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