster(trans, root, |
fa9c0d79 CM |
5113 | block_group, last_ptr, |
5114 | offset, num_bytes, | |
5115 | empty_cluster + empty_size); | |
5116 | if (ret == 0) { | |
5117 | /* | |
5118 | * now pull our allocation out of this | |
5119 | * cluster | |
5120 | */ | |
5121 | offset = btrfs_alloc_from_cluster(block_group, | |
5122 | last_ptr, num_bytes, | |
5123 | search_start); | |
5124 | if (offset) { | |
5125 | /* we found one, proceed */ | |
5126 | spin_unlock(&last_ptr->refill_lock); | |
5127 | goto checks; | |
5128 | } | |
0a24325e JB |
5129 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
5130 | && !failed_cluster_refill) { | |
817d52f8 JB |
5131 | spin_unlock(&last_ptr->refill_lock); |
5132 | ||
0a24325e | 5133 | failed_cluster_refill = true; |
817d52f8 JB |
5134 | wait_block_group_cache_progress(block_group, |
5135 | num_bytes + empty_cluster + empty_size); | |
5136 | goto have_block_group; | |
fa9c0d79 | 5137 | } |
817d52f8 | 5138 | |
fa9c0d79 CM |
5139 | /* |
5140 | * at this point we either didn't find a cluster | |
5141 | * or we weren't able to allocate a block from our | |
5142 | * cluster. Free the cluster we've been trying | |
5143 | * to use, and go to the next block group | |
5144 | */ | |
0a24325e | 5145 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 5146 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 5147 | goto loop; |
fa9c0d79 CM |
5148 | } |
5149 | ||
6226cb0a JB |
5150 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
5151 | num_bytes, empty_size); | |
1cdda9b8 JB |
5152 | /* |
5153 | * If we didn't find a chunk, and we haven't failed on this | |
5154 | * block group before, and this block group is in the middle of | |
5155 | * caching and we are ok with waiting, then go ahead and wait | |
5156 | * for progress to be made, and set failed_alloc to true. | |
5157 | * | |
5158 | * If failed_alloc is true then we've already waited on this | |
5159 | * block group once and should move on to the next block group. | |
5160 | */ | |
5161 | if (!offset && !failed_alloc && !cached && | |
5162 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 5163 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
5164 | num_bytes + empty_size); |
5165 | failed_alloc = true; | |
817d52f8 | 5166 | goto have_block_group; |
1cdda9b8 JB |
5167 | } else if (!offset) { |
5168 | goto loop; | |
817d52f8 | 5169 | } |
fa9c0d79 | 5170 | checks: |
6226cb0a | 5171 | search_start = stripe_align(root, offset); |
2552d17e | 5172 | /* move on to the next group */ |
6226cb0a JB |
5173 | if (search_start + num_bytes >= search_end) { |
5174 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 5175 | goto loop; |
6226cb0a | 5176 | } |
25179201 | 5177 | |
2552d17e JB |
5178 | /* move on to the next group */ |
5179 | if (search_start + num_bytes > | |
6226cb0a JB |
5180 | block_group->key.objectid + block_group->key.offset) { |
5181 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 5182 | goto loop; |
6226cb0a | 5183 | } |
f5a31e16 | 5184 | |
f0486c68 YZ |
5185 | ins->objectid = search_start; |
5186 | ins->offset = num_bytes; | |
2552d17e | 5187 | |
f0486c68 YZ |
5188 | if (offset < search_start) |
5189 | btrfs_add_free_space(block_group, offset, | |
5190 | search_start - offset); | |
5191 | BUG_ON(offset > search_start); | |
2552d17e | 5192 | |
b4d00d56 | 5193 | ret = btrfs_update_reserved_bytes(block_group, num_bytes, 1, |
f0486c68 YZ |
5194 | (data & BTRFS_BLOCK_GROUP_DATA)); |
5195 | if (ret == -EAGAIN) { | |
6226cb0a | 5196 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 5197 | goto loop; |
0f9dd46c | 5198 | } |
0b86a832 | 5199 | |
f0486c68 | 5200 | /* we are all good, lets return */ |
2552d17e JB |
5201 | ins->objectid = search_start; |
5202 | ins->offset = num_bytes; | |
d2fb3437 | 5203 | |
6226cb0a JB |
5204 | if (offset < search_start) |
5205 | btrfs_add_free_space(block_group, offset, | |
5206 | search_start - offset); | |
5207 | BUG_ON(offset > search_start); | |
d82a6f1d | 5208 | btrfs_put_block_group(block_group); |
2552d17e JB |
5209 | break; |
5210 | loop: | |
0a24325e | 5211 | failed_cluster_refill = false; |
1cdda9b8 | 5212 | failed_alloc = false; |
b742bb82 | 5213 | BUG_ON(index != get_block_group_index(block_group)); |
fa9c0d79 | 5214 | btrfs_put_block_group(block_group); |
2552d17e JB |
5215 | } |
5216 | up_read(&space_info->groups_sem); | |
5217 | ||
b742bb82 YZ |
5218 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
5219 | goto search; | |
5220 | ||
ccf0e725 JB |
5221 | /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for |
5222 | * for them to make caching progress. Also | |
5223 | * determine the best possible bg to cache | |
5224 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
5225 | * caching kthreads as we move along | |
817d52f8 JB |
5226 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
5227 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
5228 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
5229 | * again | |
fa9c0d79 | 5230 | */ |
723bda20 | 5231 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 5232 | index = 0; |
ccf0e725 | 5233 | if (loop == LOOP_FIND_IDEAL && found_uncached_bg) { |
817d52f8 | 5234 | found_uncached_bg = false; |
ccf0e725 | 5235 | loop++; |
bab39bf9 | 5236 | if (!ideal_cache_percent) |
817d52f8 | 5237 | goto search; |
ccf0e725 JB |
5238 | |
5239 | /* | |
5240 | * 1 of the following 2 things have happened so far | |
5241 | * | |
5242 | * 1) We found an ideal block group for caching that | |
5243 | * is mostly full and will cache quickly, so we might | |
5244 | * as well wait for it. | |
5245 | * | |
5246 | * 2) We searched for cached only and we didn't find | |
5247 | * anything, and we didn't start any caching kthreads | |
5248 | * either, so chances are we will loop through and | |
5249 | * start a couple caching kthreads, and then come back | |
5250 | * around and just wait for them. This will be slower | |
5251 | * because we will have 2 caching kthreads reading at | |
5252 | * the same time when we could have just started one | |
5253 | * and waited for it to get far enough to give us an | |
5254 | * allocation, so go ahead and go to the wait caching | |
5255 | * loop. | |
5256 | */ | |
5257 | loop = LOOP_CACHING_WAIT; | |
5258 | search_start = ideal_cache_offset; | |
5259 | ideal_cache_percent = 0; | |
5260 | goto ideal_cache; | |
5261 | } else if (loop == LOOP_FIND_IDEAL) { | |
5262 | /* | |
5263 | * Didn't find a uncached bg, wait on anything we find | |
5264 | * next. | |
5265 | */ | |
5266 | loop = LOOP_CACHING_WAIT; | |
5267 | goto search; | |
5268 | } | |
5269 | ||
723bda20 | 5270 | loop++; |
817d52f8 JB |
5271 | |
5272 | if (loop == LOOP_ALLOC_CHUNK) { | |
723bda20 JB |
5273 | if (allowed_chunk_alloc) { |
5274 | ret = do_chunk_alloc(trans, root, num_bytes + | |
5275 | 2 * 1024 * 1024, data, | |
5276 | CHUNK_ALLOC_LIMITED); | |
5277 | allowed_chunk_alloc = 0; | |
5278 | if (ret == 1) | |
5279 | done_chunk_alloc = 1; | |
5280 | } else if (!done_chunk_alloc && | |
5281 | space_info->force_alloc == | |
5282 | CHUNK_ALLOC_NO_FORCE) { | |
5283 | space_info->force_alloc = CHUNK_ALLOC_LIMITED; | |
5284 | } | |
2552d17e | 5285 | |
723bda20 JB |
5286 | /* |
5287 | * We didn't allocate a chunk, go ahead and drop the | |
5288 | * empty size and loop again. | |
5289 | */ | |
5290 | if (!done_chunk_alloc) | |
5291 | loop = LOOP_NO_EMPTY_SIZE; | |
2552d17e JB |
5292 | } |
5293 | ||
723bda20 JB |
5294 | if (loop == LOOP_NO_EMPTY_SIZE) { |
5295 | empty_size = 0; | |
5296 | empty_cluster = 0; | |
fa9c0d79 | 5297 | } |
723bda20 JB |
5298 | |
5299 | goto search; | |
2552d17e JB |
5300 | } else if (!ins->objectid) { |
5301 | ret = -ENOSPC; | |
d82a6f1d | 5302 | } else if (ins->objectid) { |
80eb234a | 5303 | ret = 0; |
be744175 | 5304 | } |
be744175 | 5305 | |
0f70abe2 | 5306 | return ret; |
fec577fb | 5307 | } |
ec44a35c | 5308 | |
9ed74f2d JB |
5309 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
5310 | int dump_block_groups) | |
0f9dd46c JB |
5311 | { |
5312 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 5313 | int index = 0; |
0f9dd46c | 5314 | |
9ed74f2d | 5315 | spin_lock(&info->lock); |
d397712b CM |
5316 | printk(KERN_INFO "space_info has %llu free, is %sfull\n", |
5317 | (unsigned long long)(info->total_bytes - info->bytes_used - | |
9ed74f2d | 5318 | info->bytes_pinned - info->bytes_reserved - |
8929ecfa | 5319 | info->bytes_readonly), |
d397712b | 5320 | (info->full) ? "" : "not "); |
8929ecfa YZ |
5321 | printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, " |
5322 | "reserved=%llu, may_use=%llu, readonly=%llu\n", | |
21380931 | 5323 | (unsigned long long)info->total_bytes, |
8929ecfa | 5324 | (unsigned long long)info->bytes_used, |
21380931 | 5325 | (unsigned long long)info->bytes_pinned, |
8929ecfa | 5326 | (unsigned long long)info->bytes_reserved, |
21380931 | 5327 | (unsigned long long)info->bytes_may_use, |
8929ecfa | 5328 | (unsigned long long)info->bytes_readonly); |
9ed74f2d JB |
5329 | spin_unlock(&info->lock); |
5330 | ||
5331 | if (!dump_block_groups) | |
5332 | return; | |
0f9dd46c | 5333 | |
80eb234a | 5334 | down_read(&info->groups_sem); |
b742bb82 YZ |
5335 | again: |
5336 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 5337 | spin_lock(&cache->lock); |
d397712b CM |
5338 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used " |
5339 | "%llu pinned %llu reserved\n", | |
5340 | (unsigned long long)cache->key.objectid, | |
5341 | (unsigned long long)cache->key.offset, | |
5342 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
5343 | (unsigned long long)cache->pinned, | |
5344 | (unsigned long long)cache->reserved); | |
0f9dd46c JB |
5345 | btrfs_dump_free_space(cache, bytes); |
5346 | spin_unlock(&cache->lock); | |
5347 | } | |
b742bb82 YZ |
5348 | if (++index < BTRFS_NR_RAID_TYPES) |
5349 | goto again; | |
80eb234a | 5350 | up_read(&info->groups_sem); |
0f9dd46c | 5351 | } |
e8569813 | 5352 | |
11833d66 YZ |
5353 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
5354 | struct btrfs_root *root, | |
5355 | u64 num_bytes, u64 min_alloc_size, | |
5356 | u64 empty_size, u64 hint_byte, | |
5357 | u64 search_end, struct btrfs_key *ins, | |
5358 | u64 data) | |
fec577fb CM |
5359 | { |
5360 | int ret; | |
fbdc762b | 5361 | u64 search_start = 0; |
925baedd | 5362 | |
6a63209f | 5363 | data = btrfs_get_alloc_profile(root, data); |
98d20f67 | 5364 | again: |
0ef3e66b CM |
5365 | /* |
5366 | * the only place that sets empty_size is btrfs_realloc_node, which | |
5367 | * is not called recursively on allocations | |
5368 | */ | |
83d3c969 | 5369 | if (empty_size || root->ref_cows) |
6324fbf3 | 5370 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
5371 | num_bytes + 2 * 1024 * 1024, data, |
5372 | CHUNK_ALLOC_NO_FORCE); | |
0b86a832 | 5373 | |
db94535d CM |
5374 | WARN_ON(num_bytes < root->sectorsize); |
5375 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
f0486c68 YZ |
5376 | search_start, search_end, hint_byte, |
5377 | ins, data); | |
3b951516 | 5378 | |
98d20f67 CM |
5379 | if (ret == -ENOSPC && num_bytes > min_alloc_size) { |
5380 | num_bytes = num_bytes >> 1; | |
0f9dd46c | 5381 | num_bytes = num_bytes & ~(root->sectorsize - 1); |
98d20f67 | 5382 | num_bytes = max(num_bytes, min_alloc_size); |
0ef3e66b | 5383 | do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 | 5384 | num_bytes, data, CHUNK_ALLOC_FORCE); |
98d20f67 CM |
5385 | goto again; |
5386 | } | |
91435650 | 5387 | if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) { |
0f9dd46c JB |
5388 | struct btrfs_space_info *sinfo; |
5389 | ||
5390 | sinfo = __find_space_info(root->fs_info, data); | |
d397712b CM |
5391 | printk(KERN_ERR "btrfs allocation failed flags %llu, " |
5392 | "wanted %llu\n", (unsigned long long)data, | |
5393 | (unsigned long long)num_bytes); | |
9ed74f2d | 5394 | dump_space_info(sinfo, num_bytes, 1); |
925baedd | 5395 | } |
0f9dd46c | 5396 | |
1abe9b8a | 5397 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
5398 | ||
0f9dd46c | 5399 | return ret; |
e6dcd2dc CM |
5400 | } |
5401 | ||
65b51a00 CM |
5402 | int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len) |
5403 | { | |
0f9dd46c | 5404 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 5405 | int ret = 0; |
0f9dd46c | 5406 | |
0f9dd46c JB |
5407 | cache = btrfs_lookup_block_group(root->fs_info, start); |
5408 | if (!cache) { | |
d397712b CM |
5409 | printk(KERN_ERR "Unable to find block group for %llu\n", |
5410 | (unsigned long long)start); | |
0f9dd46c JB |
5411 | return -ENOSPC; |
5412 | } | |
1f3c79a2 | 5413 | |
5378e607 LD |
5414 | if (btrfs_test_opt(root, DISCARD)) |
5415 | ret = btrfs_discard_extent(root, start, len, NULL); | |
1f3c79a2 | 5416 | |
0f9dd46c | 5417 | btrfs_add_free_space(cache, start, len); |
b4d00d56 | 5418 | btrfs_update_reserved_bytes(cache, len, 0, 1); |
fa9c0d79 | 5419 | btrfs_put_block_group(cache); |
817d52f8 | 5420 | |
1abe9b8a | 5421 | trace_btrfs_reserved_extent_free(root, start, len); |
5422 | ||
e6dcd2dc CM |
5423 | return ret; |
5424 | } | |
5425 | ||
5d4f98a2 YZ |
5426 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5427 | struct btrfs_root *root, | |
5428 | u64 parent, u64 root_objectid, | |
5429 | u64 flags, u64 owner, u64 offset, | |
5430 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
5431 | { |
5432 | int ret; | |
5d4f98a2 | 5433 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 5434 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 5435 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 5436 | struct btrfs_path *path; |
5d4f98a2 YZ |
5437 | struct extent_buffer *leaf; |
5438 | int type; | |
5439 | u32 size; | |
26b8003f | 5440 | |
5d4f98a2 YZ |
5441 | if (parent > 0) |
5442 | type = BTRFS_SHARED_DATA_REF_KEY; | |
5443 | else | |
5444 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 5445 | |
5d4f98a2 | 5446 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
5447 | |
5448 | path = btrfs_alloc_path(); | |
db5b493a TI |
5449 | if (!path) |
5450 | return -ENOMEM; | |
47e4bb98 | 5451 | |
b9473439 | 5452 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5453 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
5454 | ins, size); | |
ccd467d6 | 5455 | BUG_ON(ret); |
0f9dd46c | 5456 | |
5d4f98a2 YZ |
5457 | leaf = path->nodes[0]; |
5458 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 5459 | struct btrfs_extent_item); |
5d4f98a2 YZ |
5460 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
5461 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
5462 | btrfs_set_extent_flags(leaf, extent_item, | |
5463 | flags | BTRFS_EXTENT_FLAG_DATA); | |
5464 | ||
5465 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
5466 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
5467 | if (parent > 0) { | |
5468 | struct btrfs_shared_data_ref *ref; | |
5469 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
5470 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
5471 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
5472 | } else { | |
5473 | struct btrfs_extent_data_ref *ref; | |
5474 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
5475 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
5476 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
5477 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
5478 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
5479 | } | |
47e4bb98 CM |
5480 | |
5481 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 5482 | btrfs_free_path(path); |
f510cfec | 5483 | |
f0486c68 | 5484 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
f5947066 | 5485 | if (ret) { |
d397712b CM |
5486 | printk(KERN_ERR "btrfs update block group failed for %llu " |
5487 | "%llu\n", (unsigned long long)ins->objectid, | |
5488 | (unsigned long long)ins->offset); | |
f5947066 CM |
5489 | BUG(); |
5490 | } | |
e6dcd2dc CM |
5491 | return ret; |
5492 | } | |
5493 | ||
5d4f98a2 YZ |
5494 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
5495 | struct btrfs_root *root, | |
5496 | u64 parent, u64 root_objectid, | |
5497 | u64 flags, struct btrfs_disk_key *key, | |
5498 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
5499 | { |
5500 | int ret; | |
5d4f98a2 YZ |
5501 | struct btrfs_fs_info *fs_info = root->fs_info; |
5502 | struct btrfs_extent_item *extent_item; | |
5503 | struct btrfs_tree_block_info *block_info; | |
5504 | struct btrfs_extent_inline_ref *iref; | |
5505 | struct btrfs_path *path; | |
5506 | struct extent_buffer *leaf; | |
5507 | u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref); | |
1c2308f8 | 5508 | |
5d4f98a2 | 5509 | path = btrfs_alloc_path(); |
d8926bb3 MF |
5510 | if (!path) |
5511 | return -ENOMEM; | |
56bec294 | 5512 | |
5d4f98a2 YZ |
5513 | path->leave_spinning = 1; |
5514 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
5515 | ins, size); | |
56bec294 | 5516 | BUG_ON(ret); |
5d4f98a2 YZ |
5517 | |
5518 | leaf = path->nodes[0]; | |
5519 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
5520 | struct btrfs_extent_item); | |
5521 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
5522 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
5523 | btrfs_set_extent_flags(leaf, extent_item, | |
5524 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5525 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
5526 | ||
5527 | btrfs_set_tree_block_key(leaf, block_info, key); | |
5528 | btrfs_set_tree_block_level(leaf, block_info, level); | |
5529 | ||
5530 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
5531 | if (parent > 0) { | |
5532 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
5533 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
5534 | BTRFS_SHARED_BLOCK_REF_KEY); | |
5535 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
5536 | } else { | |
5537 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
5538 | BTRFS_TREE_BLOCK_REF_KEY); | |
5539 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
5540 | } | |
5541 | ||
5542 | btrfs_mark_buffer_dirty(leaf); | |
5543 | btrfs_free_path(path); | |
5544 | ||
f0486c68 | 5545 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
5d4f98a2 YZ |
5546 | if (ret) { |
5547 | printk(KERN_ERR "btrfs update block group failed for %llu " | |
5548 | "%llu\n", (unsigned long long)ins->objectid, | |
5549 | (unsigned long long)ins->offset); | |
5550 | BUG(); | |
5551 | } | |
5552 | return ret; | |
5553 | } | |
5554 | ||
5555 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5556 | struct btrfs_root *root, | |
5557 | u64 root_objectid, u64 owner, | |
5558 | u64 offset, struct btrfs_key *ins) | |
5559 | { | |
5560 | int ret; | |
5561 | ||
5562 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
5563 | ||
5564 | ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset, | |
5565 | 0, root_objectid, owner, offset, | |
5566 | BTRFS_ADD_DELAYED_EXTENT, NULL); | |
e6dcd2dc CM |
5567 | return ret; |
5568 | } | |
e02119d5 CM |
5569 | |
5570 | /* | |
5571 | * this is used by the tree logging recovery code. It records that | |
5572 | * an extent has been allocated and makes sure to clear the free | |
5573 | * space cache bits as well | |
5574 | */ | |
5d4f98a2 YZ |
5575 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5576 | struct btrfs_root *root, | |
5577 | u64 root_objectid, u64 owner, u64 offset, | |
5578 | struct btrfs_key *ins) | |
e02119d5 CM |
5579 | { |
5580 | int ret; | |
5581 | struct btrfs_block_group_cache *block_group; | |
11833d66 YZ |
5582 | struct btrfs_caching_control *caching_ctl; |
5583 | u64 start = ins->objectid; | |
5584 | u64 num_bytes = ins->offset; | |
e02119d5 | 5585 | |
e02119d5 | 5586 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
b8399dee | 5587 | cache_block_group(block_group, trans, NULL, 0); |
11833d66 | 5588 | caching_ctl = get_caching_control(block_group); |
e02119d5 | 5589 | |
11833d66 YZ |
5590 | if (!caching_ctl) { |
5591 | BUG_ON(!block_group_cache_done(block_group)); | |
5592 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
5593 | BUG_ON(ret); | |
5594 | } else { | |
5595 | mutex_lock(&caching_ctl->mutex); | |
5596 | ||
5597 | if (start >= caching_ctl->progress) { | |
5598 | ret = add_excluded_extent(root, start, num_bytes); | |
5599 | BUG_ON(ret); | |
5600 | } else if (start + num_bytes <= caching_ctl->progress) { | |
5601 | ret = btrfs_remove_free_space(block_group, | |
5602 | start, num_bytes); | |
5603 | BUG_ON(ret); | |
5604 | } else { | |
5605 | num_bytes = caching_ctl->progress - start; | |
5606 | ret = btrfs_remove_free_space(block_group, | |
5607 | start, num_bytes); | |
5608 | BUG_ON(ret); | |
5609 | ||
5610 | start = caching_ctl->progress; | |
5611 | num_bytes = ins->objectid + ins->offset - | |
5612 | caching_ctl->progress; | |
5613 | ret = add_excluded_extent(root, start, num_bytes); | |
5614 | BUG_ON(ret); | |
5615 | } | |
5616 | ||
5617 | mutex_unlock(&caching_ctl->mutex); | |
5618 | put_caching_control(caching_ctl); | |
5619 | } | |
5620 | ||
b4d00d56 | 5621 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, 1, 1); |
f0486c68 | 5622 | BUG_ON(ret); |
fa9c0d79 | 5623 | btrfs_put_block_group(block_group); |
5d4f98a2 YZ |
5624 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
5625 | 0, owner, offset, ins, 1); | |
e02119d5 CM |
5626 | return ret; |
5627 | } | |
5628 | ||
65b51a00 CM |
5629 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, |
5630 | struct btrfs_root *root, | |
4008c04a CM |
5631 | u64 bytenr, u32 blocksize, |
5632 | int level) | |
65b51a00 CM |
5633 | { |
5634 | struct extent_buffer *buf; | |
5635 | ||
5636 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
5637 | if (!buf) | |
5638 | return ERR_PTR(-ENOMEM); | |
5639 | btrfs_set_header_generation(buf, trans->transid); | |
85d4e461 | 5640 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 CM |
5641 | btrfs_tree_lock(buf); |
5642 | clean_tree_block(trans, root, buf); | |
b4ce94de CM |
5643 | |
5644 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 5645 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 5646 | |
d0c803c4 | 5647 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
8cef4e16 YZ |
5648 | /* |
5649 | * we allow two log transactions at a time, use different | |
5650 | * EXENT bit to differentiate dirty pages. | |
5651 | */ | |
5652 | if (root->log_transid % 2 == 0) | |
5653 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
5654 | buf->start + buf->len - 1, GFP_NOFS); | |
5655 | else | |
5656 | set_extent_new(&root->dirty_log_pages, buf->start, | |
5657 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 CM |
5658 | } else { |
5659 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 5660 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 5661 | } |
65b51a00 | 5662 | trans->blocks_used++; |
b4ce94de | 5663 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
5664 | return buf; |
5665 | } | |
5666 | ||
f0486c68 YZ |
5667 | static struct btrfs_block_rsv * |
5668 | use_block_rsv(struct btrfs_trans_handle *trans, | |
5669 | struct btrfs_root *root, u32 blocksize) | |
5670 | { | |
5671 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 5672 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 YZ |
5673 | int ret; |
5674 | ||
5675 | block_rsv = get_block_rsv(trans, root); | |
5676 | ||
5677 | if (block_rsv->size == 0) { | |
8bb8ab2e JB |
5678 | ret = reserve_metadata_bytes(trans, root, block_rsv, |
5679 | blocksize, 0); | |
68a82277 JB |
5680 | /* |
5681 | * If we couldn't reserve metadata bytes try and use some from | |
5682 | * the global reserve. | |
5683 | */ | |
5684 | if (ret && block_rsv != global_rsv) { | |
5685 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
5686 | if (!ret) | |
5687 | return global_rsv; | |
f0486c68 | 5688 | return ERR_PTR(ret); |
68a82277 | 5689 | } else if (ret) { |
f0486c68 | 5690 | return ERR_PTR(ret); |
68a82277 | 5691 | } |
f0486c68 YZ |
5692 | return block_rsv; |
5693 | } | |
5694 | ||
5695 | ret = block_rsv_use_bytes(block_rsv, blocksize); | |
5696 | if (!ret) | |
5697 | return block_rsv; | |
68a82277 JB |
5698 | if (ret) { |
5699 | WARN_ON(1); | |
5700 | ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize, | |
5701 | 0); | |
5702 | if (!ret) { | |
5703 | spin_lock(&block_rsv->lock); | |
5704 | block_rsv->size += blocksize; | |
5705 | spin_unlock(&block_rsv->lock); | |
5706 | return block_rsv; | |
5707 | } else if (ret && block_rsv != global_rsv) { | |
5708 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
5709 | if (!ret) | |
5710 | return global_rsv; | |
5711 | } | |
5712 | } | |
f0486c68 | 5713 | |
f0486c68 YZ |
5714 | return ERR_PTR(-ENOSPC); |
5715 | } | |
5716 | ||
5717 | static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
5718 | { | |
5719 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
5720 | block_rsv_release_bytes(block_rsv, NULL, 0); | |
5721 | } | |
5722 | ||
fec577fb | 5723 | /* |
f0486c68 YZ |
5724 | * finds a free extent and does all the dirty work required for allocation |
5725 | * returns the key for the extent through ins, and a tree buffer for | |
5726 | * the first block of the extent through buf. | |
5727 | * | |
fec577fb CM |
5728 | * returns the tree buffer or NULL. |
5729 | */ | |
5f39d397 | 5730 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
5731 | struct btrfs_root *root, u32 blocksize, |
5732 | u64 parent, u64 root_objectid, | |
5733 | struct btrfs_disk_key *key, int level, | |
5734 | u64 hint, u64 empty_size) | |
fec577fb | 5735 | { |
e2fa7227 | 5736 | struct btrfs_key ins; |
f0486c68 | 5737 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 5738 | struct extent_buffer *buf; |
f0486c68 YZ |
5739 | u64 flags = 0; |
5740 | int ret; | |
5741 | ||
fec577fb | 5742 | |
f0486c68 YZ |
5743 | block_rsv = use_block_rsv(trans, root, blocksize); |
5744 | if (IS_ERR(block_rsv)) | |
5745 | return ERR_CAST(block_rsv); | |
5746 | ||
5747 | ret = btrfs_reserve_extent(trans, root, blocksize, blocksize, | |
5748 | empty_size, hint, (u64)-1, &ins, 0); | |
fec577fb | 5749 | if (ret) { |
f0486c68 | 5750 | unuse_block_rsv(block_rsv, blocksize); |
54aa1f4d | 5751 | return ERR_PTR(ret); |
fec577fb | 5752 | } |
55c69072 | 5753 | |
4008c04a CM |
5754 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
5755 | blocksize, level); | |
f0486c68 YZ |
5756 | BUG_ON(IS_ERR(buf)); |
5757 | ||
5758 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
5759 | if (parent == 0) | |
5760 | parent = ins.objectid; | |
5761 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
5762 | } else | |
5763 | BUG_ON(parent > 0); | |
5764 | ||
5765 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
5766 | struct btrfs_delayed_extent_op *extent_op; | |
5767 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
5768 | BUG_ON(!extent_op); | |
5769 | if (key) | |
5770 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
5771 | else | |
5772 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
5773 | extent_op->flags_to_set = flags; | |
5774 | extent_op->update_key = 1; | |
5775 | extent_op->update_flags = 1; | |
5776 | extent_op->is_data = 0; | |
5777 | ||
5778 | ret = btrfs_add_delayed_tree_ref(trans, ins.objectid, | |
5779 | ins.offset, parent, root_objectid, | |
5780 | level, BTRFS_ADD_DELAYED_EXTENT, | |
5781 | extent_op); | |
5782 | BUG_ON(ret); | |
5783 | } | |
fec577fb CM |
5784 | return buf; |
5785 | } | |
a28ec197 | 5786 | |
2c47e605 YZ |
5787 | struct walk_control { |
5788 | u64 refs[BTRFS_MAX_LEVEL]; | |
5789 | u64 flags[BTRFS_MAX_LEVEL]; | |
5790 | struct btrfs_key update_progress; | |
5791 | int stage; | |
5792 | int level; | |
5793 | int shared_level; | |
5794 | int update_ref; | |
5795 | int keep_locks; | |
1c4850e2 YZ |
5796 | int reada_slot; |
5797 | int reada_count; | |
2c47e605 YZ |
5798 | }; |
5799 | ||
5800 | #define DROP_REFERENCE 1 | |
5801 | #define UPDATE_BACKREF 2 | |
5802 | ||
1c4850e2 YZ |
5803 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
5804 | struct btrfs_root *root, | |
5805 | struct walk_control *wc, | |
5806 | struct btrfs_path *path) | |
6407bf6d | 5807 | { |
1c4850e2 YZ |
5808 | u64 bytenr; |
5809 | u64 generation; | |
5810 | u64 refs; | |
94fcca9f | 5811 | u64 flags; |
5d4f98a2 | 5812 | u32 nritems; |
1c4850e2 YZ |
5813 | u32 blocksize; |
5814 | struct btrfs_key key; | |
5815 | struct extent_buffer *eb; | |
6407bf6d | 5816 | int ret; |
1c4850e2 YZ |
5817 | int slot; |
5818 | int nread = 0; | |
6407bf6d | 5819 | |
1c4850e2 YZ |
5820 | if (path->slots[wc->level] < wc->reada_slot) { |
5821 | wc->reada_count = wc->reada_count * 2 / 3; | |
5822 | wc->reada_count = max(wc->reada_count, 2); | |
5823 | } else { | |
5824 | wc->reada_count = wc->reada_count * 3 / 2; | |
5825 | wc->reada_count = min_t(int, wc->reada_count, | |
5826 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
5827 | } | |
7bb86316 | 5828 | |
1c4850e2 YZ |
5829 | eb = path->nodes[wc->level]; |
5830 | nritems = btrfs_header_nritems(eb); | |
5831 | blocksize = btrfs_level_size(root, wc->level - 1); | |
bd56b302 | 5832 | |
1c4850e2 YZ |
5833 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
5834 | if (nread >= wc->reada_count) | |
5835 | break; | |
bd56b302 | 5836 | |
2dd3e67b | 5837 | cond_resched(); |
1c4850e2 YZ |
5838 | bytenr = btrfs_node_blockptr(eb, slot); |
5839 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 5840 | |
1c4850e2 YZ |
5841 | if (slot == path->slots[wc->level]) |
5842 | goto reada; | |
5d4f98a2 | 5843 | |
1c4850e2 YZ |
5844 | if (wc->stage == UPDATE_BACKREF && |
5845 | generation <= root->root_key.offset) | |
bd56b302 CM |
5846 | continue; |
5847 | ||
94fcca9f YZ |
5848 | /* We don't lock the tree block, it's OK to be racy here */ |
5849 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, | |
5850 | &refs, &flags); | |
5851 | BUG_ON(ret); | |
5852 | BUG_ON(refs == 0); | |
5853 | ||
1c4850e2 | 5854 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
5855 | if (refs == 1) |
5856 | goto reada; | |
bd56b302 | 5857 | |
94fcca9f YZ |
5858 | if (wc->level == 1 && |
5859 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5860 | continue; | |
1c4850e2 YZ |
5861 | if (!wc->update_ref || |
5862 | generation <= root->root_key.offset) | |
5863 | continue; | |
5864 | btrfs_node_key_to_cpu(eb, &key, slot); | |
5865 | ret = btrfs_comp_cpu_keys(&key, | |
5866 | &wc->update_progress); | |
5867 | if (ret < 0) | |
5868 | continue; | |
94fcca9f YZ |
5869 | } else { |
5870 | if (wc->level == 1 && | |
5871 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5872 | continue; | |
6407bf6d | 5873 | } |
1c4850e2 YZ |
5874 | reada: |
5875 | ret = readahead_tree_block(root, bytenr, blocksize, | |
5876 | generation); | |
5877 | if (ret) | |
bd56b302 | 5878 | break; |
1c4850e2 | 5879 | nread++; |
20524f02 | 5880 | } |
1c4850e2 | 5881 | wc->reada_slot = slot; |
20524f02 | 5882 | } |
2c47e605 | 5883 | |
f82d02d9 | 5884 | /* |
2c47e605 YZ |
5885 | * hepler to process tree block while walking down the tree. |
5886 | * | |
2c47e605 YZ |
5887 | * when wc->stage == UPDATE_BACKREF, this function updates |
5888 | * back refs for pointers in the block. | |
5889 | * | |
5890 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 5891 | */ |
2c47e605 | 5892 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 5893 | struct btrfs_root *root, |
2c47e605 | 5894 | struct btrfs_path *path, |
94fcca9f | 5895 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 5896 | { |
2c47e605 YZ |
5897 | int level = wc->level; |
5898 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 5899 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
5900 | int ret; |
5901 | ||
2c47e605 YZ |
5902 | if (wc->stage == UPDATE_BACKREF && |
5903 | btrfs_header_owner(eb) != root->root_key.objectid) | |
5904 | return 1; | |
f82d02d9 | 5905 | |
2c47e605 YZ |
5906 | /* |
5907 | * when reference count of tree block is 1, it won't increase | |
5908 | * again. once full backref flag is set, we never clear it. | |
5909 | */ | |
94fcca9f YZ |
5910 | if (lookup_info && |
5911 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
5912 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
5913 | BUG_ON(!path->locks[level]); |
5914 | ret = btrfs_lookup_extent_info(trans, root, | |
5915 | eb->start, eb->len, | |
5916 | &wc->refs[level], | |
5917 | &wc->flags[level]); | |
5918 | BUG_ON(ret); | |
5919 | BUG_ON(wc->refs[level] == 0); | |
5920 | } | |
5d4f98a2 | 5921 | |
2c47e605 YZ |
5922 | if (wc->stage == DROP_REFERENCE) { |
5923 | if (wc->refs[level] > 1) | |
5924 | return 1; | |
f82d02d9 | 5925 | |
2c47e605 | 5926 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 5927 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
5928 | path->locks[level] = 0; |
5929 | } | |
5930 | return 0; | |
5931 | } | |
f82d02d9 | 5932 | |
2c47e605 YZ |
5933 | /* wc->stage == UPDATE_BACKREF */ |
5934 | if (!(wc->flags[level] & flag)) { | |
5935 | BUG_ON(!path->locks[level]); | |
5936 | ret = btrfs_inc_ref(trans, root, eb, 1); | |
f82d02d9 | 5937 | BUG_ON(ret); |
2c47e605 YZ |
5938 | ret = btrfs_dec_ref(trans, root, eb, 0); |
5939 | BUG_ON(ret); | |
5940 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, | |
5941 | eb->len, flag, 0); | |
5942 | BUG_ON(ret); | |
5943 | wc->flags[level] |= flag; | |
5944 | } | |
5945 | ||
5946 | /* | |
5947 | * the block is shared by multiple trees, so it's not good to | |
5948 | * keep the tree lock | |
5949 | */ | |
5950 | if (path->locks[level] && level > 0) { | |
bd681513 | 5951 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
5952 | path->locks[level] = 0; |
5953 | } | |
5954 | return 0; | |
5955 | } | |
5956 | ||
1c4850e2 YZ |
5957 | /* |
5958 | * hepler to process tree block pointer. | |
5959 | * | |
5960 | * when wc->stage == DROP_REFERENCE, this function checks | |
5961 | * reference count of the block pointed to. if the block | |
5962 | * is shared and we need update back refs for the subtree | |
5963 | * rooted at the block, this function changes wc->stage to | |
5964 | * UPDATE_BACKREF. if the block is shared and there is no | |
5965 | * need to update back, this function drops the reference | |
5966 | * to the block. | |
5967 | * | |
5968 | * NOTE: return value 1 means we should stop walking down. | |
5969 | */ | |
5970 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
5971 | struct btrfs_root *root, | |
5972 | struct btrfs_path *path, | |
94fcca9f | 5973 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
5974 | { |
5975 | u64 bytenr; | |
5976 | u64 generation; | |
5977 | u64 parent; | |
5978 | u32 blocksize; | |
5979 | struct btrfs_key key; | |
5980 | struct extent_buffer *next; | |
5981 | int level = wc->level; | |
5982 | int reada = 0; | |
5983 | int ret = 0; | |
5984 | ||
5985 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
5986 | path->slots[level]); | |
5987 | /* | |
5988 | * if the lower level block was created before the snapshot | |
5989 | * was created, we know there is no need to update back refs | |
5990 | * for the subtree | |
5991 | */ | |
5992 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
5993 | generation <= root->root_key.offset) { |
5994 | *lookup_info = 1; | |
1c4850e2 | 5995 | return 1; |
94fcca9f | 5996 | } |
1c4850e2 YZ |
5997 | |
5998 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
5999 | blocksize = btrfs_level_size(root, level - 1); | |
6000 | ||
6001 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
6002 | if (!next) { | |
6003 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
90d2c51d MX |
6004 | if (!next) |
6005 | return -ENOMEM; | |
1c4850e2 YZ |
6006 | reada = 1; |
6007 | } | |
6008 | btrfs_tree_lock(next); | |
6009 | btrfs_set_lock_blocking(next); | |
6010 | ||
94fcca9f YZ |
6011 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, |
6012 | &wc->refs[level - 1], | |
6013 | &wc->flags[level - 1]); | |
6014 | BUG_ON(ret); | |
6015 | BUG_ON(wc->refs[level - 1] == 0); | |
6016 | *lookup_info = 0; | |
1c4850e2 | 6017 | |
94fcca9f | 6018 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 6019 | if (wc->refs[level - 1] > 1) { |
94fcca9f YZ |
6020 | if (level == 1 && |
6021 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6022 | goto skip; | |
6023 | ||
1c4850e2 YZ |
6024 | if (!wc->update_ref || |
6025 | generation <= root->root_key.offset) | |
6026 | goto skip; | |
6027 | ||
6028 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
6029 | path->slots[level]); | |
6030 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
6031 | if (ret < 0) | |
6032 | goto skip; | |
6033 | ||
6034 | wc->stage = UPDATE_BACKREF; | |
6035 | wc->shared_level = level - 1; | |
6036 | } | |
94fcca9f YZ |
6037 | } else { |
6038 | if (level == 1 && | |
6039 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6040 | goto skip; | |
1c4850e2 YZ |
6041 | } |
6042 | ||
6043 | if (!btrfs_buffer_uptodate(next, generation)) { | |
6044 | btrfs_tree_unlock(next); | |
6045 | free_extent_buffer(next); | |
6046 | next = NULL; | |
94fcca9f | 6047 | *lookup_info = 1; |
1c4850e2 YZ |
6048 | } |
6049 | ||
6050 | if (!next) { | |
6051 | if (reada && level == 1) | |
6052 | reada_walk_down(trans, root, wc, path); | |
6053 | next = read_tree_block(root, bytenr, blocksize, generation); | |
97d9a8a4 TI |
6054 | if (!next) |
6055 | return -EIO; | |
1c4850e2 YZ |
6056 | btrfs_tree_lock(next); |
6057 | btrfs_set_lock_blocking(next); | |
6058 | } | |
6059 | ||
6060 | level--; | |
6061 | BUG_ON(level != btrfs_header_level(next)); | |
6062 | path->nodes[level] = next; | |
6063 | path->slots[level] = 0; | |
bd681513 | 6064 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
6065 | wc->level = level; |
6066 | if (wc->level == 1) | |
6067 | wc->reada_slot = 0; | |
6068 | return 0; | |
6069 | skip: | |
6070 | wc->refs[level - 1] = 0; | |
6071 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
6072 | if (wc->stage == DROP_REFERENCE) { |
6073 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
6074 | parent = path->nodes[level]->start; | |
6075 | } else { | |
6076 | BUG_ON(root->root_key.objectid != | |
6077 | btrfs_header_owner(path->nodes[level])); | |
6078 | parent = 0; | |
6079 | } | |
1c4850e2 | 6080 | |
94fcca9f YZ |
6081 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
6082 | root->root_key.objectid, level - 1, 0); | |
6083 | BUG_ON(ret); | |
1c4850e2 | 6084 | } |
1c4850e2 YZ |
6085 | btrfs_tree_unlock(next); |
6086 | free_extent_buffer(next); | |
94fcca9f | 6087 | *lookup_info = 1; |
1c4850e2 YZ |
6088 | return 1; |
6089 | } | |
6090 | ||
2c47e605 YZ |
6091 | /* |
6092 | * hepler to process tree block while walking up the tree. | |
6093 | * | |
6094 | * when wc->stage == DROP_REFERENCE, this function drops | |
6095 | * reference count on the block. | |
6096 | * | |
6097 | * when wc->stage == UPDATE_BACKREF, this function changes | |
6098 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
6099 | * to UPDATE_BACKREF previously while processing the block. | |
6100 | * | |
6101 | * NOTE: return value 1 means we should stop walking up. | |
6102 | */ | |
6103 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
6104 | struct btrfs_root *root, | |
6105 | struct btrfs_path *path, | |
6106 | struct walk_control *wc) | |
6107 | { | |
f0486c68 | 6108 | int ret; |
2c47e605 YZ |
6109 | int level = wc->level; |
6110 | struct extent_buffer *eb = path->nodes[level]; | |
6111 | u64 parent = 0; | |
6112 | ||
6113 | if (wc->stage == UPDATE_BACKREF) { | |
6114 | BUG_ON(wc->shared_level < level); | |
6115 | if (level < wc->shared_level) | |
6116 | goto out; | |
6117 | ||
2c47e605 YZ |
6118 | ret = find_next_key(path, level + 1, &wc->update_progress); |
6119 | if (ret > 0) | |
6120 | wc->update_ref = 0; | |
6121 | ||
6122 | wc->stage = DROP_REFERENCE; | |
6123 | wc->shared_level = -1; | |
6124 | path->slots[level] = 0; | |
6125 | ||
6126 | /* | |
6127 | * check reference count again if the block isn't locked. | |
6128 | * we should start walking down the tree again if reference | |
6129 | * count is one. | |
6130 | */ | |
6131 | if (!path->locks[level]) { | |
6132 | BUG_ON(level == 0); | |
6133 | btrfs_tree_lock(eb); | |
6134 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6135 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6136 | |
6137 | ret = btrfs_lookup_extent_info(trans, root, | |
6138 | eb->start, eb->len, | |
6139 | &wc->refs[level], | |
6140 | &wc->flags[level]); | |
f82d02d9 | 6141 | BUG_ON(ret); |
2c47e605 YZ |
6142 | BUG_ON(wc->refs[level] == 0); |
6143 | if (wc->refs[level] == 1) { | |
bd681513 | 6144 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6145 | return 1; |
6146 | } | |
f82d02d9 | 6147 | } |
2c47e605 | 6148 | } |
f82d02d9 | 6149 | |
2c47e605 YZ |
6150 | /* wc->stage == DROP_REFERENCE */ |
6151 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 6152 | |
2c47e605 YZ |
6153 | if (wc->refs[level] == 1) { |
6154 | if (level == 0) { | |
6155 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6156 | ret = btrfs_dec_ref(trans, root, eb, 1); | |
6157 | else | |
6158 | ret = btrfs_dec_ref(trans, root, eb, 0); | |
6159 | BUG_ON(ret); | |
6160 | } | |
6161 | /* make block locked assertion in clean_tree_block happy */ | |
6162 | if (!path->locks[level] && | |
6163 | btrfs_header_generation(eb) == trans->transid) { | |
6164 | btrfs_tree_lock(eb); | |
6165 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6166 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6167 | } |
6168 | clean_tree_block(trans, root, eb); | |
6169 | } | |
6170 | ||
6171 | if (eb == root->node) { | |
6172 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6173 | parent = eb->start; | |
6174 | else | |
6175 | BUG_ON(root->root_key.objectid != | |
6176 | btrfs_header_owner(eb)); | |
6177 | } else { | |
6178 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6179 | parent = path->nodes[level + 1]->start; | |
6180 | else | |
6181 | BUG_ON(root->root_key.objectid != | |
6182 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 6183 | } |
f82d02d9 | 6184 | |
f0486c68 | 6185 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
6186 | out: |
6187 | wc->refs[level] = 0; | |
6188 | wc->flags[level] = 0; | |
f0486c68 | 6189 | return 0; |
2c47e605 YZ |
6190 | } |
6191 | ||
6192 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
6193 | struct btrfs_root *root, | |
6194 | struct btrfs_path *path, | |
6195 | struct walk_control *wc) | |
6196 | { | |
2c47e605 | 6197 | int level = wc->level; |
94fcca9f | 6198 | int lookup_info = 1; |
2c47e605 YZ |
6199 | int ret; |
6200 | ||
6201 | while (level >= 0) { | |
94fcca9f | 6202 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
6203 | if (ret > 0) |
6204 | break; | |
6205 | ||
6206 | if (level == 0) | |
6207 | break; | |
6208 | ||
7a7965f8 YZ |
6209 | if (path->slots[level] >= |
6210 | btrfs_header_nritems(path->nodes[level])) | |
6211 | break; | |
6212 | ||
94fcca9f | 6213 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
6214 | if (ret > 0) { |
6215 | path->slots[level]++; | |
6216 | continue; | |
90d2c51d MX |
6217 | } else if (ret < 0) |
6218 | return ret; | |
1c4850e2 | 6219 | level = wc->level; |
f82d02d9 | 6220 | } |
f82d02d9 YZ |
6221 | return 0; |
6222 | } | |
6223 | ||
d397712b | 6224 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 6225 | struct btrfs_root *root, |
f82d02d9 | 6226 | struct btrfs_path *path, |
2c47e605 | 6227 | struct walk_control *wc, int max_level) |
20524f02 | 6228 | { |
2c47e605 | 6229 | int level = wc->level; |
20524f02 | 6230 | int ret; |
9f3a7427 | 6231 | |
2c47e605 YZ |
6232 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
6233 | while (level < max_level && path->nodes[level]) { | |
6234 | wc->level = level; | |
6235 | if (path->slots[level] + 1 < | |
6236 | btrfs_header_nritems(path->nodes[level])) { | |
6237 | path->slots[level]++; | |
20524f02 CM |
6238 | return 0; |
6239 | } else { | |
2c47e605 YZ |
6240 | ret = walk_up_proc(trans, root, path, wc); |
6241 | if (ret > 0) | |
6242 | return 0; | |
bd56b302 | 6243 | |
2c47e605 | 6244 | if (path->locks[level]) { |
bd681513 CM |
6245 | btrfs_tree_unlock_rw(path->nodes[level], |
6246 | path->locks[level]); | |
2c47e605 | 6247 | path->locks[level] = 0; |
f82d02d9 | 6248 | } |
2c47e605 YZ |
6249 | free_extent_buffer(path->nodes[level]); |
6250 | path->nodes[level] = NULL; | |
6251 | level++; | |
20524f02 CM |
6252 | } |
6253 | } | |
6254 | return 1; | |
6255 | } | |
6256 | ||
9aca1d51 | 6257 | /* |
2c47e605 YZ |
6258 | * drop a subvolume tree. |
6259 | * | |
6260 | * this function traverses the tree freeing any blocks that only | |
6261 | * referenced by the tree. | |
6262 | * | |
6263 | * when a shared tree block is found. this function decreases its | |
6264 | * reference count by one. if update_ref is true, this function | |
6265 | * also make sure backrefs for the shared block and all lower level | |
6266 | * blocks are properly updated. | |
9aca1d51 | 6267 | */ |
3fd0a558 YZ |
6268 | int btrfs_drop_snapshot(struct btrfs_root *root, |
6269 | struct btrfs_block_rsv *block_rsv, int update_ref) | |
20524f02 | 6270 | { |
5caf2a00 | 6271 | struct btrfs_path *path; |
2c47e605 YZ |
6272 | struct btrfs_trans_handle *trans; |
6273 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 6274 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
6275 | struct walk_control *wc; |
6276 | struct btrfs_key key; | |
6277 | int err = 0; | |
6278 | int ret; | |
6279 | int level; | |
20524f02 | 6280 | |
5caf2a00 | 6281 | path = btrfs_alloc_path(); |
38a1a919 MF |
6282 | if (!path) |
6283 | return -ENOMEM; | |
20524f02 | 6284 | |
2c47e605 | 6285 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
6286 | if (!wc) { |
6287 | btrfs_free_path(path); | |
6288 | return -ENOMEM; | |
6289 | } | |
2c47e605 | 6290 | |
a22285a6 | 6291 | trans = btrfs_start_transaction(tree_root, 0); |
98d5dc13 TI |
6292 | BUG_ON(IS_ERR(trans)); |
6293 | ||
3fd0a558 YZ |
6294 | if (block_rsv) |
6295 | trans->block_rsv = block_rsv; | |
2c47e605 | 6296 | |
9f3a7427 | 6297 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 6298 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
6299 | path->nodes[level] = btrfs_lock_root_node(root); |
6300 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 6301 | path->slots[level] = 0; |
bd681513 | 6302 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6303 | memset(&wc->update_progress, 0, |
6304 | sizeof(wc->update_progress)); | |
9f3a7427 | 6305 | } else { |
9f3a7427 | 6306 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
6307 | memcpy(&wc->update_progress, &key, |
6308 | sizeof(wc->update_progress)); | |
6309 | ||
6702ed49 | 6310 | level = root_item->drop_level; |
2c47e605 | 6311 | BUG_ON(level == 0); |
6702ed49 | 6312 | path->lowest_level = level; |
2c47e605 YZ |
6313 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6314 | path->lowest_level = 0; | |
6315 | if (ret < 0) { | |
6316 | err = ret; | |
9f3a7427 CM |
6317 | goto out; |
6318 | } | |
1c4850e2 | 6319 | WARN_ON(ret > 0); |
2c47e605 | 6320 | |
7d9eb12c CM |
6321 | /* |
6322 | * unlock our path, this is safe because only this | |
6323 | * function is allowed to delete this snapshot | |
6324 | */ | |
5d4f98a2 | 6325 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
6326 | |
6327 | level = btrfs_header_level(root->node); | |
6328 | while (1) { | |
6329 | btrfs_tree_lock(path->nodes[level]); | |
6330 | btrfs_set_lock_blocking(path->nodes[level]); | |
6331 | ||
6332 | ret = btrfs_lookup_extent_info(trans, root, | |
6333 | path->nodes[level]->start, | |
6334 | path->nodes[level]->len, | |
6335 | &wc->refs[level], | |
6336 | &wc->flags[level]); | |
6337 | BUG_ON(ret); | |
6338 | BUG_ON(wc->refs[level] == 0); | |
6339 | ||
6340 | if (level == root_item->drop_level) | |
6341 | break; | |
6342 | ||
6343 | btrfs_tree_unlock(path->nodes[level]); | |
6344 | WARN_ON(wc->refs[level] != 1); | |
6345 | level--; | |
6346 | } | |
9f3a7427 | 6347 | } |
2c47e605 YZ |
6348 | |
6349 | wc->level = level; | |
6350 | wc->shared_level = -1; | |
6351 | wc->stage = DROP_REFERENCE; | |
6352 | wc->update_ref = update_ref; | |
6353 | wc->keep_locks = 0; | |
1c4850e2 | 6354 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 6355 | |
d397712b | 6356 | while (1) { |
2c47e605 YZ |
6357 | ret = walk_down_tree(trans, root, path, wc); |
6358 | if (ret < 0) { | |
6359 | err = ret; | |
20524f02 | 6360 | break; |
2c47e605 | 6361 | } |
9aca1d51 | 6362 | |
2c47e605 YZ |
6363 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
6364 | if (ret < 0) { | |
6365 | err = ret; | |
20524f02 | 6366 | break; |
2c47e605 YZ |
6367 | } |
6368 | ||
6369 | if (ret > 0) { | |
6370 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
6371 | break; |
6372 | } | |
2c47e605 YZ |
6373 | |
6374 | if (wc->stage == DROP_REFERENCE) { | |
6375 | level = wc->level; | |
6376 | btrfs_node_key(path->nodes[level], | |
6377 | &root_item->drop_progress, | |
6378 | path->slots[level]); | |
6379 | root_item->drop_level = level; | |
6380 | } | |
6381 | ||
6382 | BUG_ON(wc->level == 0); | |
3fd0a558 | 6383 | if (btrfs_should_end_transaction(trans, tree_root)) { |
2c47e605 YZ |
6384 | ret = btrfs_update_root(trans, tree_root, |
6385 | &root->root_key, | |
6386 | root_item); | |
6387 | BUG_ON(ret); | |
6388 | ||
3fd0a558 | 6389 | btrfs_end_transaction_throttle(trans, tree_root); |
a22285a6 | 6390 | trans = btrfs_start_transaction(tree_root, 0); |
98d5dc13 | 6391 | BUG_ON(IS_ERR(trans)); |
3fd0a558 YZ |
6392 | if (block_rsv) |
6393 | trans->block_rsv = block_rsv; | |
c3e69d58 | 6394 | } |
20524f02 | 6395 | } |
b3b4aa74 | 6396 | btrfs_release_path(path); |
2c47e605 YZ |
6397 | BUG_ON(err); |
6398 | ||
6399 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
6400 | BUG_ON(ret); | |
6401 | ||
76dda93c YZ |
6402 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
6403 | ret = btrfs_find_last_root(tree_root, root->root_key.objectid, | |
6404 | NULL, NULL); | |
6405 | BUG_ON(ret < 0); | |
6406 | if (ret > 0) { | |
84cd948c JB |
6407 | /* if we fail to delete the orphan item this time |
6408 | * around, it'll get picked up the next time. | |
6409 | * | |
6410 | * The most common failure here is just -ENOENT. | |
6411 | */ | |
6412 | btrfs_del_orphan_item(trans, tree_root, | |
6413 | root->root_key.objectid); | |
76dda93c YZ |
6414 | } |
6415 | } | |
6416 | ||
6417 | if (root->in_radix) { | |
6418 | btrfs_free_fs_root(tree_root->fs_info, root); | |
6419 | } else { | |
6420 | free_extent_buffer(root->node); | |
6421 | free_extent_buffer(root->commit_root); | |
6422 | kfree(root); | |
6423 | } | |
9f3a7427 | 6424 | out: |
3fd0a558 | 6425 | btrfs_end_transaction_throttle(trans, tree_root); |
2c47e605 | 6426 | kfree(wc); |
5caf2a00 | 6427 | btrfs_free_path(path); |
2c47e605 | 6428 | return err; |
20524f02 | 6429 | } |
9078a3e1 | 6430 | |
2c47e605 YZ |
6431 | /* |
6432 | * drop subtree rooted at tree block 'node'. | |
6433 | * | |
6434 | * NOTE: this function will unlock and release tree block 'node' | |
6435 | */ | |
f82d02d9 YZ |
6436 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
6437 | struct btrfs_root *root, | |
6438 | struct extent_buffer *node, | |
6439 | struct extent_buffer *parent) | |
6440 | { | |
6441 | struct btrfs_path *path; | |
2c47e605 | 6442 | struct walk_control *wc; |
f82d02d9 YZ |
6443 | int level; |
6444 | int parent_level; | |
6445 | int ret = 0; | |
6446 | int wret; | |
6447 | ||
2c47e605 YZ |
6448 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
6449 | ||
f82d02d9 | 6450 | path = btrfs_alloc_path(); |
db5b493a TI |
6451 | if (!path) |
6452 | return -ENOMEM; | |
f82d02d9 | 6453 | |
2c47e605 | 6454 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
6455 | if (!wc) { |
6456 | btrfs_free_path(path); | |
6457 | return -ENOMEM; | |
6458 | } | |
2c47e605 | 6459 | |
b9447ef8 | 6460 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
6461 | parent_level = btrfs_header_level(parent); |
6462 | extent_buffer_get(parent); | |
6463 | path->nodes[parent_level] = parent; | |
6464 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
6465 | ||
b9447ef8 | 6466 | btrfs_assert_tree_locked(node); |
f82d02d9 | 6467 | level = btrfs_header_level(node); |
f82d02d9 YZ |
6468 | path->nodes[level] = node; |
6469 | path->slots[level] = 0; | |
bd681513 | 6470 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6471 | |
6472 | wc->refs[parent_level] = 1; | |
6473 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6474 | wc->level = level; | |
6475 | wc->shared_level = -1; | |
6476 | wc->stage = DROP_REFERENCE; | |
6477 | wc->update_ref = 0; | |
6478 | wc->keep_locks = 1; | |
1c4850e2 | 6479 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
6480 | |
6481 | while (1) { | |
2c47e605 YZ |
6482 | wret = walk_down_tree(trans, root, path, wc); |
6483 | if (wret < 0) { | |
f82d02d9 | 6484 | ret = wret; |
f82d02d9 | 6485 | break; |
2c47e605 | 6486 | } |
f82d02d9 | 6487 | |
2c47e605 | 6488 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
6489 | if (wret < 0) |
6490 | ret = wret; | |
6491 | if (wret != 0) | |
6492 | break; | |
6493 | } | |
6494 | ||
2c47e605 | 6495 | kfree(wc); |
f82d02d9 YZ |
6496 | btrfs_free_path(path); |
6497 | return ret; | |
6498 | } | |
6499 | ||
ec44a35c CM |
6500 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
6501 | { | |
6502 | u64 num_devices; | |
6503 | u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | | |
6504 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
6505 | ||
cd02dca5 CM |
6506 | /* |
6507 | * we add in the count of missing devices because we want | |
6508 | * to make sure that any RAID levels on a degraded FS | |
6509 | * continue to be honored. | |
6510 | */ | |
6511 | num_devices = root->fs_info->fs_devices->rw_devices + | |
6512 | root->fs_info->fs_devices->missing_devices; | |
6513 | ||
ec44a35c CM |
6514 | if (num_devices == 1) { |
6515 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
6516 | stripped = flags & ~stripped; | |
6517 | ||
6518 | /* turn raid0 into single device chunks */ | |
6519 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
6520 | return stripped; | |
6521 | ||
6522 | /* turn mirroring into duplication */ | |
6523 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
6524 | BTRFS_BLOCK_GROUP_RAID10)) | |
6525 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
6526 | return flags; | |
6527 | } else { | |
6528 | /* they already had raid on here, just return */ | |
ec44a35c CM |
6529 | if (flags & stripped) |
6530 | return flags; | |
6531 | ||
6532 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
6533 | stripped = flags & ~stripped; | |
6534 | ||
6535 | /* switch duplicated blocks with raid1 */ | |
6536 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
6537 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
6538 | ||
6539 | /* turn single device chunks into raid0 */ | |
6540 | return stripped | BTRFS_BLOCK_GROUP_RAID0; | |
6541 | } | |
6542 | return flags; | |
6543 | } | |
6544 | ||
199c36ea | 6545 | static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 6546 | { |
f0486c68 YZ |
6547 | struct btrfs_space_info *sinfo = cache->space_info; |
6548 | u64 num_bytes; | |
199c36ea | 6549 | u64 min_allocable_bytes; |
f0486c68 | 6550 | int ret = -ENOSPC; |
0ef3e66b | 6551 | |
f0486c68 YZ |
6552 | if (cache->ro) |
6553 | return 0; | |
c286ac48 | 6554 | |
199c36ea MX |
6555 | /* |
6556 | * We need some metadata space and system metadata space for | |
6557 | * allocating chunks in some corner cases until we force to set | |
6558 | * it to be readonly. | |
6559 | */ | |
6560 | if ((sinfo->flags & | |
6561 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
6562 | !force) | |
6563 | min_allocable_bytes = 1 * 1024 * 1024; | |
6564 | else | |
6565 | min_allocable_bytes = 0; | |
6566 | ||
f0486c68 YZ |
6567 | spin_lock(&sinfo->lock); |
6568 | spin_lock(&cache->lock); | |
6569 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
6570 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
6571 | ||
6572 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
6573 | sinfo->bytes_may_use + sinfo->bytes_readonly + | |
199c36ea MX |
6574 | cache->reserved_pinned + num_bytes + min_allocable_bytes <= |
6575 | sinfo->total_bytes) { | |
f0486c68 YZ |
6576 | sinfo->bytes_readonly += num_bytes; |
6577 | sinfo->bytes_reserved += cache->reserved_pinned; | |
6578 | cache->reserved_pinned = 0; | |
6579 | cache->ro = 1; | |
6580 | ret = 0; | |
6581 | } | |
65e5341b | 6582 | |
f0486c68 YZ |
6583 | spin_unlock(&cache->lock); |
6584 | spin_unlock(&sinfo->lock); | |
6585 | return ret; | |
6586 | } | |
7d9eb12c | 6587 | |
f0486c68 YZ |
6588 | int btrfs_set_block_group_ro(struct btrfs_root *root, |
6589 | struct btrfs_block_group_cache *cache) | |
c286ac48 | 6590 | |
f0486c68 YZ |
6591 | { |
6592 | struct btrfs_trans_handle *trans; | |
6593 | u64 alloc_flags; | |
6594 | int ret; | |
7d9eb12c | 6595 | |
f0486c68 | 6596 | BUG_ON(cache->ro); |
0ef3e66b | 6597 | |
ff5714cc | 6598 | trans = btrfs_join_transaction(root); |
f0486c68 | 6599 | BUG_ON(IS_ERR(trans)); |
5d4f98a2 | 6600 | |
f0486c68 YZ |
6601 | alloc_flags = update_block_group_flags(root, cache->flags); |
6602 | if (alloc_flags != cache->flags) | |
0e4f8f88 CM |
6603 | do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
6604 | CHUNK_ALLOC_FORCE); | |
5d4f98a2 | 6605 | |
199c36ea | 6606 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
6607 | if (!ret) |
6608 | goto out; | |
6609 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
0e4f8f88 CM |
6610 | ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
6611 | CHUNK_ALLOC_FORCE); | |
f0486c68 YZ |
6612 | if (ret < 0) |
6613 | goto out; | |
199c36ea | 6614 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
6615 | out: |
6616 | btrfs_end_transaction(trans, root); | |
6617 | return ret; | |
6618 | } | |
5d4f98a2 | 6619 | |
c87f08ca CM |
6620 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
6621 | struct btrfs_root *root, u64 type) | |
6622 | { | |
6623 | u64 alloc_flags = get_alloc_profile(root, type); | |
0e4f8f88 CM |
6624 | return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
6625 | CHUNK_ALLOC_FORCE); | |
c87f08ca CM |
6626 | } |
6627 | ||
6d07bcec MX |
6628 | /* |
6629 | * helper to account the unused space of all the readonly block group in the | |
6630 | * list. takes mirrors into account. | |
6631 | */ | |
6632 | static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list) | |
6633 | { | |
6634 | struct btrfs_block_group_cache *block_group; | |
6635 | u64 free_bytes = 0; | |
6636 | int factor; | |
6637 | ||
6638 | list_for_each_entry(block_group, groups_list, list) { | |
6639 | spin_lock(&block_group->lock); | |
6640 | ||
6641 | if (!block_group->ro) { | |
6642 | spin_unlock(&block_group->lock); | |
6643 | continue; | |
6644 | } | |
6645 | ||
6646 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
6647 | BTRFS_BLOCK_GROUP_RAID10 | | |
6648 | BTRFS_BLOCK_GROUP_DUP)) | |
6649 | factor = 2; | |
6650 | else | |
6651 | factor = 1; | |
6652 | ||
6653 | free_bytes += (block_group->key.offset - | |
6654 | btrfs_block_group_used(&block_group->item)) * | |
6655 | factor; | |
6656 | ||
6657 | spin_unlock(&block_group->lock); | |
6658 | } | |
6659 | ||
6660 | return free_bytes; | |
6661 | } | |
6662 | ||
6663 | /* | |
6664 | * helper to account the unused space of all the readonly block group in the | |
6665 | * space_info. takes mirrors into account. | |
6666 | */ | |
6667 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) | |
6668 | { | |
6669 | int i; | |
6670 | u64 free_bytes = 0; | |
6671 | ||
6672 | spin_lock(&sinfo->lock); | |
6673 | ||
6674 | for(i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
6675 | if (!list_empty(&sinfo->block_groups[i])) | |
6676 | free_bytes += __btrfs_get_ro_block_group_free_space( | |
6677 | &sinfo->block_groups[i]); | |
6678 | ||
6679 | spin_unlock(&sinfo->lock); | |
6680 | ||
6681 | return free_bytes; | |
6682 | } | |
6683 | ||
f0486c68 YZ |
6684 | int btrfs_set_block_group_rw(struct btrfs_root *root, |
6685 | struct btrfs_block_group_cache *cache) | |
5d4f98a2 | 6686 | { |
f0486c68 YZ |
6687 | struct btrfs_space_info *sinfo = cache->space_info; |
6688 | u64 num_bytes; | |
6689 | ||
6690 | BUG_ON(!cache->ro); | |
6691 | ||
6692 | spin_lock(&sinfo->lock); | |
6693 | spin_lock(&cache->lock); | |
6694 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
6695 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
6696 | sinfo->bytes_readonly -= num_bytes; | |
6697 | cache->ro = 0; | |
6698 | spin_unlock(&cache->lock); | |
6699 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
6700 | return 0; |
6701 | } | |
6702 | ||
ba1bf481 JB |
6703 | /* |
6704 | * checks to see if its even possible to relocate this block group. | |
6705 | * | |
6706 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
6707 | * ok to go ahead and try. | |
6708 | */ | |
6709 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 6710 | { |
ba1bf481 JB |
6711 | struct btrfs_block_group_cache *block_group; |
6712 | struct btrfs_space_info *space_info; | |
6713 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
6714 | struct btrfs_device *device; | |
6715 | int full = 0; | |
6716 | int ret = 0; | |
1a40e23b | 6717 | |
ba1bf481 | 6718 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 6719 | |
ba1bf481 JB |
6720 | /* odd, couldn't find the block group, leave it alone */ |
6721 | if (!block_group) | |
6722 | return -1; | |
1a40e23b | 6723 | |
ba1bf481 JB |
6724 | /* no bytes used, we're good */ |
6725 | if (!btrfs_block_group_used(&block_group->item)) | |
1a40e23b ZY |
6726 | goto out; |
6727 | ||
ba1bf481 JB |
6728 | space_info = block_group->space_info; |
6729 | spin_lock(&space_info->lock); | |
17d217fe | 6730 | |
ba1bf481 | 6731 | full = space_info->full; |
17d217fe | 6732 | |
ba1bf481 JB |
6733 | /* |
6734 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
6735 | * relocate it unless we're able to allocate a new chunk below. |
6736 | * | |
6737 | * Otherwise, we need to make sure we have room in the space to handle | |
6738 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 6739 | */ |
7ce618db CM |
6740 | if ((space_info->total_bytes != block_group->key.offset) && |
6741 | (space_info->bytes_used + space_info->bytes_reserved + | |
ba1bf481 JB |
6742 | space_info->bytes_pinned + space_info->bytes_readonly + |
6743 | btrfs_block_group_used(&block_group->item) < | |
7ce618db | 6744 | space_info->total_bytes)) { |
ba1bf481 JB |
6745 | spin_unlock(&space_info->lock); |
6746 | goto out; | |
17d217fe | 6747 | } |
ba1bf481 | 6748 | spin_unlock(&space_info->lock); |
ea8c2819 | 6749 | |
ba1bf481 JB |
6750 | /* |
6751 | * ok we don't have enough space, but maybe we have free space on our | |
6752 | * devices to allocate new chunks for relocation, so loop through our | |
6753 | * alloc devices and guess if we have enough space. However, if we | |
6754 | * were marked as full, then we know there aren't enough chunks, and we | |
6755 | * can just return. | |
6756 | */ | |
6757 | ret = -1; | |
6758 | if (full) | |
6759 | goto out; | |
ea8c2819 | 6760 | |
ba1bf481 JB |
6761 | mutex_lock(&root->fs_info->chunk_mutex); |
6762 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
6763 | u64 min_free = btrfs_block_group_used(&block_group->item); | |
7bfc837d | 6764 | u64 dev_offset; |
56bec294 | 6765 | |
ba1bf481 JB |
6766 | /* |
6767 | * check to make sure we can actually find a chunk with enough | |
6768 | * space to fit our block group in. | |
6769 | */ | |
6770 | if (device->total_bytes > device->bytes_used + min_free) { | |
6771 | ret = find_free_dev_extent(NULL, device, min_free, | |
7bfc837d | 6772 | &dev_offset, NULL); |
ba1bf481 | 6773 | if (!ret) |
73e48b27 | 6774 | break; |
ba1bf481 | 6775 | ret = -1; |
725c8463 | 6776 | } |
edbd8d4e | 6777 | } |
ba1bf481 | 6778 | mutex_unlock(&root->fs_info->chunk_mutex); |
edbd8d4e | 6779 | out: |
ba1bf481 | 6780 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
6781 | return ret; |
6782 | } | |
6783 | ||
b2950863 CH |
6784 | static int find_first_block_group(struct btrfs_root *root, |
6785 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 6786 | { |
925baedd | 6787 | int ret = 0; |
0b86a832 CM |
6788 | struct btrfs_key found_key; |
6789 | struct extent_buffer *leaf; | |
6790 | int slot; | |
edbd8d4e | 6791 | |
0b86a832 CM |
6792 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
6793 | if (ret < 0) | |
925baedd CM |
6794 | goto out; |
6795 | ||
d397712b | 6796 | while (1) { |
0b86a832 | 6797 | slot = path->slots[0]; |
edbd8d4e | 6798 | leaf = path->nodes[0]; |
0b86a832 CM |
6799 | if (slot >= btrfs_header_nritems(leaf)) { |
6800 | ret = btrfs_next_leaf(root, path); | |
6801 | if (ret == 0) | |
6802 | continue; | |
6803 | if (ret < 0) | |
925baedd | 6804 | goto out; |
0b86a832 | 6805 | break; |
edbd8d4e | 6806 | } |
0b86a832 | 6807 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 6808 | |
0b86a832 | 6809 | if (found_key.objectid >= key->objectid && |
925baedd CM |
6810 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6811 | ret = 0; | |
6812 | goto out; | |
6813 | } | |
0b86a832 | 6814 | path->slots[0]++; |
edbd8d4e | 6815 | } |
925baedd | 6816 | out: |
0b86a832 | 6817 | return ret; |
edbd8d4e CM |
6818 | } |
6819 | ||
0af3d00b JB |
6820 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
6821 | { | |
6822 | struct btrfs_block_group_cache *block_group; | |
6823 | u64 last = 0; | |
6824 | ||
6825 | while (1) { | |
6826 | struct inode *inode; | |
6827 | ||
6828 | block_group = btrfs_lookup_first_block_group(info, last); | |
6829 | while (block_group) { | |
6830 | spin_lock(&block_group->lock); | |
6831 | if (block_group->iref) | |
6832 | break; | |
6833 | spin_unlock(&block_group->lock); | |
6834 | block_group = next_block_group(info->tree_root, | |
6835 | block_group); | |
6836 | } | |
6837 | if (!block_group) { | |
6838 | if (last == 0) | |
6839 | break; | |
6840 | last = 0; | |
6841 | continue; | |
6842 | } | |
6843 | ||
6844 | inode = block_group->inode; | |
6845 | block_group->iref = 0; | |
6846 | block_group->inode = NULL; | |
6847 | spin_unlock(&block_group->lock); | |
6848 | iput(inode); | |
6849 | last = block_group->key.objectid + block_group->key.offset; | |
6850 | btrfs_put_block_group(block_group); | |
6851 | } | |
6852 | } | |
6853 | ||
1a40e23b ZY |
6854 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
6855 | { | |
6856 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 6857 | struct btrfs_space_info *space_info; |
11833d66 | 6858 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
6859 | struct rb_node *n; |
6860 | ||
11833d66 YZ |
6861 | down_write(&info->extent_commit_sem); |
6862 | while (!list_empty(&info->caching_block_groups)) { | |
6863 | caching_ctl = list_entry(info->caching_block_groups.next, | |
6864 | struct btrfs_caching_control, list); | |
6865 | list_del(&caching_ctl->list); | |
6866 | put_caching_control(caching_ctl); | |
6867 | } | |
6868 | up_write(&info->extent_commit_sem); | |
6869 | ||
1a40e23b ZY |
6870 | spin_lock(&info->block_group_cache_lock); |
6871 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
6872 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
6873 | cache_node); | |
1a40e23b ZY |
6874 | rb_erase(&block_group->cache_node, |
6875 | &info->block_group_cache_tree); | |
d899e052 YZ |
6876 | spin_unlock(&info->block_group_cache_lock); |
6877 | ||
80eb234a | 6878 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 6879 | list_del(&block_group->list); |
80eb234a | 6880 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 6881 | |
817d52f8 | 6882 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 6883 | wait_block_group_cache_done(block_group); |
817d52f8 | 6884 | |
3c14874a JB |
6885 | /* |
6886 | * We haven't cached this block group, which means we could | |
6887 | * possibly have excluded extents on this block group. | |
6888 | */ | |
6889 | if (block_group->cached == BTRFS_CACHE_NO) | |
6890 | free_excluded_extents(info->extent_root, block_group); | |
6891 | ||
817d52f8 | 6892 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 6893 | btrfs_put_block_group(block_group); |
d899e052 YZ |
6894 | |
6895 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
6896 | } |
6897 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
6898 | |
6899 | /* now that all the block groups are freed, go through and | |
6900 | * free all the space_info structs. This is only called during | |
6901 | * the final stages of unmount, and so we know nobody is | |
6902 | * using them. We call synchronize_rcu() once before we start, | |
6903 | * just to be on the safe side. | |
6904 | */ | |
6905 | synchronize_rcu(); | |
6906 | ||
8929ecfa YZ |
6907 | release_global_block_rsv(info); |
6908 | ||
4184ea7f CM |
6909 | while(!list_empty(&info->space_info)) { |
6910 | space_info = list_entry(info->space_info.next, | |
6911 | struct btrfs_space_info, | |
6912 | list); | |
f0486c68 YZ |
6913 | if (space_info->bytes_pinned > 0 || |
6914 | space_info->bytes_reserved > 0) { | |
6915 | WARN_ON(1); | |
6916 | dump_space_info(space_info, 0, 0); | |
6917 | } | |
4184ea7f CM |
6918 | list_del(&space_info->list); |
6919 | kfree(space_info); | |
6920 | } | |
1a40e23b ZY |
6921 | return 0; |
6922 | } | |
6923 | ||
b742bb82 YZ |
6924 | static void __link_block_group(struct btrfs_space_info *space_info, |
6925 | struct btrfs_block_group_cache *cache) | |
6926 | { | |
6927 | int index = get_block_group_index(cache); | |
6928 | ||
6929 | down_write(&space_info->groups_sem); | |
6930 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
6931 | up_write(&space_info->groups_sem); | |
6932 | } | |
6933 | ||
9078a3e1 CM |
6934 | int btrfs_read_block_groups(struct btrfs_root *root) |
6935 | { | |
6936 | struct btrfs_path *path; | |
6937 | int ret; | |
9078a3e1 | 6938 | struct btrfs_block_group_cache *cache; |
be744175 | 6939 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 6940 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
6941 | struct btrfs_key key; |
6942 | struct btrfs_key found_key; | |
5f39d397 | 6943 | struct extent_buffer *leaf; |
0af3d00b JB |
6944 | int need_clear = 0; |
6945 | u64 cache_gen; | |
96b5179d | 6946 | |
be744175 | 6947 | root = info->extent_root; |
9078a3e1 | 6948 | key.objectid = 0; |
0b86a832 | 6949 | key.offset = 0; |
9078a3e1 | 6950 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
6951 | path = btrfs_alloc_path(); |
6952 | if (!path) | |
6953 | return -ENOMEM; | |
026fd317 | 6954 | path->reada = 1; |
9078a3e1 | 6955 | |
0af3d00b JB |
6956 | cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy); |
6957 | if (cache_gen != 0 && | |
6958 | btrfs_super_generation(&root->fs_info->super_copy) != cache_gen) | |
6959 | need_clear = 1; | |
88c2ba3b JB |
6960 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
6961 | need_clear = 1; | |
8216ef86 JB |
6962 | if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen) |
6963 | printk(KERN_INFO "btrfs: disk space caching is enabled\n"); | |
0af3d00b | 6964 | |
d397712b | 6965 | while (1) { |
0b86a832 | 6966 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
6967 | if (ret > 0) |
6968 | break; | |
0b86a832 CM |
6969 | if (ret != 0) |
6970 | goto error; | |
5f39d397 CM |
6971 | leaf = path->nodes[0]; |
6972 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 6973 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 6974 | if (!cache) { |
0b86a832 | 6975 | ret = -ENOMEM; |
f0486c68 | 6976 | goto error; |
9078a3e1 | 6977 | } |
34d52cb6 LZ |
6978 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
6979 | GFP_NOFS); | |
6980 | if (!cache->free_space_ctl) { | |
6981 | kfree(cache); | |
6982 | ret = -ENOMEM; | |
6983 | goto error; | |
6984 | } | |
3e1ad54f | 6985 | |
d2fb3437 | 6986 | atomic_set(&cache->count, 1); |
c286ac48 | 6987 | spin_lock_init(&cache->lock); |
817d52f8 | 6988 | cache->fs_info = info; |
0f9dd46c | 6989 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 6990 | INIT_LIST_HEAD(&cache->cluster_list); |
96303081 | 6991 | |
0af3d00b JB |
6992 | if (need_clear) |
6993 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
6994 | ||
5f39d397 CM |
6995 | read_extent_buffer(leaf, &cache->item, |
6996 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
6997 | sizeof(cache->item)); | |
9078a3e1 | 6998 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 6999 | |
9078a3e1 | 7000 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 7001 | btrfs_release_path(path); |
0b86a832 | 7002 | cache->flags = btrfs_block_group_flags(&cache->item); |
817d52f8 JB |
7003 | cache->sectorsize = root->sectorsize; |
7004 | ||
34d52cb6 LZ |
7005 | btrfs_init_free_space_ctl(cache); |
7006 | ||
3c14874a JB |
7007 | /* |
7008 | * We need to exclude the super stripes now so that the space | |
7009 | * info has super bytes accounted for, otherwise we'll think | |
7010 | * we have more space than we actually do. | |
7011 | */ | |
7012 | exclude_super_stripes(root, cache); | |
7013 | ||
817d52f8 JB |
7014 | /* |
7015 | * check for two cases, either we are full, and therefore | |
7016 | * don't need to bother with the caching work since we won't | |
7017 | * find any space, or we are empty, and we can just add all | |
7018 | * the space in and be done with it. This saves us _alot_ of | |
7019 | * time, particularly in the full case. | |
7020 | */ | |
7021 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 7022 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7023 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 7024 | free_excluded_extents(root, cache); |
817d52f8 | 7025 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 7026 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
7027 | cache->cached = BTRFS_CACHE_FINISHED; |
7028 | add_new_free_space(cache, root->fs_info, | |
7029 | found_key.objectid, | |
7030 | found_key.objectid + | |
7031 | found_key.offset); | |
11833d66 | 7032 | free_excluded_extents(root, cache); |
817d52f8 | 7033 | } |
96b5179d | 7034 | |
6324fbf3 CM |
7035 | ret = update_space_info(info, cache->flags, found_key.offset, |
7036 | btrfs_block_group_used(&cache->item), | |
7037 | &space_info); | |
7038 | BUG_ON(ret); | |
7039 | cache->space_info = space_info; | |
1b2da372 | 7040 | spin_lock(&cache->space_info->lock); |
f0486c68 | 7041 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7042 | spin_unlock(&cache->space_info->lock); |
7043 | ||
b742bb82 | 7044 | __link_block_group(space_info, cache); |
0f9dd46c JB |
7045 | |
7046 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
7047 | BUG_ON(ret); | |
75ccf47d CM |
7048 | |
7049 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
2b82032c | 7050 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
199c36ea | 7051 | set_block_group_ro(cache, 1); |
9078a3e1 | 7052 | } |
b742bb82 YZ |
7053 | |
7054 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
7055 | if (!(get_alloc_profile(root, space_info->flags) & | |
7056 | (BTRFS_BLOCK_GROUP_RAID10 | | |
7057 | BTRFS_BLOCK_GROUP_RAID1 | | |
7058 | BTRFS_BLOCK_GROUP_DUP))) | |
7059 | continue; | |
7060 | /* | |
7061 | * avoid allocating from un-mirrored block group if there are | |
7062 | * mirrored block groups. | |
7063 | */ | |
7064 | list_for_each_entry(cache, &space_info->block_groups[3], list) | |
199c36ea | 7065 | set_block_group_ro(cache, 1); |
b742bb82 | 7066 | list_for_each_entry(cache, &space_info->block_groups[4], list) |
199c36ea | 7067 | set_block_group_ro(cache, 1); |
9078a3e1 | 7068 | } |
f0486c68 YZ |
7069 | |
7070 | init_global_block_rsv(info); | |
0b86a832 CM |
7071 | ret = 0; |
7072 | error: | |
9078a3e1 | 7073 | btrfs_free_path(path); |
0b86a832 | 7074 | return ret; |
9078a3e1 | 7075 | } |
6324fbf3 CM |
7076 | |
7077 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
7078 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 7079 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
7080 | u64 size) |
7081 | { | |
7082 | int ret; | |
6324fbf3 CM |
7083 | struct btrfs_root *extent_root; |
7084 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
7085 | |
7086 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 7087 | |
12fcfd22 | 7088 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 7089 | |
8f18cf13 | 7090 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
7091 | if (!cache) |
7092 | return -ENOMEM; | |
34d52cb6 LZ |
7093 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
7094 | GFP_NOFS); | |
7095 | if (!cache->free_space_ctl) { | |
7096 | kfree(cache); | |
7097 | return -ENOMEM; | |
7098 | } | |
0f9dd46c | 7099 | |
e17cade2 | 7100 | cache->key.objectid = chunk_offset; |
6324fbf3 | 7101 | cache->key.offset = size; |
d2fb3437 | 7102 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
96303081 | 7103 | cache->sectorsize = root->sectorsize; |
0af3d00b | 7104 | cache->fs_info = root->fs_info; |
96303081 | 7105 | |
d2fb3437 | 7106 | atomic_set(&cache->count, 1); |
c286ac48 | 7107 | spin_lock_init(&cache->lock); |
0f9dd46c | 7108 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7109 | INIT_LIST_HEAD(&cache->cluster_list); |
0ef3e66b | 7110 | |
34d52cb6 LZ |
7111 | btrfs_init_free_space_ctl(cache); |
7112 | ||
6324fbf3 | 7113 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
7114 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
7115 | cache->flags = type; | |
7116 | btrfs_set_block_group_flags(&cache->item, type); | |
7117 | ||
11833d66 | 7118 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7119 | cache->cached = BTRFS_CACHE_FINISHED; |
11833d66 | 7120 | exclude_super_stripes(root, cache); |
96303081 | 7121 | |
817d52f8 JB |
7122 | add_new_free_space(cache, root->fs_info, chunk_offset, |
7123 | chunk_offset + size); | |
7124 | ||
11833d66 YZ |
7125 | free_excluded_extents(root, cache); |
7126 | ||
6324fbf3 CM |
7127 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
7128 | &cache->space_info); | |
7129 | BUG_ON(ret); | |
1b2da372 JB |
7130 | |
7131 | spin_lock(&cache->space_info->lock); | |
f0486c68 | 7132 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7133 | spin_unlock(&cache->space_info->lock); |
7134 | ||
b742bb82 | 7135 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 7136 | |
0f9dd46c JB |
7137 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
7138 | BUG_ON(ret); | |
c286ac48 | 7139 | |
6324fbf3 CM |
7140 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
7141 | sizeof(cache->item)); | |
7142 | BUG_ON(ret); | |
7143 | ||
d18a2c44 | 7144 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 7145 | |
6324fbf3 CM |
7146 | return 0; |
7147 | } | |
1a40e23b ZY |
7148 | |
7149 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, | |
7150 | struct btrfs_root *root, u64 group_start) | |
7151 | { | |
7152 | struct btrfs_path *path; | |
7153 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 7154 | struct btrfs_free_cluster *cluster; |
0af3d00b | 7155 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 7156 | struct btrfs_key key; |
0af3d00b | 7157 | struct inode *inode; |
1a40e23b | 7158 | int ret; |
89a55897 | 7159 | int factor; |
1a40e23b | 7160 | |
1a40e23b ZY |
7161 | root = root->fs_info->extent_root; |
7162 | ||
7163 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
7164 | BUG_ON(!block_group); | |
c146afad | 7165 | BUG_ON(!block_group->ro); |
1a40e23b | 7166 | |
9f7c43c9 | 7167 | /* |
7168 | * Free the reserved super bytes from this block group before | |
7169 | * remove it. | |
7170 | */ | |
7171 | free_excluded_extents(root, block_group); | |
7172 | ||
1a40e23b | 7173 | memcpy(&key, &block_group->key, sizeof(key)); |
89a55897 JB |
7174 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
7175 | BTRFS_BLOCK_GROUP_RAID1 | | |
7176 | BTRFS_BLOCK_GROUP_RAID10)) | |
7177 | factor = 2; | |
7178 | else | |
7179 | factor = 1; | |
1a40e23b | 7180 | |
44fb5511 CM |
7181 | /* make sure this block group isn't part of an allocation cluster */ |
7182 | cluster = &root->fs_info->data_alloc_cluster; | |
7183 | spin_lock(&cluster->refill_lock); | |
7184 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7185 | spin_unlock(&cluster->refill_lock); | |
7186 | ||
7187 | /* | |
7188 | * make sure this block group isn't part of a metadata | |
7189 | * allocation cluster | |
7190 | */ | |
7191 | cluster = &root->fs_info->meta_alloc_cluster; | |
7192 | spin_lock(&cluster->refill_lock); | |
7193 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7194 | spin_unlock(&cluster->refill_lock); | |
7195 | ||
1a40e23b | 7196 | path = btrfs_alloc_path(); |
d8926bb3 MF |
7197 | if (!path) { |
7198 | ret = -ENOMEM; | |
7199 | goto out; | |
7200 | } | |
1a40e23b | 7201 | |
0af3d00b JB |
7202 | inode = lookup_free_space_inode(root, block_group, path); |
7203 | if (!IS_ERR(inode)) { | |
7204 | btrfs_orphan_add(trans, inode); | |
7205 | clear_nlink(inode); | |
7206 | /* One for the block groups ref */ | |
7207 | spin_lock(&block_group->lock); | |
7208 | if (block_group->iref) { | |
7209 | block_group->iref = 0; | |
7210 | block_group->inode = NULL; | |
7211 | spin_unlock(&block_group->lock); | |
7212 | iput(inode); | |
7213 | } else { | |
7214 | spin_unlock(&block_group->lock); | |
7215 | } | |
7216 | /* One for our lookup ref */ | |
7217 | iput(inode); | |
7218 | } | |
7219 | ||
7220 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
7221 | key.offset = block_group->key.objectid; | |
7222 | key.type = 0; | |
7223 | ||
7224 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
7225 | if (ret < 0) | |
7226 | goto out; | |
7227 | if (ret > 0) | |
b3b4aa74 | 7228 | btrfs_release_path(path); |
0af3d00b JB |
7229 | if (ret == 0) { |
7230 | ret = btrfs_del_item(trans, tree_root, path); | |
7231 | if (ret) | |
7232 | goto out; | |
b3b4aa74 | 7233 | btrfs_release_path(path); |
0af3d00b JB |
7234 | } |
7235 | ||
3dfdb934 | 7236 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
7237 | rb_erase(&block_group->cache_node, |
7238 | &root->fs_info->block_group_cache_tree); | |
3dfdb934 | 7239 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 7240 | |
80eb234a | 7241 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
7242 | /* |
7243 | * we must use list_del_init so people can check to see if they | |
7244 | * are still on the list after taking the semaphore | |
7245 | */ | |
7246 | list_del_init(&block_group->list); | |
80eb234a | 7247 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 7248 | |
817d52f8 | 7249 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 7250 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
7251 | |
7252 | btrfs_remove_free_space_cache(block_group); | |
7253 | ||
c146afad YZ |
7254 | spin_lock(&block_group->space_info->lock); |
7255 | block_group->space_info->total_bytes -= block_group->key.offset; | |
7256 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 7257 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
c146afad | 7258 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 7259 | |
0af3d00b JB |
7260 | memcpy(&key, &block_group->key, sizeof(key)); |
7261 | ||
283bb197 | 7262 | btrfs_clear_space_info_full(root->fs_info); |
c146afad | 7263 | |
fa9c0d79 CM |
7264 | btrfs_put_block_group(block_group); |
7265 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
7266 | |
7267 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
7268 | if (ret > 0) | |
7269 | ret = -EIO; | |
7270 | if (ret < 0) | |
7271 | goto out; | |
7272 | ||
7273 | ret = btrfs_del_item(trans, root, path); | |
7274 | out: | |
7275 | btrfs_free_path(path); | |
7276 | return ret; | |
7277 | } | |
acce952b | 7278 | |
c59021f8 | 7279 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
7280 | { | |
7281 | struct btrfs_space_info *space_info; | |
1aba86d6 | 7282 | struct btrfs_super_block *disk_super; |
7283 | u64 features; | |
7284 | u64 flags; | |
7285 | int mixed = 0; | |
c59021f8 | 7286 | int ret; |
7287 | ||
1aba86d6 | 7288 | disk_super = &fs_info->super_copy; |
7289 | if (!btrfs_super_root(disk_super)) | |
7290 | return 1; | |
c59021f8 | 7291 | |
1aba86d6 | 7292 | features = btrfs_super_incompat_flags(disk_super); |
7293 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
7294 | mixed = 1; | |
c59021f8 | 7295 | |
1aba86d6 | 7296 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
7297 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
c59021f8 | 7298 | if (ret) |
1aba86d6 | 7299 | goto out; |
c59021f8 | 7300 | |
1aba86d6 | 7301 | if (mixed) { |
7302 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
7303 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7304 | } else { | |
7305 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
7306 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7307 | if (ret) | |
7308 | goto out; | |
7309 | ||
7310 | flags = BTRFS_BLOCK_GROUP_DATA; | |
7311 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7312 | } | |
7313 | out: | |
c59021f8 | 7314 | return ret; |
7315 | } | |
7316 | ||
acce952b | 7317 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
7318 | { | |
7319 | return unpin_extent_range(root, start, end); | |
7320 | } | |
7321 | ||
7322 | int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 7323 | u64 num_bytes, u64 *actual_bytes) |
acce952b | 7324 | { |
5378e607 | 7325 | return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes); |
acce952b | 7326 | } |
f7039b1d LD |
7327 | |
7328 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) | |
7329 | { | |
7330 | struct btrfs_fs_info *fs_info = root->fs_info; | |
7331 | struct btrfs_block_group_cache *cache = NULL; | |
7332 | u64 group_trimmed; | |
7333 | u64 start; | |
7334 | u64 end; | |
7335 | u64 trimmed = 0; | |
7336 | int ret = 0; | |
7337 | ||
7338 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
7339 | ||
7340 | while (cache) { | |
7341 | if (cache->key.objectid >= (range->start + range->len)) { | |
7342 | btrfs_put_block_group(cache); | |
7343 | break; | |
7344 | } | |
7345 | ||
7346 | start = max(range->start, cache->key.objectid); | |
7347 | end = min(range->start + range->len, | |
7348 | cache->key.objectid + cache->key.offset); | |
7349 | ||
7350 | if (end - start >= range->minlen) { | |
7351 | if (!block_group_cache_done(cache)) { | |
7352 | ret = cache_block_group(cache, NULL, root, 0); | |
7353 | if (!ret) | |
7354 | wait_block_group_cache_done(cache); | |
7355 | } | |
7356 | ret = btrfs_trim_block_group(cache, | |
7357 | &group_trimmed, | |
7358 | start, | |
7359 | end, | |
7360 | range->minlen); | |
7361 | ||
7362 | trimmed += group_trimmed; | |
7363 | if (ret) { | |
7364 | btrfs_put_block_group(cache); | |
7365 | break; | |
7366 | } | |
7367 | } | |
7368 | ||
7369 | cache = next_block_group(fs_info->tree_root, cache); | |
7370 | } | |
7371 | ||
7372 | range->len = trimmed; | |
7373 | return ret; | |
7374 | } |