Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
ec6b910f | 18 | #include <linux/sched.h> |
edbd8d4e | 19 | #include <linux/pagemap.h> |
ec44a35c | 20 | #include <linux/writeback.h> |
21af804c | 21 | #include <linux/blkdev.h> |
b7a9f29f | 22 | #include <linux/sort.h> |
4184ea7f | 23 | #include <linux/rcupdate.h> |
817d52f8 | 24 | #include <linux/kthread.h> |
5a0e3ad6 | 25 | #include <linux/slab.h> |
4b4e25f2 | 26 | #include "compat.h" |
74493f7a | 27 | #include "hash.h" |
fec577fb CM |
28 | #include "ctree.h" |
29 | #include "disk-io.h" | |
30 | #include "print-tree.h" | |
e089f05c | 31 | #include "transaction.h" |
0b86a832 | 32 | #include "volumes.h" |
925baedd | 33 | #include "locking.h" |
fa9c0d79 | 34 | #include "free-space-cache.h" |
fec577fb | 35 | |
0e4f8f88 CM |
36 | /* control flags for do_chunk_alloc's force field |
37 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk | |
38 | * if we really need one. | |
39 | * | |
40 | * CHUNK_ALLOC_FORCE means it must try to allocate one | |
41 | * | |
42 | * CHUNK_ALLOC_LIMITED means to only try and allocate one | |
43 | * if we have very few chunks already allocated. This is | |
44 | * used as part of the clustering code to help make sure | |
45 | * we have a good pool of storage to cluster in, without | |
46 | * filling the FS with empty chunks | |
47 | * | |
48 | */ | |
49 | enum { | |
50 | CHUNK_ALLOC_NO_FORCE = 0, | |
51 | CHUNK_ALLOC_FORCE = 1, | |
52 | CHUNK_ALLOC_LIMITED = 2, | |
53 | }; | |
54 | ||
f3465ca4 JB |
55 | static int update_block_group(struct btrfs_trans_handle *trans, |
56 | struct btrfs_root *root, | |
f0486c68 | 57 | u64 bytenr, u64 num_bytes, int alloc); |
5d4f98a2 YZ |
58 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
59 | struct btrfs_root *root, | |
60 | u64 bytenr, u64 num_bytes, u64 parent, | |
61 | u64 root_objectid, u64 owner_objectid, | |
62 | u64 owner_offset, int refs_to_drop, | |
63 | struct btrfs_delayed_extent_op *extra_op); | |
64 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
65 | struct extent_buffer *leaf, | |
66 | struct btrfs_extent_item *ei); | |
67 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
68 | struct btrfs_root *root, | |
69 | u64 parent, u64 root_objectid, | |
70 | u64 flags, u64 owner, u64 offset, | |
71 | struct btrfs_key *ins, int ref_mod); | |
72 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
73 | struct btrfs_root *root, | |
74 | u64 parent, u64 root_objectid, | |
75 | u64 flags, struct btrfs_disk_key *key, | |
76 | int level, struct btrfs_key *ins); | |
6a63209f JB |
77 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
78 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
79 | u64 flags, int force); | |
11833d66 YZ |
80 | static int find_next_key(struct btrfs_path *path, int level, |
81 | struct btrfs_key *key); | |
9ed74f2d JB |
82 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
83 | int dump_block_groups); | |
6a63209f | 84 | |
817d52f8 JB |
85 | static noinline int |
86 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
87 | { | |
88 | smp_mb(); | |
89 | return cache->cached == BTRFS_CACHE_FINISHED; | |
90 | } | |
91 | ||
0f9dd46c JB |
92 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
93 | { | |
94 | return (cache->flags & bits) == bits; | |
95 | } | |
96 | ||
62a45b60 | 97 | static void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
98 | { |
99 | atomic_inc(&cache->count); | |
100 | } | |
101 | ||
102 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
103 | { | |
f0486c68 YZ |
104 | if (atomic_dec_and_test(&cache->count)) { |
105 | WARN_ON(cache->pinned > 0); | |
106 | WARN_ON(cache->reserved > 0); | |
107 | WARN_ON(cache->reserved_pinned > 0); | |
34d52cb6 | 108 | kfree(cache->free_space_ctl); |
11dfe35a | 109 | kfree(cache); |
f0486c68 | 110 | } |
11dfe35a JB |
111 | } |
112 | ||
0f9dd46c JB |
113 | /* |
114 | * this adds the block group to the fs_info rb tree for the block group | |
115 | * cache | |
116 | */ | |
b2950863 | 117 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
118 | struct btrfs_block_group_cache *block_group) |
119 | { | |
120 | struct rb_node **p; | |
121 | struct rb_node *parent = NULL; | |
122 | struct btrfs_block_group_cache *cache; | |
123 | ||
124 | spin_lock(&info->block_group_cache_lock); | |
125 | p = &info->block_group_cache_tree.rb_node; | |
126 | ||
127 | while (*p) { | |
128 | parent = *p; | |
129 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
130 | cache_node); | |
131 | if (block_group->key.objectid < cache->key.objectid) { | |
132 | p = &(*p)->rb_left; | |
133 | } else if (block_group->key.objectid > cache->key.objectid) { | |
134 | p = &(*p)->rb_right; | |
135 | } else { | |
136 | spin_unlock(&info->block_group_cache_lock); | |
137 | return -EEXIST; | |
138 | } | |
139 | } | |
140 | ||
141 | rb_link_node(&block_group->cache_node, parent, p); | |
142 | rb_insert_color(&block_group->cache_node, | |
143 | &info->block_group_cache_tree); | |
144 | spin_unlock(&info->block_group_cache_lock); | |
145 | ||
146 | return 0; | |
147 | } | |
148 | ||
149 | /* | |
150 | * This will return the block group at or after bytenr if contains is 0, else | |
151 | * it will return the block group that contains the bytenr | |
152 | */ | |
153 | static struct btrfs_block_group_cache * | |
154 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
155 | int contains) | |
156 | { | |
157 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
158 | struct rb_node *n; | |
159 | u64 end, start; | |
160 | ||
161 | spin_lock(&info->block_group_cache_lock); | |
162 | n = info->block_group_cache_tree.rb_node; | |
163 | ||
164 | while (n) { | |
165 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
166 | cache_node); | |
167 | end = cache->key.objectid + cache->key.offset - 1; | |
168 | start = cache->key.objectid; | |
169 | ||
170 | if (bytenr < start) { | |
171 | if (!contains && (!ret || start < ret->key.objectid)) | |
172 | ret = cache; | |
173 | n = n->rb_left; | |
174 | } else if (bytenr > start) { | |
175 | if (contains && bytenr <= end) { | |
176 | ret = cache; | |
177 | break; | |
178 | } | |
179 | n = n->rb_right; | |
180 | } else { | |
181 | ret = cache; | |
182 | break; | |
183 | } | |
184 | } | |
d2fb3437 | 185 | if (ret) |
11dfe35a | 186 | btrfs_get_block_group(ret); |
0f9dd46c JB |
187 | spin_unlock(&info->block_group_cache_lock); |
188 | ||
189 | return ret; | |
190 | } | |
191 | ||
11833d66 YZ |
192 | static int add_excluded_extent(struct btrfs_root *root, |
193 | u64 start, u64 num_bytes) | |
817d52f8 | 194 | { |
11833d66 YZ |
195 | u64 end = start + num_bytes - 1; |
196 | set_extent_bits(&root->fs_info->freed_extents[0], | |
197 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
198 | set_extent_bits(&root->fs_info->freed_extents[1], | |
199 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
200 | return 0; | |
201 | } | |
817d52f8 | 202 | |
11833d66 YZ |
203 | static void free_excluded_extents(struct btrfs_root *root, |
204 | struct btrfs_block_group_cache *cache) | |
205 | { | |
206 | u64 start, end; | |
817d52f8 | 207 | |
11833d66 YZ |
208 | start = cache->key.objectid; |
209 | end = start + cache->key.offset - 1; | |
210 | ||
211 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
212 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
213 | clear_extent_bits(&root->fs_info->freed_extents[1], | |
214 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
817d52f8 JB |
215 | } |
216 | ||
11833d66 YZ |
217 | static int exclude_super_stripes(struct btrfs_root *root, |
218 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 219 | { |
817d52f8 JB |
220 | u64 bytenr; |
221 | u64 *logical; | |
222 | int stripe_len; | |
223 | int i, nr, ret; | |
224 | ||
06b2331f YZ |
225 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
226 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
227 | cache->bytes_super += stripe_len; | |
228 | ret = add_excluded_extent(root, cache->key.objectid, | |
229 | stripe_len); | |
230 | BUG_ON(ret); | |
231 | } | |
232 | ||
817d52f8 JB |
233 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
234 | bytenr = btrfs_sb_offset(i); | |
235 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
236 | cache->key.objectid, bytenr, | |
237 | 0, &logical, &nr, &stripe_len); | |
238 | BUG_ON(ret); | |
11833d66 | 239 | |
817d52f8 | 240 | while (nr--) { |
1b2da372 | 241 | cache->bytes_super += stripe_len; |
11833d66 YZ |
242 | ret = add_excluded_extent(root, logical[nr], |
243 | stripe_len); | |
244 | BUG_ON(ret); | |
817d52f8 | 245 | } |
11833d66 | 246 | |
817d52f8 JB |
247 | kfree(logical); |
248 | } | |
817d52f8 JB |
249 | return 0; |
250 | } | |
251 | ||
11833d66 YZ |
252 | static struct btrfs_caching_control * |
253 | get_caching_control(struct btrfs_block_group_cache *cache) | |
254 | { | |
255 | struct btrfs_caching_control *ctl; | |
256 | ||
257 | spin_lock(&cache->lock); | |
258 | if (cache->cached != BTRFS_CACHE_STARTED) { | |
259 | spin_unlock(&cache->lock); | |
260 | return NULL; | |
261 | } | |
262 | ||
dde5abee JB |
263 | /* We're loading it the fast way, so we don't have a caching_ctl. */ |
264 | if (!cache->caching_ctl) { | |
265 | spin_unlock(&cache->lock); | |
11833d66 YZ |
266 | return NULL; |
267 | } | |
268 | ||
269 | ctl = cache->caching_ctl; | |
270 | atomic_inc(&ctl->count); | |
271 | spin_unlock(&cache->lock); | |
272 | return ctl; | |
273 | } | |
274 | ||
275 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
276 | { | |
277 | if (atomic_dec_and_test(&ctl->count)) | |
278 | kfree(ctl); | |
279 | } | |
280 | ||
0f9dd46c JB |
281 | /* |
282 | * this is only called by cache_block_group, since we could have freed extents | |
283 | * we need to check the pinned_extents for any extents that can't be used yet | |
284 | * since their free space will be released as soon as the transaction commits. | |
285 | */ | |
817d52f8 | 286 | static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
0f9dd46c JB |
287 | struct btrfs_fs_info *info, u64 start, u64 end) |
288 | { | |
817d52f8 | 289 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
290 | int ret; |
291 | ||
292 | while (start < end) { | |
11833d66 | 293 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 294 | &extent_start, &extent_end, |
11833d66 | 295 | EXTENT_DIRTY | EXTENT_UPTODATE); |
0f9dd46c JB |
296 | if (ret) |
297 | break; | |
298 | ||
06b2331f | 299 | if (extent_start <= start) { |
0f9dd46c JB |
300 | start = extent_end + 1; |
301 | } else if (extent_start > start && extent_start < end) { | |
302 | size = extent_start - start; | |
817d52f8 | 303 | total_added += size; |
ea6a478e JB |
304 | ret = btrfs_add_free_space(block_group, start, |
305 | size); | |
0f9dd46c JB |
306 | BUG_ON(ret); |
307 | start = extent_end + 1; | |
308 | } else { | |
309 | break; | |
310 | } | |
311 | } | |
312 | ||
313 | if (start < end) { | |
314 | size = end - start; | |
817d52f8 | 315 | total_added += size; |
ea6a478e | 316 | ret = btrfs_add_free_space(block_group, start, size); |
0f9dd46c JB |
317 | BUG_ON(ret); |
318 | } | |
319 | ||
817d52f8 | 320 | return total_added; |
0f9dd46c JB |
321 | } |
322 | ||
bab39bf9 | 323 | static noinline void caching_thread(struct btrfs_work *work) |
e37c9e69 | 324 | { |
bab39bf9 JB |
325 | struct btrfs_block_group_cache *block_group; |
326 | struct btrfs_fs_info *fs_info; | |
327 | struct btrfs_caching_control *caching_ctl; | |
328 | struct btrfs_root *extent_root; | |
e37c9e69 | 329 | struct btrfs_path *path; |
5f39d397 | 330 | struct extent_buffer *leaf; |
11833d66 | 331 | struct btrfs_key key; |
817d52f8 | 332 | u64 total_found = 0; |
11833d66 YZ |
333 | u64 last = 0; |
334 | u32 nritems; | |
335 | int ret = 0; | |
f510cfec | 336 | |
bab39bf9 JB |
337 | caching_ctl = container_of(work, struct btrfs_caching_control, work); |
338 | block_group = caching_ctl->block_group; | |
339 | fs_info = block_group->fs_info; | |
340 | extent_root = fs_info->extent_root; | |
341 | ||
e37c9e69 CM |
342 | path = btrfs_alloc_path(); |
343 | if (!path) | |
bab39bf9 | 344 | goto out; |
7d7d6068 | 345 | |
817d52f8 | 346 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 347 | |
5cd57b2c | 348 | /* |
817d52f8 JB |
349 | * We don't want to deadlock with somebody trying to allocate a new |
350 | * extent for the extent root while also trying to search the extent | |
351 | * root to add free space. So we skip locking and search the commit | |
352 | * root, since its read-only | |
5cd57b2c CM |
353 | */ |
354 | path->skip_locking = 1; | |
817d52f8 | 355 | path->search_commit_root = 1; |
026fd317 | 356 | path->reada = 1; |
817d52f8 | 357 | |
e4404d6e | 358 | key.objectid = last; |
e37c9e69 | 359 | key.offset = 0; |
11833d66 | 360 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 361 | again: |
11833d66 | 362 | mutex_lock(&caching_ctl->mutex); |
013f1b12 CM |
363 | /* need to make sure the commit_root doesn't disappear */ |
364 | down_read(&fs_info->extent_commit_sem); | |
365 | ||
11833d66 | 366 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 367 | if (ret < 0) |
ef8bbdfe | 368 | goto err; |
a512bbf8 | 369 | |
11833d66 YZ |
370 | leaf = path->nodes[0]; |
371 | nritems = btrfs_header_nritems(leaf); | |
372 | ||
d397712b | 373 | while (1) { |
7841cb28 | 374 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 375 | last = (u64)-1; |
817d52f8 | 376 | break; |
f25784b3 | 377 | } |
817d52f8 | 378 | |
11833d66 YZ |
379 | if (path->slots[0] < nritems) { |
380 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
381 | } else { | |
382 | ret = find_next_key(path, 0, &key); | |
383 | if (ret) | |
e37c9e69 | 384 | break; |
817d52f8 | 385 | |
589d8ade JB |
386 | if (need_resched() || |
387 | btrfs_next_leaf(extent_root, path)) { | |
388 | caching_ctl->progress = last; | |
ff5714cc | 389 | btrfs_release_path(path); |
589d8ade JB |
390 | up_read(&fs_info->extent_commit_sem); |
391 | mutex_unlock(&caching_ctl->mutex); | |
11833d66 | 392 | cond_resched(); |
589d8ade JB |
393 | goto again; |
394 | } | |
395 | leaf = path->nodes[0]; | |
396 | nritems = btrfs_header_nritems(leaf); | |
397 | continue; | |
11833d66 | 398 | } |
817d52f8 | 399 | |
11833d66 YZ |
400 | if (key.objectid < block_group->key.objectid) { |
401 | path->slots[0]++; | |
817d52f8 | 402 | continue; |
e37c9e69 | 403 | } |
0f9dd46c | 404 | |
e37c9e69 | 405 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 406 | block_group->key.offset) |
e37c9e69 | 407 | break; |
7d7d6068 | 408 | |
11833d66 | 409 | if (key.type == BTRFS_EXTENT_ITEM_KEY) { |
817d52f8 JB |
410 | total_found += add_new_free_space(block_group, |
411 | fs_info, last, | |
412 | key.objectid); | |
7d7d6068 | 413 | last = key.objectid + key.offset; |
817d52f8 | 414 | |
11833d66 YZ |
415 | if (total_found > (1024 * 1024 * 2)) { |
416 | total_found = 0; | |
417 | wake_up(&caching_ctl->wait); | |
418 | } | |
817d52f8 | 419 | } |
e37c9e69 CM |
420 | path->slots[0]++; |
421 | } | |
817d52f8 | 422 | ret = 0; |
e37c9e69 | 423 | |
817d52f8 JB |
424 | total_found += add_new_free_space(block_group, fs_info, last, |
425 | block_group->key.objectid + | |
426 | block_group->key.offset); | |
11833d66 | 427 | caching_ctl->progress = (u64)-1; |
817d52f8 JB |
428 | |
429 | spin_lock(&block_group->lock); | |
11833d66 | 430 | block_group->caching_ctl = NULL; |
817d52f8 JB |
431 | block_group->cached = BTRFS_CACHE_FINISHED; |
432 | spin_unlock(&block_group->lock); | |
0f9dd46c | 433 | |
54aa1f4d | 434 | err: |
e37c9e69 | 435 | btrfs_free_path(path); |
276e680d | 436 | up_read(&fs_info->extent_commit_sem); |
817d52f8 | 437 | |
11833d66 YZ |
438 | free_excluded_extents(extent_root, block_group); |
439 | ||
440 | mutex_unlock(&caching_ctl->mutex); | |
bab39bf9 | 441 | out: |
11833d66 YZ |
442 | wake_up(&caching_ctl->wait); |
443 | ||
444 | put_caching_control(caching_ctl); | |
11dfe35a | 445 | btrfs_put_block_group(block_group); |
817d52f8 JB |
446 | } |
447 | ||
9d66e233 JB |
448 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
449 | struct btrfs_trans_handle *trans, | |
b8399dee | 450 | struct btrfs_root *root, |
9d66e233 | 451 | int load_cache_only) |
817d52f8 | 452 | { |
11833d66 YZ |
453 | struct btrfs_fs_info *fs_info = cache->fs_info; |
454 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
455 | int ret = 0; |
456 | ||
11833d66 YZ |
457 | smp_mb(); |
458 | if (cache->cached != BTRFS_CACHE_NO) | |
459 | return 0; | |
460 | ||
9d66e233 JB |
461 | /* |
462 | * We can't do the read from on-disk cache during a commit since we need | |
b8399dee JB |
463 | * to have the normal tree locking. Also if we are currently trying to |
464 | * allocate blocks for the tree root we can't do the fast caching since | |
465 | * we likely hold important locks. | |
9d66e233 | 466 | */ |
f7039b1d | 467 | if (trans && (!trans->transaction->in_commit) && |
b8399dee | 468 | (root && root != root->fs_info->tree_root)) { |
9d66e233 JB |
469 | spin_lock(&cache->lock); |
470 | if (cache->cached != BTRFS_CACHE_NO) { | |
471 | spin_unlock(&cache->lock); | |
472 | return 0; | |
473 | } | |
474 | cache->cached = BTRFS_CACHE_STARTED; | |
475 | spin_unlock(&cache->lock); | |
476 | ||
477 | ret = load_free_space_cache(fs_info, cache); | |
478 | ||
479 | spin_lock(&cache->lock); | |
480 | if (ret == 1) { | |
481 | cache->cached = BTRFS_CACHE_FINISHED; | |
482 | cache->last_byte_to_unpin = (u64)-1; | |
483 | } else { | |
484 | cache->cached = BTRFS_CACHE_NO; | |
485 | } | |
486 | spin_unlock(&cache->lock); | |
3c14874a JB |
487 | if (ret == 1) { |
488 | free_excluded_extents(fs_info->extent_root, cache); | |
9d66e233 | 489 | return 0; |
3c14874a | 490 | } |
9d66e233 JB |
491 | } |
492 | ||
493 | if (load_cache_only) | |
494 | return 0; | |
495 | ||
fc0e4a31 | 496 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
11833d66 YZ |
497 | BUG_ON(!caching_ctl); |
498 | ||
499 | INIT_LIST_HEAD(&caching_ctl->list); | |
500 | mutex_init(&caching_ctl->mutex); | |
501 | init_waitqueue_head(&caching_ctl->wait); | |
502 | caching_ctl->block_group = cache; | |
503 | caching_ctl->progress = cache->key.objectid; | |
504 | /* one for caching kthread, one for caching block group list */ | |
505 | atomic_set(&caching_ctl->count, 2); | |
bab39bf9 | 506 | caching_ctl->work.func = caching_thread; |
11833d66 | 507 | |
817d52f8 JB |
508 | spin_lock(&cache->lock); |
509 | if (cache->cached != BTRFS_CACHE_NO) { | |
510 | spin_unlock(&cache->lock); | |
11833d66 YZ |
511 | kfree(caching_ctl); |
512 | return 0; | |
817d52f8 | 513 | } |
11833d66 | 514 | cache->caching_ctl = caching_ctl; |
817d52f8 JB |
515 | cache->cached = BTRFS_CACHE_STARTED; |
516 | spin_unlock(&cache->lock); | |
517 | ||
11833d66 YZ |
518 | down_write(&fs_info->extent_commit_sem); |
519 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); | |
520 | up_write(&fs_info->extent_commit_sem); | |
521 | ||
11dfe35a | 522 | btrfs_get_block_group(cache); |
11833d66 | 523 | |
bab39bf9 | 524 | btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 525 | |
ef8bbdfe | 526 | return ret; |
e37c9e69 CM |
527 | } |
528 | ||
0f9dd46c JB |
529 | /* |
530 | * return the block group that starts at or after bytenr | |
531 | */ | |
d397712b CM |
532 | static struct btrfs_block_group_cache * |
533 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 534 | { |
0f9dd46c | 535 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 536 | |
0f9dd46c | 537 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 538 | |
0f9dd46c | 539 | return cache; |
0ef3e66b CM |
540 | } |
541 | ||
0f9dd46c | 542 | /* |
9f55684c | 543 | * return the block group that contains the given bytenr |
0f9dd46c | 544 | */ |
d397712b CM |
545 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
546 | struct btrfs_fs_info *info, | |
547 | u64 bytenr) | |
be744175 | 548 | { |
0f9dd46c | 549 | struct btrfs_block_group_cache *cache; |
be744175 | 550 | |
0f9dd46c | 551 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 552 | |
0f9dd46c | 553 | return cache; |
be744175 | 554 | } |
0b86a832 | 555 | |
0f9dd46c JB |
556 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
557 | u64 flags) | |
6324fbf3 | 558 | { |
0f9dd46c | 559 | struct list_head *head = &info->space_info; |
0f9dd46c | 560 | struct btrfs_space_info *found; |
4184ea7f | 561 | |
b742bb82 YZ |
562 | flags &= BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_SYSTEM | |
563 | BTRFS_BLOCK_GROUP_METADATA; | |
564 | ||
4184ea7f CM |
565 | rcu_read_lock(); |
566 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 567 | if (found->flags & flags) { |
4184ea7f | 568 | rcu_read_unlock(); |
0f9dd46c | 569 | return found; |
4184ea7f | 570 | } |
0f9dd46c | 571 | } |
4184ea7f | 572 | rcu_read_unlock(); |
0f9dd46c | 573 | return NULL; |
6324fbf3 CM |
574 | } |
575 | ||
4184ea7f CM |
576 | /* |
577 | * after adding space to the filesystem, we need to clear the full flags | |
578 | * on all the space infos. | |
579 | */ | |
580 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
581 | { | |
582 | struct list_head *head = &info->space_info; | |
583 | struct btrfs_space_info *found; | |
584 | ||
585 | rcu_read_lock(); | |
586 | list_for_each_entry_rcu(found, head, list) | |
587 | found->full = 0; | |
588 | rcu_read_unlock(); | |
589 | } | |
590 | ||
80eb234a JB |
591 | static u64 div_factor(u64 num, int factor) |
592 | { | |
593 | if (factor == 10) | |
594 | return num; | |
595 | num *= factor; | |
596 | do_div(num, 10); | |
597 | return num; | |
598 | } | |
599 | ||
e5bc2458 CM |
600 | static u64 div_factor_fine(u64 num, int factor) |
601 | { | |
602 | if (factor == 100) | |
603 | return num; | |
604 | num *= factor; | |
605 | do_div(num, 100); | |
606 | return num; | |
607 | } | |
608 | ||
d2fb3437 YZ |
609 | u64 btrfs_find_block_group(struct btrfs_root *root, |
610 | u64 search_start, u64 search_hint, int owner) | |
cd1bc465 | 611 | { |
96b5179d | 612 | struct btrfs_block_group_cache *cache; |
cd1bc465 | 613 | u64 used; |
d2fb3437 YZ |
614 | u64 last = max(search_hint, search_start); |
615 | u64 group_start = 0; | |
31f3c99b | 616 | int full_search = 0; |
d2fb3437 | 617 | int factor = 9; |
0ef3e66b | 618 | int wrapped = 0; |
31f3c99b | 619 | again: |
e8569813 ZY |
620 | while (1) { |
621 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
0f9dd46c JB |
622 | if (!cache) |
623 | break; | |
96b5179d | 624 | |
c286ac48 | 625 | spin_lock(&cache->lock); |
96b5179d CM |
626 | last = cache->key.objectid + cache->key.offset; |
627 | used = btrfs_block_group_used(&cache->item); | |
628 | ||
d2fb3437 YZ |
629 | if ((full_search || !cache->ro) && |
630 | block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) { | |
e8569813 | 631 | if (used + cache->pinned + cache->reserved < |
d2fb3437 YZ |
632 | div_factor(cache->key.offset, factor)) { |
633 | group_start = cache->key.objectid; | |
c286ac48 | 634 | spin_unlock(&cache->lock); |
fa9c0d79 | 635 | btrfs_put_block_group(cache); |
8790d502 CM |
636 | goto found; |
637 | } | |
6324fbf3 | 638 | } |
c286ac48 | 639 | spin_unlock(&cache->lock); |
fa9c0d79 | 640 | btrfs_put_block_group(cache); |
de428b63 | 641 | cond_resched(); |
cd1bc465 | 642 | } |
0ef3e66b CM |
643 | if (!wrapped) { |
644 | last = search_start; | |
645 | wrapped = 1; | |
646 | goto again; | |
647 | } | |
648 | if (!full_search && factor < 10) { | |
be744175 | 649 | last = search_start; |
31f3c99b | 650 | full_search = 1; |
0ef3e66b | 651 | factor = 10; |
31f3c99b CM |
652 | goto again; |
653 | } | |
be744175 | 654 | found: |
d2fb3437 | 655 | return group_start; |
925baedd | 656 | } |
0f9dd46c | 657 | |
e02119d5 | 658 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 659 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
660 | { |
661 | int ret; | |
662 | struct btrfs_key key; | |
31840ae1 | 663 | struct btrfs_path *path; |
e02119d5 | 664 | |
31840ae1 | 665 | path = btrfs_alloc_path(); |
d8926bb3 MF |
666 | if (!path) |
667 | return -ENOMEM; | |
668 | ||
e02119d5 CM |
669 | key.objectid = start; |
670 | key.offset = len; | |
671 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
672 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, | |
673 | 0, 0); | |
31840ae1 | 674 | btrfs_free_path(path); |
7bb86316 CM |
675 | return ret; |
676 | } | |
677 | ||
a22285a6 YZ |
678 | /* |
679 | * helper function to lookup reference count and flags of extent. | |
680 | * | |
681 | * the head node for delayed ref is used to store the sum of all the | |
682 | * reference count modifications queued up in the rbtree. the head | |
683 | * node may also store the extent flags to set. This way you can check | |
684 | * to see what the reference count and extent flags would be if all of | |
685 | * the delayed refs are not processed. | |
686 | */ | |
687 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
688 | struct btrfs_root *root, u64 bytenr, | |
689 | u64 num_bytes, u64 *refs, u64 *flags) | |
690 | { | |
691 | struct btrfs_delayed_ref_head *head; | |
692 | struct btrfs_delayed_ref_root *delayed_refs; | |
693 | struct btrfs_path *path; | |
694 | struct btrfs_extent_item *ei; | |
695 | struct extent_buffer *leaf; | |
696 | struct btrfs_key key; | |
697 | u32 item_size; | |
698 | u64 num_refs; | |
699 | u64 extent_flags; | |
700 | int ret; | |
701 | ||
702 | path = btrfs_alloc_path(); | |
703 | if (!path) | |
704 | return -ENOMEM; | |
705 | ||
706 | key.objectid = bytenr; | |
707 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
708 | key.offset = num_bytes; | |
709 | if (!trans) { | |
710 | path->skip_locking = 1; | |
711 | path->search_commit_root = 1; | |
712 | } | |
713 | again: | |
714 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, | |
715 | &key, path, 0, 0); | |
716 | if (ret < 0) | |
717 | goto out_free; | |
718 | ||
719 | if (ret == 0) { | |
720 | leaf = path->nodes[0]; | |
721 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
722 | if (item_size >= sizeof(*ei)) { | |
723 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
724 | struct btrfs_extent_item); | |
725 | num_refs = btrfs_extent_refs(leaf, ei); | |
726 | extent_flags = btrfs_extent_flags(leaf, ei); | |
727 | } else { | |
728 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
729 | struct btrfs_extent_item_v0 *ei0; | |
730 | BUG_ON(item_size != sizeof(*ei0)); | |
731 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
732 | struct btrfs_extent_item_v0); | |
733 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
734 | /* FIXME: this isn't correct for data */ | |
735 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
736 | #else | |
737 | BUG(); | |
738 | #endif | |
739 | } | |
740 | BUG_ON(num_refs == 0); | |
741 | } else { | |
742 | num_refs = 0; | |
743 | extent_flags = 0; | |
744 | ret = 0; | |
745 | } | |
746 | ||
747 | if (!trans) | |
748 | goto out; | |
749 | ||
750 | delayed_refs = &trans->transaction->delayed_refs; | |
751 | spin_lock(&delayed_refs->lock); | |
752 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
753 | if (head) { | |
754 | if (!mutex_trylock(&head->mutex)) { | |
755 | atomic_inc(&head->node.refs); | |
756 | spin_unlock(&delayed_refs->lock); | |
757 | ||
b3b4aa74 | 758 | btrfs_release_path(path); |
a22285a6 | 759 | |
8cc33e5c DS |
760 | /* |
761 | * Mutex was contended, block until it's released and try | |
762 | * again | |
763 | */ | |
a22285a6 YZ |
764 | mutex_lock(&head->mutex); |
765 | mutex_unlock(&head->mutex); | |
766 | btrfs_put_delayed_ref(&head->node); | |
767 | goto again; | |
768 | } | |
769 | if (head->extent_op && head->extent_op->update_flags) | |
770 | extent_flags |= head->extent_op->flags_to_set; | |
771 | else | |
772 | BUG_ON(num_refs == 0); | |
773 | ||
774 | num_refs += head->node.ref_mod; | |
775 | mutex_unlock(&head->mutex); | |
776 | } | |
777 | spin_unlock(&delayed_refs->lock); | |
778 | out: | |
779 | WARN_ON(num_refs == 0); | |
780 | if (refs) | |
781 | *refs = num_refs; | |
782 | if (flags) | |
783 | *flags = extent_flags; | |
784 | out_free: | |
785 | btrfs_free_path(path); | |
786 | return ret; | |
787 | } | |
788 | ||
d8d5f3e1 CM |
789 | /* |
790 | * Back reference rules. Back refs have three main goals: | |
791 | * | |
792 | * 1) differentiate between all holders of references to an extent so that | |
793 | * when a reference is dropped we can make sure it was a valid reference | |
794 | * before freeing the extent. | |
795 | * | |
796 | * 2) Provide enough information to quickly find the holders of an extent | |
797 | * if we notice a given block is corrupted or bad. | |
798 | * | |
799 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
800 | * maintenance. This is actually the same as #2, but with a slightly | |
801 | * different use case. | |
802 | * | |
5d4f98a2 YZ |
803 | * There are two kinds of back refs. The implicit back refs is optimized |
804 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
805 | * back refs of this kind provide information about the block's owner tree | |
806 | * and the pointer's key. These information allow us to find the block by | |
807 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
808 | * referenced by their owner trees. The location of tree block is recorded | |
809 | * in the back refs. Actually the full back refs is generic, and can be | |
810 | * used in all cases the implicit back refs is used. The major shortcoming | |
811 | * of the full back refs is its overhead. Every time a tree block gets | |
812 | * COWed, we have to update back refs entry for all pointers in it. | |
813 | * | |
814 | * For a newly allocated tree block, we use implicit back refs for | |
815 | * pointers in it. This means most tree related operations only involve | |
816 | * implicit back refs. For a tree block created in old transaction, the | |
817 | * only way to drop a reference to it is COW it. So we can detect the | |
818 | * event that tree block loses its owner tree's reference and do the | |
819 | * back refs conversion. | |
820 | * | |
821 | * When a tree block is COW'd through a tree, there are four cases: | |
822 | * | |
823 | * The reference count of the block is one and the tree is the block's | |
824 | * owner tree. Nothing to do in this case. | |
825 | * | |
826 | * The reference count of the block is one and the tree is not the | |
827 | * block's owner tree. In this case, full back refs is used for pointers | |
828 | * in the block. Remove these full back refs, add implicit back refs for | |
829 | * every pointers in the new block. | |
830 | * | |
831 | * The reference count of the block is greater than one and the tree is | |
832 | * the block's owner tree. In this case, implicit back refs is used for | |
833 | * pointers in the block. Add full back refs for every pointers in the | |
834 | * block, increase lower level extents' reference counts. The original | |
835 | * implicit back refs are entailed to the new block. | |
836 | * | |
837 | * The reference count of the block is greater than one and the tree is | |
838 | * not the block's owner tree. Add implicit back refs for every pointer in | |
839 | * the new block, increase lower level extents' reference count. | |
840 | * | |
841 | * Back Reference Key composing: | |
842 | * | |
843 | * The key objectid corresponds to the first byte in the extent, | |
844 | * The key type is used to differentiate between types of back refs. | |
845 | * There are different meanings of the key offset for different types | |
846 | * of back refs. | |
847 | * | |
d8d5f3e1 CM |
848 | * File extents can be referenced by: |
849 | * | |
850 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 851 | * - different files inside a single subvolume |
d8d5f3e1 CM |
852 | * - different offsets inside a file (bookend extents in file.c) |
853 | * | |
5d4f98a2 | 854 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
855 | * |
856 | * - Objectid of the subvolume root | |
d8d5f3e1 | 857 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
858 | * - original offset in the file |
859 | * - how many bookend extents | |
d8d5f3e1 | 860 | * |
5d4f98a2 YZ |
861 | * The key offset for the implicit back refs is hash of the first |
862 | * three fields. | |
d8d5f3e1 | 863 | * |
5d4f98a2 | 864 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 865 | * |
5d4f98a2 | 866 | * - number of pointers in the tree leaf |
d8d5f3e1 | 867 | * |
5d4f98a2 YZ |
868 | * The key offset for the implicit back refs is the first byte of |
869 | * the tree leaf | |
d8d5f3e1 | 870 | * |
5d4f98a2 YZ |
871 | * When a file extent is allocated, The implicit back refs is used. |
872 | * the fields are filled in: | |
d8d5f3e1 | 873 | * |
5d4f98a2 | 874 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 875 | * |
5d4f98a2 YZ |
876 | * When a file extent is removed file truncation, we find the |
877 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 878 | * |
5d4f98a2 | 879 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 880 | * |
5d4f98a2 | 881 | * Btree extents can be referenced by: |
d8d5f3e1 | 882 | * |
5d4f98a2 | 883 | * - Different subvolumes |
d8d5f3e1 | 884 | * |
5d4f98a2 YZ |
885 | * Both the implicit back refs and the full back refs for tree blocks |
886 | * only consist of key. The key offset for the implicit back refs is | |
887 | * objectid of block's owner tree. The key offset for the full back refs | |
888 | * is the first byte of parent block. | |
d8d5f3e1 | 889 | * |
5d4f98a2 YZ |
890 | * When implicit back refs is used, information about the lowest key and |
891 | * level of the tree block are required. These information are stored in | |
892 | * tree block info structure. | |
d8d5f3e1 | 893 | */ |
31840ae1 | 894 | |
5d4f98a2 YZ |
895 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
896 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
897 | struct btrfs_root *root, | |
898 | struct btrfs_path *path, | |
899 | u64 owner, u32 extra_size) | |
7bb86316 | 900 | { |
5d4f98a2 YZ |
901 | struct btrfs_extent_item *item; |
902 | struct btrfs_extent_item_v0 *ei0; | |
903 | struct btrfs_extent_ref_v0 *ref0; | |
904 | struct btrfs_tree_block_info *bi; | |
905 | struct extent_buffer *leaf; | |
7bb86316 | 906 | struct btrfs_key key; |
5d4f98a2 YZ |
907 | struct btrfs_key found_key; |
908 | u32 new_size = sizeof(*item); | |
909 | u64 refs; | |
910 | int ret; | |
911 | ||
912 | leaf = path->nodes[0]; | |
913 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
914 | ||
915 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
916 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
917 | struct btrfs_extent_item_v0); | |
918 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
919 | ||
920 | if (owner == (u64)-1) { | |
921 | while (1) { | |
922 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
923 | ret = btrfs_next_leaf(root, path); | |
924 | if (ret < 0) | |
925 | return ret; | |
926 | BUG_ON(ret > 0); | |
927 | leaf = path->nodes[0]; | |
928 | } | |
929 | btrfs_item_key_to_cpu(leaf, &found_key, | |
930 | path->slots[0]); | |
931 | BUG_ON(key.objectid != found_key.objectid); | |
932 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
933 | path->slots[0]++; | |
934 | continue; | |
935 | } | |
936 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
937 | struct btrfs_extent_ref_v0); | |
938 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
939 | break; | |
940 | } | |
941 | } | |
b3b4aa74 | 942 | btrfs_release_path(path); |
5d4f98a2 YZ |
943 | |
944 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
945 | new_size += sizeof(*bi); | |
946 | ||
947 | new_size -= sizeof(*ei0); | |
948 | ret = btrfs_search_slot(trans, root, &key, path, | |
949 | new_size + extra_size, 1); | |
950 | if (ret < 0) | |
951 | return ret; | |
952 | BUG_ON(ret); | |
953 | ||
954 | ret = btrfs_extend_item(trans, root, path, new_size); | |
5d4f98a2 YZ |
955 | |
956 | leaf = path->nodes[0]; | |
957 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
958 | btrfs_set_extent_refs(leaf, item, refs); | |
959 | /* FIXME: get real generation */ | |
960 | btrfs_set_extent_generation(leaf, item, 0); | |
961 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
962 | btrfs_set_extent_flags(leaf, item, | |
963 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
964 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
965 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
966 | /* FIXME: get first key of the block */ | |
967 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
968 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
969 | } else { | |
970 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
971 | } | |
972 | btrfs_mark_buffer_dirty(leaf); | |
973 | return 0; | |
974 | } | |
975 | #endif | |
976 | ||
977 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
978 | { | |
979 | u32 high_crc = ~(u32)0; | |
980 | u32 low_crc = ~(u32)0; | |
981 | __le64 lenum; | |
982 | ||
983 | lenum = cpu_to_le64(root_objectid); | |
163e783e | 984 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 985 | lenum = cpu_to_le64(owner); |
163e783e | 986 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 987 | lenum = cpu_to_le64(offset); |
163e783e | 988 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
989 | |
990 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
991 | } | |
992 | ||
993 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
994 | struct btrfs_extent_data_ref *ref) | |
995 | { | |
996 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
997 | btrfs_extent_data_ref_objectid(leaf, ref), | |
998 | btrfs_extent_data_ref_offset(leaf, ref)); | |
999 | } | |
1000 | ||
1001 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1002 | struct btrfs_extent_data_ref *ref, | |
1003 | u64 root_objectid, u64 owner, u64 offset) | |
1004 | { | |
1005 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1006 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1007 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1008 | return 0; | |
1009 | return 1; | |
1010 | } | |
1011 | ||
1012 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
1013 | struct btrfs_root *root, | |
1014 | struct btrfs_path *path, | |
1015 | u64 bytenr, u64 parent, | |
1016 | u64 root_objectid, | |
1017 | u64 owner, u64 offset) | |
1018 | { | |
1019 | struct btrfs_key key; | |
1020 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1021 | struct extent_buffer *leaf; |
5d4f98a2 | 1022 | u32 nritems; |
74493f7a | 1023 | int ret; |
5d4f98a2 YZ |
1024 | int recow; |
1025 | int err = -ENOENT; | |
74493f7a | 1026 | |
31840ae1 | 1027 | key.objectid = bytenr; |
5d4f98a2 YZ |
1028 | if (parent) { |
1029 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1030 | key.offset = parent; | |
1031 | } else { | |
1032 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1033 | key.offset = hash_extent_data_ref(root_objectid, | |
1034 | owner, offset); | |
1035 | } | |
1036 | again: | |
1037 | recow = 0; | |
1038 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1039 | if (ret < 0) { | |
1040 | err = ret; | |
1041 | goto fail; | |
1042 | } | |
31840ae1 | 1043 | |
5d4f98a2 YZ |
1044 | if (parent) { |
1045 | if (!ret) | |
1046 | return 0; | |
1047 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1048 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1049 | btrfs_release_path(path); |
5d4f98a2 YZ |
1050 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1051 | if (ret < 0) { | |
1052 | err = ret; | |
1053 | goto fail; | |
1054 | } | |
1055 | if (!ret) | |
1056 | return 0; | |
1057 | #endif | |
1058 | goto fail; | |
31840ae1 ZY |
1059 | } |
1060 | ||
1061 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1062 | nritems = btrfs_header_nritems(leaf); |
1063 | while (1) { | |
1064 | if (path->slots[0] >= nritems) { | |
1065 | ret = btrfs_next_leaf(root, path); | |
1066 | if (ret < 0) | |
1067 | err = ret; | |
1068 | if (ret) | |
1069 | goto fail; | |
1070 | ||
1071 | leaf = path->nodes[0]; | |
1072 | nritems = btrfs_header_nritems(leaf); | |
1073 | recow = 1; | |
1074 | } | |
1075 | ||
1076 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1077 | if (key.objectid != bytenr || | |
1078 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1079 | goto fail; | |
1080 | ||
1081 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1082 | struct btrfs_extent_data_ref); | |
1083 | ||
1084 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1085 | owner, offset)) { | |
1086 | if (recow) { | |
b3b4aa74 | 1087 | btrfs_release_path(path); |
5d4f98a2 YZ |
1088 | goto again; |
1089 | } | |
1090 | err = 0; | |
1091 | break; | |
1092 | } | |
1093 | path->slots[0]++; | |
31840ae1 | 1094 | } |
5d4f98a2 YZ |
1095 | fail: |
1096 | return err; | |
31840ae1 ZY |
1097 | } |
1098 | ||
5d4f98a2 YZ |
1099 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
1100 | struct btrfs_root *root, | |
1101 | struct btrfs_path *path, | |
1102 | u64 bytenr, u64 parent, | |
1103 | u64 root_objectid, u64 owner, | |
1104 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
1105 | { |
1106 | struct btrfs_key key; | |
1107 | struct extent_buffer *leaf; | |
5d4f98a2 | 1108 | u32 size; |
31840ae1 ZY |
1109 | u32 num_refs; |
1110 | int ret; | |
74493f7a | 1111 | |
74493f7a | 1112 | key.objectid = bytenr; |
5d4f98a2 YZ |
1113 | if (parent) { |
1114 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1115 | key.offset = parent; | |
1116 | size = sizeof(struct btrfs_shared_data_ref); | |
1117 | } else { | |
1118 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1119 | key.offset = hash_extent_data_ref(root_objectid, | |
1120 | owner, offset); | |
1121 | size = sizeof(struct btrfs_extent_data_ref); | |
1122 | } | |
74493f7a | 1123 | |
5d4f98a2 YZ |
1124 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1125 | if (ret && ret != -EEXIST) | |
1126 | goto fail; | |
1127 | ||
1128 | leaf = path->nodes[0]; | |
1129 | if (parent) { | |
1130 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1131 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1132 | struct btrfs_shared_data_ref); |
1133 | if (ret == 0) { | |
1134 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1135 | } else { | |
1136 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1137 | num_refs += refs_to_add; | |
1138 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1139 | } |
5d4f98a2 YZ |
1140 | } else { |
1141 | struct btrfs_extent_data_ref *ref; | |
1142 | while (ret == -EEXIST) { | |
1143 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1144 | struct btrfs_extent_data_ref); | |
1145 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1146 | owner, offset)) | |
1147 | break; | |
b3b4aa74 | 1148 | btrfs_release_path(path); |
5d4f98a2 YZ |
1149 | key.offset++; |
1150 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1151 | size); | |
1152 | if (ret && ret != -EEXIST) | |
1153 | goto fail; | |
31840ae1 | 1154 | |
5d4f98a2 YZ |
1155 | leaf = path->nodes[0]; |
1156 | } | |
1157 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1158 | struct btrfs_extent_data_ref); | |
1159 | if (ret == 0) { | |
1160 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1161 | root_objectid); | |
1162 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1163 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1164 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1165 | } else { | |
1166 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1167 | num_refs += refs_to_add; | |
1168 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1169 | } |
31840ae1 | 1170 | } |
5d4f98a2 YZ |
1171 | btrfs_mark_buffer_dirty(leaf); |
1172 | ret = 0; | |
1173 | fail: | |
b3b4aa74 | 1174 | btrfs_release_path(path); |
7bb86316 | 1175 | return ret; |
74493f7a CM |
1176 | } |
1177 | ||
5d4f98a2 YZ |
1178 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
1179 | struct btrfs_root *root, | |
1180 | struct btrfs_path *path, | |
1181 | int refs_to_drop) | |
31840ae1 | 1182 | { |
5d4f98a2 YZ |
1183 | struct btrfs_key key; |
1184 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1185 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1186 | struct extent_buffer *leaf; |
5d4f98a2 | 1187 | u32 num_refs = 0; |
31840ae1 ZY |
1188 | int ret = 0; |
1189 | ||
1190 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1191 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1192 | ||
1193 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1194 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1195 | struct btrfs_extent_data_ref); | |
1196 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1197 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1198 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1199 | struct btrfs_shared_data_ref); | |
1200 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1201 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1202 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1203 | struct btrfs_extent_ref_v0 *ref0; | |
1204 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1205 | struct btrfs_extent_ref_v0); | |
1206 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1207 | #endif | |
1208 | } else { | |
1209 | BUG(); | |
1210 | } | |
1211 | ||
56bec294 CM |
1212 | BUG_ON(num_refs < refs_to_drop); |
1213 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1214 | |
31840ae1 ZY |
1215 | if (num_refs == 0) { |
1216 | ret = btrfs_del_item(trans, root, path); | |
1217 | } else { | |
5d4f98a2 YZ |
1218 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1219 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1220 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1221 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1222 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1223 | else { | |
1224 | struct btrfs_extent_ref_v0 *ref0; | |
1225 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1226 | struct btrfs_extent_ref_v0); | |
1227 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1228 | } | |
1229 | #endif | |
31840ae1 ZY |
1230 | btrfs_mark_buffer_dirty(leaf); |
1231 | } | |
31840ae1 ZY |
1232 | return ret; |
1233 | } | |
1234 | ||
5d4f98a2 YZ |
1235 | static noinline u32 extent_data_ref_count(struct btrfs_root *root, |
1236 | struct btrfs_path *path, | |
1237 | struct btrfs_extent_inline_ref *iref) | |
15916de8 | 1238 | { |
5d4f98a2 YZ |
1239 | struct btrfs_key key; |
1240 | struct extent_buffer *leaf; | |
1241 | struct btrfs_extent_data_ref *ref1; | |
1242 | struct btrfs_shared_data_ref *ref2; | |
1243 | u32 num_refs = 0; | |
1244 | ||
1245 | leaf = path->nodes[0]; | |
1246 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1247 | if (iref) { | |
1248 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1249 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1250 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1251 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1252 | } else { | |
1253 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1254 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1255 | } | |
1256 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1257 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1258 | struct btrfs_extent_data_ref); | |
1259 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1260 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1261 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1262 | struct btrfs_shared_data_ref); | |
1263 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1264 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1265 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1266 | struct btrfs_extent_ref_v0 *ref0; | |
1267 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1268 | struct btrfs_extent_ref_v0); | |
1269 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1270 | #endif |
5d4f98a2 YZ |
1271 | } else { |
1272 | WARN_ON(1); | |
1273 | } | |
1274 | return num_refs; | |
1275 | } | |
15916de8 | 1276 | |
5d4f98a2 YZ |
1277 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1278 | struct btrfs_root *root, | |
1279 | struct btrfs_path *path, | |
1280 | u64 bytenr, u64 parent, | |
1281 | u64 root_objectid) | |
1f3c79a2 | 1282 | { |
5d4f98a2 | 1283 | struct btrfs_key key; |
1f3c79a2 | 1284 | int ret; |
1f3c79a2 | 1285 | |
5d4f98a2 YZ |
1286 | key.objectid = bytenr; |
1287 | if (parent) { | |
1288 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1289 | key.offset = parent; | |
1290 | } else { | |
1291 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1292 | key.offset = root_objectid; | |
1f3c79a2 LH |
1293 | } |
1294 | ||
5d4f98a2 YZ |
1295 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1296 | if (ret > 0) | |
1297 | ret = -ENOENT; | |
1298 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1299 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1300 | btrfs_release_path(path); |
5d4f98a2 YZ |
1301 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1302 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1303 | if (ret > 0) | |
1304 | ret = -ENOENT; | |
1305 | } | |
1f3c79a2 | 1306 | #endif |
5d4f98a2 | 1307 | return ret; |
1f3c79a2 LH |
1308 | } |
1309 | ||
5d4f98a2 YZ |
1310 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1311 | struct btrfs_root *root, | |
1312 | struct btrfs_path *path, | |
1313 | u64 bytenr, u64 parent, | |
1314 | u64 root_objectid) | |
31840ae1 | 1315 | { |
5d4f98a2 | 1316 | struct btrfs_key key; |
31840ae1 | 1317 | int ret; |
31840ae1 | 1318 | |
5d4f98a2 YZ |
1319 | key.objectid = bytenr; |
1320 | if (parent) { | |
1321 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1322 | key.offset = parent; | |
1323 | } else { | |
1324 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1325 | key.offset = root_objectid; | |
1326 | } | |
1327 | ||
1328 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
b3b4aa74 | 1329 | btrfs_release_path(path); |
31840ae1 ZY |
1330 | return ret; |
1331 | } | |
1332 | ||
5d4f98a2 | 1333 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1334 | { |
5d4f98a2 YZ |
1335 | int type; |
1336 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1337 | if (parent > 0) | |
1338 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1339 | else | |
1340 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1341 | } else { | |
1342 | if (parent > 0) | |
1343 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1344 | else | |
1345 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1346 | } | |
1347 | return type; | |
31840ae1 | 1348 | } |
56bec294 | 1349 | |
2c47e605 YZ |
1350 | static int find_next_key(struct btrfs_path *path, int level, |
1351 | struct btrfs_key *key) | |
56bec294 | 1352 | |
02217ed2 | 1353 | { |
2c47e605 | 1354 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1355 | if (!path->nodes[level]) |
1356 | break; | |
5d4f98a2 YZ |
1357 | if (path->slots[level] + 1 >= |
1358 | btrfs_header_nritems(path->nodes[level])) | |
1359 | continue; | |
1360 | if (level == 0) | |
1361 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1362 | path->slots[level] + 1); | |
1363 | else | |
1364 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1365 | path->slots[level] + 1); | |
1366 | return 0; | |
1367 | } | |
1368 | return 1; | |
1369 | } | |
037e6390 | 1370 | |
5d4f98a2 YZ |
1371 | /* |
1372 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1373 | * to the address of inline back ref, and 0 is returned. | |
1374 | * | |
1375 | * if back ref isn't found, *ref_ret is set to the address where it | |
1376 | * should be inserted, and -ENOENT is returned. | |
1377 | * | |
1378 | * if insert is true and there are too many inline back refs, the path | |
1379 | * points to the extent item, and -EAGAIN is returned. | |
1380 | * | |
1381 | * NOTE: inline back refs are ordered in the same way that back ref | |
1382 | * items in the tree are ordered. | |
1383 | */ | |
1384 | static noinline_for_stack | |
1385 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1386 | struct btrfs_root *root, | |
1387 | struct btrfs_path *path, | |
1388 | struct btrfs_extent_inline_ref **ref_ret, | |
1389 | u64 bytenr, u64 num_bytes, | |
1390 | u64 parent, u64 root_objectid, | |
1391 | u64 owner, u64 offset, int insert) | |
1392 | { | |
1393 | struct btrfs_key key; | |
1394 | struct extent_buffer *leaf; | |
1395 | struct btrfs_extent_item *ei; | |
1396 | struct btrfs_extent_inline_ref *iref; | |
1397 | u64 flags; | |
1398 | u64 item_size; | |
1399 | unsigned long ptr; | |
1400 | unsigned long end; | |
1401 | int extra_size; | |
1402 | int type; | |
1403 | int want; | |
1404 | int ret; | |
1405 | int err = 0; | |
26b8003f | 1406 | |
db94535d | 1407 | key.objectid = bytenr; |
31840ae1 | 1408 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1409 | key.offset = num_bytes; |
31840ae1 | 1410 | |
5d4f98a2 YZ |
1411 | want = extent_ref_type(parent, owner); |
1412 | if (insert) { | |
1413 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1414 | path->keep_locks = 1; |
5d4f98a2 YZ |
1415 | } else |
1416 | extra_size = -1; | |
1417 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); | |
b9473439 | 1418 | if (ret < 0) { |
5d4f98a2 YZ |
1419 | err = ret; |
1420 | goto out; | |
1421 | } | |
1422 | BUG_ON(ret); | |
1423 | ||
1424 | leaf = path->nodes[0]; | |
1425 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1426 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1427 | if (item_size < sizeof(*ei)) { | |
1428 | if (!insert) { | |
1429 | err = -ENOENT; | |
1430 | goto out; | |
1431 | } | |
1432 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1433 | extra_size); | |
1434 | if (ret < 0) { | |
1435 | err = ret; | |
1436 | goto out; | |
1437 | } | |
1438 | leaf = path->nodes[0]; | |
1439 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1440 | } | |
1441 | #endif | |
1442 | BUG_ON(item_size < sizeof(*ei)); | |
1443 | ||
5d4f98a2 YZ |
1444 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1445 | flags = btrfs_extent_flags(leaf, ei); | |
1446 | ||
1447 | ptr = (unsigned long)(ei + 1); | |
1448 | end = (unsigned long)ei + item_size; | |
1449 | ||
1450 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
1451 | ptr += sizeof(struct btrfs_tree_block_info); | |
1452 | BUG_ON(ptr > end); | |
1453 | } else { | |
1454 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); | |
1455 | } | |
1456 | ||
1457 | err = -ENOENT; | |
1458 | while (1) { | |
1459 | if (ptr >= end) { | |
1460 | WARN_ON(ptr > end); | |
1461 | break; | |
1462 | } | |
1463 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1464 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1465 | if (want < type) | |
1466 | break; | |
1467 | if (want > type) { | |
1468 | ptr += btrfs_extent_inline_ref_size(type); | |
1469 | continue; | |
1470 | } | |
1471 | ||
1472 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1473 | struct btrfs_extent_data_ref *dref; | |
1474 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1475 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1476 | owner, offset)) { | |
1477 | err = 0; | |
1478 | break; | |
1479 | } | |
1480 | if (hash_extent_data_ref_item(leaf, dref) < | |
1481 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1482 | break; | |
1483 | } else { | |
1484 | u64 ref_offset; | |
1485 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1486 | if (parent > 0) { | |
1487 | if (parent == ref_offset) { | |
1488 | err = 0; | |
1489 | break; | |
1490 | } | |
1491 | if (ref_offset < parent) | |
1492 | break; | |
1493 | } else { | |
1494 | if (root_objectid == ref_offset) { | |
1495 | err = 0; | |
1496 | break; | |
1497 | } | |
1498 | if (ref_offset < root_objectid) | |
1499 | break; | |
1500 | } | |
1501 | } | |
1502 | ptr += btrfs_extent_inline_ref_size(type); | |
1503 | } | |
1504 | if (err == -ENOENT && insert) { | |
1505 | if (item_size + extra_size >= | |
1506 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1507 | err = -EAGAIN; | |
1508 | goto out; | |
1509 | } | |
1510 | /* | |
1511 | * To add new inline back ref, we have to make sure | |
1512 | * there is no corresponding back ref item. | |
1513 | * For simplicity, we just do not add new inline back | |
1514 | * ref if there is any kind of item for this block | |
1515 | */ | |
2c47e605 YZ |
1516 | if (find_next_key(path, 0, &key) == 0 && |
1517 | key.objectid == bytenr && | |
85d4198e | 1518 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1519 | err = -EAGAIN; |
1520 | goto out; | |
1521 | } | |
1522 | } | |
1523 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1524 | out: | |
85d4198e | 1525 | if (insert) { |
5d4f98a2 YZ |
1526 | path->keep_locks = 0; |
1527 | btrfs_unlock_up_safe(path, 1); | |
1528 | } | |
1529 | return err; | |
1530 | } | |
1531 | ||
1532 | /* | |
1533 | * helper to add new inline back ref | |
1534 | */ | |
1535 | static noinline_for_stack | |
1536 | int setup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1537 | struct btrfs_root *root, | |
1538 | struct btrfs_path *path, | |
1539 | struct btrfs_extent_inline_ref *iref, | |
1540 | u64 parent, u64 root_objectid, | |
1541 | u64 owner, u64 offset, int refs_to_add, | |
1542 | struct btrfs_delayed_extent_op *extent_op) | |
1543 | { | |
1544 | struct extent_buffer *leaf; | |
1545 | struct btrfs_extent_item *ei; | |
1546 | unsigned long ptr; | |
1547 | unsigned long end; | |
1548 | unsigned long item_offset; | |
1549 | u64 refs; | |
1550 | int size; | |
1551 | int type; | |
1552 | int ret; | |
1553 | ||
1554 | leaf = path->nodes[0]; | |
1555 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1556 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1557 | ||
1558 | type = extent_ref_type(parent, owner); | |
1559 | size = btrfs_extent_inline_ref_size(type); | |
1560 | ||
1561 | ret = btrfs_extend_item(trans, root, path, size); | |
5d4f98a2 YZ |
1562 | |
1563 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1564 | refs = btrfs_extent_refs(leaf, ei); | |
1565 | refs += refs_to_add; | |
1566 | btrfs_set_extent_refs(leaf, ei, refs); | |
1567 | if (extent_op) | |
1568 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1569 | ||
1570 | ptr = (unsigned long)ei + item_offset; | |
1571 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1572 | if (ptr < end - size) | |
1573 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1574 | end - size - ptr); | |
1575 | ||
1576 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1577 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1578 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1579 | struct btrfs_extent_data_ref *dref; | |
1580 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1581 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1582 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1583 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1584 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1585 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1586 | struct btrfs_shared_data_ref *sref; | |
1587 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1588 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1589 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1590 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1591 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1592 | } else { | |
1593 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1594 | } | |
1595 | btrfs_mark_buffer_dirty(leaf); | |
1596 | return 0; | |
1597 | } | |
1598 | ||
1599 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1600 | struct btrfs_root *root, | |
1601 | struct btrfs_path *path, | |
1602 | struct btrfs_extent_inline_ref **ref_ret, | |
1603 | u64 bytenr, u64 num_bytes, u64 parent, | |
1604 | u64 root_objectid, u64 owner, u64 offset) | |
1605 | { | |
1606 | int ret; | |
1607 | ||
1608 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1609 | bytenr, num_bytes, parent, | |
1610 | root_objectid, owner, offset, 0); | |
1611 | if (ret != -ENOENT) | |
54aa1f4d | 1612 | return ret; |
5d4f98a2 | 1613 | |
b3b4aa74 | 1614 | btrfs_release_path(path); |
5d4f98a2 YZ |
1615 | *ref_ret = NULL; |
1616 | ||
1617 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1618 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1619 | root_objectid); | |
1620 | } else { | |
1621 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1622 | root_objectid, owner, offset); | |
b9473439 | 1623 | } |
5d4f98a2 YZ |
1624 | return ret; |
1625 | } | |
31840ae1 | 1626 | |
5d4f98a2 YZ |
1627 | /* |
1628 | * helper to update/remove inline back ref | |
1629 | */ | |
1630 | static noinline_for_stack | |
1631 | int update_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1632 | struct btrfs_root *root, | |
1633 | struct btrfs_path *path, | |
1634 | struct btrfs_extent_inline_ref *iref, | |
1635 | int refs_to_mod, | |
1636 | struct btrfs_delayed_extent_op *extent_op) | |
1637 | { | |
1638 | struct extent_buffer *leaf; | |
1639 | struct btrfs_extent_item *ei; | |
1640 | struct btrfs_extent_data_ref *dref = NULL; | |
1641 | struct btrfs_shared_data_ref *sref = NULL; | |
1642 | unsigned long ptr; | |
1643 | unsigned long end; | |
1644 | u32 item_size; | |
1645 | int size; | |
1646 | int type; | |
1647 | int ret; | |
1648 | u64 refs; | |
1649 | ||
1650 | leaf = path->nodes[0]; | |
1651 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1652 | refs = btrfs_extent_refs(leaf, ei); | |
1653 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1654 | refs += refs_to_mod; | |
1655 | btrfs_set_extent_refs(leaf, ei, refs); | |
1656 | if (extent_op) | |
1657 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1658 | ||
1659 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1660 | ||
1661 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1662 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1663 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1664 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1665 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1666 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1667 | } else { | |
1668 | refs = 1; | |
1669 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1670 | } |
31840ae1 | 1671 | |
5d4f98a2 YZ |
1672 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1673 | refs += refs_to_mod; | |
1674 | ||
1675 | if (refs > 0) { | |
1676 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1677 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1678 | else | |
1679 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1680 | } else { | |
1681 | size = btrfs_extent_inline_ref_size(type); | |
1682 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1683 | ptr = (unsigned long)iref; | |
1684 | end = (unsigned long)ei + item_size; | |
1685 | if (ptr + size < end) | |
1686 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1687 | end - ptr - size); | |
1688 | item_size -= size; | |
1689 | ret = btrfs_truncate_item(trans, root, path, item_size, 1); | |
5d4f98a2 YZ |
1690 | } |
1691 | btrfs_mark_buffer_dirty(leaf); | |
1692 | return 0; | |
1693 | } | |
1694 | ||
1695 | static noinline_for_stack | |
1696 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1697 | struct btrfs_root *root, | |
1698 | struct btrfs_path *path, | |
1699 | u64 bytenr, u64 num_bytes, u64 parent, | |
1700 | u64 root_objectid, u64 owner, | |
1701 | u64 offset, int refs_to_add, | |
1702 | struct btrfs_delayed_extent_op *extent_op) | |
1703 | { | |
1704 | struct btrfs_extent_inline_ref *iref; | |
1705 | int ret; | |
1706 | ||
1707 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1708 | bytenr, num_bytes, parent, | |
1709 | root_objectid, owner, offset, 1); | |
1710 | if (ret == 0) { | |
1711 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
1712 | ret = update_inline_extent_backref(trans, root, path, iref, | |
1713 | refs_to_add, extent_op); | |
1714 | } else if (ret == -ENOENT) { | |
1715 | ret = setup_inline_extent_backref(trans, root, path, iref, | |
1716 | parent, root_objectid, | |
1717 | owner, offset, refs_to_add, | |
1718 | extent_op); | |
771ed689 | 1719 | } |
5d4f98a2 YZ |
1720 | return ret; |
1721 | } | |
31840ae1 | 1722 | |
5d4f98a2 YZ |
1723 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1724 | struct btrfs_root *root, | |
1725 | struct btrfs_path *path, | |
1726 | u64 bytenr, u64 parent, u64 root_objectid, | |
1727 | u64 owner, u64 offset, int refs_to_add) | |
1728 | { | |
1729 | int ret; | |
1730 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1731 | BUG_ON(refs_to_add != 1); | |
1732 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1733 | parent, root_objectid); | |
1734 | } else { | |
1735 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1736 | parent, root_objectid, | |
1737 | owner, offset, refs_to_add); | |
1738 | } | |
1739 | return ret; | |
1740 | } | |
56bec294 | 1741 | |
5d4f98a2 YZ |
1742 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1743 | struct btrfs_root *root, | |
1744 | struct btrfs_path *path, | |
1745 | struct btrfs_extent_inline_ref *iref, | |
1746 | int refs_to_drop, int is_data) | |
1747 | { | |
1748 | int ret; | |
b9473439 | 1749 | |
5d4f98a2 YZ |
1750 | BUG_ON(!is_data && refs_to_drop != 1); |
1751 | if (iref) { | |
1752 | ret = update_inline_extent_backref(trans, root, path, iref, | |
1753 | -refs_to_drop, NULL); | |
1754 | } else if (is_data) { | |
1755 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop); | |
1756 | } else { | |
1757 | ret = btrfs_del_item(trans, root, path); | |
1758 | } | |
1759 | return ret; | |
1760 | } | |
1761 | ||
5378e607 | 1762 | static int btrfs_issue_discard(struct block_device *bdev, |
5d4f98a2 YZ |
1763 | u64 start, u64 len) |
1764 | { | |
5378e607 | 1765 | return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0); |
5d4f98a2 | 1766 | } |
5d4f98a2 YZ |
1767 | |
1768 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 1769 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1770 | { |
5d4f98a2 | 1771 | int ret; |
5378e607 | 1772 | u64 discarded_bytes = 0; |
5d4f98a2 YZ |
1773 | struct btrfs_multi_bio *multi = NULL; |
1774 | ||
e244a0ae | 1775 | |
5d4f98a2 | 1776 | /* Tell the block device(s) that the sectors can be discarded */ |
5378e607 LD |
1777 | ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD, |
1778 | bytenr, &num_bytes, &multi, 0); | |
5d4f98a2 YZ |
1779 | if (!ret) { |
1780 | struct btrfs_bio_stripe *stripe = multi->stripes; | |
1781 | int i; | |
1782 | ||
5d4f98a2 YZ |
1783 | |
1784 | for (i = 0; i < multi->num_stripes; i++, stripe++) { | |
d5e2003c JB |
1785 | if (!stripe->dev->can_discard) |
1786 | continue; | |
1787 | ||
5378e607 LD |
1788 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1789 | stripe->physical, | |
1790 | stripe->length); | |
1791 | if (!ret) | |
1792 | discarded_bytes += stripe->length; | |
1793 | else if (ret != -EOPNOTSUPP) | |
1794 | break; | |
d5e2003c JB |
1795 | |
1796 | /* | |
1797 | * Just in case we get back EOPNOTSUPP for some reason, | |
1798 | * just ignore the return value so we don't screw up | |
1799 | * people calling discard_extent. | |
1800 | */ | |
1801 | ret = 0; | |
5d4f98a2 YZ |
1802 | } |
1803 | kfree(multi); | |
1804 | } | |
5378e607 LD |
1805 | |
1806 | if (actual_bytes) | |
1807 | *actual_bytes = discarded_bytes; | |
1808 | ||
5d4f98a2 YZ |
1809 | |
1810 | return ret; | |
5d4f98a2 YZ |
1811 | } |
1812 | ||
1813 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1814 | struct btrfs_root *root, | |
1815 | u64 bytenr, u64 num_bytes, u64 parent, | |
1816 | u64 root_objectid, u64 owner, u64 offset) | |
1817 | { | |
1818 | int ret; | |
1819 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && | |
1820 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
1821 | ||
1822 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1823 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes, | |
1824 | parent, root_objectid, (int)owner, | |
1825 | BTRFS_ADD_DELAYED_REF, NULL); | |
1826 | } else { | |
1827 | ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes, | |
1828 | parent, root_objectid, owner, offset, | |
1829 | BTRFS_ADD_DELAYED_REF, NULL); | |
1830 | } | |
1831 | return ret; | |
1832 | } | |
1833 | ||
1834 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1835 | struct btrfs_root *root, | |
1836 | u64 bytenr, u64 num_bytes, | |
1837 | u64 parent, u64 root_objectid, | |
1838 | u64 owner, u64 offset, int refs_to_add, | |
1839 | struct btrfs_delayed_extent_op *extent_op) | |
1840 | { | |
1841 | struct btrfs_path *path; | |
1842 | struct extent_buffer *leaf; | |
1843 | struct btrfs_extent_item *item; | |
1844 | u64 refs; | |
1845 | int ret; | |
1846 | int err = 0; | |
1847 | ||
1848 | path = btrfs_alloc_path(); | |
1849 | if (!path) | |
1850 | return -ENOMEM; | |
1851 | ||
1852 | path->reada = 1; | |
1853 | path->leave_spinning = 1; | |
1854 | /* this will setup the path even if it fails to insert the back ref */ | |
1855 | ret = insert_inline_extent_backref(trans, root->fs_info->extent_root, | |
1856 | path, bytenr, num_bytes, parent, | |
1857 | root_objectid, owner, offset, | |
1858 | refs_to_add, extent_op); | |
1859 | if (ret == 0) | |
1860 | goto out; | |
1861 | ||
1862 | if (ret != -EAGAIN) { | |
1863 | err = ret; | |
1864 | goto out; | |
1865 | } | |
1866 | ||
1867 | leaf = path->nodes[0]; | |
1868 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1869 | refs = btrfs_extent_refs(leaf, item); | |
1870 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1871 | if (extent_op) | |
1872 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1873 | |
5d4f98a2 | 1874 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 1875 | btrfs_release_path(path); |
56bec294 CM |
1876 | |
1877 | path->reada = 1; | |
b9473439 CM |
1878 | path->leave_spinning = 1; |
1879 | ||
56bec294 CM |
1880 | /* now insert the actual backref */ |
1881 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
1882 | path, bytenr, parent, root_objectid, |
1883 | owner, offset, refs_to_add); | |
56bec294 | 1884 | BUG_ON(ret); |
5d4f98a2 | 1885 | out: |
56bec294 | 1886 | btrfs_free_path(path); |
5d4f98a2 | 1887 | return err; |
56bec294 CM |
1888 | } |
1889 | ||
5d4f98a2 YZ |
1890 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
1891 | struct btrfs_root *root, | |
1892 | struct btrfs_delayed_ref_node *node, | |
1893 | struct btrfs_delayed_extent_op *extent_op, | |
1894 | int insert_reserved) | |
56bec294 | 1895 | { |
5d4f98a2 YZ |
1896 | int ret = 0; |
1897 | struct btrfs_delayed_data_ref *ref; | |
1898 | struct btrfs_key ins; | |
1899 | u64 parent = 0; | |
1900 | u64 ref_root = 0; | |
1901 | u64 flags = 0; | |
1902 | ||
1903 | ins.objectid = node->bytenr; | |
1904 | ins.offset = node->num_bytes; | |
1905 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1906 | ||
1907 | ref = btrfs_delayed_node_to_data_ref(node); | |
1908 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) | |
1909 | parent = ref->parent; | |
1910 | else | |
1911 | ref_root = ref->root; | |
1912 | ||
1913 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
1914 | if (extent_op) { | |
1915 | BUG_ON(extent_op->update_key); | |
1916 | flags |= extent_op->flags_to_set; | |
1917 | } | |
1918 | ret = alloc_reserved_file_extent(trans, root, | |
1919 | parent, ref_root, flags, | |
1920 | ref->objectid, ref->offset, | |
1921 | &ins, node->ref_mod); | |
5d4f98a2 YZ |
1922 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
1923 | ret = __btrfs_inc_extent_ref(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 if (node->action == BTRFS_DROP_DELAYED_REF) { | |
1929 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
1930 | node->num_bytes, parent, | |
1931 | ref_root, ref->objectid, | |
1932 | ref->offset, node->ref_mod, | |
1933 | extent_op); | |
1934 | } else { | |
1935 | BUG(); | |
1936 | } | |
1937 | return ret; | |
1938 | } | |
1939 | ||
1940 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1941 | struct extent_buffer *leaf, | |
1942 | struct btrfs_extent_item *ei) | |
1943 | { | |
1944 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1945 | if (extent_op->update_flags) { | |
1946 | flags |= extent_op->flags_to_set; | |
1947 | btrfs_set_extent_flags(leaf, ei, flags); | |
1948 | } | |
1949 | ||
1950 | if (extent_op->update_key) { | |
1951 | struct btrfs_tree_block_info *bi; | |
1952 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
1953 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
1954 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
1955 | } | |
1956 | } | |
1957 | ||
1958 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
1959 | struct btrfs_root *root, | |
1960 | struct btrfs_delayed_ref_node *node, | |
1961 | struct btrfs_delayed_extent_op *extent_op) | |
1962 | { | |
1963 | struct btrfs_key key; | |
1964 | struct btrfs_path *path; | |
1965 | struct btrfs_extent_item *ei; | |
1966 | struct extent_buffer *leaf; | |
1967 | u32 item_size; | |
56bec294 | 1968 | int ret; |
5d4f98a2 YZ |
1969 | int err = 0; |
1970 | ||
1971 | path = btrfs_alloc_path(); | |
1972 | if (!path) | |
1973 | return -ENOMEM; | |
1974 | ||
1975 | key.objectid = node->bytenr; | |
1976 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1977 | key.offset = node->num_bytes; | |
1978 | ||
1979 | path->reada = 1; | |
1980 | path->leave_spinning = 1; | |
1981 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
1982 | path, 0, 1); | |
1983 | if (ret < 0) { | |
1984 | err = ret; | |
1985 | goto out; | |
1986 | } | |
1987 | if (ret > 0) { | |
1988 | err = -EIO; | |
1989 | goto out; | |
1990 | } | |
1991 | ||
1992 | leaf = path->nodes[0]; | |
1993 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1994 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1995 | if (item_size < sizeof(*ei)) { | |
1996 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
1997 | path, (u64)-1, 0); | |
1998 | if (ret < 0) { | |
1999 | err = ret; | |
2000 | goto out; | |
2001 | } | |
2002 | leaf = path->nodes[0]; | |
2003 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2004 | } | |
2005 | #endif | |
2006 | BUG_ON(item_size < sizeof(*ei)); | |
2007 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2008 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2009 | |
5d4f98a2 YZ |
2010 | btrfs_mark_buffer_dirty(leaf); |
2011 | out: | |
2012 | btrfs_free_path(path); | |
2013 | return err; | |
56bec294 CM |
2014 | } |
2015 | ||
5d4f98a2 YZ |
2016 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2017 | struct btrfs_root *root, | |
2018 | struct btrfs_delayed_ref_node *node, | |
2019 | struct btrfs_delayed_extent_op *extent_op, | |
2020 | int insert_reserved) | |
56bec294 CM |
2021 | { |
2022 | int ret = 0; | |
5d4f98a2 YZ |
2023 | struct btrfs_delayed_tree_ref *ref; |
2024 | struct btrfs_key ins; | |
2025 | u64 parent = 0; | |
2026 | u64 ref_root = 0; | |
56bec294 | 2027 | |
5d4f98a2 YZ |
2028 | ins.objectid = node->bytenr; |
2029 | ins.offset = node->num_bytes; | |
2030 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
56bec294 | 2031 | |
5d4f98a2 YZ |
2032 | ref = btrfs_delayed_node_to_tree_ref(node); |
2033 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2034 | parent = ref->parent; | |
2035 | else | |
2036 | ref_root = ref->root; | |
2037 | ||
2038 | BUG_ON(node->ref_mod != 1); | |
2039 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
2040 | BUG_ON(!extent_op || !extent_op->update_flags || | |
2041 | !extent_op->update_key); | |
2042 | ret = alloc_reserved_tree_block(trans, root, | |
2043 | parent, ref_root, | |
2044 | extent_op->flags_to_set, | |
2045 | &extent_op->key, | |
2046 | ref->level, &ins); | |
5d4f98a2 YZ |
2047 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2048 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
2049 | node->num_bytes, parent, ref_root, | |
2050 | ref->level, 0, 1, extent_op); | |
2051 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
2052 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
2053 | node->num_bytes, parent, ref_root, | |
2054 | ref->level, 0, 1, extent_op); | |
2055 | } else { | |
2056 | BUG(); | |
2057 | } | |
56bec294 CM |
2058 | return ret; |
2059 | } | |
2060 | ||
2061 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 YZ |
2062 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2063 | struct btrfs_root *root, | |
2064 | struct btrfs_delayed_ref_node *node, | |
2065 | struct btrfs_delayed_extent_op *extent_op, | |
2066 | int insert_reserved) | |
56bec294 CM |
2067 | { |
2068 | int ret; | |
5d4f98a2 | 2069 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2070 | struct btrfs_delayed_ref_head *head; |
2071 | /* | |
2072 | * we've hit the end of the chain and we were supposed | |
2073 | * to insert this extent into the tree. But, it got | |
2074 | * deleted before we ever needed to insert it, so all | |
2075 | * we have to do is clean up the accounting | |
2076 | */ | |
5d4f98a2 YZ |
2077 | BUG_ON(extent_op); |
2078 | head = btrfs_delayed_node_to_head(node); | |
56bec294 | 2079 | if (insert_reserved) { |
f0486c68 YZ |
2080 | btrfs_pin_extent(root, node->bytenr, |
2081 | node->num_bytes, 1); | |
5d4f98a2 YZ |
2082 | if (head->is_data) { |
2083 | ret = btrfs_del_csums(trans, root, | |
2084 | node->bytenr, | |
2085 | node->num_bytes); | |
2086 | BUG_ON(ret); | |
2087 | } | |
56bec294 | 2088 | } |
56bec294 CM |
2089 | mutex_unlock(&head->mutex); |
2090 | return 0; | |
2091 | } | |
2092 | ||
5d4f98a2 YZ |
2093 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2094 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2095 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2096 | insert_reserved); | |
2097 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2098 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2099 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2100 | insert_reserved); | |
2101 | else | |
2102 | BUG(); | |
2103 | return ret; | |
56bec294 CM |
2104 | } |
2105 | ||
2106 | static noinline struct btrfs_delayed_ref_node * | |
2107 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
2108 | { | |
2109 | struct rb_node *node; | |
2110 | struct btrfs_delayed_ref_node *ref; | |
2111 | int action = BTRFS_ADD_DELAYED_REF; | |
2112 | again: | |
2113 | /* | |
2114 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2115 | * this prevents ref count from going down to zero when | |
2116 | * there still are pending delayed ref. | |
2117 | */ | |
2118 | node = rb_prev(&head->node.rb_node); | |
2119 | while (1) { | |
2120 | if (!node) | |
2121 | break; | |
2122 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2123 | rb_node); | |
2124 | if (ref->bytenr != head->node.bytenr) | |
2125 | break; | |
5d4f98a2 | 2126 | if (ref->action == action) |
56bec294 CM |
2127 | return ref; |
2128 | node = rb_prev(node); | |
2129 | } | |
2130 | if (action == BTRFS_ADD_DELAYED_REF) { | |
2131 | action = BTRFS_DROP_DELAYED_REF; | |
2132 | goto again; | |
2133 | } | |
2134 | return NULL; | |
2135 | } | |
2136 | ||
c3e69d58 CM |
2137 | static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, |
2138 | struct btrfs_root *root, | |
2139 | struct list_head *cluster) | |
56bec294 | 2140 | { |
56bec294 CM |
2141 | struct btrfs_delayed_ref_root *delayed_refs; |
2142 | struct btrfs_delayed_ref_node *ref; | |
2143 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2144 | struct btrfs_delayed_extent_op *extent_op; |
56bec294 | 2145 | int ret; |
c3e69d58 | 2146 | int count = 0; |
56bec294 | 2147 | int must_insert_reserved = 0; |
56bec294 CM |
2148 | |
2149 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2150 | while (1) { |
2151 | if (!locked_ref) { | |
c3e69d58 CM |
2152 | /* pick a new head ref from the cluster list */ |
2153 | if (list_empty(cluster)) | |
56bec294 | 2154 | break; |
56bec294 | 2155 | |
c3e69d58 CM |
2156 | locked_ref = list_entry(cluster->next, |
2157 | struct btrfs_delayed_ref_head, cluster); | |
2158 | ||
2159 | /* grab the lock that says we are going to process | |
2160 | * all the refs for this head */ | |
2161 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
2162 | ||
2163 | /* | |
2164 | * we may have dropped the spin lock to get the head | |
2165 | * mutex lock, and that might have given someone else | |
2166 | * time to free the head. If that's true, it has been | |
2167 | * removed from our list and we can move on. | |
2168 | */ | |
2169 | if (ret == -EAGAIN) { | |
2170 | locked_ref = NULL; | |
2171 | count++; | |
2172 | continue; | |
56bec294 CM |
2173 | } |
2174 | } | |
a28ec197 | 2175 | |
56bec294 CM |
2176 | /* |
2177 | * record the must insert reserved flag before we | |
2178 | * drop the spin lock. | |
2179 | */ | |
2180 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2181 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2182 | |
5d4f98a2 YZ |
2183 | extent_op = locked_ref->extent_op; |
2184 | locked_ref->extent_op = NULL; | |
2185 | ||
56bec294 CM |
2186 | /* |
2187 | * locked_ref is the head node, so we have to go one | |
2188 | * node back for any delayed ref updates | |
2189 | */ | |
56bec294 CM |
2190 | ref = select_delayed_ref(locked_ref); |
2191 | if (!ref) { | |
2192 | /* All delayed refs have been processed, Go ahead | |
2193 | * and send the head node to run_one_delayed_ref, | |
2194 | * so that any accounting fixes can happen | |
2195 | */ | |
2196 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2197 | |
2198 | if (extent_op && must_insert_reserved) { | |
2199 | kfree(extent_op); | |
2200 | extent_op = NULL; | |
2201 | } | |
2202 | ||
2203 | if (extent_op) { | |
2204 | spin_unlock(&delayed_refs->lock); | |
2205 | ||
2206 | ret = run_delayed_extent_op(trans, root, | |
2207 | ref, extent_op); | |
2208 | BUG_ON(ret); | |
2209 | kfree(extent_op); | |
2210 | ||
2211 | cond_resched(); | |
2212 | spin_lock(&delayed_refs->lock); | |
2213 | continue; | |
2214 | } | |
2215 | ||
c3e69d58 | 2216 | list_del_init(&locked_ref->cluster); |
56bec294 CM |
2217 | locked_ref = NULL; |
2218 | } | |
02217ed2 | 2219 | |
56bec294 CM |
2220 | ref->in_tree = 0; |
2221 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
2222 | delayed_refs->num_entries--; | |
5d4f98a2 | 2223 | |
56bec294 | 2224 | spin_unlock(&delayed_refs->lock); |
925baedd | 2225 | |
5d4f98a2 | 2226 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 CM |
2227 | must_insert_reserved); |
2228 | BUG_ON(ret); | |
eb099670 | 2229 | |
5d4f98a2 YZ |
2230 | btrfs_put_delayed_ref(ref); |
2231 | kfree(extent_op); | |
c3e69d58 | 2232 | count++; |
5d4f98a2 | 2233 | |
c3e69d58 CM |
2234 | cond_resched(); |
2235 | spin_lock(&delayed_refs->lock); | |
2236 | } | |
2237 | return count; | |
2238 | } | |
2239 | ||
2240 | /* | |
2241 | * this starts processing the delayed reference count updates and | |
2242 | * extent insertions we have queued up so far. count can be | |
2243 | * 0, which means to process everything in the tree at the start | |
2244 | * of the run (but not newly added entries), or it can be some target | |
2245 | * number you'd like to process. | |
2246 | */ | |
2247 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2248 | struct btrfs_root *root, unsigned long count) | |
2249 | { | |
2250 | struct rb_node *node; | |
2251 | struct btrfs_delayed_ref_root *delayed_refs; | |
2252 | struct btrfs_delayed_ref_node *ref; | |
2253 | struct list_head cluster; | |
2254 | int ret; | |
2255 | int run_all = count == (unsigned long)-1; | |
2256 | int run_most = 0; | |
2257 | ||
2258 | if (root == root->fs_info->extent_root) | |
2259 | root = root->fs_info->tree_root; | |
2260 | ||
2261 | delayed_refs = &trans->transaction->delayed_refs; | |
2262 | INIT_LIST_HEAD(&cluster); | |
2263 | again: | |
2264 | spin_lock(&delayed_refs->lock); | |
2265 | if (count == 0) { | |
2266 | count = delayed_refs->num_entries * 2; | |
2267 | run_most = 1; | |
2268 | } | |
2269 | while (1) { | |
2270 | if (!(run_all || run_most) && | |
2271 | delayed_refs->num_heads_ready < 64) | |
2272 | break; | |
eb099670 | 2273 | |
56bec294 | 2274 | /* |
c3e69d58 CM |
2275 | * go find something we can process in the rbtree. We start at |
2276 | * the beginning of the tree, and then build a cluster | |
2277 | * of refs to process starting at the first one we are able to | |
2278 | * lock | |
56bec294 | 2279 | */ |
c3e69d58 CM |
2280 | ret = btrfs_find_ref_cluster(trans, &cluster, |
2281 | delayed_refs->run_delayed_start); | |
2282 | if (ret) | |
56bec294 CM |
2283 | break; |
2284 | ||
c3e69d58 CM |
2285 | ret = run_clustered_refs(trans, root, &cluster); |
2286 | BUG_ON(ret < 0); | |
2287 | ||
2288 | count -= min_t(unsigned long, ret, count); | |
2289 | ||
2290 | if (count == 0) | |
2291 | break; | |
eb099670 | 2292 | } |
c3e69d58 | 2293 | |
56bec294 | 2294 | if (run_all) { |
56bec294 | 2295 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 2296 | if (!node) |
56bec294 | 2297 | goto out; |
c3e69d58 | 2298 | count = (unsigned long)-1; |
e9d0b13b | 2299 | |
56bec294 CM |
2300 | while (node) { |
2301 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2302 | rb_node); | |
2303 | if (btrfs_delayed_ref_is_head(ref)) { | |
2304 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 2305 | |
56bec294 CM |
2306 | head = btrfs_delayed_node_to_head(ref); |
2307 | atomic_inc(&ref->refs); | |
2308 | ||
2309 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2310 | /* |
2311 | * Mutex was contended, block until it's | |
2312 | * released and try again | |
2313 | */ | |
56bec294 CM |
2314 | mutex_lock(&head->mutex); |
2315 | mutex_unlock(&head->mutex); | |
2316 | ||
2317 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2318 | cond_resched(); |
56bec294 CM |
2319 | goto again; |
2320 | } | |
2321 | node = rb_next(node); | |
2322 | } | |
2323 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
2324 | schedule_timeout(1); |
2325 | goto again; | |
5f39d397 | 2326 | } |
54aa1f4d | 2327 | out: |
c3e69d58 | 2328 | spin_unlock(&delayed_refs->lock); |
a28ec197 CM |
2329 | return 0; |
2330 | } | |
2331 | ||
5d4f98a2 YZ |
2332 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2333 | struct btrfs_root *root, | |
2334 | u64 bytenr, u64 num_bytes, u64 flags, | |
2335 | int is_data) | |
2336 | { | |
2337 | struct btrfs_delayed_extent_op *extent_op; | |
2338 | int ret; | |
2339 | ||
2340 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
2341 | if (!extent_op) | |
2342 | return -ENOMEM; | |
2343 | ||
2344 | extent_op->flags_to_set = flags; | |
2345 | extent_op->update_flags = 1; | |
2346 | extent_op->update_key = 0; | |
2347 | extent_op->is_data = is_data ? 1 : 0; | |
2348 | ||
2349 | ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op); | |
2350 | if (ret) | |
2351 | kfree(extent_op); | |
2352 | return ret; | |
2353 | } | |
2354 | ||
2355 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2356 | struct btrfs_root *root, | |
2357 | struct btrfs_path *path, | |
2358 | u64 objectid, u64 offset, u64 bytenr) | |
2359 | { | |
2360 | struct btrfs_delayed_ref_head *head; | |
2361 | struct btrfs_delayed_ref_node *ref; | |
2362 | struct btrfs_delayed_data_ref *data_ref; | |
2363 | struct btrfs_delayed_ref_root *delayed_refs; | |
2364 | struct rb_node *node; | |
2365 | int ret = 0; | |
2366 | ||
2367 | ret = -ENOENT; | |
2368 | delayed_refs = &trans->transaction->delayed_refs; | |
2369 | spin_lock(&delayed_refs->lock); | |
2370 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2371 | if (!head) | |
2372 | goto out; | |
2373 | ||
2374 | if (!mutex_trylock(&head->mutex)) { | |
2375 | atomic_inc(&head->node.refs); | |
2376 | spin_unlock(&delayed_refs->lock); | |
2377 | ||
b3b4aa74 | 2378 | btrfs_release_path(path); |
5d4f98a2 | 2379 | |
8cc33e5c DS |
2380 | /* |
2381 | * Mutex was contended, block until it's released and let | |
2382 | * caller try again | |
2383 | */ | |
5d4f98a2 YZ |
2384 | mutex_lock(&head->mutex); |
2385 | mutex_unlock(&head->mutex); | |
2386 | btrfs_put_delayed_ref(&head->node); | |
2387 | return -EAGAIN; | |
2388 | } | |
2389 | ||
2390 | node = rb_prev(&head->node.rb_node); | |
2391 | if (!node) | |
2392 | goto out_unlock; | |
2393 | ||
2394 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2395 | ||
2396 | if (ref->bytenr != bytenr) | |
2397 | goto out_unlock; | |
2398 | ||
2399 | ret = 1; | |
2400 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2401 | goto out_unlock; | |
2402 | ||
2403 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2404 | ||
2405 | node = rb_prev(node); | |
2406 | if (node) { | |
2407 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2408 | if (ref->bytenr == bytenr) | |
2409 | goto out_unlock; | |
2410 | } | |
2411 | ||
2412 | if (data_ref->root != root->root_key.objectid || | |
2413 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2414 | goto out_unlock; | |
2415 | ||
2416 | ret = 0; | |
2417 | out_unlock: | |
2418 | mutex_unlock(&head->mutex); | |
2419 | out: | |
2420 | spin_unlock(&delayed_refs->lock); | |
2421 | return ret; | |
2422 | } | |
2423 | ||
2424 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2425 | struct btrfs_root *root, | |
2426 | struct btrfs_path *path, | |
2427 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2428 | { |
2429 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2430 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2431 | struct btrfs_extent_data_ref *ref; |
2432 | struct btrfs_extent_inline_ref *iref; | |
2433 | struct btrfs_extent_item *ei; | |
f321e491 | 2434 | struct btrfs_key key; |
5d4f98a2 | 2435 | u32 item_size; |
be20aa9d | 2436 | int ret; |
925baedd | 2437 | |
be20aa9d | 2438 | key.objectid = bytenr; |
31840ae1 | 2439 | key.offset = (u64)-1; |
f321e491 | 2440 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2441 | |
be20aa9d CM |
2442 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2443 | if (ret < 0) | |
2444 | goto out; | |
2445 | BUG_ON(ret == 0); | |
80ff3856 YZ |
2446 | |
2447 | ret = -ENOENT; | |
2448 | if (path->slots[0] == 0) | |
31840ae1 | 2449 | goto out; |
be20aa9d | 2450 | |
31840ae1 | 2451 | path->slots[0]--; |
f321e491 | 2452 | leaf = path->nodes[0]; |
5d4f98a2 | 2453 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2454 | |
5d4f98a2 | 2455 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2456 | goto out; |
f321e491 | 2457 | |
5d4f98a2 YZ |
2458 | ret = 1; |
2459 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2460 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2461 | if (item_size < sizeof(*ei)) { | |
2462 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2463 | goto out; | |
2464 | } | |
2465 | #endif | |
2466 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2467 | |
5d4f98a2 YZ |
2468 | if (item_size != sizeof(*ei) + |
2469 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2470 | goto out; | |
be20aa9d | 2471 | |
5d4f98a2 YZ |
2472 | if (btrfs_extent_generation(leaf, ei) <= |
2473 | btrfs_root_last_snapshot(&root->root_item)) | |
2474 | goto out; | |
2475 | ||
2476 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2477 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2478 | BTRFS_EXTENT_DATA_REF_KEY) | |
2479 | goto out; | |
2480 | ||
2481 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2482 | if (btrfs_extent_refs(leaf, ei) != | |
2483 | btrfs_extent_data_ref_count(leaf, ref) || | |
2484 | btrfs_extent_data_ref_root(leaf, ref) != | |
2485 | root->root_key.objectid || | |
2486 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2487 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2488 | goto out; | |
2489 | ||
2490 | ret = 0; | |
2491 | out: | |
2492 | return ret; | |
2493 | } | |
2494 | ||
2495 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2496 | struct btrfs_root *root, | |
2497 | u64 objectid, u64 offset, u64 bytenr) | |
2498 | { | |
2499 | struct btrfs_path *path; | |
2500 | int ret; | |
2501 | int ret2; | |
2502 | ||
2503 | path = btrfs_alloc_path(); | |
2504 | if (!path) | |
2505 | return -ENOENT; | |
2506 | ||
2507 | do { | |
2508 | ret = check_committed_ref(trans, root, path, objectid, | |
2509 | offset, bytenr); | |
2510 | if (ret && ret != -ENOENT) | |
f321e491 | 2511 | goto out; |
80ff3856 | 2512 | |
5d4f98a2 YZ |
2513 | ret2 = check_delayed_ref(trans, root, path, objectid, |
2514 | offset, bytenr); | |
2515 | } while (ret2 == -EAGAIN); | |
2516 | ||
2517 | if (ret2 && ret2 != -ENOENT) { | |
2518 | ret = ret2; | |
2519 | goto out; | |
f321e491 | 2520 | } |
5d4f98a2 YZ |
2521 | |
2522 | if (ret != -ENOENT || ret2 != -ENOENT) | |
2523 | ret = 0; | |
be20aa9d | 2524 | out: |
80ff3856 | 2525 | btrfs_free_path(path); |
f0486c68 YZ |
2526 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2527 | WARN_ON(ret > 0); | |
f321e491 | 2528 | return ret; |
be20aa9d | 2529 | } |
c5739bba | 2530 | |
5d4f98a2 | 2531 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2532 | struct btrfs_root *root, |
5d4f98a2 YZ |
2533 | struct extent_buffer *buf, |
2534 | int full_backref, int inc) | |
31840ae1 ZY |
2535 | { |
2536 | u64 bytenr; | |
5d4f98a2 YZ |
2537 | u64 num_bytes; |
2538 | u64 parent; | |
31840ae1 | 2539 | u64 ref_root; |
31840ae1 | 2540 | u32 nritems; |
31840ae1 ZY |
2541 | struct btrfs_key key; |
2542 | struct btrfs_file_extent_item *fi; | |
2543 | int i; | |
2544 | int level; | |
2545 | int ret = 0; | |
31840ae1 | 2546 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
5d4f98a2 | 2547 | u64, u64, u64, u64, u64, u64); |
31840ae1 ZY |
2548 | |
2549 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
2550 | nritems = btrfs_header_nritems(buf); |
2551 | level = btrfs_header_level(buf); | |
2552 | ||
5d4f98a2 YZ |
2553 | if (!root->ref_cows && level == 0) |
2554 | return 0; | |
31840ae1 | 2555 | |
5d4f98a2 YZ |
2556 | if (inc) |
2557 | process_func = btrfs_inc_extent_ref; | |
2558 | else | |
2559 | process_func = btrfs_free_extent; | |
31840ae1 | 2560 | |
5d4f98a2 YZ |
2561 | if (full_backref) |
2562 | parent = buf->start; | |
2563 | else | |
2564 | parent = 0; | |
2565 | ||
2566 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2567 | if (level == 0) { |
5d4f98a2 | 2568 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
2569 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
2570 | continue; | |
5d4f98a2 | 2571 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2572 | struct btrfs_file_extent_item); |
2573 | if (btrfs_file_extent_type(buf, fi) == | |
2574 | BTRFS_FILE_EXTENT_INLINE) | |
2575 | continue; | |
2576 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2577 | if (bytenr == 0) | |
2578 | continue; | |
5d4f98a2 YZ |
2579 | |
2580 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2581 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2582 | ret = process_func(trans, root, bytenr, num_bytes, | |
2583 | parent, ref_root, key.objectid, | |
2584 | key.offset); | |
31840ae1 ZY |
2585 | if (ret) |
2586 | goto fail; | |
2587 | } else { | |
5d4f98a2 YZ |
2588 | bytenr = btrfs_node_blockptr(buf, i); |
2589 | num_bytes = btrfs_level_size(root, level - 1); | |
2590 | ret = process_func(trans, root, bytenr, num_bytes, | |
2591 | parent, ref_root, level - 1, 0); | |
31840ae1 ZY |
2592 | if (ret) |
2593 | goto fail; | |
2594 | } | |
2595 | } | |
2596 | return 0; | |
2597 | fail: | |
5d4f98a2 YZ |
2598 | BUG(); |
2599 | return ret; | |
2600 | } | |
2601 | ||
2602 | int btrfs_inc_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, 1); | |
2606 | } | |
2607 | ||
2608 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
2609 | struct extent_buffer *buf, int full_backref) | |
2610 | { | |
2611 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); | |
31840ae1 ZY |
2612 | } |
2613 | ||
9078a3e1 CM |
2614 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2615 | struct btrfs_root *root, | |
2616 | struct btrfs_path *path, | |
2617 | struct btrfs_block_group_cache *cache) | |
2618 | { | |
2619 | int ret; | |
9078a3e1 | 2620 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
2621 | unsigned long bi; |
2622 | struct extent_buffer *leaf; | |
9078a3e1 | 2623 | |
9078a3e1 | 2624 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
2625 | if (ret < 0) |
2626 | goto fail; | |
9078a3e1 | 2627 | BUG_ON(ret); |
5f39d397 CM |
2628 | |
2629 | leaf = path->nodes[0]; | |
2630 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2631 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
2632 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 2633 | btrfs_release_path(path); |
54aa1f4d | 2634 | fail: |
9078a3e1 CM |
2635 | if (ret) |
2636 | return ret; | |
9078a3e1 CM |
2637 | return 0; |
2638 | ||
2639 | } | |
2640 | ||
4a8c9a62 YZ |
2641 | static struct btrfs_block_group_cache * |
2642 | next_block_group(struct btrfs_root *root, | |
2643 | struct btrfs_block_group_cache *cache) | |
2644 | { | |
2645 | struct rb_node *node; | |
2646 | spin_lock(&root->fs_info->block_group_cache_lock); | |
2647 | node = rb_next(&cache->cache_node); | |
2648 | btrfs_put_block_group(cache); | |
2649 | if (node) { | |
2650 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
2651 | cache_node); | |
11dfe35a | 2652 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
2653 | } else |
2654 | cache = NULL; | |
2655 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
2656 | return cache; | |
2657 | } | |
2658 | ||
0af3d00b JB |
2659 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
2660 | struct btrfs_trans_handle *trans, | |
2661 | struct btrfs_path *path) | |
2662 | { | |
2663 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
2664 | struct inode *inode = NULL; | |
2665 | u64 alloc_hint = 0; | |
2b20982e | 2666 | int dcs = BTRFS_DC_ERROR; |
0af3d00b JB |
2667 | int num_pages = 0; |
2668 | int retries = 0; | |
2669 | int ret = 0; | |
2670 | ||
2671 | /* | |
2672 | * If this block group is smaller than 100 megs don't bother caching the | |
2673 | * block group. | |
2674 | */ | |
2675 | if (block_group->key.offset < (100 * 1024 * 1024)) { | |
2676 | spin_lock(&block_group->lock); | |
2677 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
2678 | spin_unlock(&block_group->lock); | |
2679 | return 0; | |
2680 | } | |
2681 | ||
2682 | again: | |
2683 | inode = lookup_free_space_inode(root, block_group, path); | |
2684 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
2685 | ret = PTR_ERR(inode); | |
b3b4aa74 | 2686 | btrfs_release_path(path); |
0af3d00b JB |
2687 | goto out; |
2688 | } | |
2689 | ||
2690 | if (IS_ERR(inode)) { | |
2691 | BUG_ON(retries); | |
2692 | retries++; | |
2693 | ||
2694 | if (block_group->ro) | |
2695 | goto out_free; | |
2696 | ||
2697 | ret = create_free_space_inode(root, trans, block_group, path); | |
2698 | if (ret) | |
2699 | goto out_free; | |
2700 | goto again; | |
2701 | } | |
2702 | ||
2703 | /* | |
2704 | * We want to set the generation to 0, that way if anything goes wrong | |
2705 | * from here on out we know not to trust this cache when we load up next | |
2706 | * time. | |
2707 | */ | |
2708 | BTRFS_I(inode)->generation = 0; | |
2709 | ret = btrfs_update_inode(trans, root, inode); | |
2710 | WARN_ON(ret); | |
2711 | ||
2712 | if (i_size_read(inode) > 0) { | |
2713 | ret = btrfs_truncate_free_space_cache(root, trans, path, | |
2714 | inode); | |
2715 | if (ret) | |
2716 | goto out_put; | |
2717 | } | |
2718 | ||
2719 | spin_lock(&block_group->lock); | |
2720 | if (block_group->cached != BTRFS_CACHE_FINISHED) { | |
2b20982e JB |
2721 | /* We're not cached, don't bother trying to write stuff out */ |
2722 | dcs = BTRFS_DC_WRITTEN; | |
0af3d00b JB |
2723 | spin_unlock(&block_group->lock); |
2724 | goto out_put; | |
2725 | } | |
2726 | spin_unlock(&block_group->lock); | |
2727 | ||
2728 | num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024); | |
2729 | if (!num_pages) | |
2730 | num_pages = 1; | |
2731 | ||
2732 | /* | |
2733 | * Just to make absolutely sure we have enough space, we're going to | |
2734 | * preallocate 12 pages worth of space for each block group. In | |
2735 | * practice we ought to use at most 8, but we need extra space so we can | |
2736 | * add our header and have a terminator between the extents and the | |
2737 | * bitmaps. | |
2738 | */ | |
2739 | num_pages *= 16; | |
2740 | num_pages *= PAGE_CACHE_SIZE; | |
2741 | ||
2742 | ret = btrfs_check_data_free_space(inode, num_pages); | |
2743 | if (ret) | |
2744 | goto out_put; | |
2745 | ||
2746 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
2747 | num_pages, num_pages, | |
2748 | &alloc_hint); | |
2b20982e JB |
2749 | if (!ret) |
2750 | dcs = BTRFS_DC_SETUP; | |
0af3d00b JB |
2751 | btrfs_free_reserved_data_space(inode, num_pages); |
2752 | out_put: | |
2753 | iput(inode); | |
2754 | out_free: | |
b3b4aa74 | 2755 | btrfs_release_path(path); |
0af3d00b JB |
2756 | out: |
2757 | spin_lock(&block_group->lock); | |
2b20982e | 2758 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
2759 | spin_unlock(&block_group->lock); |
2760 | ||
2761 | return ret; | |
2762 | } | |
2763 | ||
96b5179d CM |
2764 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
2765 | struct btrfs_root *root) | |
9078a3e1 | 2766 | { |
4a8c9a62 | 2767 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 2768 | int err = 0; |
9078a3e1 | 2769 | struct btrfs_path *path; |
96b5179d | 2770 | u64 last = 0; |
9078a3e1 CM |
2771 | |
2772 | path = btrfs_alloc_path(); | |
2773 | if (!path) | |
2774 | return -ENOMEM; | |
2775 | ||
0af3d00b JB |
2776 | again: |
2777 | while (1) { | |
2778 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
2779 | while (cache) { | |
2780 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
2781 | break; | |
2782 | cache = next_block_group(root, cache); | |
2783 | } | |
2784 | if (!cache) { | |
2785 | if (last == 0) | |
2786 | break; | |
2787 | last = 0; | |
2788 | continue; | |
2789 | } | |
2790 | err = cache_save_setup(cache, trans, path); | |
2791 | last = cache->key.objectid + cache->key.offset; | |
2792 | btrfs_put_block_group(cache); | |
2793 | } | |
2794 | ||
d397712b | 2795 | while (1) { |
4a8c9a62 YZ |
2796 | if (last == 0) { |
2797 | err = btrfs_run_delayed_refs(trans, root, | |
2798 | (unsigned long)-1); | |
2799 | BUG_ON(err); | |
0f9dd46c | 2800 | } |
54aa1f4d | 2801 | |
4a8c9a62 YZ |
2802 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
2803 | while (cache) { | |
0af3d00b JB |
2804 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) { |
2805 | btrfs_put_block_group(cache); | |
2806 | goto again; | |
2807 | } | |
2808 | ||
4a8c9a62 YZ |
2809 | if (cache->dirty) |
2810 | break; | |
2811 | cache = next_block_group(root, cache); | |
2812 | } | |
2813 | if (!cache) { | |
2814 | if (last == 0) | |
2815 | break; | |
2816 | last = 0; | |
2817 | continue; | |
2818 | } | |
0f9dd46c | 2819 | |
0cb59c99 JB |
2820 | if (cache->disk_cache_state == BTRFS_DC_SETUP) |
2821 | cache->disk_cache_state = BTRFS_DC_NEED_WRITE; | |
e8569813 | 2822 | cache->dirty = 0; |
4a8c9a62 | 2823 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 2824 | |
4a8c9a62 YZ |
2825 | err = write_one_cache_group(trans, root, path, cache); |
2826 | BUG_ON(err); | |
2827 | btrfs_put_block_group(cache); | |
9078a3e1 | 2828 | } |
4a8c9a62 | 2829 | |
0cb59c99 JB |
2830 | while (1) { |
2831 | /* | |
2832 | * I don't think this is needed since we're just marking our | |
2833 | * preallocated extent as written, but just in case it can't | |
2834 | * hurt. | |
2835 | */ | |
2836 | if (last == 0) { | |
2837 | err = btrfs_run_delayed_refs(trans, root, | |
2838 | (unsigned long)-1); | |
2839 | BUG_ON(err); | |
2840 | } | |
2841 | ||
2842 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
2843 | while (cache) { | |
2844 | /* | |
2845 | * Really this shouldn't happen, but it could if we | |
2846 | * couldn't write the entire preallocated extent and | |
2847 | * splitting the extent resulted in a new block. | |
2848 | */ | |
2849 | if (cache->dirty) { | |
2850 | btrfs_put_block_group(cache); | |
2851 | goto again; | |
2852 | } | |
2853 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
2854 | break; | |
2855 | cache = next_block_group(root, cache); | |
2856 | } | |
2857 | if (!cache) { | |
2858 | if (last == 0) | |
2859 | break; | |
2860 | last = 0; | |
2861 | continue; | |
2862 | } | |
2863 | ||
2864 | btrfs_write_out_cache(root, trans, cache, path); | |
2865 | ||
2866 | /* | |
2867 | * If we didn't have an error then the cache state is still | |
2868 | * NEED_WRITE, so we can set it to WRITTEN. | |
2869 | */ | |
2870 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
2871 | cache->disk_cache_state = BTRFS_DC_WRITTEN; | |
2872 | last = cache->key.objectid + cache->key.offset; | |
2873 | btrfs_put_block_group(cache); | |
2874 | } | |
2875 | ||
9078a3e1 | 2876 | btrfs_free_path(path); |
4a8c9a62 | 2877 | return 0; |
9078a3e1 CM |
2878 | } |
2879 | ||
d2fb3437 YZ |
2880 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
2881 | { | |
2882 | struct btrfs_block_group_cache *block_group; | |
2883 | int readonly = 0; | |
2884 | ||
2885 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
2886 | if (!block_group || block_group->ro) | |
2887 | readonly = 1; | |
2888 | if (block_group) | |
fa9c0d79 | 2889 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
2890 | return readonly; |
2891 | } | |
2892 | ||
593060d7 CM |
2893 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
2894 | u64 total_bytes, u64 bytes_used, | |
2895 | struct btrfs_space_info **space_info) | |
2896 | { | |
2897 | struct btrfs_space_info *found; | |
b742bb82 YZ |
2898 | int i; |
2899 | int factor; | |
2900 | ||
2901 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
2902 | BTRFS_BLOCK_GROUP_RAID10)) | |
2903 | factor = 2; | |
2904 | else | |
2905 | factor = 1; | |
593060d7 CM |
2906 | |
2907 | found = __find_space_info(info, flags); | |
2908 | if (found) { | |
25179201 | 2909 | spin_lock(&found->lock); |
593060d7 | 2910 | found->total_bytes += total_bytes; |
89a55897 | 2911 | found->disk_total += total_bytes * factor; |
593060d7 | 2912 | found->bytes_used += bytes_used; |
b742bb82 | 2913 | found->disk_used += bytes_used * factor; |
8f18cf13 | 2914 | found->full = 0; |
25179201 | 2915 | spin_unlock(&found->lock); |
593060d7 CM |
2916 | *space_info = found; |
2917 | return 0; | |
2918 | } | |
c146afad | 2919 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
2920 | if (!found) |
2921 | return -ENOMEM; | |
2922 | ||
b742bb82 YZ |
2923 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
2924 | INIT_LIST_HEAD(&found->block_groups[i]); | |
80eb234a | 2925 | init_rwsem(&found->groups_sem); |
0f9dd46c | 2926 | spin_lock_init(&found->lock); |
b742bb82 YZ |
2927 | found->flags = flags & (BTRFS_BLOCK_GROUP_DATA | |
2928 | BTRFS_BLOCK_GROUP_SYSTEM | | |
2929 | BTRFS_BLOCK_GROUP_METADATA); | |
593060d7 | 2930 | found->total_bytes = total_bytes; |
89a55897 | 2931 | found->disk_total = total_bytes * factor; |
593060d7 | 2932 | found->bytes_used = bytes_used; |
b742bb82 | 2933 | found->disk_used = bytes_used * factor; |
593060d7 | 2934 | found->bytes_pinned = 0; |
e8569813 | 2935 | found->bytes_reserved = 0; |
c146afad | 2936 | found->bytes_readonly = 0; |
f0486c68 | 2937 | found->bytes_may_use = 0; |
593060d7 | 2938 | found->full = 0; |
0e4f8f88 | 2939 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 2940 | found->chunk_alloc = 0; |
fdb5effd JB |
2941 | found->flush = 0; |
2942 | init_waitqueue_head(&found->wait); | |
593060d7 | 2943 | *space_info = found; |
4184ea7f | 2944 | list_add_rcu(&found->list, &info->space_info); |
593060d7 CM |
2945 | return 0; |
2946 | } | |
2947 | ||
8790d502 CM |
2948 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
2949 | { | |
2950 | u64 extra_flags = flags & (BTRFS_BLOCK_GROUP_RAID0 | | |
611f0e00 | 2951 | BTRFS_BLOCK_GROUP_RAID1 | |
321aecc6 | 2952 | BTRFS_BLOCK_GROUP_RAID10 | |
611f0e00 | 2953 | BTRFS_BLOCK_GROUP_DUP); |
8790d502 CM |
2954 | if (extra_flags) { |
2955 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
2956 | fs_info->avail_data_alloc_bits |= extra_flags; | |
2957 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
2958 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
2959 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
2960 | fs_info->avail_system_alloc_bits |= extra_flags; | |
2961 | } | |
2962 | } | |
593060d7 | 2963 | |
2b82032c | 2964 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 2965 | { |
cd02dca5 CM |
2966 | /* |
2967 | * we add in the count of missing devices because we want | |
2968 | * to make sure that any RAID levels on a degraded FS | |
2969 | * continue to be honored. | |
2970 | */ | |
2971 | u64 num_devices = root->fs_info->fs_devices->rw_devices + | |
2972 | root->fs_info->fs_devices->missing_devices; | |
a061fc8d CM |
2973 | |
2974 | if (num_devices == 1) | |
2975 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
2976 | if (num_devices < 4) | |
2977 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
2978 | ||
ec44a35c CM |
2979 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
2980 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 2981 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 2982 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 2983 | } |
ec44a35c CM |
2984 | |
2985 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 2986 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 2987 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 2988 | } |
ec44a35c CM |
2989 | |
2990 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
2991 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
2992 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
2993 | (flags & BTRFS_BLOCK_GROUP_DUP))) | |
2994 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; | |
2995 | return flags; | |
2996 | } | |
2997 | ||
b742bb82 | 2998 | static u64 get_alloc_profile(struct btrfs_root *root, u64 flags) |
6a63209f | 2999 | { |
b742bb82 YZ |
3000 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3001 | flags |= root->fs_info->avail_data_alloc_bits & | |
3002 | root->fs_info->data_alloc_profile; | |
3003 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3004 | flags |= root->fs_info->avail_system_alloc_bits & | |
3005 | root->fs_info->system_alloc_profile; | |
3006 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3007 | flags |= root->fs_info->avail_metadata_alloc_bits & | |
3008 | root->fs_info->metadata_alloc_profile; | |
3009 | return btrfs_reduce_alloc_profile(root, flags); | |
6a63209f JB |
3010 | } |
3011 | ||
6d07bcec | 3012 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 3013 | { |
b742bb82 | 3014 | u64 flags; |
9ed74f2d | 3015 | |
b742bb82 YZ |
3016 | if (data) |
3017 | flags = BTRFS_BLOCK_GROUP_DATA; | |
3018 | else if (root == root->fs_info->chunk_root) | |
3019 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 3020 | else |
b742bb82 | 3021 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3022 | |
b742bb82 | 3023 | return get_alloc_profile(root, flags); |
6a63209f | 3024 | } |
9ed74f2d | 3025 | |
6a63209f JB |
3026 | void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode) |
3027 | { | |
6a63209f | 3028 | BTRFS_I(inode)->space_info = __find_space_info(root->fs_info, |
f0486c68 | 3029 | BTRFS_BLOCK_GROUP_DATA); |
9ed74f2d JB |
3030 | } |
3031 | ||
6a63209f | 3032 | /* |
6a63209f JB |
3033 | * This will check the space that the inode allocates from to make sure we have |
3034 | * enough space for bytes. | |
6a63209f | 3035 | */ |
0ca1f7ce | 3036 | int btrfs_check_data_free_space(struct inode *inode, u64 bytes) |
6a63209f | 3037 | { |
6a63209f | 3038 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 3039 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ab6e2410 | 3040 | u64 used; |
0af3d00b | 3041 | int ret = 0, committed = 0, alloc_chunk = 1; |
6a63209f | 3042 | |
6a63209f JB |
3043 | /* make sure bytes are sectorsize aligned */ |
3044 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
6a63209f | 3045 | |
82d5902d LZ |
3046 | if (root == root->fs_info->tree_root || |
3047 | BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) { | |
0af3d00b JB |
3048 | alloc_chunk = 0; |
3049 | committed = 1; | |
3050 | } | |
3051 | ||
6a63209f | 3052 | data_sinfo = BTRFS_I(inode)->space_info; |
33b4d47f CM |
3053 | if (!data_sinfo) |
3054 | goto alloc; | |
9ed74f2d | 3055 | |
6a63209f JB |
3056 | again: |
3057 | /* make sure we have enough space to handle the data first */ | |
3058 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
3059 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
3060 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
3061 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
3062 | |
3063 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 3064 | struct btrfs_trans_handle *trans; |
9ed74f2d | 3065 | |
6a63209f JB |
3066 | /* |
3067 | * if we don't have enough free bytes in this space then we need | |
3068 | * to alloc a new chunk. | |
3069 | */ | |
0af3d00b | 3070 | if (!data_sinfo->full && alloc_chunk) { |
6a63209f | 3071 | u64 alloc_target; |
9ed74f2d | 3072 | |
0e4f8f88 | 3073 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 3074 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 3075 | alloc: |
6a63209f | 3076 | alloc_target = btrfs_get_alloc_profile(root, 1); |
7a7eaa40 | 3077 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3078 | if (IS_ERR(trans)) |
3079 | return PTR_ERR(trans); | |
9ed74f2d | 3080 | |
6a63209f JB |
3081 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
3082 | bytes + 2 * 1024 * 1024, | |
0e4f8f88 CM |
3083 | alloc_target, |
3084 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 3085 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
3086 | if (ret < 0) { |
3087 | if (ret != -ENOSPC) | |
3088 | return ret; | |
3089 | else | |
3090 | goto commit_trans; | |
3091 | } | |
9ed74f2d | 3092 | |
33b4d47f CM |
3093 | if (!data_sinfo) { |
3094 | btrfs_set_inode_space_info(root, inode); | |
3095 | data_sinfo = BTRFS_I(inode)->space_info; | |
3096 | } | |
6a63209f JB |
3097 | goto again; |
3098 | } | |
f2bb8f5c JB |
3099 | |
3100 | /* | |
3101 | * If we have less pinned bytes than we want to allocate then | |
3102 | * don't bother committing the transaction, it won't help us. | |
3103 | */ | |
3104 | if (data_sinfo->bytes_pinned < bytes) | |
3105 | committed = 1; | |
6a63209f | 3106 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3107 | |
4e06bdd6 | 3108 | /* commit the current transaction and try again */ |
d52a5b5f | 3109 | commit_trans: |
a4abeea4 JB |
3110 | if (!committed && |
3111 | !atomic_read(&root->fs_info->open_ioctl_trans)) { | |
4e06bdd6 | 3112 | committed = 1; |
7a7eaa40 | 3113 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3114 | if (IS_ERR(trans)) |
3115 | return PTR_ERR(trans); | |
4e06bdd6 JB |
3116 | ret = btrfs_commit_transaction(trans, root); |
3117 | if (ret) | |
3118 | return ret; | |
3119 | goto again; | |
3120 | } | |
9ed74f2d | 3121 | |
6a63209f JB |
3122 | return -ENOSPC; |
3123 | } | |
3124 | data_sinfo->bytes_may_use += bytes; | |
6a63209f | 3125 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3126 | |
9ed74f2d | 3127 | return 0; |
9ed74f2d | 3128 | } |
6a63209f | 3129 | |
6a63209f | 3130 | /* |
0ca1f7ce YZ |
3131 | * called when we are clearing an delalloc extent from the |
3132 | * inode's io_tree or there was an error for whatever reason | |
3133 | * after calling btrfs_check_data_free_space | |
6a63209f | 3134 | */ |
0ca1f7ce | 3135 | void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes) |
e3ccfa98 | 3136 | { |
0ca1f7ce | 3137 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6a63209f | 3138 | struct btrfs_space_info *data_sinfo; |
e3ccfa98 | 3139 | |
6a63209f JB |
3140 | /* make sure bytes are sectorsize aligned */ |
3141 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
e3ccfa98 | 3142 | |
6a63209f JB |
3143 | data_sinfo = BTRFS_I(inode)->space_info; |
3144 | spin_lock(&data_sinfo->lock); | |
3145 | data_sinfo->bytes_may_use -= bytes; | |
6a63209f | 3146 | spin_unlock(&data_sinfo->lock); |
e3ccfa98 JB |
3147 | } |
3148 | ||
97e728d4 | 3149 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 3150 | { |
97e728d4 JB |
3151 | struct list_head *head = &info->space_info; |
3152 | struct btrfs_space_info *found; | |
e3ccfa98 | 3153 | |
97e728d4 JB |
3154 | rcu_read_lock(); |
3155 | list_for_each_entry_rcu(found, head, list) { | |
3156 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 3157 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 3158 | } |
97e728d4 | 3159 | rcu_read_unlock(); |
e3ccfa98 JB |
3160 | } |
3161 | ||
e5bc2458 | 3162 | static int should_alloc_chunk(struct btrfs_root *root, |
0e4f8f88 CM |
3163 | struct btrfs_space_info *sinfo, u64 alloc_bytes, |
3164 | int force) | |
32c00aff | 3165 | { |
424499db | 3166 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 3167 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 3168 | u64 thresh; |
e3ccfa98 | 3169 | |
0e4f8f88 CM |
3170 | if (force == CHUNK_ALLOC_FORCE) |
3171 | return 1; | |
3172 | ||
3173 | /* | |
3174 | * in limited mode, we want to have some free space up to | |
3175 | * about 1% of the FS size. | |
3176 | */ | |
3177 | if (force == CHUNK_ALLOC_LIMITED) { | |
3178 | thresh = btrfs_super_total_bytes(&root->fs_info->super_copy); | |
3179 | thresh = max_t(u64, 64 * 1024 * 1024, | |
3180 | div_factor_fine(thresh, 1)); | |
3181 | ||
3182 | if (num_bytes - num_allocated < thresh) | |
3183 | return 1; | |
3184 | } | |
3185 | ||
3186 | /* | |
3187 | * we have two similar checks here, one based on percentage | |
3188 | * and once based on a hard number of 256MB. The idea | |
3189 | * is that if we have a good amount of free | |
3190 | * room, don't allocate a chunk. A good mount is | |
3191 | * less than 80% utilized of the chunks we have allocated, | |
3192 | * or more than 256MB free | |
3193 | */ | |
3194 | if (num_allocated + alloc_bytes + 256 * 1024 * 1024 < num_bytes) | |
424499db | 3195 | return 0; |
e3ccfa98 | 3196 | |
0e4f8f88 | 3197 | if (num_allocated + alloc_bytes < div_factor(num_bytes, 8)) |
424499db | 3198 | return 0; |
32c00aff | 3199 | |
e5bc2458 | 3200 | thresh = btrfs_super_total_bytes(&root->fs_info->super_copy); |
0e4f8f88 CM |
3201 | |
3202 | /* 256MB or 5% of the FS */ | |
e5bc2458 CM |
3203 | thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 5)); |
3204 | ||
3205 | if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 3)) | |
14ed0ca6 | 3206 | return 0; |
424499db | 3207 | return 1; |
32c00aff JB |
3208 | } |
3209 | ||
6324fbf3 CM |
3210 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
3211 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 3212 | u64 flags, int force) |
9ed74f2d | 3213 | { |
6324fbf3 | 3214 | struct btrfs_space_info *space_info; |
97e728d4 | 3215 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 3216 | int wait_for_alloc = 0; |
9ed74f2d | 3217 | int ret = 0; |
9ed74f2d | 3218 | |
2b82032c | 3219 | flags = btrfs_reduce_alloc_profile(extent_root, flags); |
ec44a35c | 3220 | |
6324fbf3 | 3221 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
3222 | if (!space_info) { |
3223 | ret = update_space_info(extent_root->fs_info, flags, | |
3224 | 0, 0, &space_info); | |
3225 | BUG_ON(ret); | |
9ed74f2d | 3226 | } |
6324fbf3 | 3227 | BUG_ON(!space_info); |
9ed74f2d | 3228 | |
6d74119f | 3229 | again: |
25179201 | 3230 | spin_lock(&space_info->lock); |
9ed74f2d | 3231 | if (space_info->force_alloc) |
0e4f8f88 | 3232 | force = space_info->force_alloc; |
25179201 JB |
3233 | if (space_info->full) { |
3234 | spin_unlock(&space_info->lock); | |
6d74119f | 3235 | return 0; |
9ed74f2d JB |
3236 | } |
3237 | ||
0e4f8f88 | 3238 | if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) { |
25179201 | 3239 | spin_unlock(&space_info->lock); |
6d74119f JB |
3240 | return 0; |
3241 | } else if (space_info->chunk_alloc) { | |
3242 | wait_for_alloc = 1; | |
3243 | } else { | |
3244 | space_info->chunk_alloc = 1; | |
9ed74f2d | 3245 | } |
0e4f8f88 | 3246 | |
25179201 | 3247 | spin_unlock(&space_info->lock); |
9ed74f2d | 3248 | |
6d74119f JB |
3249 | mutex_lock(&fs_info->chunk_mutex); |
3250 | ||
3251 | /* | |
3252 | * The chunk_mutex is held throughout the entirety of a chunk | |
3253 | * allocation, so once we've acquired the chunk_mutex we know that the | |
3254 | * other guy is done and we need to recheck and see if we should | |
3255 | * allocate. | |
3256 | */ | |
3257 | if (wait_for_alloc) { | |
3258 | mutex_unlock(&fs_info->chunk_mutex); | |
3259 | wait_for_alloc = 0; | |
3260 | goto again; | |
3261 | } | |
3262 | ||
67377734 JB |
3263 | /* |
3264 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3265 | * allocating mixed chunks instead of individual chunks. | |
3266 | */ | |
3267 | if (btrfs_mixed_space_info(space_info)) | |
3268 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3269 | ||
97e728d4 JB |
3270 | /* |
3271 | * if we're doing a data chunk, go ahead and make sure that | |
3272 | * we keep a reasonable number of metadata chunks allocated in the | |
3273 | * FS as well. | |
3274 | */ | |
9ed74f2d | 3275 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
3276 | fs_info->data_chunk_allocations++; |
3277 | if (!(fs_info->data_chunk_allocations % | |
3278 | fs_info->metadata_ratio)) | |
3279 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
3280 | } |
3281 | ||
2b82032c | 3282 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
92b8e897 MF |
3283 | if (ret < 0 && ret != -ENOSPC) |
3284 | goto out; | |
3285 | ||
9ed74f2d | 3286 | spin_lock(&space_info->lock); |
9ed74f2d | 3287 | if (ret) |
6324fbf3 | 3288 | space_info->full = 1; |
424499db YZ |
3289 | else |
3290 | ret = 1; | |
6d74119f | 3291 | |
0e4f8f88 | 3292 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3293 | space_info->chunk_alloc = 0; |
9ed74f2d | 3294 | spin_unlock(&space_info->lock); |
92b8e897 | 3295 | out: |
c146afad | 3296 | mutex_unlock(&extent_root->fs_info->chunk_mutex); |
0f9dd46c | 3297 | return ret; |
6324fbf3 | 3298 | } |
9ed74f2d | 3299 | |
9ed74f2d | 3300 | /* |
5da9d01b | 3301 | * shrink metadata reservation for delalloc |
9ed74f2d | 3302 | */ |
5da9d01b | 3303 | static int shrink_delalloc(struct btrfs_trans_handle *trans, |
0019f10d | 3304 | struct btrfs_root *root, u64 to_reclaim, int sync) |
5da9d01b | 3305 | { |
0ca1f7ce | 3306 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 3307 | struct btrfs_space_info *space_info; |
5da9d01b YZ |
3308 | u64 reserved; |
3309 | u64 max_reclaim; | |
3310 | u64 reclaimed = 0; | |
b1953bce | 3311 | long time_left; |
bf9022e0 | 3312 | int nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT; |
b1953bce | 3313 | int loops = 0; |
36e39c40 | 3314 | unsigned long progress; |
5da9d01b | 3315 | |
0ca1f7ce | 3316 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 3317 | space_info = block_rsv->space_info; |
bf9022e0 CM |
3318 | |
3319 | smp_mb(); | |
0019f10d | 3320 | reserved = space_info->bytes_reserved; |
36e39c40 | 3321 | progress = space_info->reservation_progress; |
5da9d01b YZ |
3322 | |
3323 | if (reserved == 0) | |
3324 | return 0; | |
c4f675cd | 3325 | |
fdb5effd JB |
3326 | smp_mb(); |
3327 | if (root->fs_info->delalloc_bytes == 0) { | |
3328 | if (trans) | |
3329 | return 0; | |
3330 | btrfs_wait_ordered_extents(root, 0, 0); | |
3331 | return 0; | |
3332 | } | |
3333 | ||
5da9d01b YZ |
3334 | max_reclaim = min(reserved, to_reclaim); |
3335 | ||
b1953bce | 3336 | while (loops < 1024) { |
bf9022e0 CM |
3337 | /* have the flusher threads jump in and do some IO */ |
3338 | smp_mb(); | |
3339 | nr_pages = min_t(unsigned long, nr_pages, | |
3340 | root->fs_info->delalloc_bytes >> PAGE_CACHE_SHIFT); | |
3341 | writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages); | |
5da9d01b | 3342 | |
0019f10d | 3343 | spin_lock(&space_info->lock); |
36e39c40 | 3344 | if (reserved > space_info->bytes_reserved) |
0019f10d JB |
3345 | reclaimed += reserved - space_info->bytes_reserved; |
3346 | reserved = space_info->bytes_reserved; | |
3347 | spin_unlock(&space_info->lock); | |
5da9d01b | 3348 | |
36e39c40 CM |
3349 | loops++; |
3350 | ||
5da9d01b YZ |
3351 | if (reserved == 0 || reclaimed >= max_reclaim) |
3352 | break; | |
3353 | ||
3354 | if (trans && trans->transaction->blocked) | |
3355 | return -EAGAIN; | |
bf9022e0 | 3356 | |
36e39c40 | 3357 | time_left = schedule_timeout_interruptible(1); |
b1953bce JB |
3358 | |
3359 | /* We were interrupted, exit */ | |
3360 | if (time_left) | |
3361 | break; | |
3362 | ||
36e39c40 CM |
3363 | /* we've kicked the IO a few times, if anything has been freed, |
3364 | * exit. There is no sense in looping here for a long time | |
3365 | * when we really need to commit the transaction, or there are | |
3366 | * just too many writers without enough free space | |
3367 | */ | |
3368 | ||
3369 | if (loops > 3) { | |
3370 | smp_mb(); | |
3371 | if (progress != space_info->reservation_progress) | |
3372 | break; | |
3373 | } | |
bf9022e0 | 3374 | |
5da9d01b | 3375 | } |
fdb5effd JB |
3376 | if (reclaimed >= to_reclaim && !trans) |
3377 | btrfs_wait_ordered_extents(root, 0, 0); | |
5da9d01b YZ |
3378 | return reclaimed >= to_reclaim; |
3379 | } | |
3380 | ||
8bb8ab2e JB |
3381 | /* |
3382 | * Retries tells us how many times we've called reserve_metadata_bytes. The | |
3383 | * idea is if this is the first call (retries == 0) then we will add to our | |
3384 | * reserved count if we can't make the allocation in order to hold our place | |
3385 | * while we go and try and free up space. That way for retries > 1 we don't try | |
3386 | * and add space, we just check to see if the amount of unused space is >= the | |
3387 | * total space, meaning that our reservation is valid. | |
3388 | * | |
3389 | * However if we don't intend to retry this reservation, pass -1 as retries so | |
3390 | * that it short circuits this logic. | |
3391 | */ | |
3392 | static int reserve_metadata_bytes(struct btrfs_trans_handle *trans, | |
3393 | struct btrfs_root *root, | |
3394 | struct btrfs_block_rsv *block_rsv, | |
3395 | u64 orig_bytes, int flush) | |
9ed74f2d | 3396 | { |
f0486c68 | 3397 | struct btrfs_space_info *space_info = block_rsv->space_info; |
8bb8ab2e JB |
3398 | u64 unused; |
3399 | u64 num_bytes = orig_bytes; | |
3400 | int retries = 0; | |
3401 | int ret = 0; | |
38227933 | 3402 | bool committed = false; |
fdb5effd | 3403 | bool flushing = false; |
9ed74f2d | 3404 | |
8bb8ab2e | 3405 | again: |
fdb5effd | 3406 | ret = 0; |
8bb8ab2e | 3407 | spin_lock(&space_info->lock); |
fdb5effd JB |
3408 | /* |
3409 | * We only want to wait if somebody other than us is flushing and we are | |
3410 | * actually alloed to flush. | |
3411 | */ | |
3412 | while (flush && !flushing && space_info->flush) { | |
3413 | spin_unlock(&space_info->lock); | |
3414 | /* | |
3415 | * If we have a trans handle we can't wait because the flusher | |
3416 | * may have to commit the transaction, which would mean we would | |
3417 | * deadlock since we are waiting for the flusher to finish, but | |
3418 | * hold the current transaction open. | |
3419 | */ | |
3420 | if (trans) | |
3421 | return -EAGAIN; | |
3422 | ret = wait_event_interruptible(space_info->wait, | |
3423 | !space_info->flush); | |
3424 | /* Must have been interrupted, return */ | |
3425 | if (ret) | |
3426 | return -EINTR; | |
3427 | ||
3428 | spin_lock(&space_info->lock); | |
3429 | } | |
3430 | ||
3431 | ret = -ENOSPC; | |
8bb8ab2e JB |
3432 | unused = space_info->bytes_used + space_info->bytes_reserved + |
3433 | space_info->bytes_pinned + space_info->bytes_readonly + | |
3434 | space_info->bytes_may_use; | |
9ed74f2d | 3435 | |
8bb8ab2e JB |
3436 | /* |
3437 | * The idea here is that we've not already over-reserved the block group | |
3438 | * then we can go ahead and save our reservation first and then start | |
3439 | * flushing if we need to. Otherwise if we've already overcommitted | |
3440 | * lets start flushing stuff first and then come back and try to make | |
3441 | * our reservation. | |
3442 | */ | |
3443 | if (unused <= space_info->total_bytes) { | |
6f334348 | 3444 | unused = space_info->total_bytes - unused; |
8bb8ab2e | 3445 | if (unused >= num_bytes) { |
fdb5effd | 3446 | space_info->bytes_reserved += orig_bytes; |
8bb8ab2e JB |
3447 | ret = 0; |
3448 | } else { | |
3449 | /* | |
3450 | * Ok set num_bytes to orig_bytes since we aren't | |
3451 | * overocmmitted, this way we only try and reclaim what | |
3452 | * we need. | |
3453 | */ | |
3454 | num_bytes = orig_bytes; | |
3455 | } | |
3456 | } else { | |
3457 | /* | |
3458 | * Ok we're over committed, set num_bytes to the overcommitted | |
3459 | * amount plus the amount of bytes that we need for this | |
3460 | * reservation. | |
3461 | */ | |
3462 | num_bytes = unused - space_info->total_bytes + | |
3463 | (orig_bytes * (retries + 1)); | |
3464 | } | |
9ed74f2d | 3465 | |
8bb8ab2e JB |
3466 | /* |
3467 | * Couldn't make our reservation, save our place so while we're trying | |
3468 | * to reclaim space we can actually use it instead of somebody else | |
3469 | * stealing it from us. | |
3470 | */ | |
fdb5effd JB |
3471 | if (ret && flush) { |
3472 | flushing = true; | |
3473 | space_info->flush = 1; | |
8bb8ab2e | 3474 | } |
9ed74f2d | 3475 | |
f0486c68 | 3476 | spin_unlock(&space_info->lock); |
9ed74f2d | 3477 | |
fdb5effd | 3478 | if (!ret || !flush) |
8bb8ab2e | 3479 | goto out; |
f0486c68 | 3480 | |
8bb8ab2e JB |
3481 | /* |
3482 | * We do synchronous shrinking since we don't actually unreserve | |
3483 | * metadata until after the IO is completed. | |
3484 | */ | |
3485 | ret = shrink_delalloc(trans, root, num_bytes, 1); | |
fdb5effd | 3486 | if (ret < 0) |
8bb8ab2e | 3487 | goto out; |
f0486c68 | 3488 | |
75c195a2 CM |
3489 | ret = 0; |
3490 | ||
8bb8ab2e JB |
3491 | /* |
3492 | * So if we were overcommitted it's possible that somebody else flushed | |
3493 | * out enough space and we simply didn't have enough space to reclaim, | |
3494 | * so go back around and try again. | |
3495 | */ | |
3496 | if (retries < 2) { | |
3497 | retries++; | |
3498 | goto again; | |
3499 | } | |
f0486c68 | 3500 | |
8bb8ab2e JB |
3501 | /* |
3502 | * Not enough space to be reclaimed, don't bother committing the | |
3503 | * transaction. | |
3504 | */ | |
fdb5effd | 3505 | spin_lock(&space_info->lock); |
8bb8ab2e JB |
3506 | if (space_info->bytes_pinned < orig_bytes) |
3507 | ret = -ENOSPC; | |
3508 | spin_unlock(&space_info->lock); | |
3509 | if (ret) | |
3510 | goto out; | |
f0486c68 | 3511 | |
8bb8ab2e | 3512 | ret = -EAGAIN; |
75c195a2 | 3513 | if (trans) |
8bb8ab2e | 3514 | goto out; |
f0486c68 | 3515 | |
8bb8ab2e | 3516 | ret = -ENOSPC; |
75c195a2 CM |
3517 | if (committed) |
3518 | goto out; | |
3519 | ||
7a7eaa40 | 3520 | trans = btrfs_join_transaction(root); |
8bb8ab2e JB |
3521 | if (IS_ERR(trans)) |
3522 | goto out; | |
3523 | ret = btrfs_commit_transaction(trans, root); | |
38227933 JB |
3524 | if (!ret) { |
3525 | trans = NULL; | |
3526 | committed = true; | |
8bb8ab2e | 3527 | goto again; |
38227933 | 3528 | } |
8bb8ab2e JB |
3529 | |
3530 | out: | |
fdb5effd | 3531 | if (flushing) { |
8bb8ab2e | 3532 | spin_lock(&space_info->lock); |
fdb5effd JB |
3533 | space_info->flush = 0; |
3534 | wake_up_all(&space_info->wait); | |
8bb8ab2e | 3535 | spin_unlock(&space_info->lock); |
f0486c68 | 3536 | } |
f0486c68 YZ |
3537 | return ret; |
3538 | } | |
3539 | ||
3540 | static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans, | |
3541 | struct btrfs_root *root) | |
3542 | { | |
3543 | struct btrfs_block_rsv *block_rsv; | |
3544 | if (root->ref_cows) | |
3545 | block_rsv = trans->block_rsv; | |
3546 | else | |
3547 | block_rsv = root->block_rsv; | |
3548 | ||
3549 | if (!block_rsv) | |
3550 | block_rsv = &root->fs_info->empty_block_rsv; | |
3551 | ||
3552 | return block_rsv; | |
3553 | } | |
3554 | ||
3555 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
3556 | u64 num_bytes) | |
3557 | { | |
3558 | int ret = -ENOSPC; | |
3559 | spin_lock(&block_rsv->lock); | |
3560 | if (block_rsv->reserved >= num_bytes) { | |
3561 | block_rsv->reserved -= num_bytes; | |
3562 | if (block_rsv->reserved < block_rsv->size) | |
3563 | block_rsv->full = 0; | |
3564 | ret = 0; | |
3565 | } | |
3566 | spin_unlock(&block_rsv->lock); | |
3567 | return ret; | |
3568 | } | |
3569 | ||
3570 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
3571 | u64 num_bytes, int update_size) | |
3572 | { | |
3573 | spin_lock(&block_rsv->lock); | |
3574 | block_rsv->reserved += num_bytes; | |
3575 | if (update_size) | |
3576 | block_rsv->size += num_bytes; | |
3577 | else if (block_rsv->reserved >= block_rsv->size) | |
3578 | block_rsv->full = 1; | |
3579 | spin_unlock(&block_rsv->lock); | |
3580 | } | |
3581 | ||
62a45b60 DS |
3582 | static void block_rsv_release_bytes(struct btrfs_block_rsv *block_rsv, |
3583 | struct btrfs_block_rsv *dest, u64 num_bytes) | |
f0486c68 YZ |
3584 | { |
3585 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
3586 | ||
3587 | spin_lock(&block_rsv->lock); | |
3588 | if (num_bytes == (u64)-1) | |
3589 | num_bytes = block_rsv->size; | |
3590 | block_rsv->size -= num_bytes; | |
3591 | if (block_rsv->reserved >= block_rsv->size) { | |
3592 | num_bytes = block_rsv->reserved - block_rsv->size; | |
3593 | block_rsv->reserved = block_rsv->size; | |
3594 | block_rsv->full = 1; | |
3595 | } else { | |
3596 | num_bytes = 0; | |
3597 | } | |
3598 | spin_unlock(&block_rsv->lock); | |
3599 | ||
3600 | if (num_bytes > 0) { | |
3601 | if (dest) { | |
e9e22899 JB |
3602 | spin_lock(&dest->lock); |
3603 | if (!dest->full) { | |
3604 | u64 bytes_to_add; | |
3605 | ||
3606 | bytes_to_add = dest->size - dest->reserved; | |
3607 | bytes_to_add = min(num_bytes, bytes_to_add); | |
3608 | dest->reserved += bytes_to_add; | |
3609 | if (dest->reserved >= dest->size) | |
3610 | dest->full = 1; | |
3611 | num_bytes -= bytes_to_add; | |
3612 | } | |
3613 | spin_unlock(&dest->lock); | |
3614 | } | |
3615 | if (num_bytes) { | |
f0486c68 YZ |
3616 | spin_lock(&space_info->lock); |
3617 | space_info->bytes_reserved -= num_bytes; | |
36e39c40 | 3618 | space_info->reservation_progress++; |
f0486c68 | 3619 | spin_unlock(&space_info->lock); |
4e06bdd6 | 3620 | } |
9ed74f2d | 3621 | } |
f0486c68 | 3622 | } |
4e06bdd6 | 3623 | |
f0486c68 YZ |
3624 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
3625 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
3626 | { | |
3627 | int ret; | |
9ed74f2d | 3628 | |
f0486c68 YZ |
3629 | ret = block_rsv_use_bytes(src, num_bytes); |
3630 | if (ret) | |
3631 | return ret; | |
9ed74f2d | 3632 | |
f0486c68 | 3633 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
3634 | return 0; |
3635 | } | |
3636 | ||
f0486c68 | 3637 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv) |
9ed74f2d | 3638 | { |
f0486c68 YZ |
3639 | memset(rsv, 0, sizeof(*rsv)); |
3640 | spin_lock_init(&rsv->lock); | |
3641 | atomic_set(&rsv->usage, 1); | |
3642 | rsv->priority = 6; | |
3643 | INIT_LIST_HEAD(&rsv->list); | |
3644 | } | |
3645 | ||
3646 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root) | |
3647 | { | |
3648 | struct btrfs_block_rsv *block_rsv; | |
3649 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 3650 | |
f0486c68 YZ |
3651 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
3652 | if (!block_rsv) | |
3653 | return NULL; | |
9ed74f2d | 3654 | |
f0486c68 | 3655 | btrfs_init_block_rsv(block_rsv); |
f0486c68 YZ |
3656 | block_rsv->space_info = __find_space_info(fs_info, |
3657 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
3658 | return block_rsv; |
3659 | } | |
9ed74f2d | 3660 | |
f0486c68 YZ |
3661 | void btrfs_free_block_rsv(struct btrfs_root *root, |
3662 | struct btrfs_block_rsv *rsv) | |
3663 | { | |
3664 | if (rsv && atomic_dec_and_test(&rsv->usage)) { | |
3665 | btrfs_block_rsv_release(root, rsv, (u64)-1); | |
3666 | if (!rsv->durable) | |
3667 | kfree(rsv); | |
3668 | } | |
9ed74f2d JB |
3669 | } |
3670 | ||
3671 | /* | |
f0486c68 YZ |
3672 | * make the block_rsv struct be able to capture freed space. |
3673 | * the captured space will re-add to the the block_rsv struct | |
3674 | * after transaction commit | |
9ed74f2d | 3675 | */ |
f0486c68 YZ |
3676 | void btrfs_add_durable_block_rsv(struct btrfs_fs_info *fs_info, |
3677 | struct btrfs_block_rsv *block_rsv) | |
9ed74f2d | 3678 | { |
f0486c68 YZ |
3679 | block_rsv->durable = 1; |
3680 | mutex_lock(&fs_info->durable_block_rsv_mutex); | |
3681 | list_add_tail(&block_rsv->list, &fs_info->durable_block_rsv_list); | |
3682 | mutex_unlock(&fs_info->durable_block_rsv_mutex); | |
3683 | } | |
9ed74f2d | 3684 | |
f0486c68 YZ |
3685 | int btrfs_block_rsv_add(struct btrfs_trans_handle *trans, |
3686 | struct btrfs_root *root, | |
3687 | struct btrfs_block_rsv *block_rsv, | |
8bb8ab2e | 3688 | u64 num_bytes) |
f0486c68 YZ |
3689 | { |
3690 | int ret; | |
9ed74f2d | 3691 | |
f0486c68 YZ |
3692 | if (num_bytes == 0) |
3693 | return 0; | |
8bb8ab2e JB |
3694 | |
3695 | ret = reserve_metadata_bytes(trans, root, block_rsv, num_bytes, 1); | |
f0486c68 YZ |
3696 | if (!ret) { |
3697 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
3698 | return 0; | |
3699 | } | |
9ed74f2d | 3700 | |
f0486c68 YZ |
3701 | return ret; |
3702 | } | |
9ed74f2d | 3703 | |
f0486c68 YZ |
3704 | int btrfs_block_rsv_check(struct btrfs_trans_handle *trans, |
3705 | struct btrfs_root *root, | |
3706 | struct btrfs_block_rsv *block_rsv, | |
3707 | u64 min_reserved, int min_factor) | |
3708 | { | |
3709 | u64 num_bytes = 0; | |
3710 | int commit_trans = 0; | |
3711 | int ret = -ENOSPC; | |
9ed74f2d | 3712 | |
f0486c68 YZ |
3713 | if (!block_rsv) |
3714 | return 0; | |
9ed74f2d | 3715 | |
f0486c68 YZ |
3716 | spin_lock(&block_rsv->lock); |
3717 | if (min_factor > 0) | |
3718 | num_bytes = div_factor(block_rsv->size, min_factor); | |
3719 | if (min_reserved > num_bytes) | |
3720 | num_bytes = min_reserved; | |
9ed74f2d | 3721 | |
f0486c68 YZ |
3722 | if (block_rsv->reserved >= num_bytes) { |
3723 | ret = 0; | |
3724 | } else { | |
3725 | num_bytes -= block_rsv->reserved; | |
3726 | if (block_rsv->durable && | |
3727 | block_rsv->freed[0] + block_rsv->freed[1] >= num_bytes) | |
3728 | commit_trans = 1; | |
3729 | } | |
3730 | spin_unlock(&block_rsv->lock); | |
3731 | if (!ret) | |
3732 | return 0; | |
3733 | ||
3734 | if (block_rsv->refill_used) { | |
8bb8ab2e JB |
3735 | ret = reserve_metadata_bytes(trans, root, block_rsv, |
3736 | num_bytes, 0); | |
f0486c68 YZ |
3737 | if (!ret) { |
3738 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
3739 | return 0; | |
4e06bdd6 | 3740 | } |
f0486c68 | 3741 | } |
9ed74f2d | 3742 | |
f0486c68 YZ |
3743 | if (commit_trans) { |
3744 | if (trans) | |
3745 | return -EAGAIN; | |
7a7eaa40 | 3746 | trans = btrfs_join_transaction(root); |
f0486c68 YZ |
3747 | BUG_ON(IS_ERR(trans)); |
3748 | ret = btrfs_commit_transaction(trans, root); | |
3749 | return 0; | |
6a63209f | 3750 | } |
9ed74f2d | 3751 | |
f0486c68 YZ |
3752 | return -ENOSPC; |
3753 | } | |
3754 | ||
3755 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, | |
3756 | struct btrfs_block_rsv *dst_rsv, | |
3757 | u64 num_bytes) | |
3758 | { | |
3759 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
3760 | } | |
3761 | ||
3762 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
3763 | struct btrfs_block_rsv *block_rsv, | |
3764 | u64 num_bytes) | |
3765 | { | |
3766 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
3767 | if (global_rsv->full || global_rsv == block_rsv || | |
3768 | block_rsv->space_info != global_rsv->space_info) | |
3769 | global_rsv = NULL; | |
3770 | block_rsv_release_bytes(block_rsv, global_rsv, num_bytes); | |
6a63209f JB |
3771 | } |
3772 | ||
3773 | /* | |
8929ecfa YZ |
3774 | * helper to calculate size of global block reservation. |
3775 | * the desired value is sum of space used by extent tree, | |
3776 | * checksum tree and root tree | |
6a63209f | 3777 | */ |
8929ecfa | 3778 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 3779 | { |
8929ecfa YZ |
3780 | struct btrfs_space_info *sinfo; |
3781 | u64 num_bytes; | |
3782 | u64 meta_used; | |
3783 | u64 data_used; | |
3784 | int csum_size = btrfs_super_csum_size(&fs_info->super_copy); | |
6a63209f | 3785 | |
8929ecfa YZ |
3786 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
3787 | spin_lock(&sinfo->lock); | |
3788 | data_used = sinfo->bytes_used; | |
3789 | spin_unlock(&sinfo->lock); | |
33b4d47f | 3790 | |
8929ecfa YZ |
3791 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
3792 | spin_lock(&sinfo->lock); | |
6d48755d JB |
3793 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
3794 | data_used = 0; | |
8929ecfa YZ |
3795 | meta_used = sinfo->bytes_used; |
3796 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 3797 | |
8929ecfa YZ |
3798 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
3799 | csum_size * 2; | |
3800 | num_bytes += div64_u64(data_used + meta_used, 50); | |
4e06bdd6 | 3801 | |
8929ecfa YZ |
3802 | if (num_bytes * 3 > meta_used) |
3803 | num_bytes = div64_u64(meta_used, 3); | |
ab6e2410 | 3804 | |
8929ecfa YZ |
3805 | return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10); |
3806 | } | |
6a63209f | 3807 | |
8929ecfa YZ |
3808 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
3809 | { | |
3810 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
3811 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
3812 | u64 num_bytes; | |
6a63209f | 3813 | |
8929ecfa | 3814 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 3815 | |
8929ecfa YZ |
3816 | spin_lock(&block_rsv->lock); |
3817 | spin_lock(&sinfo->lock); | |
4e06bdd6 | 3818 | |
8929ecfa | 3819 | block_rsv->size = num_bytes; |
4e06bdd6 | 3820 | |
8929ecfa | 3821 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
6d48755d JB |
3822 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
3823 | sinfo->bytes_may_use; | |
8929ecfa YZ |
3824 | |
3825 | if (sinfo->total_bytes > num_bytes) { | |
3826 | num_bytes = sinfo->total_bytes - num_bytes; | |
3827 | block_rsv->reserved += num_bytes; | |
3828 | sinfo->bytes_reserved += num_bytes; | |
6a63209f | 3829 | } |
6a63209f | 3830 | |
8929ecfa YZ |
3831 | if (block_rsv->reserved >= block_rsv->size) { |
3832 | num_bytes = block_rsv->reserved - block_rsv->size; | |
3833 | sinfo->bytes_reserved -= num_bytes; | |
36e39c40 | 3834 | sinfo->reservation_progress++; |
8929ecfa YZ |
3835 | block_rsv->reserved = block_rsv->size; |
3836 | block_rsv->full = 1; | |
3837 | } | |
182608c8 | 3838 | |
8929ecfa YZ |
3839 | spin_unlock(&sinfo->lock); |
3840 | spin_unlock(&block_rsv->lock); | |
6a63209f JB |
3841 | } |
3842 | ||
f0486c68 | 3843 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 3844 | { |
f0486c68 | 3845 | struct btrfs_space_info *space_info; |
6a63209f | 3846 | |
f0486c68 YZ |
3847 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
3848 | fs_info->chunk_block_rsv.space_info = space_info; | |
3849 | fs_info->chunk_block_rsv.priority = 10; | |
6a63209f | 3850 | |
f0486c68 | 3851 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa YZ |
3852 | fs_info->global_block_rsv.space_info = space_info; |
3853 | fs_info->global_block_rsv.priority = 10; | |
3854 | fs_info->global_block_rsv.refill_used = 1; | |
3855 | fs_info->delalloc_block_rsv.space_info = space_info; | |
f0486c68 YZ |
3856 | fs_info->trans_block_rsv.space_info = space_info; |
3857 | fs_info->empty_block_rsv.space_info = space_info; | |
3858 | fs_info->empty_block_rsv.priority = 10; | |
3859 | ||
8929ecfa YZ |
3860 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
3861 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
3862 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
3863 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 3864 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa YZ |
3865 | |
3866 | btrfs_add_durable_block_rsv(fs_info, &fs_info->global_block_rsv); | |
3867 | ||
3868 | btrfs_add_durable_block_rsv(fs_info, &fs_info->delalloc_block_rsv); | |
3869 | ||
3870 | update_global_block_rsv(fs_info); | |
6a63209f JB |
3871 | } |
3872 | ||
8929ecfa | 3873 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 3874 | { |
8929ecfa YZ |
3875 | block_rsv_release_bytes(&fs_info->global_block_rsv, NULL, (u64)-1); |
3876 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); | |
3877 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
3878 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
3879 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
3880 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
3881 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
f0486c68 | 3882 | } |
6a63209f | 3883 | |
fcb80c2a JB |
3884 | int btrfs_truncate_reserve_metadata(struct btrfs_trans_handle *trans, |
3885 | struct btrfs_root *root, | |
3886 | struct btrfs_block_rsv *rsv) | |
3887 | { | |
3888 | struct btrfs_block_rsv *trans_rsv = &root->fs_info->trans_block_rsv; | |
3889 | u64 num_bytes; | |
3890 | int ret; | |
3891 | ||
3892 | /* | |
3893 | * Truncate should be freeing data, but give us 2 items just in case it | |
3894 | * needs to use some space. We may want to be smarter about this in the | |
3895 | * future. | |
3896 | */ | |
ff5714cc | 3897 | num_bytes = btrfs_calc_trans_metadata_size(root, 2); |
fcb80c2a JB |
3898 | |
3899 | /* We already have enough bytes, just return */ | |
3900 | if (rsv->reserved >= num_bytes) | |
3901 | return 0; | |
3902 | ||
3903 | num_bytes -= rsv->reserved; | |
3904 | ||
3905 | /* | |
3906 | * You should have reserved enough space before hand to do this, so this | |
3907 | * should not fail. | |
3908 | */ | |
3909 | ret = block_rsv_migrate_bytes(trans_rsv, rsv, num_bytes); | |
3910 | BUG_ON(ret); | |
3911 | ||
3912 | return 0; | |
3913 | } | |
3914 | ||
a22285a6 YZ |
3915 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
3916 | struct btrfs_root *root) | |
6a63209f | 3917 | { |
a22285a6 YZ |
3918 | if (!trans->bytes_reserved) |
3919 | return; | |
6a63209f | 3920 | |
a22285a6 YZ |
3921 | BUG_ON(trans->block_rsv != &root->fs_info->trans_block_rsv); |
3922 | btrfs_block_rsv_release(root, trans->block_rsv, | |
3923 | trans->bytes_reserved); | |
3924 | trans->bytes_reserved = 0; | |
3925 | } | |
6a63209f | 3926 | |
d68fc57b YZ |
3927 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
3928 | struct inode *inode) | |
3929 | { | |
3930 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3931 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
3932 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
3933 | ||
3934 | /* | |
fcb80c2a JB |
3935 | * We need to hold space in order to delete our orphan item once we've |
3936 | * added it, so this takes the reservation so we can release it later | |
3937 | * when we are truly done with the orphan item. | |
d68fc57b | 3938 | */ |
ff5714cc | 3939 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
d68fc57b | 3940 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
6a63209f JB |
3941 | } |
3942 | ||
d68fc57b | 3943 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 3944 | { |
d68fc57b | 3945 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 3946 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
d68fc57b YZ |
3947 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
3948 | } | |
97e728d4 | 3949 | |
a22285a6 YZ |
3950 | int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans, |
3951 | struct btrfs_pending_snapshot *pending) | |
3952 | { | |
3953 | struct btrfs_root *root = pending->root; | |
3954 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
3955 | struct btrfs_block_rsv *dst_rsv = &pending->block_rsv; | |
3956 | /* | |
3957 | * two for root back/forward refs, two for directory entries | |
3958 | * and one for root of the snapshot. | |
3959 | */ | |
16cdcec7 | 3960 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5); |
a22285a6 YZ |
3961 | dst_rsv->space_info = src_rsv->space_info; |
3962 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
97e728d4 JB |
3963 | } |
3964 | ||
9e0baf60 JB |
3965 | static unsigned drop_outstanding_extent(struct inode *inode) |
3966 | { | |
3967 | unsigned dropped_extents = 0; | |
3968 | ||
3969 | spin_lock(&BTRFS_I(inode)->lock); | |
3970 | BUG_ON(!BTRFS_I(inode)->outstanding_extents); | |
3971 | BTRFS_I(inode)->outstanding_extents--; | |
3972 | ||
3973 | /* | |
3974 | * If we have more or the same amount of outsanding extents than we have | |
3975 | * reserved then we need to leave the reserved extents count alone. | |
3976 | */ | |
3977 | if (BTRFS_I(inode)->outstanding_extents >= | |
3978 | BTRFS_I(inode)->reserved_extents) | |
3979 | goto out; | |
3980 | ||
3981 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
3982 | BTRFS_I(inode)->outstanding_extents; | |
3983 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
3984 | out: | |
3985 | spin_unlock(&BTRFS_I(inode)->lock); | |
3986 | return dropped_extents; | |
3987 | } | |
3988 | ||
0ca1f7ce | 3989 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes) |
6324fbf3 | 3990 | { |
0ca1f7ce YZ |
3991 | return num_bytes >>= 3; |
3992 | } | |
c146afad | 3993 | |
0ca1f7ce YZ |
3994 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
3995 | { | |
3996 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
3997 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 JB |
3998 | u64 to_reserve = 0; |
3999 | unsigned nr_extents = 0; | |
0ca1f7ce | 4000 | int ret; |
6324fbf3 | 4001 | |
0ca1f7ce YZ |
4002 | if (btrfs_transaction_in_commit(root->fs_info)) |
4003 | schedule_timeout(1); | |
ec44a35c | 4004 | |
0ca1f7ce | 4005 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 4006 | |
9e0baf60 JB |
4007 | spin_lock(&BTRFS_I(inode)->lock); |
4008 | BTRFS_I(inode)->outstanding_extents++; | |
4009 | ||
4010 | if (BTRFS_I(inode)->outstanding_extents > | |
4011 | BTRFS_I(inode)->reserved_extents) { | |
4012 | nr_extents = BTRFS_I(inode)->outstanding_extents - | |
4013 | BTRFS_I(inode)->reserved_extents; | |
4014 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
57a45ced | 4015 | |
16cdcec7 | 4016 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); |
593060d7 | 4017 | } |
9e0baf60 | 4018 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 4019 | |
0ca1f7ce | 4020 | to_reserve += calc_csum_metadata_size(inode, num_bytes); |
8bb8ab2e | 4021 | ret = reserve_metadata_bytes(NULL, root, block_rsv, to_reserve, 1); |
9e0baf60 JB |
4022 | if (ret) { |
4023 | unsigned dropped; | |
4024 | /* | |
4025 | * We don't need the return value since our reservation failed, | |
4026 | * we just need to clean up our counter. | |
4027 | */ | |
4028 | dropped = drop_outstanding_extent(inode); | |
4029 | WARN_ON(dropped > 1); | |
0ca1f7ce | 4030 | return ret; |
9e0baf60 | 4031 | } |
25179201 | 4032 | |
0ca1f7ce YZ |
4033 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
4034 | ||
0ca1f7ce YZ |
4035 | return 0; |
4036 | } | |
4037 | ||
4038 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) | |
4039 | { | |
4040 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
4041 | u64 to_free = 0; |
4042 | unsigned dropped; | |
0ca1f7ce YZ |
4043 | |
4044 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
9e0baf60 | 4045 | dropped = drop_outstanding_extent(inode); |
97e728d4 | 4046 | |
0ca1f7ce | 4047 | to_free = calc_csum_metadata_size(inode, num_bytes); |
9e0baf60 JB |
4048 | if (dropped > 0) |
4049 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce YZ |
4050 | |
4051 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, | |
4052 | to_free); | |
4053 | } | |
4054 | ||
4055 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes) | |
4056 | { | |
4057 | int ret; | |
4058 | ||
4059 | ret = btrfs_check_data_free_space(inode, num_bytes); | |
d397712b | 4060 | if (ret) |
0ca1f7ce YZ |
4061 | return ret; |
4062 | ||
4063 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes); | |
4064 | if (ret) { | |
4065 | btrfs_free_reserved_data_space(inode, num_bytes); | |
4066 | return ret; | |
4067 | } | |
4068 | ||
4069 | return 0; | |
4070 | } | |
4071 | ||
4072 | void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes) | |
4073 | { | |
4074 | btrfs_delalloc_release_metadata(inode, num_bytes); | |
4075 | btrfs_free_reserved_data_space(inode, num_bytes); | |
6324fbf3 CM |
4076 | } |
4077 | ||
9078a3e1 CM |
4078 | static int update_block_group(struct btrfs_trans_handle *trans, |
4079 | struct btrfs_root *root, | |
f0486c68 | 4080 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 4081 | { |
0af3d00b | 4082 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 4083 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 4084 | u64 total = num_bytes; |
9078a3e1 | 4085 | u64 old_val; |
db94535d | 4086 | u64 byte_in_group; |
0af3d00b | 4087 | int factor; |
3e1ad54f | 4088 | |
5d4f98a2 YZ |
4089 | /* block accounting for super block */ |
4090 | spin_lock(&info->delalloc_lock); | |
4091 | old_val = btrfs_super_bytes_used(&info->super_copy); | |
4092 | if (alloc) | |
4093 | old_val += num_bytes; | |
4094 | else | |
4095 | old_val -= num_bytes; | |
4096 | btrfs_set_super_bytes_used(&info->super_copy, old_val); | |
5d4f98a2 YZ |
4097 | spin_unlock(&info->delalloc_lock); |
4098 | ||
d397712b | 4099 | while (total) { |
db94535d | 4100 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 4101 | if (!cache) |
9078a3e1 | 4102 | return -1; |
b742bb82 YZ |
4103 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
4104 | BTRFS_BLOCK_GROUP_RAID1 | | |
4105 | BTRFS_BLOCK_GROUP_RAID10)) | |
4106 | factor = 2; | |
4107 | else | |
4108 | factor = 1; | |
9d66e233 JB |
4109 | /* |
4110 | * If this block group has free space cache written out, we | |
4111 | * need to make sure to load it if we are removing space. This | |
4112 | * is because we need the unpinning stage to actually add the | |
4113 | * space back to the block group, otherwise we will leak space. | |
4114 | */ | |
4115 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
b8399dee | 4116 | cache_block_group(cache, trans, NULL, 1); |
0af3d00b | 4117 | |
db94535d CM |
4118 | byte_in_group = bytenr - cache->key.objectid; |
4119 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 4120 | |
25179201 | 4121 | spin_lock(&cache->space_info->lock); |
c286ac48 | 4122 | spin_lock(&cache->lock); |
0af3d00b JB |
4123 | |
4124 | if (btrfs_super_cache_generation(&info->super_copy) != 0 && | |
4125 | cache->disk_cache_state < BTRFS_DC_CLEAR) | |
4126 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
4127 | ||
0f9dd46c | 4128 | cache->dirty = 1; |
9078a3e1 | 4129 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 4130 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 4131 | if (alloc) { |
db94535d | 4132 | old_val += num_bytes; |
11833d66 YZ |
4133 | btrfs_set_block_group_used(&cache->item, old_val); |
4134 | cache->reserved -= num_bytes; | |
11833d66 | 4135 | cache->space_info->bytes_reserved -= num_bytes; |
36e39c40 | 4136 | cache->space_info->reservation_progress++; |
b742bb82 YZ |
4137 | cache->space_info->bytes_used += num_bytes; |
4138 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 4139 | spin_unlock(&cache->lock); |
25179201 | 4140 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 4141 | } else { |
db94535d | 4142 | old_val -= num_bytes; |
c286ac48 | 4143 | btrfs_set_block_group_used(&cache->item, old_val); |
f0486c68 YZ |
4144 | cache->pinned += num_bytes; |
4145 | cache->space_info->bytes_pinned += num_bytes; | |
6324fbf3 | 4146 | cache->space_info->bytes_used -= num_bytes; |
b742bb82 | 4147 | cache->space_info->disk_used -= num_bytes * factor; |
c286ac48 | 4148 | spin_unlock(&cache->lock); |
25179201 | 4149 | spin_unlock(&cache->space_info->lock); |
1f3c79a2 | 4150 | |
f0486c68 YZ |
4151 | set_extent_dirty(info->pinned_extents, |
4152 | bytenr, bytenr + num_bytes - 1, | |
4153 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 4154 | } |
fa9c0d79 | 4155 | btrfs_put_block_group(cache); |
db94535d CM |
4156 | total -= num_bytes; |
4157 | bytenr += num_bytes; | |
9078a3e1 CM |
4158 | } |
4159 | return 0; | |
4160 | } | |
6324fbf3 | 4161 | |
a061fc8d CM |
4162 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
4163 | { | |
0f9dd46c | 4164 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 4165 | u64 bytenr; |
0f9dd46c JB |
4166 | |
4167 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); | |
4168 | if (!cache) | |
a061fc8d | 4169 | return 0; |
0f9dd46c | 4170 | |
d2fb3437 | 4171 | bytenr = cache->key.objectid; |
fa9c0d79 | 4172 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
4173 | |
4174 | return bytenr; | |
a061fc8d CM |
4175 | } |
4176 | ||
f0486c68 YZ |
4177 | static int pin_down_extent(struct btrfs_root *root, |
4178 | struct btrfs_block_group_cache *cache, | |
4179 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 4180 | { |
11833d66 YZ |
4181 | spin_lock(&cache->space_info->lock); |
4182 | spin_lock(&cache->lock); | |
4183 | cache->pinned += num_bytes; | |
4184 | cache->space_info->bytes_pinned += num_bytes; | |
4185 | if (reserved) { | |
4186 | cache->reserved -= num_bytes; | |
4187 | cache->space_info->bytes_reserved -= num_bytes; | |
36e39c40 | 4188 | cache->space_info->reservation_progress++; |
11833d66 YZ |
4189 | } |
4190 | spin_unlock(&cache->lock); | |
4191 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 4192 | |
f0486c68 YZ |
4193 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
4194 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
4195 | return 0; | |
4196 | } | |
68b38550 | 4197 | |
f0486c68 YZ |
4198 | /* |
4199 | * this function must be called within transaction | |
4200 | */ | |
4201 | int btrfs_pin_extent(struct btrfs_root *root, | |
4202 | u64 bytenr, u64 num_bytes, int reserved) | |
4203 | { | |
4204 | struct btrfs_block_group_cache *cache; | |
68b38550 | 4205 | |
f0486c68 YZ |
4206 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
4207 | BUG_ON(!cache); | |
4208 | ||
4209 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
4210 | ||
4211 | btrfs_put_block_group(cache); | |
11833d66 YZ |
4212 | return 0; |
4213 | } | |
4214 | ||
f0486c68 YZ |
4215 | /* |
4216 | * update size of reserved extents. this function may return -EAGAIN | |
4217 | * if 'reserve' is true or 'sinfo' is false. | |
4218 | */ | |
b4d00d56 LD |
4219 | int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
4220 | u64 num_bytes, int reserve, int sinfo) | |
11833d66 | 4221 | { |
f0486c68 YZ |
4222 | int ret = 0; |
4223 | if (sinfo) { | |
4224 | struct btrfs_space_info *space_info = cache->space_info; | |
4225 | spin_lock(&space_info->lock); | |
4226 | spin_lock(&cache->lock); | |
4227 | if (reserve) { | |
4228 | if (cache->ro) { | |
4229 | ret = -EAGAIN; | |
4230 | } else { | |
4231 | cache->reserved += num_bytes; | |
4232 | space_info->bytes_reserved += num_bytes; | |
4233 | } | |
4234 | } else { | |
4235 | if (cache->ro) | |
4236 | space_info->bytes_readonly += num_bytes; | |
4237 | cache->reserved -= num_bytes; | |
4238 | space_info->bytes_reserved -= num_bytes; | |
36e39c40 | 4239 | space_info->reservation_progress++; |
f0486c68 YZ |
4240 | } |
4241 | spin_unlock(&cache->lock); | |
4242 | spin_unlock(&space_info->lock); | |
11833d66 | 4243 | } else { |
f0486c68 YZ |
4244 | spin_lock(&cache->lock); |
4245 | if (cache->ro) { | |
4246 | ret = -EAGAIN; | |
4247 | } else { | |
4248 | if (reserve) | |
4249 | cache->reserved += num_bytes; | |
4250 | else | |
4251 | cache->reserved -= num_bytes; | |
4252 | } | |
4253 | spin_unlock(&cache->lock); | |
324ae4df | 4254 | } |
f0486c68 | 4255 | return ret; |
324ae4df | 4256 | } |
9078a3e1 | 4257 | |
11833d66 YZ |
4258 | int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
4259 | struct btrfs_root *root) | |
e8569813 | 4260 | { |
e8569813 | 4261 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
4262 | struct btrfs_caching_control *next; |
4263 | struct btrfs_caching_control *caching_ctl; | |
4264 | struct btrfs_block_group_cache *cache; | |
e8569813 | 4265 | |
11833d66 | 4266 | down_write(&fs_info->extent_commit_sem); |
25179201 | 4267 | |
11833d66 YZ |
4268 | list_for_each_entry_safe(caching_ctl, next, |
4269 | &fs_info->caching_block_groups, list) { | |
4270 | cache = caching_ctl->block_group; | |
4271 | if (block_group_cache_done(cache)) { | |
4272 | cache->last_byte_to_unpin = (u64)-1; | |
4273 | list_del_init(&caching_ctl->list); | |
4274 | put_caching_control(caching_ctl); | |
e8569813 | 4275 | } else { |
11833d66 | 4276 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 4277 | } |
e8569813 | 4278 | } |
11833d66 YZ |
4279 | |
4280 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
4281 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
4282 | else | |
4283 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
4284 | ||
4285 | up_write(&fs_info->extent_commit_sem); | |
8929ecfa YZ |
4286 | |
4287 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
4288 | return 0; |
4289 | } | |
4290 | ||
11833d66 | 4291 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
ccd467d6 | 4292 | { |
11833d66 YZ |
4293 | struct btrfs_fs_info *fs_info = root->fs_info; |
4294 | struct btrfs_block_group_cache *cache = NULL; | |
4295 | u64 len; | |
ccd467d6 | 4296 | |
11833d66 YZ |
4297 | while (start <= end) { |
4298 | if (!cache || | |
4299 | start >= cache->key.objectid + cache->key.offset) { | |
4300 | if (cache) | |
4301 | btrfs_put_block_group(cache); | |
4302 | cache = btrfs_lookup_block_group(fs_info, start); | |
4303 | BUG_ON(!cache); | |
4304 | } | |
4305 | ||
4306 | len = cache->key.objectid + cache->key.offset - start; | |
4307 | len = min(len, end + 1 - start); | |
4308 | ||
4309 | if (start < cache->last_byte_to_unpin) { | |
4310 | len = min(len, cache->last_byte_to_unpin - start); | |
4311 | btrfs_add_free_space(cache, start, len); | |
4312 | } | |
4313 | ||
f0486c68 YZ |
4314 | start += len; |
4315 | ||
11833d66 YZ |
4316 | spin_lock(&cache->space_info->lock); |
4317 | spin_lock(&cache->lock); | |
4318 | cache->pinned -= len; | |
4319 | cache->space_info->bytes_pinned -= len; | |
f0486c68 YZ |
4320 | if (cache->ro) { |
4321 | cache->space_info->bytes_readonly += len; | |
4322 | } else if (cache->reserved_pinned > 0) { | |
4323 | len = min(len, cache->reserved_pinned); | |
4324 | cache->reserved_pinned -= len; | |
4325 | cache->space_info->bytes_reserved += len; | |
4326 | } | |
11833d66 YZ |
4327 | spin_unlock(&cache->lock); |
4328 | spin_unlock(&cache->space_info->lock); | |
ccd467d6 | 4329 | } |
11833d66 YZ |
4330 | |
4331 | if (cache) | |
4332 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
4333 | return 0; |
4334 | } | |
4335 | ||
4336 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 4337 | struct btrfs_root *root) |
a28ec197 | 4338 | { |
11833d66 YZ |
4339 | struct btrfs_fs_info *fs_info = root->fs_info; |
4340 | struct extent_io_tree *unpin; | |
f0486c68 YZ |
4341 | struct btrfs_block_rsv *block_rsv; |
4342 | struct btrfs_block_rsv *next_rsv; | |
1a5bc167 CM |
4343 | u64 start; |
4344 | u64 end; | |
f0486c68 | 4345 | int idx; |
a28ec197 | 4346 | int ret; |
a28ec197 | 4347 | |
11833d66 YZ |
4348 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
4349 | unpin = &fs_info->freed_extents[1]; | |
4350 | else | |
4351 | unpin = &fs_info->freed_extents[0]; | |
4352 | ||
d397712b | 4353 | while (1) { |
1a5bc167 CM |
4354 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
4355 | EXTENT_DIRTY); | |
4356 | if (ret) | |
a28ec197 | 4357 | break; |
1f3c79a2 | 4358 | |
5378e607 LD |
4359 | if (btrfs_test_opt(root, DISCARD)) |
4360 | ret = btrfs_discard_extent(root, start, | |
4361 | end + 1 - start, NULL); | |
1f3c79a2 | 4362 | |
1a5bc167 | 4363 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
11833d66 | 4364 | unpin_extent_range(root, start, end); |
b9473439 | 4365 | cond_resched(); |
a28ec197 | 4366 | } |
817d52f8 | 4367 | |
f0486c68 YZ |
4368 | mutex_lock(&fs_info->durable_block_rsv_mutex); |
4369 | list_for_each_entry_safe(block_rsv, next_rsv, | |
4370 | &fs_info->durable_block_rsv_list, list) { | |
444528b3 | 4371 | |
f0486c68 YZ |
4372 | idx = trans->transid & 0x1; |
4373 | if (block_rsv->freed[idx] > 0) { | |
4374 | block_rsv_add_bytes(block_rsv, | |
4375 | block_rsv->freed[idx], 0); | |
4376 | block_rsv->freed[idx] = 0; | |
4377 | } | |
4378 | if (atomic_read(&block_rsv->usage) == 0) { | |
4379 | btrfs_block_rsv_release(root, block_rsv, (u64)-1); | |
31840ae1 | 4380 | |
f0486c68 YZ |
4381 | if (block_rsv->freed[0] == 0 && |
4382 | block_rsv->freed[1] == 0) { | |
4383 | list_del_init(&block_rsv->list); | |
4384 | kfree(block_rsv); | |
4385 | } | |
4386 | } else { | |
4387 | btrfs_block_rsv_release(root, block_rsv, 0); | |
8ef97622 | 4388 | } |
f4b9aa8d | 4389 | } |
f0486c68 | 4390 | mutex_unlock(&fs_info->durable_block_rsv_mutex); |
31840ae1 | 4391 | |
e20d96d6 CM |
4392 | return 0; |
4393 | } | |
4394 | ||
5d4f98a2 YZ |
4395 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
4396 | struct btrfs_root *root, | |
4397 | u64 bytenr, u64 num_bytes, u64 parent, | |
4398 | u64 root_objectid, u64 owner_objectid, | |
4399 | u64 owner_offset, int refs_to_drop, | |
4400 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 4401 | { |
e2fa7227 | 4402 | struct btrfs_key key; |
5d4f98a2 | 4403 | struct btrfs_path *path; |
1261ec42 CM |
4404 | struct btrfs_fs_info *info = root->fs_info; |
4405 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 4406 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
4407 | struct btrfs_extent_item *ei; |
4408 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 4409 | int ret; |
5d4f98a2 | 4410 | int is_data; |
952fccac CM |
4411 | int extent_slot = 0; |
4412 | int found_extent = 0; | |
4413 | int num_to_del = 1; | |
5d4f98a2 YZ |
4414 | u32 item_size; |
4415 | u64 refs; | |
037e6390 | 4416 | |
5caf2a00 | 4417 | path = btrfs_alloc_path(); |
54aa1f4d CM |
4418 | if (!path) |
4419 | return -ENOMEM; | |
5f26f772 | 4420 | |
3c12ac72 | 4421 | path->reada = 1; |
b9473439 | 4422 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4423 | |
4424 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
4425 | BUG_ON(!is_data && refs_to_drop != 1); | |
4426 | ||
4427 | ret = lookup_extent_backref(trans, extent_root, path, &iref, | |
4428 | bytenr, num_bytes, parent, | |
4429 | root_objectid, owner_objectid, | |
4430 | owner_offset); | |
7bb86316 | 4431 | if (ret == 0) { |
952fccac | 4432 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
4433 | while (extent_slot >= 0) { |
4434 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 4435 | extent_slot); |
5d4f98a2 | 4436 | if (key.objectid != bytenr) |
952fccac | 4437 | break; |
5d4f98a2 YZ |
4438 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
4439 | key.offset == num_bytes) { | |
952fccac CM |
4440 | found_extent = 1; |
4441 | break; | |
4442 | } | |
4443 | if (path->slots[0] - extent_slot > 5) | |
4444 | break; | |
5d4f98a2 | 4445 | extent_slot--; |
952fccac | 4446 | } |
5d4f98a2 YZ |
4447 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
4448 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
4449 | if (found_extent && item_size < sizeof(*ei)) | |
4450 | found_extent = 0; | |
4451 | #endif | |
31840ae1 | 4452 | if (!found_extent) { |
5d4f98a2 | 4453 | BUG_ON(iref); |
56bec294 | 4454 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
4455 | NULL, refs_to_drop, |
4456 | is_data); | |
31840ae1 | 4457 | BUG_ON(ret); |
b3b4aa74 | 4458 | btrfs_release_path(path); |
b9473439 | 4459 | path->leave_spinning = 1; |
5d4f98a2 YZ |
4460 | |
4461 | key.objectid = bytenr; | |
4462 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
4463 | key.offset = num_bytes; | |
4464 | ||
31840ae1 ZY |
4465 | ret = btrfs_search_slot(trans, extent_root, |
4466 | &key, path, -1, 1); | |
f3465ca4 JB |
4467 | if (ret) { |
4468 | printk(KERN_ERR "umm, got %d back from search" | |
d397712b CM |
4469 | ", was looking for %llu\n", ret, |
4470 | (unsigned long long)bytenr); | |
b783e62d JB |
4471 | if (ret > 0) |
4472 | btrfs_print_leaf(extent_root, | |
4473 | path->nodes[0]); | |
f3465ca4 | 4474 | } |
31840ae1 ZY |
4475 | BUG_ON(ret); |
4476 | extent_slot = path->slots[0]; | |
4477 | } | |
7bb86316 CM |
4478 | } else { |
4479 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
4480 | WARN_ON(1); | |
d397712b | 4481 | printk(KERN_ERR "btrfs unable to find ref byte nr %llu " |
5d4f98a2 | 4482 | "parent %llu root %llu owner %llu offset %llu\n", |
d397712b | 4483 | (unsigned long long)bytenr, |
56bec294 | 4484 | (unsigned long long)parent, |
d397712b | 4485 | (unsigned long long)root_objectid, |
5d4f98a2 YZ |
4486 | (unsigned long long)owner_objectid, |
4487 | (unsigned long long)owner_offset); | |
7bb86316 | 4488 | } |
5f39d397 CM |
4489 | |
4490 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
4491 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
4492 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
4493 | if (item_size < sizeof(*ei)) { | |
4494 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
4495 | ret = convert_extent_item_v0(trans, extent_root, path, | |
4496 | owner_objectid, 0); | |
4497 | BUG_ON(ret < 0); | |
4498 | ||
b3b4aa74 | 4499 | btrfs_release_path(path); |
5d4f98a2 YZ |
4500 | path->leave_spinning = 1; |
4501 | ||
4502 | key.objectid = bytenr; | |
4503 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
4504 | key.offset = num_bytes; | |
4505 | ||
4506 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
4507 | -1, 1); | |
4508 | if (ret) { | |
4509 | printk(KERN_ERR "umm, got %d back from search" | |
4510 | ", was looking for %llu\n", ret, | |
4511 | (unsigned long long)bytenr); | |
4512 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
4513 | } | |
4514 | BUG_ON(ret); | |
4515 | extent_slot = path->slots[0]; | |
4516 | leaf = path->nodes[0]; | |
4517 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
4518 | } | |
4519 | #endif | |
4520 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 4521 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 4522 | struct btrfs_extent_item); |
5d4f98a2 YZ |
4523 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { |
4524 | struct btrfs_tree_block_info *bi; | |
4525 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
4526 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
4527 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
4528 | } | |
56bec294 | 4529 | |
5d4f98a2 | 4530 | refs = btrfs_extent_refs(leaf, ei); |
56bec294 CM |
4531 | BUG_ON(refs < refs_to_drop); |
4532 | refs -= refs_to_drop; | |
5f39d397 | 4533 | |
5d4f98a2 YZ |
4534 | if (refs > 0) { |
4535 | if (extent_op) | |
4536 | __run_delayed_extent_op(extent_op, leaf, ei); | |
4537 | /* | |
4538 | * In the case of inline back ref, reference count will | |
4539 | * be updated by remove_extent_backref | |
952fccac | 4540 | */ |
5d4f98a2 YZ |
4541 | if (iref) { |
4542 | BUG_ON(!found_extent); | |
4543 | } else { | |
4544 | btrfs_set_extent_refs(leaf, ei, refs); | |
4545 | btrfs_mark_buffer_dirty(leaf); | |
4546 | } | |
4547 | if (found_extent) { | |
4548 | ret = remove_extent_backref(trans, extent_root, path, | |
4549 | iref, refs_to_drop, | |
4550 | is_data); | |
952fccac CM |
4551 | BUG_ON(ret); |
4552 | } | |
5d4f98a2 | 4553 | } else { |
5d4f98a2 YZ |
4554 | if (found_extent) { |
4555 | BUG_ON(is_data && refs_to_drop != | |
4556 | extent_data_ref_count(root, path, iref)); | |
4557 | if (iref) { | |
4558 | BUG_ON(path->slots[0] != extent_slot); | |
4559 | } else { | |
4560 | BUG_ON(path->slots[0] != extent_slot + 1); | |
4561 | path->slots[0] = extent_slot; | |
4562 | num_to_del = 2; | |
4563 | } | |
78fae27e | 4564 | } |
b9473439 | 4565 | |
952fccac CM |
4566 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
4567 | num_to_del); | |
31840ae1 | 4568 | BUG_ON(ret); |
b3b4aa74 | 4569 | btrfs_release_path(path); |
21af804c | 4570 | |
5d4f98a2 | 4571 | if (is_data) { |
459931ec CM |
4572 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
4573 | BUG_ON(ret); | |
d57e62b8 CM |
4574 | } else { |
4575 | invalidate_mapping_pages(info->btree_inode->i_mapping, | |
4576 | bytenr >> PAGE_CACHE_SHIFT, | |
4577 | (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT); | |
459931ec CM |
4578 | } |
4579 | ||
f0486c68 | 4580 | ret = update_block_group(trans, root, bytenr, num_bytes, 0); |
dcbdd4dc | 4581 | BUG_ON(ret); |
a28ec197 | 4582 | } |
5caf2a00 | 4583 | btrfs_free_path(path); |
a28ec197 CM |
4584 | return ret; |
4585 | } | |
4586 | ||
1887be66 | 4587 | /* |
f0486c68 | 4588 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
4589 | * delayed ref for that extent as well. This searches the delayed ref tree for |
4590 | * a given extent, and if there are no other delayed refs to be processed, it | |
4591 | * removes it from the tree. | |
4592 | */ | |
4593 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
4594 | struct btrfs_root *root, u64 bytenr) | |
4595 | { | |
4596 | struct btrfs_delayed_ref_head *head; | |
4597 | struct btrfs_delayed_ref_root *delayed_refs; | |
4598 | struct btrfs_delayed_ref_node *ref; | |
4599 | struct rb_node *node; | |
f0486c68 | 4600 | int ret = 0; |
1887be66 CM |
4601 | |
4602 | delayed_refs = &trans->transaction->delayed_refs; | |
4603 | spin_lock(&delayed_refs->lock); | |
4604 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
4605 | if (!head) | |
4606 | goto out; | |
4607 | ||
4608 | node = rb_prev(&head->node.rb_node); | |
4609 | if (!node) | |
4610 | goto out; | |
4611 | ||
4612 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
4613 | ||
4614 | /* there are still entries for this ref, we can't drop it */ | |
4615 | if (ref->bytenr == bytenr) | |
4616 | goto out; | |
4617 | ||
5d4f98a2 YZ |
4618 | if (head->extent_op) { |
4619 | if (!head->must_insert_reserved) | |
4620 | goto out; | |
4621 | kfree(head->extent_op); | |
4622 | head->extent_op = NULL; | |
4623 | } | |
4624 | ||
1887be66 CM |
4625 | /* |
4626 | * waiting for the lock here would deadlock. If someone else has it | |
4627 | * locked they are already in the process of dropping it anyway | |
4628 | */ | |
4629 | if (!mutex_trylock(&head->mutex)) | |
4630 | goto out; | |
4631 | ||
4632 | /* | |
4633 | * at this point we have a head with no other entries. Go | |
4634 | * ahead and process it. | |
4635 | */ | |
4636 | head->node.in_tree = 0; | |
4637 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 4638 | |
1887be66 CM |
4639 | delayed_refs->num_entries--; |
4640 | ||
4641 | /* | |
4642 | * we don't take a ref on the node because we're removing it from the | |
4643 | * tree, so we just steal the ref the tree was holding. | |
4644 | */ | |
c3e69d58 CM |
4645 | delayed_refs->num_heads--; |
4646 | if (list_empty(&head->cluster)) | |
4647 | delayed_refs->num_heads_ready--; | |
4648 | ||
4649 | list_del_init(&head->cluster); | |
1887be66 CM |
4650 | spin_unlock(&delayed_refs->lock); |
4651 | ||
f0486c68 YZ |
4652 | BUG_ON(head->extent_op); |
4653 | if (head->must_insert_reserved) | |
4654 | ret = 1; | |
4655 | ||
4656 | mutex_unlock(&head->mutex); | |
1887be66 | 4657 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 4658 | return ret; |
1887be66 CM |
4659 | out: |
4660 | spin_unlock(&delayed_refs->lock); | |
4661 | return 0; | |
4662 | } | |
4663 | ||
f0486c68 YZ |
4664 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
4665 | struct btrfs_root *root, | |
4666 | struct extent_buffer *buf, | |
4667 | u64 parent, int last_ref) | |
4668 | { | |
4669 | struct btrfs_block_rsv *block_rsv; | |
4670 | struct btrfs_block_group_cache *cache = NULL; | |
4671 | int ret; | |
4672 | ||
4673 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
4674 | ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len, | |
4675 | parent, root->root_key.objectid, | |
4676 | btrfs_header_level(buf), | |
4677 | BTRFS_DROP_DELAYED_REF, NULL); | |
4678 | BUG_ON(ret); | |
4679 | } | |
4680 | ||
4681 | if (!last_ref) | |
4682 | return; | |
4683 | ||
4684 | block_rsv = get_block_rsv(trans, root); | |
4685 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); | |
3bf84a5a YZ |
4686 | if (block_rsv->space_info != cache->space_info) |
4687 | goto out; | |
f0486c68 YZ |
4688 | |
4689 | if (btrfs_header_generation(buf) == trans->transid) { | |
4690 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
4691 | ret = check_ref_cleanup(trans, root, buf->start); | |
4692 | if (!ret) | |
4693 | goto pin; | |
4694 | } | |
4695 | ||
4696 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
4697 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
4698 | goto pin; | |
4699 | } | |
4700 | ||
4701 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
4702 | ||
4703 | btrfs_add_free_space(cache, buf->start, buf->len); | |
b4d00d56 | 4704 | ret = btrfs_update_reserved_bytes(cache, buf->len, 0, 0); |
f0486c68 YZ |
4705 | if (ret == -EAGAIN) { |
4706 | /* block group became read-only */ | |
b4d00d56 | 4707 | btrfs_update_reserved_bytes(cache, buf->len, 0, 1); |
f0486c68 YZ |
4708 | goto out; |
4709 | } | |
4710 | ||
4711 | ret = 1; | |
4712 | spin_lock(&block_rsv->lock); | |
4713 | if (block_rsv->reserved < block_rsv->size) { | |
4714 | block_rsv->reserved += buf->len; | |
4715 | ret = 0; | |
4716 | } | |
4717 | spin_unlock(&block_rsv->lock); | |
4718 | ||
4719 | if (ret) { | |
4720 | spin_lock(&cache->space_info->lock); | |
4721 | cache->space_info->bytes_reserved -= buf->len; | |
36e39c40 | 4722 | cache->space_info->reservation_progress++; |
f0486c68 YZ |
4723 | spin_unlock(&cache->space_info->lock); |
4724 | } | |
4725 | goto out; | |
4726 | } | |
4727 | pin: | |
4728 | if (block_rsv->durable && !cache->ro) { | |
4729 | ret = 0; | |
4730 | spin_lock(&cache->lock); | |
4731 | if (!cache->ro) { | |
4732 | cache->reserved_pinned += buf->len; | |
4733 | ret = 1; | |
4734 | } | |
4735 | spin_unlock(&cache->lock); | |
4736 | ||
4737 | if (ret) { | |
4738 | spin_lock(&block_rsv->lock); | |
4739 | block_rsv->freed[trans->transid & 0x1] += buf->len; | |
4740 | spin_unlock(&block_rsv->lock); | |
4741 | } | |
4742 | } | |
4743 | out: | |
a826d6dc JB |
4744 | /* |
4745 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
4746 | * anymore. | |
4747 | */ | |
4748 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
4749 | btrfs_put_block_group(cache); |
4750 | } | |
4751 | ||
925baedd | 4752 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
31840ae1 ZY |
4753 | struct btrfs_root *root, |
4754 | u64 bytenr, u64 num_bytes, u64 parent, | |
5d4f98a2 | 4755 | u64 root_objectid, u64 owner, u64 offset) |
925baedd CM |
4756 | { |
4757 | int ret; | |
4758 | ||
56bec294 CM |
4759 | /* |
4760 | * tree log blocks never actually go into the extent allocation | |
4761 | * tree, just update pinning info and exit early. | |
56bec294 | 4762 | */ |
5d4f98a2 YZ |
4763 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
4764 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 4765 | /* unlocks the pinned mutex */ |
11833d66 | 4766 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 4767 | ret = 0; |
5d4f98a2 YZ |
4768 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
4769 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes, | |
4770 | parent, root_objectid, (int)owner, | |
4771 | BTRFS_DROP_DELAYED_REF, NULL); | |
1887be66 | 4772 | BUG_ON(ret); |
5d4f98a2 YZ |
4773 | } else { |
4774 | ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes, | |
4775 | parent, root_objectid, owner, | |
4776 | offset, BTRFS_DROP_DELAYED_REF, NULL); | |
4777 | BUG_ON(ret); | |
56bec294 | 4778 | } |
925baedd CM |
4779 | return ret; |
4780 | } | |
4781 | ||
87ee04eb CM |
4782 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
4783 | { | |
4784 | u64 mask = ((u64)root->stripesize - 1); | |
4785 | u64 ret = (val + mask) & ~mask; | |
4786 | return ret; | |
4787 | } | |
4788 | ||
817d52f8 JB |
4789 | /* |
4790 | * when we wait for progress in the block group caching, its because | |
4791 | * our allocation attempt failed at least once. So, we must sleep | |
4792 | * and let some progress happen before we try again. | |
4793 | * | |
4794 | * This function will sleep at least once waiting for new free space to | |
4795 | * show up, and then it will check the block group free space numbers | |
4796 | * for our min num_bytes. Another option is to have it go ahead | |
4797 | * and look in the rbtree for a free extent of a given size, but this | |
4798 | * is a good start. | |
4799 | */ | |
4800 | static noinline int | |
4801 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
4802 | u64 num_bytes) | |
4803 | { | |
11833d66 | 4804 | struct btrfs_caching_control *caching_ctl; |
817d52f8 JB |
4805 | DEFINE_WAIT(wait); |
4806 | ||
11833d66 YZ |
4807 | caching_ctl = get_caching_control(cache); |
4808 | if (!caching_ctl) | |
817d52f8 | 4809 | return 0; |
817d52f8 | 4810 | |
11833d66 | 4811 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 4812 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
4813 | |
4814 | put_caching_control(caching_ctl); | |
4815 | return 0; | |
4816 | } | |
4817 | ||
4818 | static noinline int | |
4819 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
4820 | { | |
4821 | struct btrfs_caching_control *caching_ctl; | |
4822 | DEFINE_WAIT(wait); | |
4823 | ||
4824 | caching_ctl = get_caching_control(cache); | |
4825 | if (!caching_ctl) | |
4826 | return 0; | |
4827 | ||
4828 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
4829 | ||
4830 | put_caching_control(caching_ctl); | |
817d52f8 JB |
4831 | return 0; |
4832 | } | |
4833 | ||
b742bb82 YZ |
4834 | static int get_block_group_index(struct btrfs_block_group_cache *cache) |
4835 | { | |
4836 | int index; | |
4837 | if (cache->flags & BTRFS_BLOCK_GROUP_RAID10) | |
4838 | index = 0; | |
4839 | else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1) | |
4840 | index = 1; | |
4841 | else if (cache->flags & BTRFS_BLOCK_GROUP_DUP) | |
4842 | index = 2; | |
4843 | else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0) | |
4844 | index = 3; | |
4845 | else | |
4846 | index = 4; | |
4847 | return index; | |
4848 | } | |
4849 | ||
817d52f8 | 4850 | enum btrfs_loop_type { |
ccf0e725 | 4851 | LOOP_FIND_IDEAL = 0, |
817d52f8 JB |
4852 | LOOP_CACHING_NOWAIT = 1, |
4853 | LOOP_CACHING_WAIT = 2, | |
4854 | LOOP_ALLOC_CHUNK = 3, | |
4855 | LOOP_NO_EMPTY_SIZE = 4, | |
4856 | }; | |
4857 | ||
fec577fb CM |
4858 | /* |
4859 | * walks the btree of allocated extents and find a hole of a given size. | |
4860 | * The key ins is changed to record the hole: | |
4861 | * ins->objectid == block start | |
62e2749e | 4862 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
4863 | * ins->offset == number of blocks |
4864 | * Any available blocks before search_start are skipped. | |
4865 | */ | |
d397712b | 4866 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
4867 | struct btrfs_root *orig_root, |
4868 | u64 num_bytes, u64 empty_size, | |
4869 | u64 search_start, u64 search_end, | |
4870 | u64 hint_byte, struct btrfs_key *ins, | |
e0f54067 | 4871 | u64 data) |
fec577fb | 4872 | { |
80eb234a | 4873 | int ret = 0; |
d397712b | 4874 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 4875 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 4876 | struct btrfs_block_group_cache *block_group = NULL; |
239b14b3 | 4877 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 4878 | int allowed_chunk_alloc = 0; |
ccf0e725 | 4879 | int done_chunk_alloc = 0; |
80eb234a | 4880 | struct btrfs_space_info *space_info; |
fa9c0d79 CM |
4881 | int last_ptr_loop = 0; |
4882 | int loop = 0; | |
f0486c68 | 4883 | int index = 0; |
817d52f8 | 4884 | bool found_uncached_bg = false; |
0a24325e | 4885 | bool failed_cluster_refill = false; |
1cdda9b8 | 4886 | bool failed_alloc = false; |
67377734 | 4887 | bool use_cluster = true; |
ccf0e725 JB |
4888 | u64 ideal_cache_percent = 0; |
4889 | u64 ideal_cache_offset = 0; | |
fec577fb | 4890 | |
db94535d | 4891 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 4892 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
4893 | ins->objectid = 0; |
4894 | ins->offset = 0; | |
b1a4d965 | 4895 | |
2552d17e | 4896 | space_info = __find_space_info(root->fs_info, data); |
1b1d1f66 | 4897 | if (!space_info) { |
e0f54067 | 4898 | printk(KERN_ERR "No space info for %llu\n", data); |
1b1d1f66 JB |
4899 | return -ENOSPC; |
4900 | } | |
2552d17e | 4901 | |
67377734 JB |
4902 | /* |
4903 | * If the space info is for both data and metadata it means we have a | |
4904 | * small filesystem and we can't use the clustering stuff. | |
4905 | */ | |
4906 | if (btrfs_mixed_space_info(space_info)) | |
4907 | use_cluster = false; | |
4908 | ||
0ef3e66b CM |
4909 | if (orig_root->ref_cows || empty_size) |
4910 | allowed_chunk_alloc = 1; | |
4911 | ||
67377734 | 4912 | if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) { |
fa9c0d79 | 4913 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
4914 | if (!btrfs_test_opt(root, SSD)) |
4915 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
4916 | } |
4917 | ||
67377734 JB |
4918 | if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster && |
4919 | btrfs_test_opt(root, SSD)) { | |
fa9c0d79 CM |
4920 | last_ptr = &root->fs_info->data_alloc_cluster; |
4921 | } | |
0f9dd46c | 4922 | |
239b14b3 | 4923 | if (last_ptr) { |
fa9c0d79 CM |
4924 | spin_lock(&last_ptr->lock); |
4925 | if (last_ptr->block_group) | |
4926 | hint_byte = last_ptr->window_start; | |
4927 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 4928 | } |
fa9c0d79 | 4929 | |
a061fc8d | 4930 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 4931 | search_start = max(search_start, hint_byte); |
0b86a832 | 4932 | |
817d52f8 | 4933 | if (!last_ptr) |
fa9c0d79 | 4934 | empty_cluster = 0; |
fa9c0d79 | 4935 | |
2552d17e | 4936 | if (search_start == hint_byte) { |
ccf0e725 | 4937 | ideal_cache: |
2552d17e JB |
4938 | block_group = btrfs_lookup_block_group(root->fs_info, |
4939 | search_start); | |
817d52f8 JB |
4940 | /* |
4941 | * we don't want to use the block group if it doesn't match our | |
4942 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
4943 | * |
4944 | * However if we are re-searching with an ideal block group | |
4945 | * picked out then we don't care that the block group is cached. | |
817d52f8 JB |
4946 | */ |
4947 | if (block_group && block_group_bits(block_group, data) && | |
ccf0e725 JB |
4948 | (block_group->cached != BTRFS_CACHE_NO || |
4949 | search_start == ideal_cache_offset)) { | |
2552d17e | 4950 | down_read(&space_info->groups_sem); |
44fb5511 CM |
4951 | if (list_empty(&block_group->list) || |
4952 | block_group->ro) { | |
4953 | /* | |
4954 | * someone is removing this block group, | |
4955 | * we can't jump into the have_block_group | |
4956 | * target because our list pointers are not | |
4957 | * valid | |
4958 | */ | |
4959 | btrfs_put_block_group(block_group); | |
4960 | up_read(&space_info->groups_sem); | |
ccf0e725 | 4961 | } else { |
b742bb82 | 4962 | index = get_block_group_index(block_group); |
44fb5511 | 4963 | goto have_block_group; |
ccf0e725 | 4964 | } |
2552d17e | 4965 | } else if (block_group) { |
fa9c0d79 | 4966 | btrfs_put_block_group(block_group); |
2552d17e | 4967 | } |
42e70e7a | 4968 | } |
2552d17e | 4969 | search: |
80eb234a | 4970 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
4971 | list_for_each_entry(block_group, &space_info->block_groups[index], |
4972 | list) { | |
6226cb0a | 4973 | u64 offset; |
817d52f8 | 4974 | int cached; |
8a1413a2 | 4975 | |
11dfe35a | 4976 | btrfs_get_block_group(block_group); |
2552d17e | 4977 | search_start = block_group->key.objectid; |
42e70e7a | 4978 | |
83a50de9 CM |
4979 | /* |
4980 | * this can happen if we end up cycling through all the | |
4981 | * raid types, but we want to make sure we only allocate | |
4982 | * for the proper type. | |
4983 | */ | |
4984 | if (!block_group_bits(block_group, data)) { | |
4985 | u64 extra = BTRFS_BLOCK_GROUP_DUP | | |
4986 | BTRFS_BLOCK_GROUP_RAID1 | | |
4987 | BTRFS_BLOCK_GROUP_RAID10; | |
4988 | ||
4989 | /* | |
4990 | * if they asked for extra copies and this block group | |
4991 | * doesn't provide them, bail. This does allow us to | |
4992 | * fill raid0 from raid1. | |
4993 | */ | |
4994 | if ((data & extra) && !(block_group->flags & extra)) | |
4995 | goto loop; | |
4996 | } | |
4997 | ||
2552d17e | 4998 | have_block_group: |
817d52f8 | 4999 | if (unlikely(block_group->cached == BTRFS_CACHE_NO)) { |
ccf0e725 JB |
5000 | u64 free_percent; |
5001 | ||
b8399dee JB |
5002 | ret = cache_block_group(block_group, trans, |
5003 | orig_root, 1); | |
9d66e233 JB |
5004 | if (block_group->cached == BTRFS_CACHE_FINISHED) |
5005 | goto have_block_group; | |
5006 | ||
ccf0e725 JB |
5007 | free_percent = btrfs_block_group_used(&block_group->item); |
5008 | free_percent *= 100; | |
5009 | free_percent = div64_u64(free_percent, | |
5010 | block_group->key.offset); | |
5011 | free_percent = 100 - free_percent; | |
5012 | if (free_percent > ideal_cache_percent && | |
5013 | likely(!block_group->ro)) { | |
5014 | ideal_cache_offset = block_group->key.objectid; | |
5015 | ideal_cache_percent = free_percent; | |
5016 | } | |
5017 | ||
817d52f8 | 5018 | /* |
bab39bf9 JB |
5019 | * The caching workers are limited to 2 threads, so we |
5020 | * can queue as much work as we care to. | |
817d52f8 | 5021 | */ |
bab39bf9 | 5022 | if (loop > LOOP_FIND_IDEAL) { |
b8399dee JB |
5023 | ret = cache_block_group(block_group, trans, |
5024 | orig_root, 0); | |
817d52f8 | 5025 | BUG_ON(ret); |
2552d17e | 5026 | } |
817d52f8 JB |
5027 | found_uncached_bg = true; |
5028 | ||
ccf0e725 JB |
5029 | /* |
5030 | * If loop is set for cached only, try the next block | |
5031 | * group. | |
5032 | */ | |
5033 | if (loop == LOOP_FIND_IDEAL) | |
817d52f8 JB |
5034 | goto loop; |
5035 | } | |
5036 | ||
ccf0e725 JB |
5037 | cached = block_group_cache_done(block_group); |
5038 | if (unlikely(!cached)) | |
5039 | found_uncached_bg = true; | |
5040 | ||
ea6a478e | 5041 | if (unlikely(block_group->ro)) |
2552d17e | 5042 | goto loop; |
0f9dd46c | 5043 | |
ff5714cc | 5044 | spin_lock(&block_group->free_space_ctl->tree_lock); |
cca1c81f | 5045 | if (cached && |
ff5714cc CM |
5046 | block_group->free_space_ctl->free_space < |
5047 | num_bytes + empty_size) { | |
5048 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
cca1c81f JB |
5049 | goto loop; |
5050 | } | |
ff5714cc | 5051 | spin_unlock(&block_group->free_space_ctl->tree_lock); |
cca1c81f | 5052 | |
0a24325e JB |
5053 | /* |
5054 | * Ok we want to try and use the cluster allocator, so lets look | |
5055 | * there, unless we are on LOOP_NO_EMPTY_SIZE, since we will | |
5056 | * have tried the cluster allocator plenty of times at this | |
5057 | * point and not have found anything, so we are likely way too | |
5058 | * fragmented for the clustering stuff to find anything, so lets | |
5059 | * just skip it and let the allocator find whatever block it can | |
5060 | * find | |
5061 | */ | |
5062 | if (last_ptr && loop < LOOP_NO_EMPTY_SIZE) { | |
fa9c0d79 CM |
5063 | /* |
5064 | * the refill lock keeps out other | |
5065 | * people trying to start a new cluster | |
5066 | */ | |
5067 | spin_lock(&last_ptr->refill_lock); | |
44fb5511 CM |
5068 | if (last_ptr->block_group && |
5069 | (last_ptr->block_group->ro || | |
5070 | !block_group_bits(last_ptr->block_group, data))) { | |
5071 | offset = 0; | |
5072 | goto refill_cluster; | |
5073 | } | |
5074 | ||
fa9c0d79 CM |
5075 | offset = btrfs_alloc_from_cluster(block_group, last_ptr, |
5076 | num_bytes, search_start); | |
5077 | if (offset) { | |
5078 | /* we have a block, we're done */ | |
5079 | spin_unlock(&last_ptr->refill_lock); | |
5080 | goto checks; | |
5081 | } | |
5082 | ||
5083 | spin_lock(&last_ptr->lock); | |
5084 | /* | |
5085 | * whoops, this cluster doesn't actually point to | |
5086 | * this block group. Get a ref on the block | |
5087 | * group is does point to and try again | |
5088 | */ | |
5089 | if (!last_ptr_loop && last_ptr->block_group && | |
ff1f2b44 | 5090 | last_ptr->block_group != block_group && |
5091 | index <= | |
5092 | get_block_group_index(last_ptr->block_group)) { | |
fa9c0d79 CM |
5093 | |
5094 | btrfs_put_block_group(block_group); | |
5095 | block_group = last_ptr->block_group; | |
11dfe35a | 5096 | btrfs_get_block_group(block_group); |
fa9c0d79 CM |
5097 | spin_unlock(&last_ptr->lock); |
5098 | spin_unlock(&last_ptr->refill_lock); | |
5099 | ||
5100 | last_ptr_loop = 1; | |
5101 | search_start = block_group->key.objectid; | |
44fb5511 CM |
5102 | /* |
5103 | * we know this block group is properly | |
5104 | * in the list because | |
5105 | * btrfs_remove_block_group, drops the | |
5106 | * cluster before it removes the block | |
5107 | * group from the list | |
5108 | */ | |
fa9c0d79 CM |
5109 | goto have_block_group; |
5110 | } | |
5111 | spin_unlock(&last_ptr->lock); | |
44fb5511 | 5112 | refill_cluster: |
fa9c0d79 CM |
5113 | /* |
5114 | * this cluster didn't work out, free it and | |
5115 | * start over | |
5116 | */ | |
5117 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
5118 | ||
5119 | last_ptr_loop = 0; | |
5120 | ||
5121 | /* allocate a cluster in this block group */ | |
451d7585 | 5122 | ret = btrfs_find_space_cluster(trans, root, |
fa9c0d79 CM |
5123 | block_group, last_ptr, |
5124 | offset, num_bytes, | |
5125 | empty_cluster + empty_size); | |
5126 | if (ret == 0) { | |
5127 | /* | |
5128 | * now pull our allocation out of this | |
5129 | * cluster | |
5130 | */ | |
5131 | offset = btrfs_alloc_from_cluster(block_group, | |
5132 | last_ptr, num_bytes, | |
5133 | search_start); | |
5134 | if (offset) { | |
5135 | /* we found one, proceed */ | |
5136 | spin_unlock(&last_ptr->refill_lock); | |
5137 | goto checks; | |
5138 | } | |
0a24325e JB |
5139 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
5140 | && !failed_cluster_refill) { | |
817d52f8 JB |
5141 | spin_unlock(&last_ptr->refill_lock); |
5142 | ||
0a24325e | 5143 | failed_cluster_refill = true; |
817d52f8 JB |
5144 | wait_block_group_cache_progress(block_group, |
5145 | num_bytes + empty_cluster + empty_size); | |
5146 | goto have_block_group; | |
fa9c0d79 | 5147 | } |
817d52f8 | 5148 | |
fa9c0d79 CM |
5149 | /* |
5150 | * at this point we either didn't find a cluster | |
5151 | * or we weren't able to allocate a block from our | |
5152 | * cluster. Free the cluster we've been trying | |
5153 | * to use, and go to the next block group | |
5154 | */ | |
0a24325e | 5155 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 5156 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 5157 | goto loop; |
fa9c0d79 CM |
5158 | } |
5159 | ||
6226cb0a JB |
5160 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
5161 | num_bytes, empty_size); | |
1cdda9b8 JB |
5162 | /* |
5163 | * If we didn't find a chunk, and we haven't failed on this | |
5164 | * block group before, and this block group is in the middle of | |
5165 | * caching and we are ok with waiting, then go ahead and wait | |
5166 | * for progress to be made, and set failed_alloc to true. | |
5167 | * | |
5168 | * If failed_alloc is true then we've already waited on this | |
5169 | * block group once and should move on to the next block group. | |
5170 | */ | |
5171 | if (!offset && !failed_alloc && !cached && | |
5172 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 5173 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
5174 | num_bytes + empty_size); |
5175 | failed_alloc = true; | |
817d52f8 | 5176 | goto have_block_group; |
1cdda9b8 JB |
5177 | } else if (!offset) { |
5178 | goto loop; | |
817d52f8 | 5179 | } |
fa9c0d79 | 5180 | checks: |
6226cb0a | 5181 | search_start = stripe_align(root, offset); |
2552d17e | 5182 | /* move on to the next group */ |
6226cb0a JB |
5183 | if (search_start + num_bytes >= search_end) { |
5184 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 5185 | goto loop; |
6226cb0a | 5186 | } |
25179201 | 5187 | |
2552d17e JB |
5188 | /* move on to the next group */ |
5189 | if (search_start + num_bytes > | |
6226cb0a JB |
5190 | block_group->key.objectid + block_group->key.offset) { |
5191 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 5192 | goto loop; |
6226cb0a | 5193 | } |
f5a31e16 | 5194 | |
f0486c68 YZ |
5195 | ins->objectid = search_start; |
5196 | ins->offset = num_bytes; | |
2552d17e | 5197 | |
f0486c68 YZ |
5198 | if (offset < search_start) |
5199 | btrfs_add_free_space(block_group, offset, | |
5200 | search_start - offset); | |
5201 | BUG_ON(offset > search_start); | |
2552d17e | 5202 | |
b4d00d56 | 5203 | ret = btrfs_update_reserved_bytes(block_group, num_bytes, 1, |
f0486c68 YZ |
5204 | (data & BTRFS_BLOCK_GROUP_DATA)); |
5205 | if (ret == -EAGAIN) { | |
6226cb0a | 5206 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 5207 | goto loop; |
0f9dd46c | 5208 | } |
0b86a832 | 5209 | |
f0486c68 | 5210 | /* we are all good, lets return */ |
2552d17e JB |
5211 | ins->objectid = search_start; |
5212 | ins->offset = num_bytes; | |
d2fb3437 | 5213 | |
6226cb0a JB |
5214 | if (offset < search_start) |
5215 | btrfs_add_free_space(block_group, offset, | |
5216 | search_start - offset); | |
5217 | BUG_ON(offset > search_start); | |
d82a6f1d | 5218 | btrfs_put_block_group(block_group); |
2552d17e JB |
5219 | break; |
5220 | loop: | |
0a24325e | 5221 | failed_cluster_refill = false; |
1cdda9b8 | 5222 | failed_alloc = false; |
b742bb82 | 5223 | BUG_ON(index != get_block_group_index(block_group)); |
fa9c0d79 | 5224 | btrfs_put_block_group(block_group); |
2552d17e JB |
5225 | } |
5226 | up_read(&space_info->groups_sem); | |
5227 | ||
b742bb82 YZ |
5228 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
5229 | goto search; | |
5230 | ||
ccf0e725 JB |
5231 | /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for |
5232 | * for them to make caching progress. Also | |
5233 | * determine the best possible bg to cache | |
5234 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
5235 | * caching kthreads as we move along | |
817d52f8 JB |
5236 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
5237 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
5238 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
5239 | * again | |
fa9c0d79 | 5240 | */ |
723bda20 | 5241 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 5242 | index = 0; |
ccf0e725 | 5243 | if (loop == LOOP_FIND_IDEAL && found_uncached_bg) { |
817d52f8 | 5244 | found_uncached_bg = false; |
ccf0e725 | 5245 | loop++; |
bab39bf9 | 5246 | if (!ideal_cache_percent) |
817d52f8 | 5247 | goto search; |
ccf0e725 JB |
5248 | |
5249 | /* | |
5250 | * 1 of the following 2 things have happened so far | |
5251 | * | |
5252 | * 1) We found an ideal block group for caching that | |
5253 | * is mostly full and will cache quickly, so we might | |
5254 | * as well wait for it. | |
5255 | * | |
5256 | * 2) We searched for cached only and we didn't find | |
5257 | * anything, and we didn't start any caching kthreads | |
5258 | * either, so chances are we will loop through and | |
5259 | * start a couple caching kthreads, and then come back | |
5260 | * around and just wait for them. This will be slower | |
5261 | * because we will have 2 caching kthreads reading at | |
5262 | * the same time when we could have just started one | |
5263 | * and waited for it to get far enough to give us an | |
5264 | * allocation, so go ahead and go to the wait caching | |
5265 | * loop. | |
5266 | */ | |
5267 | loop = LOOP_CACHING_WAIT; | |
5268 | search_start = ideal_cache_offset; | |
5269 | ideal_cache_percent = 0; | |
5270 | goto ideal_cache; | |
5271 | } else if (loop == LOOP_FIND_IDEAL) { | |
5272 | /* | |
5273 | * Didn't find a uncached bg, wait on anything we find | |
5274 | * next. | |
5275 | */ | |
5276 | loop = LOOP_CACHING_WAIT; | |
5277 | goto search; | |
5278 | } | |
5279 | ||
723bda20 | 5280 | loop++; |
817d52f8 JB |
5281 | |
5282 | if (loop == LOOP_ALLOC_CHUNK) { | |
723bda20 JB |
5283 | if (allowed_chunk_alloc) { |
5284 | ret = do_chunk_alloc(trans, root, num_bytes + | |
5285 | 2 * 1024 * 1024, data, | |
5286 | CHUNK_ALLOC_LIMITED); | |
5287 | allowed_chunk_alloc = 0; | |
5288 | if (ret == 1) | |
5289 | done_chunk_alloc = 1; | |
5290 | } else if (!done_chunk_alloc && | |
5291 | space_info->force_alloc == | |
5292 | CHUNK_ALLOC_NO_FORCE) { | |
5293 | space_info->force_alloc = CHUNK_ALLOC_LIMITED; | |
5294 | } | |
2552d17e | 5295 | |
723bda20 JB |
5296 | /* |
5297 | * We didn't allocate a chunk, go ahead and drop the | |
5298 | * empty size and loop again. | |
5299 | */ | |
5300 | if (!done_chunk_alloc) | |
5301 | loop = LOOP_NO_EMPTY_SIZE; | |
2552d17e JB |
5302 | } |
5303 | ||
723bda20 JB |
5304 | if (loop == LOOP_NO_EMPTY_SIZE) { |
5305 | empty_size = 0; | |
5306 | empty_cluster = 0; | |
fa9c0d79 | 5307 | } |
723bda20 JB |
5308 | |
5309 | goto search; | |
2552d17e JB |
5310 | } else if (!ins->objectid) { |
5311 | ret = -ENOSPC; | |
d82a6f1d | 5312 | } else if (ins->objectid) { |
80eb234a | 5313 | ret = 0; |
be744175 | 5314 | } |
be744175 | 5315 | |
0f70abe2 | 5316 | return ret; |
fec577fb | 5317 | } |
ec44a35c | 5318 | |
9ed74f2d JB |
5319 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
5320 | int dump_block_groups) | |
0f9dd46c JB |
5321 | { |
5322 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 5323 | int index = 0; |
0f9dd46c | 5324 | |
9ed74f2d | 5325 | spin_lock(&info->lock); |
d397712b CM |
5326 | printk(KERN_INFO "space_info has %llu free, is %sfull\n", |
5327 | (unsigned long long)(info->total_bytes - info->bytes_used - | |
9ed74f2d | 5328 | info->bytes_pinned - info->bytes_reserved - |
8929ecfa | 5329 | info->bytes_readonly), |
d397712b | 5330 | (info->full) ? "" : "not "); |
8929ecfa YZ |
5331 | printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, " |
5332 | "reserved=%llu, may_use=%llu, readonly=%llu\n", | |
21380931 | 5333 | (unsigned long long)info->total_bytes, |
8929ecfa | 5334 | (unsigned long long)info->bytes_used, |
21380931 | 5335 | (unsigned long long)info->bytes_pinned, |
8929ecfa | 5336 | (unsigned long long)info->bytes_reserved, |
21380931 | 5337 | (unsigned long long)info->bytes_may_use, |
8929ecfa | 5338 | (unsigned long long)info->bytes_readonly); |
9ed74f2d JB |
5339 | spin_unlock(&info->lock); |
5340 | ||
5341 | if (!dump_block_groups) | |
5342 | return; | |
0f9dd46c | 5343 | |
80eb234a | 5344 | down_read(&info->groups_sem); |
b742bb82 YZ |
5345 | again: |
5346 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 5347 | spin_lock(&cache->lock); |
d397712b CM |
5348 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used " |
5349 | "%llu pinned %llu reserved\n", | |
5350 | (unsigned long long)cache->key.objectid, | |
5351 | (unsigned long long)cache->key.offset, | |
5352 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
5353 | (unsigned long long)cache->pinned, | |
5354 | (unsigned long long)cache->reserved); | |
0f9dd46c JB |
5355 | btrfs_dump_free_space(cache, bytes); |
5356 | spin_unlock(&cache->lock); | |
5357 | } | |
b742bb82 YZ |
5358 | if (++index < BTRFS_NR_RAID_TYPES) |
5359 | goto again; | |
80eb234a | 5360 | up_read(&info->groups_sem); |
0f9dd46c | 5361 | } |
e8569813 | 5362 | |
11833d66 YZ |
5363 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
5364 | struct btrfs_root *root, | |
5365 | u64 num_bytes, u64 min_alloc_size, | |
5366 | u64 empty_size, u64 hint_byte, | |
5367 | u64 search_end, struct btrfs_key *ins, | |
5368 | u64 data) | |
fec577fb CM |
5369 | { |
5370 | int ret; | |
fbdc762b | 5371 | u64 search_start = 0; |
925baedd | 5372 | |
6a63209f | 5373 | data = btrfs_get_alloc_profile(root, data); |
98d20f67 | 5374 | again: |
0ef3e66b CM |
5375 | /* |
5376 | * the only place that sets empty_size is btrfs_realloc_node, which | |
5377 | * is not called recursively on allocations | |
5378 | */ | |
83d3c969 | 5379 | if (empty_size || root->ref_cows) |
6324fbf3 | 5380 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
5381 | num_bytes + 2 * 1024 * 1024, data, |
5382 | CHUNK_ALLOC_NO_FORCE); | |
0b86a832 | 5383 | |
db94535d CM |
5384 | WARN_ON(num_bytes < root->sectorsize); |
5385 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
f0486c68 YZ |
5386 | search_start, search_end, hint_byte, |
5387 | ins, data); | |
3b951516 | 5388 | |
98d20f67 CM |
5389 | if (ret == -ENOSPC && num_bytes > min_alloc_size) { |
5390 | num_bytes = num_bytes >> 1; | |
0f9dd46c | 5391 | num_bytes = num_bytes & ~(root->sectorsize - 1); |
98d20f67 | 5392 | num_bytes = max(num_bytes, min_alloc_size); |
0ef3e66b | 5393 | do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 | 5394 | num_bytes, data, CHUNK_ALLOC_FORCE); |
98d20f67 CM |
5395 | goto again; |
5396 | } | |
91435650 | 5397 | if (ret == -ENOSPC && btrfs_test_opt(root, ENOSPC_DEBUG)) { |
0f9dd46c JB |
5398 | struct btrfs_space_info *sinfo; |
5399 | ||
5400 | sinfo = __find_space_info(root->fs_info, data); | |
d397712b CM |
5401 | printk(KERN_ERR "btrfs allocation failed flags %llu, " |
5402 | "wanted %llu\n", (unsigned long long)data, | |
5403 | (unsigned long long)num_bytes); | |
9ed74f2d | 5404 | dump_space_info(sinfo, num_bytes, 1); |
925baedd | 5405 | } |
0f9dd46c | 5406 | |
1abe9b8a | 5407 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
5408 | ||
0f9dd46c | 5409 | return ret; |
e6dcd2dc CM |
5410 | } |
5411 | ||
65b51a00 CM |
5412 | int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len) |
5413 | { | |
0f9dd46c | 5414 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 5415 | int ret = 0; |
0f9dd46c | 5416 | |
0f9dd46c JB |
5417 | cache = btrfs_lookup_block_group(root->fs_info, start); |
5418 | if (!cache) { | |
d397712b CM |
5419 | printk(KERN_ERR "Unable to find block group for %llu\n", |
5420 | (unsigned long long)start); | |
0f9dd46c JB |
5421 | return -ENOSPC; |
5422 | } | |
1f3c79a2 | 5423 | |
5378e607 LD |
5424 | if (btrfs_test_opt(root, DISCARD)) |
5425 | ret = btrfs_discard_extent(root, start, len, NULL); | |
1f3c79a2 | 5426 | |
0f9dd46c | 5427 | btrfs_add_free_space(cache, start, len); |
b4d00d56 | 5428 | btrfs_update_reserved_bytes(cache, len, 0, 1); |
fa9c0d79 | 5429 | btrfs_put_block_group(cache); |
817d52f8 | 5430 | |
1abe9b8a | 5431 | trace_btrfs_reserved_extent_free(root, start, len); |
5432 | ||
e6dcd2dc CM |
5433 | return ret; |
5434 | } | |
5435 | ||
5d4f98a2 YZ |
5436 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5437 | struct btrfs_root *root, | |
5438 | u64 parent, u64 root_objectid, | |
5439 | u64 flags, u64 owner, u64 offset, | |
5440 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
5441 | { |
5442 | int ret; | |
5d4f98a2 | 5443 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 5444 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 5445 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 5446 | struct btrfs_path *path; |
5d4f98a2 YZ |
5447 | struct extent_buffer *leaf; |
5448 | int type; | |
5449 | u32 size; | |
26b8003f | 5450 | |
5d4f98a2 YZ |
5451 | if (parent > 0) |
5452 | type = BTRFS_SHARED_DATA_REF_KEY; | |
5453 | else | |
5454 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 5455 | |
5d4f98a2 | 5456 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
5457 | |
5458 | path = btrfs_alloc_path(); | |
db5b493a TI |
5459 | if (!path) |
5460 | return -ENOMEM; | |
47e4bb98 | 5461 | |
b9473439 | 5462 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5463 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
5464 | ins, size); | |
ccd467d6 | 5465 | BUG_ON(ret); |
0f9dd46c | 5466 | |
5d4f98a2 YZ |
5467 | leaf = path->nodes[0]; |
5468 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 5469 | struct btrfs_extent_item); |
5d4f98a2 YZ |
5470 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
5471 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
5472 | btrfs_set_extent_flags(leaf, extent_item, | |
5473 | flags | BTRFS_EXTENT_FLAG_DATA); | |
5474 | ||
5475 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
5476 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
5477 | if (parent > 0) { | |
5478 | struct btrfs_shared_data_ref *ref; | |
5479 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
5480 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
5481 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
5482 | } else { | |
5483 | struct btrfs_extent_data_ref *ref; | |
5484 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
5485 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
5486 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
5487 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
5488 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
5489 | } | |
47e4bb98 CM |
5490 | |
5491 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 5492 | btrfs_free_path(path); |
f510cfec | 5493 | |
f0486c68 | 5494 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
f5947066 | 5495 | if (ret) { |
d397712b CM |
5496 | printk(KERN_ERR "btrfs update block group failed for %llu " |
5497 | "%llu\n", (unsigned long long)ins->objectid, | |
5498 | (unsigned long long)ins->offset); | |
f5947066 CM |
5499 | BUG(); |
5500 | } | |
e6dcd2dc CM |
5501 | return ret; |
5502 | } | |
5503 | ||
5d4f98a2 YZ |
5504 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
5505 | struct btrfs_root *root, | |
5506 | u64 parent, u64 root_objectid, | |
5507 | u64 flags, struct btrfs_disk_key *key, | |
5508 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
5509 | { |
5510 | int ret; | |
5d4f98a2 YZ |
5511 | struct btrfs_fs_info *fs_info = root->fs_info; |
5512 | struct btrfs_extent_item *extent_item; | |
5513 | struct btrfs_tree_block_info *block_info; | |
5514 | struct btrfs_extent_inline_ref *iref; | |
5515 | struct btrfs_path *path; | |
5516 | struct extent_buffer *leaf; | |
5517 | u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref); | |
1c2308f8 | 5518 | |
5d4f98a2 | 5519 | path = btrfs_alloc_path(); |
d8926bb3 MF |
5520 | if (!path) |
5521 | return -ENOMEM; | |
56bec294 | 5522 | |
5d4f98a2 YZ |
5523 | path->leave_spinning = 1; |
5524 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
5525 | ins, size); | |
56bec294 | 5526 | BUG_ON(ret); |
5d4f98a2 YZ |
5527 | |
5528 | leaf = path->nodes[0]; | |
5529 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
5530 | struct btrfs_extent_item); | |
5531 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
5532 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
5533 | btrfs_set_extent_flags(leaf, extent_item, | |
5534 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5535 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
5536 | ||
5537 | btrfs_set_tree_block_key(leaf, block_info, key); | |
5538 | btrfs_set_tree_block_level(leaf, block_info, level); | |
5539 | ||
5540 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
5541 | if (parent > 0) { | |
5542 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
5543 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
5544 | BTRFS_SHARED_BLOCK_REF_KEY); | |
5545 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
5546 | } else { | |
5547 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
5548 | BTRFS_TREE_BLOCK_REF_KEY); | |
5549 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
5550 | } | |
5551 | ||
5552 | btrfs_mark_buffer_dirty(leaf); | |
5553 | btrfs_free_path(path); | |
5554 | ||
f0486c68 | 5555 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
5d4f98a2 YZ |
5556 | if (ret) { |
5557 | printk(KERN_ERR "btrfs update block group failed for %llu " | |
5558 | "%llu\n", (unsigned long long)ins->objectid, | |
5559 | (unsigned long long)ins->offset); | |
5560 | BUG(); | |
5561 | } | |
5562 | return ret; | |
5563 | } | |
5564 | ||
5565 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5566 | struct btrfs_root *root, | |
5567 | u64 root_objectid, u64 owner, | |
5568 | u64 offset, struct btrfs_key *ins) | |
5569 | { | |
5570 | int ret; | |
5571 | ||
5572 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
5573 | ||
5574 | ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset, | |
5575 | 0, root_objectid, owner, offset, | |
5576 | BTRFS_ADD_DELAYED_EXTENT, NULL); | |
e6dcd2dc CM |
5577 | return ret; |
5578 | } | |
e02119d5 CM |
5579 | |
5580 | /* | |
5581 | * this is used by the tree logging recovery code. It records that | |
5582 | * an extent has been allocated and makes sure to clear the free | |
5583 | * space cache bits as well | |
5584 | */ | |
5d4f98a2 YZ |
5585 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5586 | struct btrfs_root *root, | |
5587 | u64 root_objectid, u64 owner, u64 offset, | |
5588 | struct btrfs_key *ins) | |
e02119d5 CM |
5589 | { |
5590 | int ret; | |
5591 | struct btrfs_block_group_cache *block_group; | |
11833d66 YZ |
5592 | struct btrfs_caching_control *caching_ctl; |
5593 | u64 start = ins->objectid; | |
5594 | u64 num_bytes = ins->offset; | |
e02119d5 | 5595 | |
e02119d5 | 5596 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
b8399dee | 5597 | cache_block_group(block_group, trans, NULL, 0); |
11833d66 | 5598 | caching_ctl = get_caching_control(block_group); |
e02119d5 | 5599 | |
11833d66 YZ |
5600 | if (!caching_ctl) { |
5601 | BUG_ON(!block_group_cache_done(block_group)); | |
5602 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
5603 | BUG_ON(ret); | |
5604 | } else { | |
5605 | mutex_lock(&caching_ctl->mutex); | |
5606 | ||
5607 | if (start >= caching_ctl->progress) { | |
5608 | ret = add_excluded_extent(root, start, num_bytes); | |
5609 | BUG_ON(ret); | |
5610 | } else if (start + num_bytes <= caching_ctl->progress) { | |
5611 | ret = btrfs_remove_free_space(block_group, | |
5612 | start, num_bytes); | |
5613 | BUG_ON(ret); | |
5614 | } else { | |
5615 | num_bytes = caching_ctl->progress - start; | |
5616 | ret = btrfs_remove_free_space(block_group, | |
5617 | start, num_bytes); | |
5618 | BUG_ON(ret); | |
5619 | ||
5620 | start = caching_ctl->progress; | |
5621 | num_bytes = ins->objectid + ins->offset - | |
5622 | caching_ctl->progress; | |
5623 | ret = add_excluded_extent(root, start, num_bytes); | |
5624 | BUG_ON(ret); | |
5625 | } | |
5626 | ||
5627 | mutex_unlock(&caching_ctl->mutex); | |
5628 | put_caching_control(caching_ctl); | |
5629 | } | |
5630 | ||
b4d00d56 | 5631 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, 1, 1); |
f0486c68 | 5632 | BUG_ON(ret); |
fa9c0d79 | 5633 | btrfs_put_block_group(block_group); |
5d4f98a2 YZ |
5634 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
5635 | 0, owner, offset, ins, 1); | |
e02119d5 CM |
5636 | return ret; |
5637 | } | |
5638 | ||
65b51a00 CM |
5639 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, |
5640 | struct btrfs_root *root, | |
4008c04a CM |
5641 | u64 bytenr, u32 blocksize, |
5642 | int level) | |
65b51a00 CM |
5643 | { |
5644 | struct extent_buffer *buf; | |
5645 | ||
5646 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
5647 | if (!buf) | |
5648 | return ERR_PTR(-ENOMEM); | |
5649 | btrfs_set_header_generation(buf, trans->transid); | |
85d4e461 | 5650 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 CM |
5651 | btrfs_tree_lock(buf); |
5652 | clean_tree_block(trans, root, buf); | |
b4ce94de CM |
5653 | |
5654 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 5655 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 5656 | |
d0c803c4 | 5657 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
8cef4e16 YZ |
5658 | /* |
5659 | * we allow two log transactions at a time, use different | |
5660 | * EXENT bit to differentiate dirty pages. | |
5661 | */ | |
5662 | if (root->log_transid % 2 == 0) | |
5663 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
5664 | buf->start + buf->len - 1, GFP_NOFS); | |
5665 | else | |
5666 | set_extent_new(&root->dirty_log_pages, buf->start, | |
5667 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 CM |
5668 | } else { |
5669 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 5670 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 5671 | } |
65b51a00 | 5672 | trans->blocks_used++; |
b4ce94de | 5673 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
5674 | return buf; |
5675 | } | |
5676 | ||
f0486c68 YZ |
5677 | static struct btrfs_block_rsv * |
5678 | use_block_rsv(struct btrfs_trans_handle *trans, | |
5679 | struct btrfs_root *root, u32 blocksize) | |
5680 | { | |
5681 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 5682 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 YZ |
5683 | int ret; |
5684 | ||
5685 | block_rsv = get_block_rsv(trans, root); | |
5686 | ||
5687 | if (block_rsv->size == 0) { | |
8bb8ab2e JB |
5688 | ret = reserve_metadata_bytes(trans, root, block_rsv, |
5689 | blocksize, 0); | |
68a82277 JB |
5690 | /* |
5691 | * If we couldn't reserve metadata bytes try and use some from | |
5692 | * the global reserve. | |
5693 | */ | |
5694 | if (ret && block_rsv != global_rsv) { | |
5695 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
5696 | if (!ret) | |
5697 | return global_rsv; | |
f0486c68 | 5698 | return ERR_PTR(ret); |
68a82277 | 5699 | } else if (ret) { |
f0486c68 | 5700 | return ERR_PTR(ret); |
68a82277 | 5701 | } |
f0486c68 YZ |
5702 | return block_rsv; |
5703 | } | |
5704 | ||
5705 | ret = block_rsv_use_bytes(block_rsv, blocksize); | |
5706 | if (!ret) | |
5707 | return block_rsv; | |
68a82277 JB |
5708 | if (ret) { |
5709 | WARN_ON(1); | |
5710 | ret = reserve_metadata_bytes(trans, root, block_rsv, blocksize, | |
5711 | 0); | |
5712 | if (!ret) { | |
5713 | spin_lock(&block_rsv->lock); | |
5714 | block_rsv->size += blocksize; | |
5715 | spin_unlock(&block_rsv->lock); | |
5716 | return block_rsv; | |
5717 | } else if (ret && block_rsv != global_rsv) { | |
5718 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
5719 | if (!ret) | |
5720 | return global_rsv; | |
5721 | } | |
5722 | } | |
f0486c68 | 5723 | |
f0486c68 YZ |
5724 | return ERR_PTR(-ENOSPC); |
5725 | } | |
5726 | ||
5727 | static void unuse_block_rsv(struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
5728 | { | |
5729 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
5730 | block_rsv_release_bytes(block_rsv, NULL, 0); | |
5731 | } | |
5732 | ||
fec577fb | 5733 | /* |
f0486c68 YZ |
5734 | * finds a free extent and does all the dirty work required for allocation |
5735 | * returns the key for the extent through ins, and a tree buffer for | |
5736 | * the first block of the extent through buf. | |
5737 | * | |
fec577fb CM |
5738 | * returns the tree buffer or NULL. |
5739 | */ | |
5f39d397 | 5740 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
5741 | struct btrfs_root *root, u32 blocksize, |
5742 | u64 parent, u64 root_objectid, | |
5743 | struct btrfs_disk_key *key, int level, | |
5744 | u64 hint, u64 empty_size) | |
fec577fb | 5745 | { |
e2fa7227 | 5746 | struct btrfs_key ins; |
f0486c68 | 5747 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 5748 | struct extent_buffer *buf; |
f0486c68 YZ |
5749 | u64 flags = 0; |
5750 | int ret; | |
5751 | ||
fec577fb | 5752 | |
f0486c68 YZ |
5753 | block_rsv = use_block_rsv(trans, root, blocksize); |
5754 | if (IS_ERR(block_rsv)) | |
5755 | return ERR_CAST(block_rsv); | |
5756 | ||
5757 | ret = btrfs_reserve_extent(trans, root, blocksize, blocksize, | |
5758 | empty_size, hint, (u64)-1, &ins, 0); | |
fec577fb | 5759 | if (ret) { |
f0486c68 | 5760 | unuse_block_rsv(block_rsv, blocksize); |
54aa1f4d | 5761 | return ERR_PTR(ret); |
fec577fb | 5762 | } |
55c69072 | 5763 | |
4008c04a CM |
5764 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
5765 | blocksize, level); | |
f0486c68 YZ |
5766 | BUG_ON(IS_ERR(buf)); |
5767 | ||
5768 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
5769 | if (parent == 0) | |
5770 | parent = ins.objectid; | |
5771 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
5772 | } else | |
5773 | BUG_ON(parent > 0); | |
5774 | ||
5775 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
5776 | struct btrfs_delayed_extent_op *extent_op; | |
5777 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
5778 | BUG_ON(!extent_op); | |
5779 | if (key) | |
5780 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
5781 | else | |
5782 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
5783 | extent_op->flags_to_set = flags; | |
5784 | extent_op->update_key = 1; | |
5785 | extent_op->update_flags = 1; | |
5786 | extent_op->is_data = 0; | |
5787 | ||
5788 | ret = btrfs_add_delayed_tree_ref(trans, ins.objectid, | |
5789 | ins.offset, parent, root_objectid, | |
5790 | level, BTRFS_ADD_DELAYED_EXTENT, | |
5791 | extent_op); | |
5792 | BUG_ON(ret); | |
5793 | } | |
fec577fb CM |
5794 | return buf; |
5795 | } | |
a28ec197 | 5796 | |
2c47e605 YZ |
5797 | struct walk_control { |
5798 | u64 refs[BTRFS_MAX_LEVEL]; | |
5799 | u64 flags[BTRFS_MAX_LEVEL]; | |
5800 | struct btrfs_key update_progress; | |
5801 | int stage; | |
5802 | int level; | |
5803 | int shared_level; | |
5804 | int update_ref; | |
5805 | int keep_locks; | |
1c4850e2 YZ |
5806 | int reada_slot; |
5807 | int reada_count; | |
2c47e605 YZ |
5808 | }; |
5809 | ||
5810 | #define DROP_REFERENCE 1 | |
5811 | #define UPDATE_BACKREF 2 | |
5812 | ||
1c4850e2 YZ |
5813 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
5814 | struct btrfs_root *root, | |
5815 | struct walk_control *wc, | |
5816 | struct btrfs_path *path) | |
6407bf6d | 5817 | { |
1c4850e2 YZ |
5818 | u64 bytenr; |
5819 | u64 generation; | |
5820 | u64 refs; | |
94fcca9f | 5821 | u64 flags; |
5d4f98a2 | 5822 | u32 nritems; |
1c4850e2 YZ |
5823 | u32 blocksize; |
5824 | struct btrfs_key key; | |
5825 | struct extent_buffer *eb; | |
6407bf6d | 5826 | int ret; |
1c4850e2 YZ |
5827 | int slot; |
5828 | int nread = 0; | |
6407bf6d | 5829 | |
1c4850e2 YZ |
5830 | if (path->slots[wc->level] < wc->reada_slot) { |
5831 | wc->reada_count = wc->reada_count * 2 / 3; | |
5832 | wc->reada_count = max(wc->reada_count, 2); | |
5833 | } else { | |
5834 | wc->reada_count = wc->reada_count * 3 / 2; | |
5835 | wc->reada_count = min_t(int, wc->reada_count, | |
5836 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
5837 | } | |
7bb86316 | 5838 | |
1c4850e2 YZ |
5839 | eb = path->nodes[wc->level]; |
5840 | nritems = btrfs_header_nritems(eb); | |
5841 | blocksize = btrfs_level_size(root, wc->level - 1); | |
bd56b302 | 5842 | |
1c4850e2 YZ |
5843 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
5844 | if (nread >= wc->reada_count) | |
5845 | break; | |
bd56b302 | 5846 | |
2dd3e67b | 5847 | cond_resched(); |
1c4850e2 YZ |
5848 | bytenr = btrfs_node_blockptr(eb, slot); |
5849 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 5850 | |
1c4850e2 YZ |
5851 | if (slot == path->slots[wc->level]) |
5852 | goto reada; | |
5d4f98a2 | 5853 | |
1c4850e2 YZ |
5854 | if (wc->stage == UPDATE_BACKREF && |
5855 | generation <= root->root_key.offset) | |
bd56b302 CM |
5856 | continue; |
5857 | ||
94fcca9f YZ |
5858 | /* We don't lock the tree block, it's OK to be racy here */ |
5859 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, | |
5860 | &refs, &flags); | |
5861 | BUG_ON(ret); | |
5862 | BUG_ON(refs == 0); | |
5863 | ||
1c4850e2 | 5864 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
5865 | if (refs == 1) |
5866 | goto reada; | |
bd56b302 | 5867 | |
94fcca9f YZ |
5868 | if (wc->level == 1 && |
5869 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5870 | continue; | |
1c4850e2 YZ |
5871 | if (!wc->update_ref || |
5872 | generation <= root->root_key.offset) | |
5873 | continue; | |
5874 | btrfs_node_key_to_cpu(eb, &key, slot); | |
5875 | ret = btrfs_comp_cpu_keys(&key, | |
5876 | &wc->update_progress); | |
5877 | if (ret < 0) | |
5878 | continue; | |
94fcca9f YZ |
5879 | } else { |
5880 | if (wc->level == 1 && | |
5881 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
5882 | continue; | |
6407bf6d | 5883 | } |
1c4850e2 YZ |
5884 | reada: |
5885 | ret = readahead_tree_block(root, bytenr, blocksize, | |
5886 | generation); | |
5887 | if (ret) | |
bd56b302 | 5888 | break; |
1c4850e2 | 5889 | nread++; |
20524f02 | 5890 | } |
1c4850e2 | 5891 | wc->reada_slot = slot; |
20524f02 | 5892 | } |
2c47e605 | 5893 | |
f82d02d9 | 5894 | /* |
2c47e605 YZ |
5895 | * hepler to process tree block while walking down the tree. |
5896 | * | |
2c47e605 YZ |
5897 | * when wc->stage == UPDATE_BACKREF, this function updates |
5898 | * back refs for pointers in the block. | |
5899 | * | |
5900 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 5901 | */ |
2c47e605 | 5902 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 5903 | struct btrfs_root *root, |
2c47e605 | 5904 | struct btrfs_path *path, |
94fcca9f | 5905 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 5906 | { |
2c47e605 YZ |
5907 | int level = wc->level; |
5908 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 5909 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
5910 | int ret; |
5911 | ||
2c47e605 YZ |
5912 | if (wc->stage == UPDATE_BACKREF && |
5913 | btrfs_header_owner(eb) != root->root_key.objectid) | |
5914 | return 1; | |
f82d02d9 | 5915 | |
2c47e605 YZ |
5916 | /* |
5917 | * when reference count of tree block is 1, it won't increase | |
5918 | * again. once full backref flag is set, we never clear it. | |
5919 | */ | |
94fcca9f YZ |
5920 | if (lookup_info && |
5921 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
5922 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
5923 | BUG_ON(!path->locks[level]); |
5924 | ret = btrfs_lookup_extent_info(trans, root, | |
5925 | eb->start, eb->len, | |
5926 | &wc->refs[level], | |
5927 | &wc->flags[level]); | |
5928 | BUG_ON(ret); | |
5929 | BUG_ON(wc->refs[level] == 0); | |
5930 | } | |
5d4f98a2 | 5931 | |
2c47e605 YZ |
5932 | if (wc->stage == DROP_REFERENCE) { |
5933 | if (wc->refs[level] > 1) | |
5934 | return 1; | |
f82d02d9 | 5935 | |
2c47e605 | 5936 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 5937 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
5938 | path->locks[level] = 0; |
5939 | } | |
5940 | return 0; | |
5941 | } | |
f82d02d9 | 5942 | |
2c47e605 YZ |
5943 | /* wc->stage == UPDATE_BACKREF */ |
5944 | if (!(wc->flags[level] & flag)) { | |
5945 | BUG_ON(!path->locks[level]); | |
5946 | ret = btrfs_inc_ref(trans, root, eb, 1); | |
f82d02d9 | 5947 | BUG_ON(ret); |
2c47e605 YZ |
5948 | ret = btrfs_dec_ref(trans, root, eb, 0); |
5949 | BUG_ON(ret); | |
5950 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, | |
5951 | eb->len, flag, 0); | |
5952 | BUG_ON(ret); | |
5953 | wc->flags[level] |= flag; | |
5954 | } | |
5955 | ||
5956 | /* | |
5957 | * the block is shared by multiple trees, so it's not good to | |
5958 | * keep the tree lock | |
5959 | */ | |
5960 | if (path->locks[level] && level > 0) { | |
bd681513 | 5961 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
5962 | path->locks[level] = 0; |
5963 | } | |
5964 | return 0; | |
5965 | } | |
5966 | ||
1c4850e2 YZ |
5967 | /* |
5968 | * hepler to process tree block pointer. | |
5969 | * | |
5970 | * when wc->stage == DROP_REFERENCE, this function checks | |
5971 | * reference count of the block pointed to. if the block | |
5972 | * is shared and we need update back refs for the subtree | |
5973 | * rooted at the block, this function changes wc->stage to | |
5974 | * UPDATE_BACKREF. if the block is shared and there is no | |
5975 | * need to update back, this function drops the reference | |
5976 | * to the block. | |
5977 | * | |
5978 | * NOTE: return value 1 means we should stop walking down. | |
5979 | */ | |
5980 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
5981 | struct btrfs_root *root, | |
5982 | struct btrfs_path *path, | |
94fcca9f | 5983 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
5984 | { |
5985 | u64 bytenr; | |
5986 | u64 generation; | |
5987 | u64 parent; | |
5988 | u32 blocksize; | |
5989 | struct btrfs_key key; | |
5990 | struct extent_buffer *next; | |
5991 | int level = wc->level; | |
5992 | int reada = 0; | |
5993 | int ret = 0; | |
5994 | ||
5995 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
5996 | path->slots[level]); | |
5997 | /* | |
5998 | * if the lower level block was created before the snapshot | |
5999 | * was created, we know there is no need to update back refs | |
6000 | * for the subtree | |
6001 | */ | |
6002 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
6003 | generation <= root->root_key.offset) { |
6004 | *lookup_info = 1; | |
1c4850e2 | 6005 | return 1; |
94fcca9f | 6006 | } |
1c4850e2 YZ |
6007 | |
6008 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
6009 | blocksize = btrfs_level_size(root, level - 1); | |
6010 | ||
6011 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
6012 | if (!next) { | |
6013 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
90d2c51d MX |
6014 | if (!next) |
6015 | return -ENOMEM; | |
1c4850e2 YZ |
6016 | reada = 1; |
6017 | } | |
6018 | btrfs_tree_lock(next); | |
6019 | btrfs_set_lock_blocking(next); | |
6020 | ||
94fcca9f YZ |
6021 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, |
6022 | &wc->refs[level - 1], | |
6023 | &wc->flags[level - 1]); | |
6024 | BUG_ON(ret); | |
6025 | BUG_ON(wc->refs[level - 1] == 0); | |
6026 | *lookup_info = 0; | |
1c4850e2 | 6027 | |
94fcca9f | 6028 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 6029 | if (wc->refs[level - 1] > 1) { |
94fcca9f YZ |
6030 | if (level == 1 && |
6031 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6032 | goto skip; | |
6033 | ||
1c4850e2 YZ |
6034 | if (!wc->update_ref || |
6035 | generation <= root->root_key.offset) | |
6036 | goto skip; | |
6037 | ||
6038 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
6039 | path->slots[level]); | |
6040 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
6041 | if (ret < 0) | |
6042 | goto skip; | |
6043 | ||
6044 | wc->stage = UPDATE_BACKREF; | |
6045 | wc->shared_level = level - 1; | |
6046 | } | |
94fcca9f YZ |
6047 | } else { |
6048 | if (level == 1 && | |
6049 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6050 | goto skip; | |
1c4850e2 YZ |
6051 | } |
6052 | ||
6053 | if (!btrfs_buffer_uptodate(next, generation)) { | |
6054 | btrfs_tree_unlock(next); | |
6055 | free_extent_buffer(next); | |
6056 | next = NULL; | |
94fcca9f | 6057 | *lookup_info = 1; |
1c4850e2 YZ |
6058 | } |
6059 | ||
6060 | if (!next) { | |
6061 | if (reada && level == 1) | |
6062 | reada_walk_down(trans, root, wc, path); | |
6063 | next = read_tree_block(root, bytenr, blocksize, generation); | |
97d9a8a4 TI |
6064 | if (!next) |
6065 | return -EIO; | |
1c4850e2 YZ |
6066 | btrfs_tree_lock(next); |
6067 | btrfs_set_lock_blocking(next); | |
6068 | } | |
6069 | ||
6070 | level--; | |
6071 | BUG_ON(level != btrfs_header_level(next)); | |
6072 | path->nodes[level] = next; | |
6073 | path->slots[level] = 0; | |
bd681513 | 6074 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
6075 | wc->level = level; |
6076 | if (wc->level == 1) | |
6077 | wc->reada_slot = 0; | |
6078 | return 0; | |
6079 | skip: | |
6080 | wc->refs[level - 1] = 0; | |
6081 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
6082 | if (wc->stage == DROP_REFERENCE) { |
6083 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
6084 | parent = path->nodes[level]->start; | |
6085 | } else { | |
6086 | BUG_ON(root->root_key.objectid != | |
6087 | btrfs_header_owner(path->nodes[level])); | |
6088 | parent = 0; | |
6089 | } | |
1c4850e2 | 6090 | |
94fcca9f YZ |
6091 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
6092 | root->root_key.objectid, level - 1, 0); | |
6093 | BUG_ON(ret); | |
1c4850e2 | 6094 | } |
1c4850e2 YZ |
6095 | btrfs_tree_unlock(next); |
6096 | free_extent_buffer(next); | |
94fcca9f | 6097 | *lookup_info = 1; |
1c4850e2 YZ |
6098 | return 1; |
6099 | } | |
6100 | ||
2c47e605 YZ |
6101 | /* |
6102 | * hepler to process tree block while walking up the tree. | |
6103 | * | |
6104 | * when wc->stage == DROP_REFERENCE, this function drops | |
6105 | * reference count on the block. | |
6106 | * | |
6107 | * when wc->stage == UPDATE_BACKREF, this function changes | |
6108 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
6109 | * to UPDATE_BACKREF previously while processing the block. | |
6110 | * | |
6111 | * NOTE: return value 1 means we should stop walking up. | |
6112 | */ | |
6113 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
6114 | struct btrfs_root *root, | |
6115 | struct btrfs_path *path, | |
6116 | struct walk_control *wc) | |
6117 | { | |
f0486c68 | 6118 | int ret; |
2c47e605 YZ |
6119 | int level = wc->level; |
6120 | struct extent_buffer *eb = path->nodes[level]; | |
6121 | u64 parent = 0; | |
6122 | ||
6123 | if (wc->stage == UPDATE_BACKREF) { | |
6124 | BUG_ON(wc->shared_level < level); | |
6125 | if (level < wc->shared_level) | |
6126 | goto out; | |
6127 | ||
2c47e605 YZ |
6128 | ret = find_next_key(path, level + 1, &wc->update_progress); |
6129 | if (ret > 0) | |
6130 | wc->update_ref = 0; | |
6131 | ||
6132 | wc->stage = DROP_REFERENCE; | |
6133 | wc->shared_level = -1; | |
6134 | path->slots[level] = 0; | |
6135 | ||
6136 | /* | |
6137 | * check reference count again if the block isn't locked. | |
6138 | * we should start walking down the tree again if reference | |
6139 | * count is one. | |
6140 | */ | |
6141 | if (!path->locks[level]) { | |
6142 | BUG_ON(level == 0); | |
6143 | btrfs_tree_lock(eb); | |
6144 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6145 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6146 | |
6147 | ret = btrfs_lookup_extent_info(trans, root, | |
6148 | eb->start, eb->len, | |
6149 | &wc->refs[level], | |
6150 | &wc->flags[level]); | |
f82d02d9 | 6151 | BUG_ON(ret); |
2c47e605 YZ |
6152 | BUG_ON(wc->refs[level] == 0); |
6153 | if (wc->refs[level] == 1) { | |
bd681513 | 6154 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6155 | return 1; |
6156 | } | |
f82d02d9 | 6157 | } |
2c47e605 | 6158 | } |
f82d02d9 | 6159 | |
2c47e605 YZ |
6160 | /* wc->stage == DROP_REFERENCE */ |
6161 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 6162 | |
2c47e605 YZ |
6163 | if (wc->refs[level] == 1) { |
6164 | if (level == 0) { | |
6165 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6166 | ret = btrfs_dec_ref(trans, root, eb, 1); | |
6167 | else | |
6168 | ret = btrfs_dec_ref(trans, root, eb, 0); | |
6169 | BUG_ON(ret); | |
6170 | } | |
6171 | /* make block locked assertion in clean_tree_block happy */ | |
6172 | if (!path->locks[level] && | |
6173 | btrfs_header_generation(eb) == trans->transid) { | |
6174 | btrfs_tree_lock(eb); | |
6175 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6176 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6177 | } |
6178 | clean_tree_block(trans, root, eb); | |
6179 | } | |
6180 | ||
6181 | if (eb == root->node) { | |
6182 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6183 | parent = eb->start; | |
6184 | else | |
6185 | BUG_ON(root->root_key.objectid != | |
6186 | btrfs_header_owner(eb)); | |
6187 | } else { | |
6188 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6189 | parent = path->nodes[level + 1]->start; | |
6190 | else | |
6191 | BUG_ON(root->root_key.objectid != | |
6192 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 6193 | } |
f82d02d9 | 6194 | |
f0486c68 | 6195 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
6196 | out: |
6197 | wc->refs[level] = 0; | |
6198 | wc->flags[level] = 0; | |
f0486c68 | 6199 | return 0; |
2c47e605 YZ |
6200 | } |
6201 | ||
6202 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
6203 | struct btrfs_root *root, | |
6204 | struct btrfs_path *path, | |
6205 | struct walk_control *wc) | |
6206 | { | |
2c47e605 | 6207 | int level = wc->level; |
94fcca9f | 6208 | int lookup_info = 1; |
2c47e605 YZ |
6209 | int ret; |
6210 | ||
6211 | while (level >= 0) { | |
94fcca9f | 6212 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
6213 | if (ret > 0) |
6214 | break; | |
6215 | ||
6216 | if (level == 0) | |
6217 | break; | |
6218 | ||
7a7965f8 YZ |
6219 | if (path->slots[level] >= |
6220 | btrfs_header_nritems(path->nodes[level])) | |
6221 | break; | |
6222 | ||
94fcca9f | 6223 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
6224 | if (ret > 0) { |
6225 | path->slots[level]++; | |
6226 | continue; | |
90d2c51d MX |
6227 | } else if (ret < 0) |
6228 | return ret; | |
1c4850e2 | 6229 | level = wc->level; |
f82d02d9 | 6230 | } |
f82d02d9 YZ |
6231 | return 0; |
6232 | } | |
6233 | ||
d397712b | 6234 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 6235 | struct btrfs_root *root, |
f82d02d9 | 6236 | struct btrfs_path *path, |
2c47e605 | 6237 | struct walk_control *wc, int max_level) |
20524f02 | 6238 | { |
2c47e605 | 6239 | int level = wc->level; |
20524f02 | 6240 | int ret; |
9f3a7427 | 6241 | |
2c47e605 YZ |
6242 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
6243 | while (level < max_level && path->nodes[level]) { | |
6244 | wc->level = level; | |
6245 | if (path->slots[level] + 1 < | |
6246 | btrfs_header_nritems(path->nodes[level])) { | |
6247 | path->slots[level]++; | |
20524f02 CM |
6248 | return 0; |
6249 | } else { | |
2c47e605 YZ |
6250 | ret = walk_up_proc(trans, root, path, wc); |
6251 | if (ret > 0) | |
6252 | return 0; | |
bd56b302 | 6253 | |
2c47e605 | 6254 | if (path->locks[level]) { |
bd681513 CM |
6255 | btrfs_tree_unlock_rw(path->nodes[level], |
6256 | path->locks[level]); | |
2c47e605 | 6257 | path->locks[level] = 0; |
f82d02d9 | 6258 | } |
2c47e605 YZ |
6259 | free_extent_buffer(path->nodes[level]); |
6260 | path->nodes[level] = NULL; | |
6261 | level++; | |
20524f02 CM |
6262 | } |
6263 | } | |
6264 | return 1; | |
6265 | } | |
6266 | ||
9aca1d51 | 6267 | /* |
2c47e605 YZ |
6268 | * drop a subvolume tree. |
6269 | * | |
6270 | * this function traverses the tree freeing any blocks that only | |
6271 | * referenced by the tree. | |
6272 | * | |
6273 | * when a shared tree block is found. this function decreases its | |
6274 | * reference count by one. if update_ref is true, this function | |
6275 | * also make sure backrefs for the shared block and all lower level | |
6276 | * blocks are properly updated. | |
9aca1d51 | 6277 | */ |
cb1b69f4 TI |
6278 | void btrfs_drop_snapshot(struct btrfs_root *root, |
6279 | struct btrfs_block_rsv *block_rsv, int update_ref) | |
20524f02 | 6280 | { |
5caf2a00 | 6281 | struct btrfs_path *path; |
2c47e605 YZ |
6282 | struct btrfs_trans_handle *trans; |
6283 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 6284 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
6285 | struct walk_control *wc; |
6286 | struct btrfs_key key; | |
6287 | int err = 0; | |
6288 | int ret; | |
6289 | int level; | |
20524f02 | 6290 | |
5caf2a00 | 6291 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
6292 | if (!path) { |
6293 | err = -ENOMEM; | |
6294 | goto out; | |
6295 | } | |
20524f02 | 6296 | |
2c47e605 | 6297 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
6298 | if (!wc) { |
6299 | btrfs_free_path(path); | |
cb1b69f4 TI |
6300 | err = -ENOMEM; |
6301 | goto out; | |
38a1a919 | 6302 | } |
2c47e605 | 6303 | |
a22285a6 | 6304 | trans = btrfs_start_transaction(tree_root, 0); |
98d5dc13 TI |
6305 | BUG_ON(IS_ERR(trans)); |
6306 | ||
3fd0a558 YZ |
6307 | if (block_rsv) |
6308 | trans->block_rsv = block_rsv; | |
2c47e605 | 6309 | |
9f3a7427 | 6310 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 6311 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
6312 | path->nodes[level] = btrfs_lock_root_node(root); |
6313 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 6314 | path->slots[level] = 0; |
bd681513 | 6315 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6316 | memset(&wc->update_progress, 0, |
6317 | sizeof(wc->update_progress)); | |
9f3a7427 | 6318 | } else { |
9f3a7427 | 6319 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
6320 | memcpy(&wc->update_progress, &key, |
6321 | sizeof(wc->update_progress)); | |
6322 | ||
6702ed49 | 6323 | level = root_item->drop_level; |
2c47e605 | 6324 | BUG_ON(level == 0); |
6702ed49 | 6325 | path->lowest_level = level; |
2c47e605 YZ |
6326 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6327 | path->lowest_level = 0; | |
6328 | if (ret < 0) { | |
6329 | err = ret; | |
cb1b69f4 | 6330 | goto out_free; |
9f3a7427 | 6331 | } |
1c4850e2 | 6332 | WARN_ON(ret > 0); |
2c47e605 | 6333 | |
7d9eb12c CM |
6334 | /* |
6335 | * unlock our path, this is safe because only this | |
6336 | * function is allowed to delete this snapshot | |
6337 | */ | |
5d4f98a2 | 6338 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
6339 | |
6340 | level = btrfs_header_level(root->node); | |
6341 | while (1) { | |
6342 | btrfs_tree_lock(path->nodes[level]); | |
6343 | btrfs_set_lock_blocking(path->nodes[level]); | |
6344 | ||
6345 | ret = btrfs_lookup_extent_info(trans, root, | |
6346 | path->nodes[level]->start, | |
6347 | path->nodes[level]->len, | |
6348 | &wc->refs[level], | |
6349 | &wc->flags[level]); | |
6350 | BUG_ON(ret); | |
6351 | BUG_ON(wc->refs[level] == 0); | |
6352 | ||
6353 | if (level == root_item->drop_level) | |
6354 | break; | |
6355 | ||
6356 | btrfs_tree_unlock(path->nodes[level]); | |
6357 | WARN_ON(wc->refs[level] != 1); | |
6358 | level--; | |
6359 | } | |
9f3a7427 | 6360 | } |
2c47e605 YZ |
6361 | |
6362 | wc->level = level; | |
6363 | wc->shared_level = -1; | |
6364 | wc->stage = DROP_REFERENCE; | |
6365 | wc->update_ref = update_ref; | |
6366 | wc->keep_locks = 0; | |
1c4850e2 | 6367 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 6368 | |
d397712b | 6369 | while (1) { |
2c47e605 YZ |
6370 | ret = walk_down_tree(trans, root, path, wc); |
6371 | if (ret < 0) { | |
6372 | err = ret; | |
20524f02 | 6373 | break; |
2c47e605 | 6374 | } |
9aca1d51 | 6375 | |
2c47e605 YZ |
6376 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
6377 | if (ret < 0) { | |
6378 | err = ret; | |
20524f02 | 6379 | break; |
2c47e605 YZ |
6380 | } |
6381 | ||
6382 | if (ret > 0) { | |
6383 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
6384 | break; |
6385 | } | |
2c47e605 YZ |
6386 | |
6387 | if (wc->stage == DROP_REFERENCE) { | |
6388 | level = wc->level; | |
6389 | btrfs_node_key(path->nodes[level], | |
6390 | &root_item->drop_progress, | |
6391 | path->slots[level]); | |
6392 | root_item->drop_level = level; | |
6393 | } | |
6394 | ||
6395 | BUG_ON(wc->level == 0); | |
3fd0a558 | 6396 | if (btrfs_should_end_transaction(trans, tree_root)) { |
2c47e605 YZ |
6397 | ret = btrfs_update_root(trans, tree_root, |
6398 | &root->root_key, | |
6399 | root_item); | |
6400 | BUG_ON(ret); | |
6401 | ||
3fd0a558 | 6402 | btrfs_end_transaction_throttle(trans, tree_root); |
a22285a6 | 6403 | trans = btrfs_start_transaction(tree_root, 0); |
98d5dc13 | 6404 | BUG_ON(IS_ERR(trans)); |
3fd0a558 YZ |
6405 | if (block_rsv) |
6406 | trans->block_rsv = block_rsv; | |
c3e69d58 | 6407 | } |
20524f02 | 6408 | } |
b3b4aa74 | 6409 | btrfs_release_path(path); |
2c47e605 YZ |
6410 | BUG_ON(err); |
6411 | ||
6412 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
6413 | BUG_ON(ret); | |
6414 | ||
76dda93c YZ |
6415 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
6416 | ret = btrfs_find_last_root(tree_root, root->root_key.objectid, | |
6417 | NULL, NULL); | |
6418 | BUG_ON(ret < 0); | |
6419 | if (ret > 0) { | |
84cd948c JB |
6420 | /* if we fail to delete the orphan item this time |
6421 | * around, it'll get picked up the next time. | |
6422 | * | |
6423 | * The most common failure here is just -ENOENT. | |
6424 | */ | |
6425 | btrfs_del_orphan_item(trans, tree_root, | |
6426 | root->root_key.objectid); | |
76dda93c YZ |
6427 | } |
6428 | } | |
6429 | ||
6430 | if (root->in_radix) { | |
6431 | btrfs_free_fs_root(tree_root->fs_info, root); | |
6432 | } else { | |
6433 | free_extent_buffer(root->node); | |
6434 | free_extent_buffer(root->commit_root); | |
6435 | kfree(root); | |
6436 | } | |
cb1b69f4 | 6437 | out_free: |
3fd0a558 | 6438 | btrfs_end_transaction_throttle(trans, tree_root); |
2c47e605 | 6439 | kfree(wc); |
5caf2a00 | 6440 | btrfs_free_path(path); |
cb1b69f4 TI |
6441 | out: |
6442 | if (err) | |
6443 | btrfs_std_error(root->fs_info, err); | |
6444 | return; | |
20524f02 | 6445 | } |
9078a3e1 | 6446 | |
2c47e605 YZ |
6447 | /* |
6448 | * drop subtree rooted at tree block 'node'. | |
6449 | * | |
6450 | * NOTE: this function will unlock and release tree block 'node' | |
6451 | */ | |
f82d02d9 YZ |
6452 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
6453 | struct btrfs_root *root, | |
6454 | struct extent_buffer *node, | |
6455 | struct extent_buffer *parent) | |
6456 | { | |
6457 | struct btrfs_path *path; | |
2c47e605 | 6458 | struct walk_control *wc; |
f82d02d9 YZ |
6459 | int level; |
6460 | int parent_level; | |
6461 | int ret = 0; | |
6462 | int wret; | |
6463 | ||
2c47e605 YZ |
6464 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
6465 | ||
f82d02d9 | 6466 | path = btrfs_alloc_path(); |
db5b493a TI |
6467 | if (!path) |
6468 | return -ENOMEM; | |
f82d02d9 | 6469 | |
2c47e605 | 6470 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
6471 | if (!wc) { |
6472 | btrfs_free_path(path); | |
6473 | return -ENOMEM; | |
6474 | } | |
2c47e605 | 6475 | |
b9447ef8 | 6476 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
6477 | parent_level = btrfs_header_level(parent); |
6478 | extent_buffer_get(parent); | |
6479 | path->nodes[parent_level] = parent; | |
6480 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
6481 | ||
b9447ef8 | 6482 | btrfs_assert_tree_locked(node); |
f82d02d9 | 6483 | level = btrfs_header_level(node); |
f82d02d9 YZ |
6484 | path->nodes[level] = node; |
6485 | path->slots[level] = 0; | |
bd681513 | 6486 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6487 | |
6488 | wc->refs[parent_level] = 1; | |
6489 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6490 | wc->level = level; | |
6491 | wc->shared_level = -1; | |
6492 | wc->stage = DROP_REFERENCE; | |
6493 | wc->update_ref = 0; | |
6494 | wc->keep_locks = 1; | |
1c4850e2 | 6495 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
6496 | |
6497 | while (1) { | |
2c47e605 YZ |
6498 | wret = walk_down_tree(trans, root, path, wc); |
6499 | if (wret < 0) { | |
f82d02d9 | 6500 | ret = wret; |
f82d02d9 | 6501 | break; |
2c47e605 | 6502 | } |
f82d02d9 | 6503 | |
2c47e605 | 6504 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
6505 | if (wret < 0) |
6506 | ret = wret; | |
6507 | if (wret != 0) | |
6508 | break; | |
6509 | } | |
6510 | ||
2c47e605 | 6511 | kfree(wc); |
f82d02d9 YZ |
6512 | btrfs_free_path(path); |
6513 | return ret; | |
6514 | } | |
6515 | ||
ec44a35c CM |
6516 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
6517 | { | |
6518 | u64 num_devices; | |
6519 | u64 stripped = BTRFS_BLOCK_GROUP_RAID0 | | |
6520 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
6521 | ||
cd02dca5 CM |
6522 | /* |
6523 | * we add in the count of missing devices because we want | |
6524 | * to make sure that any RAID levels on a degraded FS | |
6525 | * continue to be honored. | |
6526 | */ | |
6527 | num_devices = root->fs_info->fs_devices->rw_devices + | |
6528 | root->fs_info->fs_devices->missing_devices; | |
6529 | ||
ec44a35c CM |
6530 | if (num_devices == 1) { |
6531 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
6532 | stripped = flags & ~stripped; | |
6533 | ||
6534 | /* turn raid0 into single device chunks */ | |
6535 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
6536 | return stripped; | |
6537 | ||
6538 | /* turn mirroring into duplication */ | |
6539 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
6540 | BTRFS_BLOCK_GROUP_RAID10)) | |
6541 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
6542 | return flags; | |
6543 | } else { | |
6544 | /* they already had raid on here, just return */ | |
ec44a35c CM |
6545 | if (flags & stripped) |
6546 | return flags; | |
6547 | ||
6548 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
6549 | stripped = flags & ~stripped; | |
6550 | ||
6551 | /* switch duplicated blocks with raid1 */ | |
6552 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
6553 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
6554 | ||
6555 | /* turn single device chunks into raid0 */ | |
6556 | return stripped | BTRFS_BLOCK_GROUP_RAID0; | |
6557 | } | |
6558 | return flags; | |
6559 | } | |
6560 | ||
199c36ea | 6561 | static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 6562 | { |
f0486c68 YZ |
6563 | struct btrfs_space_info *sinfo = cache->space_info; |
6564 | u64 num_bytes; | |
199c36ea | 6565 | u64 min_allocable_bytes; |
f0486c68 | 6566 | int ret = -ENOSPC; |
0ef3e66b | 6567 | |
c286ac48 | 6568 | |
199c36ea MX |
6569 | /* |
6570 | * We need some metadata space and system metadata space for | |
6571 | * allocating chunks in some corner cases until we force to set | |
6572 | * it to be readonly. | |
6573 | */ | |
6574 | if ((sinfo->flags & | |
6575 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
6576 | !force) | |
6577 | min_allocable_bytes = 1 * 1024 * 1024; | |
6578 | else | |
6579 | min_allocable_bytes = 0; | |
6580 | ||
f0486c68 YZ |
6581 | spin_lock(&sinfo->lock); |
6582 | spin_lock(&cache->lock); | |
61cfea9b W |
6583 | |
6584 | if (cache->ro) { | |
6585 | ret = 0; | |
6586 | goto out; | |
6587 | } | |
6588 | ||
f0486c68 YZ |
6589 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
6590 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
6591 | ||
6592 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
6593 | sinfo->bytes_may_use + sinfo->bytes_readonly + | |
199c36ea MX |
6594 | cache->reserved_pinned + num_bytes + min_allocable_bytes <= |
6595 | sinfo->total_bytes) { | |
f0486c68 YZ |
6596 | sinfo->bytes_readonly += num_bytes; |
6597 | sinfo->bytes_reserved += cache->reserved_pinned; | |
6598 | cache->reserved_pinned = 0; | |
6599 | cache->ro = 1; | |
6600 | ret = 0; | |
6601 | } | |
61cfea9b | 6602 | out: |
f0486c68 YZ |
6603 | spin_unlock(&cache->lock); |
6604 | spin_unlock(&sinfo->lock); | |
6605 | return ret; | |
6606 | } | |
7d9eb12c | 6607 | |
f0486c68 YZ |
6608 | int btrfs_set_block_group_ro(struct btrfs_root *root, |
6609 | struct btrfs_block_group_cache *cache) | |
c286ac48 | 6610 | |
f0486c68 YZ |
6611 | { |
6612 | struct btrfs_trans_handle *trans; | |
6613 | u64 alloc_flags; | |
6614 | int ret; | |
7d9eb12c | 6615 | |
f0486c68 | 6616 | BUG_ON(cache->ro); |
0ef3e66b | 6617 | |
ff5714cc | 6618 | trans = btrfs_join_transaction(root); |
f0486c68 | 6619 | BUG_ON(IS_ERR(trans)); |
5d4f98a2 | 6620 | |
f0486c68 YZ |
6621 | alloc_flags = update_block_group_flags(root, cache->flags); |
6622 | if (alloc_flags != cache->flags) | |
0e4f8f88 CM |
6623 | do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
6624 | CHUNK_ALLOC_FORCE); | |
5d4f98a2 | 6625 | |
199c36ea | 6626 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
6627 | if (!ret) |
6628 | goto out; | |
6629 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
0e4f8f88 CM |
6630 | ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
6631 | CHUNK_ALLOC_FORCE); | |
f0486c68 YZ |
6632 | if (ret < 0) |
6633 | goto out; | |
199c36ea | 6634 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
6635 | out: |
6636 | btrfs_end_transaction(trans, root); | |
6637 | return ret; | |
6638 | } | |
5d4f98a2 | 6639 | |
c87f08ca CM |
6640 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
6641 | struct btrfs_root *root, u64 type) | |
6642 | { | |
6643 | u64 alloc_flags = get_alloc_profile(root, type); | |
0e4f8f88 CM |
6644 | return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
6645 | CHUNK_ALLOC_FORCE); | |
c87f08ca CM |
6646 | } |
6647 | ||
6d07bcec MX |
6648 | /* |
6649 | * helper to account the unused space of all the readonly block group in the | |
6650 | * list. takes mirrors into account. | |
6651 | */ | |
6652 | static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list) | |
6653 | { | |
6654 | struct btrfs_block_group_cache *block_group; | |
6655 | u64 free_bytes = 0; | |
6656 | int factor; | |
6657 | ||
6658 | list_for_each_entry(block_group, groups_list, list) { | |
6659 | spin_lock(&block_group->lock); | |
6660 | ||
6661 | if (!block_group->ro) { | |
6662 | spin_unlock(&block_group->lock); | |
6663 | continue; | |
6664 | } | |
6665 | ||
6666 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
6667 | BTRFS_BLOCK_GROUP_RAID10 | | |
6668 | BTRFS_BLOCK_GROUP_DUP)) | |
6669 | factor = 2; | |
6670 | else | |
6671 | factor = 1; | |
6672 | ||
6673 | free_bytes += (block_group->key.offset - | |
6674 | btrfs_block_group_used(&block_group->item)) * | |
6675 | factor; | |
6676 | ||
6677 | spin_unlock(&block_group->lock); | |
6678 | } | |
6679 | ||
6680 | return free_bytes; | |
6681 | } | |
6682 | ||
6683 | /* | |
6684 | * helper to account the unused space of all the readonly block group in the | |
6685 | * space_info. takes mirrors into account. | |
6686 | */ | |
6687 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) | |
6688 | { | |
6689 | int i; | |
6690 | u64 free_bytes = 0; | |
6691 | ||
6692 | spin_lock(&sinfo->lock); | |
6693 | ||
6694 | for(i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
6695 | if (!list_empty(&sinfo->block_groups[i])) | |
6696 | free_bytes += __btrfs_get_ro_block_group_free_space( | |
6697 | &sinfo->block_groups[i]); | |
6698 | ||
6699 | spin_unlock(&sinfo->lock); | |
6700 | ||
6701 | return free_bytes; | |
6702 | } | |
6703 | ||
f0486c68 YZ |
6704 | int btrfs_set_block_group_rw(struct btrfs_root *root, |
6705 | struct btrfs_block_group_cache *cache) | |
5d4f98a2 | 6706 | { |
f0486c68 YZ |
6707 | struct btrfs_space_info *sinfo = cache->space_info; |
6708 | u64 num_bytes; | |
6709 | ||
6710 | BUG_ON(!cache->ro); | |
6711 | ||
6712 | spin_lock(&sinfo->lock); | |
6713 | spin_lock(&cache->lock); | |
6714 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
6715 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
6716 | sinfo->bytes_readonly -= num_bytes; | |
6717 | cache->ro = 0; | |
6718 | spin_unlock(&cache->lock); | |
6719 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
6720 | return 0; |
6721 | } | |
6722 | ||
ba1bf481 JB |
6723 | /* |
6724 | * checks to see if its even possible to relocate this block group. | |
6725 | * | |
6726 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
6727 | * ok to go ahead and try. | |
6728 | */ | |
6729 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 6730 | { |
ba1bf481 JB |
6731 | struct btrfs_block_group_cache *block_group; |
6732 | struct btrfs_space_info *space_info; | |
6733 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
6734 | struct btrfs_device *device; | |
cdcb725c | 6735 | u64 min_free; |
6719db6a JB |
6736 | u64 dev_min = 1; |
6737 | u64 dev_nr = 0; | |
cdcb725c | 6738 | int index; |
ba1bf481 JB |
6739 | int full = 0; |
6740 | int ret = 0; | |
1a40e23b | 6741 | |
ba1bf481 | 6742 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 6743 | |
ba1bf481 JB |
6744 | /* odd, couldn't find the block group, leave it alone */ |
6745 | if (!block_group) | |
6746 | return -1; | |
1a40e23b | 6747 | |
cdcb725c | 6748 | min_free = btrfs_block_group_used(&block_group->item); |
6749 | ||
ba1bf481 | 6750 | /* no bytes used, we're good */ |
cdcb725c | 6751 | if (!min_free) |
1a40e23b ZY |
6752 | goto out; |
6753 | ||
ba1bf481 JB |
6754 | space_info = block_group->space_info; |
6755 | spin_lock(&space_info->lock); | |
17d217fe | 6756 | |
ba1bf481 | 6757 | full = space_info->full; |
17d217fe | 6758 | |
ba1bf481 JB |
6759 | /* |
6760 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
6761 | * relocate it unless we're able to allocate a new chunk below. |
6762 | * | |
6763 | * Otherwise, we need to make sure we have room in the space to handle | |
6764 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 6765 | */ |
7ce618db | 6766 | if ((space_info->total_bytes != block_group->key.offset) && |
cdcb725c | 6767 | (space_info->bytes_used + space_info->bytes_reserved + |
6768 | space_info->bytes_pinned + space_info->bytes_readonly + | |
6769 | min_free < space_info->total_bytes)) { | |
ba1bf481 JB |
6770 | spin_unlock(&space_info->lock); |
6771 | goto out; | |
17d217fe | 6772 | } |
ba1bf481 | 6773 | spin_unlock(&space_info->lock); |
ea8c2819 | 6774 | |
ba1bf481 JB |
6775 | /* |
6776 | * ok we don't have enough space, but maybe we have free space on our | |
6777 | * devices to allocate new chunks for relocation, so loop through our | |
6778 | * alloc devices and guess if we have enough space. However, if we | |
6779 | * were marked as full, then we know there aren't enough chunks, and we | |
6780 | * can just return. | |
6781 | */ | |
6782 | ret = -1; | |
6783 | if (full) | |
6784 | goto out; | |
ea8c2819 | 6785 | |
cdcb725c | 6786 | /* |
6787 | * index: | |
6788 | * 0: raid10 | |
6789 | * 1: raid1 | |
6790 | * 2: dup | |
6791 | * 3: raid0 | |
6792 | * 4: single | |
6793 | */ | |
6794 | index = get_block_group_index(block_group); | |
6795 | if (index == 0) { | |
6796 | dev_min = 4; | |
6719db6a JB |
6797 | /* Divide by 2 */ |
6798 | min_free >>= 1; | |
cdcb725c | 6799 | } else if (index == 1) { |
6800 | dev_min = 2; | |
6801 | } else if (index == 2) { | |
6719db6a JB |
6802 | /* Multiply by 2 */ |
6803 | min_free <<= 1; | |
cdcb725c | 6804 | } else if (index == 3) { |
6805 | dev_min = fs_devices->rw_devices; | |
6719db6a | 6806 | do_div(min_free, dev_min); |
cdcb725c | 6807 | } |
6808 | ||
ba1bf481 JB |
6809 | mutex_lock(&root->fs_info->chunk_mutex); |
6810 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7bfc837d | 6811 | u64 dev_offset; |
56bec294 | 6812 | |
ba1bf481 JB |
6813 | /* |
6814 | * check to make sure we can actually find a chunk with enough | |
6815 | * space to fit our block group in. | |
6816 | */ | |
6817 | if (device->total_bytes > device->bytes_used + min_free) { | |
6818 | ret = find_free_dev_extent(NULL, device, min_free, | |
7bfc837d | 6819 | &dev_offset, NULL); |
ba1bf481 | 6820 | if (!ret) |
cdcb725c | 6821 | dev_nr++; |
6822 | ||
6823 | if (dev_nr >= dev_min) | |
73e48b27 | 6824 | break; |
cdcb725c | 6825 | |
ba1bf481 | 6826 | ret = -1; |
725c8463 | 6827 | } |
edbd8d4e | 6828 | } |
ba1bf481 | 6829 | mutex_unlock(&root->fs_info->chunk_mutex); |
edbd8d4e | 6830 | out: |
ba1bf481 | 6831 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
6832 | return ret; |
6833 | } | |
6834 | ||
b2950863 CH |
6835 | static int find_first_block_group(struct btrfs_root *root, |
6836 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 6837 | { |
925baedd | 6838 | int ret = 0; |
0b86a832 CM |
6839 | struct btrfs_key found_key; |
6840 | struct extent_buffer *leaf; | |
6841 | int slot; | |
edbd8d4e | 6842 | |
0b86a832 CM |
6843 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
6844 | if (ret < 0) | |
925baedd CM |
6845 | goto out; |
6846 | ||
d397712b | 6847 | while (1) { |
0b86a832 | 6848 | slot = path->slots[0]; |
edbd8d4e | 6849 | leaf = path->nodes[0]; |
0b86a832 CM |
6850 | if (slot >= btrfs_header_nritems(leaf)) { |
6851 | ret = btrfs_next_leaf(root, path); | |
6852 | if (ret == 0) | |
6853 | continue; | |
6854 | if (ret < 0) | |
925baedd | 6855 | goto out; |
0b86a832 | 6856 | break; |
edbd8d4e | 6857 | } |
0b86a832 | 6858 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 6859 | |
0b86a832 | 6860 | if (found_key.objectid >= key->objectid && |
925baedd CM |
6861 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6862 | ret = 0; | |
6863 | goto out; | |
6864 | } | |
0b86a832 | 6865 | path->slots[0]++; |
edbd8d4e | 6866 | } |
925baedd | 6867 | out: |
0b86a832 | 6868 | return ret; |
edbd8d4e CM |
6869 | } |
6870 | ||
0af3d00b JB |
6871 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
6872 | { | |
6873 | struct btrfs_block_group_cache *block_group; | |
6874 | u64 last = 0; | |
6875 | ||
6876 | while (1) { | |
6877 | struct inode *inode; | |
6878 | ||
6879 | block_group = btrfs_lookup_first_block_group(info, last); | |
6880 | while (block_group) { | |
6881 | spin_lock(&block_group->lock); | |
6882 | if (block_group->iref) | |
6883 | break; | |
6884 | spin_unlock(&block_group->lock); | |
6885 | block_group = next_block_group(info->tree_root, | |
6886 | block_group); | |
6887 | } | |
6888 | if (!block_group) { | |
6889 | if (last == 0) | |
6890 | break; | |
6891 | last = 0; | |
6892 | continue; | |
6893 | } | |
6894 | ||
6895 | inode = block_group->inode; | |
6896 | block_group->iref = 0; | |
6897 | block_group->inode = NULL; | |
6898 | spin_unlock(&block_group->lock); | |
6899 | iput(inode); | |
6900 | last = block_group->key.objectid + block_group->key.offset; | |
6901 | btrfs_put_block_group(block_group); | |
6902 | } | |
6903 | } | |
6904 | ||
1a40e23b ZY |
6905 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
6906 | { | |
6907 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 6908 | struct btrfs_space_info *space_info; |
11833d66 | 6909 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
6910 | struct rb_node *n; |
6911 | ||
11833d66 YZ |
6912 | down_write(&info->extent_commit_sem); |
6913 | while (!list_empty(&info->caching_block_groups)) { | |
6914 | caching_ctl = list_entry(info->caching_block_groups.next, | |
6915 | struct btrfs_caching_control, list); | |
6916 | list_del(&caching_ctl->list); | |
6917 | put_caching_control(caching_ctl); | |
6918 | } | |
6919 | up_write(&info->extent_commit_sem); | |
6920 | ||
1a40e23b ZY |
6921 | spin_lock(&info->block_group_cache_lock); |
6922 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
6923 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
6924 | cache_node); | |
1a40e23b ZY |
6925 | rb_erase(&block_group->cache_node, |
6926 | &info->block_group_cache_tree); | |
d899e052 YZ |
6927 | spin_unlock(&info->block_group_cache_lock); |
6928 | ||
80eb234a | 6929 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 6930 | list_del(&block_group->list); |
80eb234a | 6931 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 6932 | |
817d52f8 | 6933 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 6934 | wait_block_group_cache_done(block_group); |
817d52f8 | 6935 | |
3c14874a JB |
6936 | /* |
6937 | * We haven't cached this block group, which means we could | |
6938 | * possibly have excluded extents on this block group. | |
6939 | */ | |
6940 | if (block_group->cached == BTRFS_CACHE_NO) | |
6941 | free_excluded_extents(info->extent_root, block_group); | |
6942 | ||
817d52f8 | 6943 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 6944 | btrfs_put_block_group(block_group); |
d899e052 YZ |
6945 | |
6946 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
6947 | } |
6948 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
6949 | |
6950 | /* now that all the block groups are freed, go through and | |
6951 | * free all the space_info structs. This is only called during | |
6952 | * the final stages of unmount, and so we know nobody is | |
6953 | * using them. We call synchronize_rcu() once before we start, | |
6954 | * just to be on the safe side. | |
6955 | */ | |
6956 | synchronize_rcu(); | |
6957 | ||
8929ecfa YZ |
6958 | release_global_block_rsv(info); |
6959 | ||
4184ea7f CM |
6960 | while(!list_empty(&info->space_info)) { |
6961 | space_info = list_entry(info->space_info.next, | |
6962 | struct btrfs_space_info, | |
6963 | list); | |
f0486c68 YZ |
6964 | if (space_info->bytes_pinned > 0 || |
6965 | space_info->bytes_reserved > 0) { | |
6966 | WARN_ON(1); | |
6967 | dump_space_info(space_info, 0, 0); | |
6968 | } | |
4184ea7f CM |
6969 | list_del(&space_info->list); |
6970 | kfree(space_info); | |
6971 | } | |
1a40e23b ZY |
6972 | return 0; |
6973 | } | |
6974 | ||
b742bb82 YZ |
6975 | static void __link_block_group(struct btrfs_space_info *space_info, |
6976 | struct btrfs_block_group_cache *cache) | |
6977 | { | |
6978 | int index = get_block_group_index(cache); | |
6979 | ||
6980 | down_write(&space_info->groups_sem); | |
6981 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
6982 | up_write(&space_info->groups_sem); | |
6983 | } | |
6984 | ||
9078a3e1 CM |
6985 | int btrfs_read_block_groups(struct btrfs_root *root) |
6986 | { | |
6987 | struct btrfs_path *path; | |
6988 | int ret; | |
9078a3e1 | 6989 | struct btrfs_block_group_cache *cache; |
be744175 | 6990 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 6991 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
6992 | struct btrfs_key key; |
6993 | struct btrfs_key found_key; | |
5f39d397 | 6994 | struct extent_buffer *leaf; |
0af3d00b JB |
6995 | int need_clear = 0; |
6996 | u64 cache_gen; | |
96b5179d | 6997 | |
be744175 | 6998 | root = info->extent_root; |
9078a3e1 | 6999 | key.objectid = 0; |
0b86a832 | 7000 | key.offset = 0; |
9078a3e1 | 7001 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
7002 | path = btrfs_alloc_path(); |
7003 | if (!path) | |
7004 | return -ENOMEM; | |
026fd317 | 7005 | path->reada = 1; |
9078a3e1 | 7006 | |
0af3d00b JB |
7007 | cache_gen = btrfs_super_cache_generation(&root->fs_info->super_copy); |
7008 | if (cache_gen != 0 && | |
7009 | btrfs_super_generation(&root->fs_info->super_copy) != cache_gen) | |
7010 | need_clear = 1; | |
88c2ba3b JB |
7011 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
7012 | need_clear = 1; | |
8216ef86 JB |
7013 | if (!btrfs_test_opt(root, SPACE_CACHE) && cache_gen) |
7014 | printk(KERN_INFO "btrfs: disk space caching is enabled\n"); | |
0af3d00b | 7015 | |
d397712b | 7016 | while (1) { |
0b86a832 | 7017 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
7018 | if (ret > 0) |
7019 | break; | |
0b86a832 CM |
7020 | if (ret != 0) |
7021 | goto error; | |
5f39d397 CM |
7022 | leaf = path->nodes[0]; |
7023 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 7024 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 7025 | if (!cache) { |
0b86a832 | 7026 | ret = -ENOMEM; |
f0486c68 | 7027 | goto error; |
9078a3e1 | 7028 | } |
34d52cb6 LZ |
7029 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
7030 | GFP_NOFS); | |
7031 | if (!cache->free_space_ctl) { | |
7032 | kfree(cache); | |
7033 | ret = -ENOMEM; | |
7034 | goto error; | |
7035 | } | |
3e1ad54f | 7036 | |
d2fb3437 | 7037 | atomic_set(&cache->count, 1); |
c286ac48 | 7038 | spin_lock_init(&cache->lock); |
817d52f8 | 7039 | cache->fs_info = info; |
0f9dd46c | 7040 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7041 | INIT_LIST_HEAD(&cache->cluster_list); |
96303081 | 7042 | |
0af3d00b JB |
7043 | if (need_clear) |
7044 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
7045 | ||
5f39d397 CM |
7046 | read_extent_buffer(leaf, &cache->item, |
7047 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
7048 | sizeof(cache->item)); | |
9078a3e1 | 7049 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 7050 | |
9078a3e1 | 7051 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 7052 | btrfs_release_path(path); |
0b86a832 | 7053 | cache->flags = btrfs_block_group_flags(&cache->item); |
817d52f8 JB |
7054 | cache->sectorsize = root->sectorsize; |
7055 | ||
34d52cb6 LZ |
7056 | btrfs_init_free_space_ctl(cache); |
7057 | ||
3c14874a JB |
7058 | /* |
7059 | * We need to exclude the super stripes now so that the space | |
7060 | * info has super bytes accounted for, otherwise we'll think | |
7061 | * we have more space than we actually do. | |
7062 | */ | |
7063 | exclude_super_stripes(root, cache); | |
7064 | ||
817d52f8 JB |
7065 | /* |
7066 | * check for two cases, either we are full, and therefore | |
7067 | * don't need to bother with the caching work since we won't | |
7068 | * find any space, or we are empty, and we can just add all | |
7069 | * the space in and be done with it. This saves us _alot_ of | |
7070 | * time, particularly in the full case. | |
7071 | */ | |
7072 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 7073 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7074 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 7075 | free_excluded_extents(root, cache); |
817d52f8 | 7076 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 7077 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
7078 | cache->cached = BTRFS_CACHE_FINISHED; |
7079 | add_new_free_space(cache, root->fs_info, | |
7080 | found_key.objectid, | |
7081 | found_key.objectid + | |
7082 | found_key.offset); | |
11833d66 | 7083 | free_excluded_extents(root, cache); |
817d52f8 | 7084 | } |
96b5179d | 7085 | |
6324fbf3 CM |
7086 | ret = update_space_info(info, cache->flags, found_key.offset, |
7087 | btrfs_block_group_used(&cache->item), | |
7088 | &space_info); | |
7089 | BUG_ON(ret); | |
7090 | cache->space_info = space_info; | |
1b2da372 | 7091 | spin_lock(&cache->space_info->lock); |
f0486c68 | 7092 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7093 | spin_unlock(&cache->space_info->lock); |
7094 | ||
b742bb82 | 7095 | __link_block_group(space_info, cache); |
0f9dd46c JB |
7096 | |
7097 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
7098 | BUG_ON(ret); | |
75ccf47d CM |
7099 | |
7100 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
2b82032c | 7101 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
199c36ea | 7102 | set_block_group_ro(cache, 1); |
9078a3e1 | 7103 | } |
b742bb82 YZ |
7104 | |
7105 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
7106 | if (!(get_alloc_profile(root, space_info->flags) & | |
7107 | (BTRFS_BLOCK_GROUP_RAID10 | | |
7108 | BTRFS_BLOCK_GROUP_RAID1 | | |
7109 | BTRFS_BLOCK_GROUP_DUP))) | |
7110 | continue; | |
7111 | /* | |
7112 | * avoid allocating from un-mirrored block group if there are | |
7113 | * mirrored block groups. | |
7114 | */ | |
7115 | list_for_each_entry(cache, &space_info->block_groups[3], list) | |
199c36ea | 7116 | set_block_group_ro(cache, 1); |
b742bb82 | 7117 | list_for_each_entry(cache, &space_info->block_groups[4], list) |
199c36ea | 7118 | set_block_group_ro(cache, 1); |
9078a3e1 | 7119 | } |
f0486c68 YZ |
7120 | |
7121 | init_global_block_rsv(info); | |
0b86a832 CM |
7122 | ret = 0; |
7123 | error: | |
9078a3e1 | 7124 | btrfs_free_path(path); |
0b86a832 | 7125 | return ret; |
9078a3e1 | 7126 | } |
6324fbf3 CM |
7127 | |
7128 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
7129 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 7130 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
7131 | u64 size) |
7132 | { | |
7133 | int ret; | |
6324fbf3 CM |
7134 | struct btrfs_root *extent_root; |
7135 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
7136 | |
7137 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 7138 | |
12fcfd22 | 7139 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 7140 | |
8f18cf13 | 7141 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
7142 | if (!cache) |
7143 | return -ENOMEM; | |
34d52cb6 LZ |
7144 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
7145 | GFP_NOFS); | |
7146 | if (!cache->free_space_ctl) { | |
7147 | kfree(cache); | |
7148 | return -ENOMEM; | |
7149 | } | |
0f9dd46c | 7150 | |
e17cade2 | 7151 | cache->key.objectid = chunk_offset; |
6324fbf3 | 7152 | cache->key.offset = size; |
d2fb3437 | 7153 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
96303081 | 7154 | cache->sectorsize = root->sectorsize; |
0af3d00b | 7155 | cache->fs_info = root->fs_info; |
96303081 | 7156 | |
d2fb3437 | 7157 | atomic_set(&cache->count, 1); |
c286ac48 | 7158 | spin_lock_init(&cache->lock); |
0f9dd46c | 7159 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7160 | INIT_LIST_HEAD(&cache->cluster_list); |
0ef3e66b | 7161 | |
34d52cb6 LZ |
7162 | btrfs_init_free_space_ctl(cache); |
7163 | ||
6324fbf3 | 7164 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
7165 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
7166 | cache->flags = type; | |
7167 | btrfs_set_block_group_flags(&cache->item, type); | |
7168 | ||
11833d66 | 7169 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7170 | cache->cached = BTRFS_CACHE_FINISHED; |
11833d66 | 7171 | exclude_super_stripes(root, cache); |
96303081 | 7172 | |
817d52f8 JB |
7173 | add_new_free_space(cache, root->fs_info, chunk_offset, |
7174 | chunk_offset + size); | |
7175 | ||
11833d66 YZ |
7176 | free_excluded_extents(root, cache); |
7177 | ||
6324fbf3 CM |
7178 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
7179 | &cache->space_info); | |
7180 | BUG_ON(ret); | |
1b2da372 JB |
7181 | |
7182 | spin_lock(&cache->space_info->lock); | |
f0486c68 | 7183 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7184 | spin_unlock(&cache->space_info->lock); |
7185 | ||
b742bb82 | 7186 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 7187 | |
0f9dd46c JB |
7188 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
7189 | BUG_ON(ret); | |
c286ac48 | 7190 | |
6324fbf3 CM |
7191 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
7192 | sizeof(cache->item)); | |
7193 | BUG_ON(ret); | |
7194 | ||
d18a2c44 | 7195 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 7196 | |
6324fbf3 CM |
7197 | return 0; |
7198 | } | |
1a40e23b ZY |
7199 | |
7200 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, | |
7201 | struct btrfs_root *root, u64 group_start) | |
7202 | { | |
7203 | struct btrfs_path *path; | |
7204 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 7205 | struct btrfs_free_cluster *cluster; |
0af3d00b | 7206 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 7207 | struct btrfs_key key; |
0af3d00b | 7208 | struct inode *inode; |
1a40e23b | 7209 | int ret; |
89a55897 | 7210 | int factor; |
1a40e23b | 7211 | |
1a40e23b ZY |
7212 | root = root->fs_info->extent_root; |
7213 | ||
7214 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
7215 | BUG_ON(!block_group); | |
c146afad | 7216 | BUG_ON(!block_group->ro); |
1a40e23b | 7217 | |
9f7c43c9 | 7218 | /* |
7219 | * Free the reserved super bytes from this block group before | |
7220 | * remove it. | |
7221 | */ | |
7222 | free_excluded_extents(root, block_group); | |
7223 | ||
1a40e23b | 7224 | memcpy(&key, &block_group->key, sizeof(key)); |
89a55897 JB |
7225 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
7226 | BTRFS_BLOCK_GROUP_RAID1 | | |
7227 | BTRFS_BLOCK_GROUP_RAID10)) | |
7228 | factor = 2; | |
7229 | else | |
7230 | factor = 1; | |
1a40e23b | 7231 | |
44fb5511 CM |
7232 | /* make sure this block group isn't part of an allocation cluster */ |
7233 | cluster = &root->fs_info->data_alloc_cluster; | |
7234 | spin_lock(&cluster->refill_lock); | |
7235 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7236 | spin_unlock(&cluster->refill_lock); | |
7237 | ||
7238 | /* | |
7239 | * make sure this block group isn't part of a metadata | |
7240 | * allocation cluster | |
7241 | */ | |
7242 | cluster = &root->fs_info->meta_alloc_cluster; | |
7243 | spin_lock(&cluster->refill_lock); | |
7244 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7245 | spin_unlock(&cluster->refill_lock); | |
7246 | ||
1a40e23b | 7247 | path = btrfs_alloc_path(); |
d8926bb3 MF |
7248 | if (!path) { |
7249 | ret = -ENOMEM; | |
7250 | goto out; | |
7251 | } | |
1a40e23b | 7252 | |
0af3d00b JB |
7253 | inode = lookup_free_space_inode(root, block_group, path); |
7254 | if (!IS_ERR(inode)) { | |
b532402e TI |
7255 | ret = btrfs_orphan_add(trans, inode); |
7256 | BUG_ON(ret); | |
0af3d00b JB |
7257 | clear_nlink(inode); |
7258 | /* One for the block groups ref */ | |
7259 | spin_lock(&block_group->lock); | |
7260 | if (block_group->iref) { | |
7261 | block_group->iref = 0; | |
7262 | block_group->inode = NULL; | |
7263 | spin_unlock(&block_group->lock); | |
7264 | iput(inode); | |
7265 | } else { | |
7266 | spin_unlock(&block_group->lock); | |
7267 | } | |
7268 | /* One for our lookup ref */ | |
7269 | iput(inode); | |
7270 | } | |
7271 | ||
7272 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
7273 | key.offset = block_group->key.objectid; | |
7274 | key.type = 0; | |
7275 | ||
7276 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
7277 | if (ret < 0) | |
7278 | goto out; | |
7279 | if (ret > 0) | |
b3b4aa74 | 7280 | btrfs_release_path(path); |
0af3d00b JB |
7281 | if (ret == 0) { |
7282 | ret = btrfs_del_item(trans, tree_root, path); | |
7283 | if (ret) | |
7284 | goto out; | |
b3b4aa74 | 7285 | btrfs_release_path(path); |
0af3d00b JB |
7286 | } |
7287 | ||
3dfdb934 | 7288 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
7289 | rb_erase(&block_group->cache_node, |
7290 | &root->fs_info->block_group_cache_tree); | |
3dfdb934 | 7291 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 7292 | |
80eb234a | 7293 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
7294 | /* |
7295 | * we must use list_del_init so people can check to see if they | |
7296 | * are still on the list after taking the semaphore | |
7297 | */ | |
7298 | list_del_init(&block_group->list); | |
80eb234a | 7299 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 7300 | |
817d52f8 | 7301 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 7302 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
7303 | |
7304 | btrfs_remove_free_space_cache(block_group); | |
7305 | ||
c146afad YZ |
7306 | spin_lock(&block_group->space_info->lock); |
7307 | block_group->space_info->total_bytes -= block_group->key.offset; | |
7308 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 7309 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
c146afad | 7310 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 7311 | |
0af3d00b JB |
7312 | memcpy(&key, &block_group->key, sizeof(key)); |
7313 | ||
283bb197 | 7314 | btrfs_clear_space_info_full(root->fs_info); |
c146afad | 7315 | |
fa9c0d79 CM |
7316 | btrfs_put_block_group(block_group); |
7317 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
7318 | |
7319 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
7320 | if (ret > 0) | |
7321 | ret = -EIO; | |
7322 | if (ret < 0) | |
7323 | goto out; | |
7324 | ||
7325 | ret = btrfs_del_item(trans, root, path); | |
7326 | out: | |
7327 | btrfs_free_path(path); | |
7328 | return ret; | |
7329 | } | |
acce952b | 7330 | |
c59021f8 | 7331 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
7332 | { | |
7333 | struct btrfs_space_info *space_info; | |
1aba86d6 | 7334 | struct btrfs_super_block *disk_super; |
7335 | u64 features; | |
7336 | u64 flags; | |
7337 | int mixed = 0; | |
c59021f8 | 7338 | int ret; |
7339 | ||
1aba86d6 | 7340 | disk_super = &fs_info->super_copy; |
7341 | if (!btrfs_super_root(disk_super)) | |
7342 | return 1; | |
c59021f8 | 7343 | |
1aba86d6 | 7344 | features = btrfs_super_incompat_flags(disk_super); |
7345 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
7346 | mixed = 1; | |
c59021f8 | 7347 | |
1aba86d6 | 7348 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
7349 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
c59021f8 | 7350 | if (ret) |
1aba86d6 | 7351 | goto out; |
c59021f8 | 7352 | |
1aba86d6 | 7353 | if (mixed) { |
7354 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
7355 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7356 | } else { | |
7357 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
7358 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7359 | if (ret) | |
7360 | goto out; | |
7361 | ||
7362 | flags = BTRFS_BLOCK_GROUP_DATA; | |
7363 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
7364 | } | |
7365 | out: | |
c59021f8 | 7366 | return ret; |
7367 | } | |
7368 | ||
acce952b | 7369 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
7370 | { | |
7371 | return unpin_extent_range(root, start, end); | |
7372 | } | |
7373 | ||
7374 | int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 7375 | u64 num_bytes, u64 *actual_bytes) |
acce952b | 7376 | { |
5378e607 | 7377 | return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes); |
acce952b | 7378 | } |
f7039b1d LD |
7379 | |
7380 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) | |
7381 | { | |
7382 | struct btrfs_fs_info *fs_info = root->fs_info; | |
7383 | struct btrfs_block_group_cache *cache = NULL; | |
7384 | u64 group_trimmed; | |
7385 | u64 start; | |
7386 | u64 end; | |
7387 | u64 trimmed = 0; | |
7388 | int ret = 0; | |
7389 | ||
7390 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
7391 | ||
7392 | while (cache) { | |
7393 | if (cache->key.objectid >= (range->start + range->len)) { | |
7394 | btrfs_put_block_group(cache); | |
7395 | break; | |
7396 | } | |
7397 | ||
7398 | start = max(range->start, cache->key.objectid); | |
7399 | end = min(range->start + range->len, | |
7400 | cache->key.objectid + cache->key.offset); | |
7401 | ||
7402 | if (end - start >= range->minlen) { | |
7403 | if (!block_group_cache_done(cache)) { | |
7404 | ret = cache_block_group(cache, NULL, root, 0); | |
7405 | if (!ret) | |
7406 | wait_block_group_cache_done(cache); | |
7407 | } | |
7408 | ret = btrfs_trim_block_group(cache, | |
7409 | &group_trimmed, | |
7410 | start, | |
7411 | end, | |
7412 | range->minlen); | |
7413 | ||
7414 | trimmed += group_trimmed; | |
7415 | if (ret) { | |
7416 | btrfs_put_block_group(cache); | |
7417 | break; | |
7418 | } | |
7419 | } | |
7420 | ||
7421 | cache = next_block_group(fs_info->tree_root, cache); | |
7422 | } | |
7423 | ||
7424 | range->len = trimmed; | |
7425 | return ret; | |
7426 | } |