Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
ec6b910f | 18 | #include <linux/sched.h> |
edbd8d4e | 19 | #include <linux/pagemap.h> |
ec44a35c | 20 | #include <linux/writeback.h> |
21af804c | 21 | #include <linux/blkdev.h> |
b7a9f29f | 22 | #include <linux/sort.h> |
4184ea7f | 23 | #include <linux/rcupdate.h> |
817d52f8 | 24 | #include <linux/kthread.h> |
5a0e3ad6 | 25 | #include <linux/slab.h> |
dff51cd1 | 26 | #include <linux/ratelimit.h> |
4b4e25f2 | 27 | #include "compat.h" |
74493f7a | 28 | #include "hash.h" |
fec577fb CM |
29 | #include "ctree.h" |
30 | #include "disk-io.h" | |
31 | #include "print-tree.h" | |
e089f05c | 32 | #include "transaction.h" |
0b86a832 | 33 | #include "volumes.h" |
925baedd | 34 | #include "locking.h" |
fa9c0d79 | 35 | #include "free-space-cache.h" |
fec577fb | 36 | |
709c0486 AJ |
37 | #undef SCRAMBLE_DELAYED_REFS |
38 | ||
9e622d6b MX |
39 | /* |
40 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
41 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
42 | * if we really need one. | |
43 | * | |
0e4f8f88 CM |
44 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
45 | * if we have very few chunks already allocated. This is | |
46 | * used as part of the clustering code to help make sure | |
47 | * we have a good pool of storage to cluster in, without | |
48 | * filling the FS with empty chunks | |
49 | * | |
9e622d6b MX |
50 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
51 | * | |
0e4f8f88 CM |
52 | */ |
53 | enum { | |
54 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
55 | CHUNK_ALLOC_LIMITED = 1, |
56 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
57 | }; |
58 | ||
fb25e914 JB |
59 | /* |
60 | * Control how reservations are dealt with. | |
61 | * | |
62 | * RESERVE_FREE - freeing a reservation. | |
63 | * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for | |
64 | * ENOSPC accounting | |
65 | * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update | |
66 | * bytes_may_use as the ENOSPC accounting is done elsewhere | |
67 | */ | |
68 | enum { | |
69 | RESERVE_FREE = 0, | |
70 | RESERVE_ALLOC = 1, | |
71 | RESERVE_ALLOC_NO_ACCOUNT = 2, | |
72 | }; | |
73 | ||
f3465ca4 JB |
74 | static int update_block_group(struct btrfs_trans_handle *trans, |
75 | struct btrfs_root *root, | |
f0486c68 | 76 | u64 bytenr, u64 num_bytes, int alloc); |
5d4f98a2 YZ |
77 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
78 | struct btrfs_root *root, | |
79 | u64 bytenr, u64 num_bytes, u64 parent, | |
80 | u64 root_objectid, u64 owner_objectid, | |
81 | u64 owner_offset, int refs_to_drop, | |
82 | struct btrfs_delayed_extent_op *extra_op); | |
83 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
84 | struct extent_buffer *leaf, | |
85 | struct btrfs_extent_item *ei); | |
86 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
87 | struct btrfs_root *root, | |
88 | u64 parent, u64 root_objectid, | |
89 | u64 flags, u64 owner, u64 offset, | |
90 | struct btrfs_key *ins, int ref_mod); | |
91 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
92 | struct btrfs_root *root, | |
93 | u64 parent, u64 root_objectid, | |
94 | u64 flags, struct btrfs_disk_key *key, | |
95 | int level, struct btrfs_key *ins); | |
6a63209f JB |
96 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
97 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
98 | u64 flags, int force); | |
11833d66 YZ |
99 | static int find_next_key(struct btrfs_path *path, int level, |
100 | struct btrfs_key *key); | |
9ed74f2d JB |
101 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
102 | int dump_block_groups); | |
fb25e914 JB |
103 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
104 | u64 num_bytes, int reserve); | |
6a63209f | 105 | |
817d52f8 JB |
106 | static noinline int |
107 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
108 | { | |
109 | smp_mb(); | |
110 | return cache->cached == BTRFS_CACHE_FINISHED; | |
111 | } | |
112 | ||
0f9dd46c JB |
113 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
114 | { | |
115 | return (cache->flags & bits) == bits; | |
116 | } | |
117 | ||
62a45b60 | 118 | static void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
119 | { |
120 | atomic_inc(&cache->count); | |
121 | } | |
122 | ||
123 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
124 | { | |
f0486c68 YZ |
125 | if (atomic_dec_and_test(&cache->count)) { |
126 | WARN_ON(cache->pinned > 0); | |
127 | WARN_ON(cache->reserved > 0); | |
34d52cb6 | 128 | kfree(cache->free_space_ctl); |
11dfe35a | 129 | kfree(cache); |
f0486c68 | 130 | } |
11dfe35a JB |
131 | } |
132 | ||
0f9dd46c JB |
133 | /* |
134 | * this adds the block group to the fs_info rb tree for the block group | |
135 | * cache | |
136 | */ | |
b2950863 | 137 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
138 | struct btrfs_block_group_cache *block_group) |
139 | { | |
140 | struct rb_node **p; | |
141 | struct rb_node *parent = NULL; | |
142 | struct btrfs_block_group_cache *cache; | |
143 | ||
144 | spin_lock(&info->block_group_cache_lock); | |
145 | p = &info->block_group_cache_tree.rb_node; | |
146 | ||
147 | while (*p) { | |
148 | parent = *p; | |
149 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
150 | cache_node); | |
151 | if (block_group->key.objectid < cache->key.objectid) { | |
152 | p = &(*p)->rb_left; | |
153 | } else if (block_group->key.objectid > cache->key.objectid) { | |
154 | p = &(*p)->rb_right; | |
155 | } else { | |
156 | spin_unlock(&info->block_group_cache_lock); | |
157 | return -EEXIST; | |
158 | } | |
159 | } | |
160 | ||
161 | rb_link_node(&block_group->cache_node, parent, p); | |
162 | rb_insert_color(&block_group->cache_node, | |
163 | &info->block_group_cache_tree); | |
164 | spin_unlock(&info->block_group_cache_lock); | |
165 | ||
166 | return 0; | |
167 | } | |
168 | ||
169 | /* | |
170 | * This will return the block group at or after bytenr if contains is 0, else | |
171 | * it will return the block group that contains the bytenr | |
172 | */ | |
173 | static struct btrfs_block_group_cache * | |
174 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
175 | int contains) | |
176 | { | |
177 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
178 | struct rb_node *n; | |
179 | u64 end, start; | |
180 | ||
181 | spin_lock(&info->block_group_cache_lock); | |
182 | n = info->block_group_cache_tree.rb_node; | |
183 | ||
184 | while (n) { | |
185 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
186 | cache_node); | |
187 | end = cache->key.objectid + cache->key.offset - 1; | |
188 | start = cache->key.objectid; | |
189 | ||
190 | if (bytenr < start) { | |
191 | if (!contains && (!ret || start < ret->key.objectid)) | |
192 | ret = cache; | |
193 | n = n->rb_left; | |
194 | } else if (bytenr > start) { | |
195 | if (contains && bytenr <= end) { | |
196 | ret = cache; | |
197 | break; | |
198 | } | |
199 | n = n->rb_right; | |
200 | } else { | |
201 | ret = cache; | |
202 | break; | |
203 | } | |
204 | } | |
d2fb3437 | 205 | if (ret) |
11dfe35a | 206 | btrfs_get_block_group(ret); |
0f9dd46c JB |
207 | spin_unlock(&info->block_group_cache_lock); |
208 | ||
209 | return ret; | |
210 | } | |
211 | ||
11833d66 YZ |
212 | static int add_excluded_extent(struct btrfs_root *root, |
213 | u64 start, u64 num_bytes) | |
817d52f8 | 214 | { |
11833d66 YZ |
215 | u64 end = start + num_bytes - 1; |
216 | set_extent_bits(&root->fs_info->freed_extents[0], | |
217 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
218 | set_extent_bits(&root->fs_info->freed_extents[1], | |
219 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
220 | return 0; | |
221 | } | |
817d52f8 | 222 | |
11833d66 YZ |
223 | static void free_excluded_extents(struct btrfs_root *root, |
224 | struct btrfs_block_group_cache *cache) | |
225 | { | |
226 | u64 start, end; | |
817d52f8 | 227 | |
11833d66 YZ |
228 | start = cache->key.objectid; |
229 | end = start + cache->key.offset - 1; | |
230 | ||
231 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
232 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
233 | clear_extent_bits(&root->fs_info->freed_extents[1], | |
234 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
817d52f8 JB |
235 | } |
236 | ||
11833d66 YZ |
237 | static int exclude_super_stripes(struct btrfs_root *root, |
238 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 239 | { |
817d52f8 JB |
240 | u64 bytenr; |
241 | u64 *logical; | |
242 | int stripe_len; | |
243 | int i, nr, ret; | |
244 | ||
06b2331f YZ |
245 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
246 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
247 | cache->bytes_super += stripe_len; | |
248 | ret = add_excluded_extent(root, cache->key.objectid, | |
249 | stripe_len); | |
79787eaa | 250 | BUG_ON(ret); /* -ENOMEM */ |
06b2331f YZ |
251 | } |
252 | ||
817d52f8 JB |
253 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
254 | bytenr = btrfs_sb_offset(i); | |
255 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
256 | cache->key.objectid, bytenr, | |
257 | 0, &logical, &nr, &stripe_len); | |
79787eaa | 258 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 | 259 | |
817d52f8 | 260 | while (nr--) { |
1b2da372 | 261 | cache->bytes_super += stripe_len; |
11833d66 YZ |
262 | ret = add_excluded_extent(root, logical[nr], |
263 | stripe_len); | |
79787eaa | 264 | BUG_ON(ret); /* -ENOMEM */ |
817d52f8 | 265 | } |
11833d66 | 266 | |
817d52f8 JB |
267 | kfree(logical); |
268 | } | |
817d52f8 JB |
269 | return 0; |
270 | } | |
271 | ||
11833d66 YZ |
272 | static struct btrfs_caching_control * |
273 | get_caching_control(struct btrfs_block_group_cache *cache) | |
274 | { | |
275 | struct btrfs_caching_control *ctl; | |
276 | ||
277 | spin_lock(&cache->lock); | |
278 | if (cache->cached != BTRFS_CACHE_STARTED) { | |
279 | spin_unlock(&cache->lock); | |
280 | return NULL; | |
281 | } | |
282 | ||
dde5abee JB |
283 | /* We're loading it the fast way, so we don't have a caching_ctl. */ |
284 | if (!cache->caching_ctl) { | |
285 | spin_unlock(&cache->lock); | |
11833d66 YZ |
286 | return NULL; |
287 | } | |
288 | ||
289 | ctl = cache->caching_ctl; | |
290 | atomic_inc(&ctl->count); | |
291 | spin_unlock(&cache->lock); | |
292 | return ctl; | |
293 | } | |
294 | ||
295 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
296 | { | |
297 | if (atomic_dec_and_test(&ctl->count)) | |
298 | kfree(ctl); | |
299 | } | |
300 | ||
0f9dd46c JB |
301 | /* |
302 | * this is only called by cache_block_group, since we could have freed extents | |
303 | * we need to check the pinned_extents for any extents that can't be used yet | |
304 | * since their free space will be released as soon as the transaction commits. | |
305 | */ | |
817d52f8 | 306 | static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
0f9dd46c JB |
307 | struct btrfs_fs_info *info, u64 start, u64 end) |
308 | { | |
817d52f8 | 309 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
310 | int ret; |
311 | ||
312 | while (start < end) { | |
11833d66 | 313 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 314 | &extent_start, &extent_end, |
11833d66 | 315 | EXTENT_DIRTY | EXTENT_UPTODATE); |
0f9dd46c JB |
316 | if (ret) |
317 | break; | |
318 | ||
06b2331f | 319 | if (extent_start <= start) { |
0f9dd46c JB |
320 | start = extent_end + 1; |
321 | } else if (extent_start > start && extent_start < end) { | |
322 | size = extent_start - start; | |
817d52f8 | 323 | total_added += size; |
ea6a478e JB |
324 | ret = btrfs_add_free_space(block_group, start, |
325 | size); | |
79787eaa | 326 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
327 | start = extent_end + 1; |
328 | } else { | |
329 | break; | |
330 | } | |
331 | } | |
332 | ||
333 | if (start < end) { | |
334 | size = end - start; | |
817d52f8 | 335 | total_added += size; |
ea6a478e | 336 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 337 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
338 | } |
339 | ||
817d52f8 | 340 | return total_added; |
0f9dd46c JB |
341 | } |
342 | ||
bab39bf9 | 343 | static noinline void caching_thread(struct btrfs_work *work) |
e37c9e69 | 344 | { |
bab39bf9 JB |
345 | struct btrfs_block_group_cache *block_group; |
346 | struct btrfs_fs_info *fs_info; | |
347 | struct btrfs_caching_control *caching_ctl; | |
348 | struct btrfs_root *extent_root; | |
e37c9e69 | 349 | struct btrfs_path *path; |
5f39d397 | 350 | struct extent_buffer *leaf; |
11833d66 | 351 | struct btrfs_key key; |
817d52f8 | 352 | u64 total_found = 0; |
11833d66 YZ |
353 | u64 last = 0; |
354 | u32 nritems; | |
355 | int ret = 0; | |
f510cfec | 356 | |
bab39bf9 JB |
357 | caching_ctl = container_of(work, struct btrfs_caching_control, work); |
358 | block_group = caching_ctl->block_group; | |
359 | fs_info = block_group->fs_info; | |
360 | extent_root = fs_info->extent_root; | |
361 | ||
e37c9e69 CM |
362 | path = btrfs_alloc_path(); |
363 | if (!path) | |
bab39bf9 | 364 | goto out; |
7d7d6068 | 365 | |
817d52f8 | 366 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 367 | |
5cd57b2c | 368 | /* |
817d52f8 JB |
369 | * We don't want to deadlock with somebody trying to allocate a new |
370 | * extent for the extent root while also trying to search the extent | |
371 | * root to add free space. So we skip locking and search the commit | |
372 | * root, since its read-only | |
5cd57b2c CM |
373 | */ |
374 | path->skip_locking = 1; | |
817d52f8 | 375 | path->search_commit_root = 1; |
026fd317 | 376 | path->reada = 1; |
817d52f8 | 377 | |
e4404d6e | 378 | key.objectid = last; |
e37c9e69 | 379 | key.offset = 0; |
11833d66 | 380 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 381 | again: |
11833d66 | 382 | mutex_lock(&caching_ctl->mutex); |
013f1b12 CM |
383 | /* need to make sure the commit_root doesn't disappear */ |
384 | down_read(&fs_info->extent_commit_sem); | |
385 | ||
11833d66 | 386 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 387 | if (ret < 0) |
ef8bbdfe | 388 | goto err; |
a512bbf8 | 389 | |
11833d66 YZ |
390 | leaf = path->nodes[0]; |
391 | nritems = btrfs_header_nritems(leaf); | |
392 | ||
d397712b | 393 | while (1) { |
7841cb28 | 394 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 395 | last = (u64)-1; |
817d52f8 | 396 | break; |
f25784b3 | 397 | } |
817d52f8 | 398 | |
11833d66 YZ |
399 | if (path->slots[0] < nritems) { |
400 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
401 | } else { | |
402 | ret = find_next_key(path, 0, &key); | |
403 | if (ret) | |
e37c9e69 | 404 | break; |
817d52f8 | 405 | |
589d8ade JB |
406 | if (need_resched() || |
407 | btrfs_next_leaf(extent_root, path)) { | |
408 | caching_ctl->progress = last; | |
ff5714cc | 409 | btrfs_release_path(path); |
589d8ade JB |
410 | up_read(&fs_info->extent_commit_sem); |
411 | mutex_unlock(&caching_ctl->mutex); | |
11833d66 | 412 | cond_resched(); |
589d8ade JB |
413 | goto again; |
414 | } | |
415 | leaf = path->nodes[0]; | |
416 | nritems = btrfs_header_nritems(leaf); | |
417 | continue; | |
11833d66 | 418 | } |
817d52f8 | 419 | |
11833d66 YZ |
420 | if (key.objectid < block_group->key.objectid) { |
421 | path->slots[0]++; | |
817d52f8 | 422 | continue; |
e37c9e69 | 423 | } |
0f9dd46c | 424 | |
e37c9e69 | 425 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 426 | block_group->key.offset) |
e37c9e69 | 427 | break; |
7d7d6068 | 428 | |
11833d66 | 429 | if (key.type == BTRFS_EXTENT_ITEM_KEY) { |
817d52f8 JB |
430 | total_found += add_new_free_space(block_group, |
431 | fs_info, last, | |
432 | key.objectid); | |
7d7d6068 | 433 | last = key.objectid + key.offset; |
817d52f8 | 434 | |
11833d66 YZ |
435 | if (total_found > (1024 * 1024 * 2)) { |
436 | total_found = 0; | |
437 | wake_up(&caching_ctl->wait); | |
438 | } | |
817d52f8 | 439 | } |
e37c9e69 CM |
440 | path->slots[0]++; |
441 | } | |
817d52f8 | 442 | ret = 0; |
e37c9e69 | 443 | |
817d52f8 JB |
444 | total_found += add_new_free_space(block_group, fs_info, last, |
445 | block_group->key.objectid + | |
446 | block_group->key.offset); | |
11833d66 | 447 | caching_ctl->progress = (u64)-1; |
817d52f8 JB |
448 | |
449 | spin_lock(&block_group->lock); | |
11833d66 | 450 | block_group->caching_ctl = NULL; |
817d52f8 JB |
451 | block_group->cached = BTRFS_CACHE_FINISHED; |
452 | spin_unlock(&block_group->lock); | |
0f9dd46c | 453 | |
54aa1f4d | 454 | err: |
e37c9e69 | 455 | btrfs_free_path(path); |
276e680d | 456 | up_read(&fs_info->extent_commit_sem); |
817d52f8 | 457 | |
11833d66 YZ |
458 | free_excluded_extents(extent_root, block_group); |
459 | ||
460 | mutex_unlock(&caching_ctl->mutex); | |
bab39bf9 | 461 | out: |
11833d66 YZ |
462 | wake_up(&caching_ctl->wait); |
463 | ||
464 | put_caching_control(caching_ctl); | |
11dfe35a | 465 | btrfs_put_block_group(block_group); |
817d52f8 JB |
466 | } |
467 | ||
9d66e233 JB |
468 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
469 | struct btrfs_trans_handle *trans, | |
b8399dee | 470 | struct btrfs_root *root, |
9d66e233 | 471 | int load_cache_only) |
817d52f8 | 472 | { |
291c7d2f | 473 | DEFINE_WAIT(wait); |
11833d66 YZ |
474 | struct btrfs_fs_info *fs_info = cache->fs_info; |
475 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
476 | int ret = 0; |
477 | ||
291c7d2f | 478 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
479 | if (!caching_ctl) |
480 | return -ENOMEM; | |
291c7d2f JB |
481 | |
482 | INIT_LIST_HEAD(&caching_ctl->list); | |
483 | mutex_init(&caching_ctl->mutex); | |
484 | init_waitqueue_head(&caching_ctl->wait); | |
485 | caching_ctl->block_group = cache; | |
486 | caching_ctl->progress = cache->key.objectid; | |
487 | atomic_set(&caching_ctl->count, 1); | |
488 | caching_ctl->work.func = caching_thread; | |
489 | ||
490 | spin_lock(&cache->lock); | |
491 | /* | |
492 | * This should be a rare occasion, but this could happen I think in the | |
493 | * case where one thread starts to load the space cache info, and then | |
494 | * some other thread starts a transaction commit which tries to do an | |
495 | * allocation while the other thread is still loading the space cache | |
496 | * info. The previous loop should have kept us from choosing this block | |
497 | * group, but if we've moved to the state where we will wait on caching | |
498 | * block groups we need to first check if we're doing a fast load here, | |
499 | * so we can wait for it to finish, otherwise we could end up allocating | |
500 | * from a block group who's cache gets evicted for one reason or | |
501 | * another. | |
502 | */ | |
503 | while (cache->cached == BTRFS_CACHE_FAST) { | |
504 | struct btrfs_caching_control *ctl; | |
505 | ||
506 | ctl = cache->caching_ctl; | |
507 | atomic_inc(&ctl->count); | |
508 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); | |
509 | spin_unlock(&cache->lock); | |
510 | ||
511 | schedule(); | |
512 | ||
513 | finish_wait(&ctl->wait, &wait); | |
514 | put_caching_control(ctl); | |
515 | spin_lock(&cache->lock); | |
516 | } | |
517 | ||
518 | if (cache->cached != BTRFS_CACHE_NO) { | |
519 | spin_unlock(&cache->lock); | |
520 | kfree(caching_ctl); | |
11833d66 | 521 | return 0; |
291c7d2f JB |
522 | } |
523 | WARN_ON(cache->caching_ctl); | |
524 | cache->caching_ctl = caching_ctl; | |
525 | cache->cached = BTRFS_CACHE_FAST; | |
526 | spin_unlock(&cache->lock); | |
11833d66 | 527 | |
9d66e233 JB |
528 | /* |
529 | * We can't do the read from on-disk cache during a commit since we need | |
b8399dee JB |
530 | * to have the normal tree locking. Also if we are currently trying to |
531 | * allocate blocks for the tree root we can't do the fast caching since | |
532 | * we likely hold important locks. | |
9d66e233 | 533 | */ |
d53ba474 | 534 | if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) { |
9d66e233 JB |
535 | ret = load_free_space_cache(fs_info, cache); |
536 | ||
537 | spin_lock(&cache->lock); | |
538 | if (ret == 1) { | |
291c7d2f | 539 | cache->caching_ctl = NULL; |
9d66e233 JB |
540 | cache->cached = BTRFS_CACHE_FINISHED; |
541 | cache->last_byte_to_unpin = (u64)-1; | |
542 | } else { | |
291c7d2f JB |
543 | if (load_cache_only) { |
544 | cache->caching_ctl = NULL; | |
545 | cache->cached = BTRFS_CACHE_NO; | |
546 | } else { | |
547 | cache->cached = BTRFS_CACHE_STARTED; | |
548 | } | |
9d66e233 JB |
549 | } |
550 | spin_unlock(&cache->lock); | |
291c7d2f | 551 | wake_up(&caching_ctl->wait); |
3c14874a | 552 | if (ret == 1) { |
291c7d2f | 553 | put_caching_control(caching_ctl); |
3c14874a | 554 | free_excluded_extents(fs_info->extent_root, cache); |
9d66e233 | 555 | return 0; |
3c14874a | 556 | } |
291c7d2f JB |
557 | } else { |
558 | /* | |
559 | * We are not going to do the fast caching, set cached to the | |
560 | * appropriate value and wakeup any waiters. | |
561 | */ | |
562 | spin_lock(&cache->lock); | |
563 | if (load_cache_only) { | |
564 | cache->caching_ctl = NULL; | |
565 | cache->cached = BTRFS_CACHE_NO; | |
566 | } else { | |
567 | cache->cached = BTRFS_CACHE_STARTED; | |
568 | } | |
569 | spin_unlock(&cache->lock); | |
570 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
571 | } |
572 | ||
291c7d2f JB |
573 | if (load_cache_only) { |
574 | put_caching_control(caching_ctl); | |
11833d66 | 575 | return 0; |
817d52f8 | 576 | } |
817d52f8 | 577 | |
11833d66 | 578 | down_write(&fs_info->extent_commit_sem); |
291c7d2f | 579 | atomic_inc(&caching_ctl->count); |
11833d66 YZ |
580 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
581 | up_write(&fs_info->extent_commit_sem); | |
582 | ||
11dfe35a | 583 | btrfs_get_block_group(cache); |
11833d66 | 584 | |
bab39bf9 | 585 | btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 586 | |
ef8bbdfe | 587 | return ret; |
e37c9e69 CM |
588 | } |
589 | ||
0f9dd46c JB |
590 | /* |
591 | * return the block group that starts at or after bytenr | |
592 | */ | |
d397712b CM |
593 | static struct btrfs_block_group_cache * |
594 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 595 | { |
0f9dd46c | 596 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 597 | |
0f9dd46c | 598 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 599 | |
0f9dd46c | 600 | return cache; |
0ef3e66b CM |
601 | } |
602 | ||
0f9dd46c | 603 | /* |
9f55684c | 604 | * return the block group that contains the given bytenr |
0f9dd46c | 605 | */ |
d397712b CM |
606 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
607 | struct btrfs_fs_info *info, | |
608 | u64 bytenr) | |
be744175 | 609 | { |
0f9dd46c | 610 | struct btrfs_block_group_cache *cache; |
be744175 | 611 | |
0f9dd46c | 612 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 613 | |
0f9dd46c | 614 | return cache; |
be744175 | 615 | } |
0b86a832 | 616 | |
0f9dd46c JB |
617 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
618 | u64 flags) | |
6324fbf3 | 619 | { |
0f9dd46c | 620 | struct list_head *head = &info->space_info; |
0f9dd46c | 621 | struct btrfs_space_info *found; |
4184ea7f | 622 | |
52ba6929 | 623 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 624 | |
4184ea7f CM |
625 | rcu_read_lock(); |
626 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 627 | if (found->flags & flags) { |
4184ea7f | 628 | rcu_read_unlock(); |
0f9dd46c | 629 | return found; |
4184ea7f | 630 | } |
0f9dd46c | 631 | } |
4184ea7f | 632 | rcu_read_unlock(); |
0f9dd46c | 633 | return NULL; |
6324fbf3 CM |
634 | } |
635 | ||
4184ea7f CM |
636 | /* |
637 | * after adding space to the filesystem, we need to clear the full flags | |
638 | * on all the space infos. | |
639 | */ | |
640 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
641 | { | |
642 | struct list_head *head = &info->space_info; | |
643 | struct btrfs_space_info *found; | |
644 | ||
645 | rcu_read_lock(); | |
646 | list_for_each_entry_rcu(found, head, list) | |
647 | found->full = 0; | |
648 | rcu_read_unlock(); | |
649 | } | |
650 | ||
80eb234a JB |
651 | static u64 div_factor(u64 num, int factor) |
652 | { | |
653 | if (factor == 10) | |
654 | return num; | |
655 | num *= factor; | |
656 | do_div(num, 10); | |
657 | return num; | |
658 | } | |
659 | ||
e5bc2458 CM |
660 | static u64 div_factor_fine(u64 num, int factor) |
661 | { | |
662 | if (factor == 100) | |
663 | return num; | |
664 | num *= factor; | |
665 | do_div(num, 100); | |
666 | return num; | |
667 | } | |
668 | ||
d2fb3437 YZ |
669 | u64 btrfs_find_block_group(struct btrfs_root *root, |
670 | u64 search_start, u64 search_hint, int owner) | |
cd1bc465 | 671 | { |
96b5179d | 672 | struct btrfs_block_group_cache *cache; |
cd1bc465 | 673 | u64 used; |
d2fb3437 YZ |
674 | u64 last = max(search_hint, search_start); |
675 | u64 group_start = 0; | |
31f3c99b | 676 | int full_search = 0; |
d2fb3437 | 677 | int factor = 9; |
0ef3e66b | 678 | int wrapped = 0; |
31f3c99b | 679 | again: |
e8569813 ZY |
680 | while (1) { |
681 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
0f9dd46c JB |
682 | if (!cache) |
683 | break; | |
96b5179d | 684 | |
c286ac48 | 685 | spin_lock(&cache->lock); |
96b5179d CM |
686 | last = cache->key.objectid + cache->key.offset; |
687 | used = btrfs_block_group_used(&cache->item); | |
688 | ||
d2fb3437 YZ |
689 | if ((full_search || !cache->ro) && |
690 | block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) { | |
e8569813 | 691 | if (used + cache->pinned + cache->reserved < |
d2fb3437 YZ |
692 | div_factor(cache->key.offset, factor)) { |
693 | group_start = cache->key.objectid; | |
c286ac48 | 694 | spin_unlock(&cache->lock); |
fa9c0d79 | 695 | btrfs_put_block_group(cache); |
8790d502 CM |
696 | goto found; |
697 | } | |
6324fbf3 | 698 | } |
c286ac48 | 699 | spin_unlock(&cache->lock); |
fa9c0d79 | 700 | btrfs_put_block_group(cache); |
de428b63 | 701 | cond_resched(); |
cd1bc465 | 702 | } |
0ef3e66b CM |
703 | if (!wrapped) { |
704 | last = search_start; | |
705 | wrapped = 1; | |
706 | goto again; | |
707 | } | |
708 | if (!full_search && factor < 10) { | |
be744175 | 709 | last = search_start; |
31f3c99b | 710 | full_search = 1; |
0ef3e66b | 711 | factor = 10; |
31f3c99b CM |
712 | goto again; |
713 | } | |
be744175 | 714 | found: |
d2fb3437 | 715 | return group_start; |
925baedd | 716 | } |
0f9dd46c | 717 | |
e02119d5 | 718 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 719 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
720 | { |
721 | int ret; | |
722 | struct btrfs_key key; | |
31840ae1 | 723 | struct btrfs_path *path; |
e02119d5 | 724 | |
31840ae1 | 725 | path = btrfs_alloc_path(); |
d8926bb3 MF |
726 | if (!path) |
727 | return -ENOMEM; | |
728 | ||
e02119d5 CM |
729 | key.objectid = start; |
730 | key.offset = len; | |
731 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
732 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, | |
733 | 0, 0); | |
31840ae1 | 734 | btrfs_free_path(path); |
7bb86316 CM |
735 | return ret; |
736 | } | |
737 | ||
a22285a6 YZ |
738 | /* |
739 | * helper function to lookup reference count and flags of extent. | |
740 | * | |
741 | * the head node for delayed ref is used to store the sum of all the | |
742 | * reference count modifications queued up in the rbtree. the head | |
743 | * node may also store the extent flags to set. This way you can check | |
744 | * to see what the reference count and extent flags would be if all of | |
745 | * the delayed refs are not processed. | |
746 | */ | |
747 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
748 | struct btrfs_root *root, u64 bytenr, | |
749 | u64 num_bytes, u64 *refs, u64 *flags) | |
750 | { | |
751 | struct btrfs_delayed_ref_head *head; | |
752 | struct btrfs_delayed_ref_root *delayed_refs; | |
753 | struct btrfs_path *path; | |
754 | struct btrfs_extent_item *ei; | |
755 | struct extent_buffer *leaf; | |
756 | struct btrfs_key key; | |
757 | u32 item_size; | |
758 | u64 num_refs; | |
759 | u64 extent_flags; | |
760 | int ret; | |
761 | ||
762 | path = btrfs_alloc_path(); | |
763 | if (!path) | |
764 | return -ENOMEM; | |
765 | ||
766 | key.objectid = bytenr; | |
767 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
768 | key.offset = num_bytes; | |
769 | if (!trans) { | |
770 | path->skip_locking = 1; | |
771 | path->search_commit_root = 1; | |
772 | } | |
773 | again: | |
774 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, | |
775 | &key, path, 0, 0); | |
776 | if (ret < 0) | |
777 | goto out_free; | |
778 | ||
779 | if (ret == 0) { | |
780 | leaf = path->nodes[0]; | |
781 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
782 | if (item_size >= sizeof(*ei)) { | |
783 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
784 | struct btrfs_extent_item); | |
785 | num_refs = btrfs_extent_refs(leaf, ei); | |
786 | extent_flags = btrfs_extent_flags(leaf, ei); | |
787 | } else { | |
788 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
789 | struct btrfs_extent_item_v0 *ei0; | |
790 | BUG_ON(item_size != sizeof(*ei0)); | |
791 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
792 | struct btrfs_extent_item_v0); | |
793 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
794 | /* FIXME: this isn't correct for data */ | |
795 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
796 | #else | |
797 | BUG(); | |
798 | #endif | |
799 | } | |
800 | BUG_ON(num_refs == 0); | |
801 | } else { | |
802 | num_refs = 0; | |
803 | extent_flags = 0; | |
804 | ret = 0; | |
805 | } | |
806 | ||
807 | if (!trans) | |
808 | goto out; | |
809 | ||
810 | delayed_refs = &trans->transaction->delayed_refs; | |
811 | spin_lock(&delayed_refs->lock); | |
812 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
813 | if (head) { | |
814 | if (!mutex_trylock(&head->mutex)) { | |
815 | atomic_inc(&head->node.refs); | |
816 | spin_unlock(&delayed_refs->lock); | |
817 | ||
b3b4aa74 | 818 | btrfs_release_path(path); |
a22285a6 | 819 | |
8cc33e5c DS |
820 | /* |
821 | * Mutex was contended, block until it's released and try | |
822 | * again | |
823 | */ | |
a22285a6 YZ |
824 | mutex_lock(&head->mutex); |
825 | mutex_unlock(&head->mutex); | |
826 | btrfs_put_delayed_ref(&head->node); | |
827 | goto again; | |
828 | } | |
829 | if (head->extent_op && head->extent_op->update_flags) | |
830 | extent_flags |= head->extent_op->flags_to_set; | |
831 | else | |
832 | BUG_ON(num_refs == 0); | |
833 | ||
834 | num_refs += head->node.ref_mod; | |
835 | mutex_unlock(&head->mutex); | |
836 | } | |
837 | spin_unlock(&delayed_refs->lock); | |
838 | out: | |
839 | WARN_ON(num_refs == 0); | |
840 | if (refs) | |
841 | *refs = num_refs; | |
842 | if (flags) | |
843 | *flags = extent_flags; | |
844 | out_free: | |
845 | btrfs_free_path(path); | |
846 | return ret; | |
847 | } | |
848 | ||
d8d5f3e1 CM |
849 | /* |
850 | * Back reference rules. Back refs have three main goals: | |
851 | * | |
852 | * 1) differentiate between all holders of references to an extent so that | |
853 | * when a reference is dropped we can make sure it was a valid reference | |
854 | * before freeing the extent. | |
855 | * | |
856 | * 2) Provide enough information to quickly find the holders of an extent | |
857 | * if we notice a given block is corrupted or bad. | |
858 | * | |
859 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
860 | * maintenance. This is actually the same as #2, but with a slightly | |
861 | * different use case. | |
862 | * | |
5d4f98a2 YZ |
863 | * There are two kinds of back refs. The implicit back refs is optimized |
864 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
865 | * back refs of this kind provide information about the block's owner tree | |
866 | * and the pointer's key. These information allow us to find the block by | |
867 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
868 | * referenced by their owner trees. The location of tree block is recorded | |
869 | * in the back refs. Actually the full back refs is generic, and can be | |
870 | * used in all cases the implicit back refs is used. The major shortcoming | |
871 | * of the full back refs is its overhead. Every time a tree block gets | |
872 | * COWed, we have to update back refs entry for all pointers in it. | |
873 | * | |
874 | * For a newly allocated tree block, we use implicit back refs for | |
875 | * pointers in it. This means most tree related operations only involve | |
876 | * implicit back refs. For a tree block created in old transaction, the | |
877 | * only way to drop a reference to it is COW it. So we can detect the | |
878 | * event that tree block loses its owner tree's reference and do the | |
879 | * back refs conversion. | |
880 | * | |
881 | * When a tree block is COW'd through a tree, there are four cases: | |
882 | * | |
883 | * The reference count of the block is one and the tree is the block's | |
884 | * owner tree. Nothing to do in this case. | |
885 | * | |
886 | * The reference count of the block is one and the tree is not the | |
887 | * block's owner tree. In this case, full back refs is used for pointers | |
888 | * in the block. Remove these full back refs, add implicit back refs for | |
889 | * every pointers in the new block. | |
890 | * | |
891 | * The reference count of the block is greater than one and the tree is | |
892 | * the block's owner tree. In this case, implicit back refs is used for | |
893 | * pointers in the block. Add full back refs for every pointers in the | |
894 | * block, increase lower level extents' reference counts. The original | |
895 | * implicit back refs are entailed to the new block. | |
896 | * | |
897 | * The reference count of the block is greater than one and the tree is | |
898 | * not the block's owner tree. Add implicit back refs for every pointer in | |
899 | * the new block, increase lower level extents' reference count. | |
900 | * | |
901 | * Back Reference Key composing: | |
902 | * | |
903 | * The key objectid corresponds to the first byte in the extent, | |
904 | * The key type is used to differentiate between types of back refs. | |
905 | * There are different meanings of the key offset for different types | |
906 | * of back refs. | |
907 | * | |
d8d5f3e1 CM |
908 | * File extents can be referenced by: |
909 | * | |
910 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 911 | * - different files inside a single subvolume |
d8d5f3e1 CM |
912 | * - different offsets inside a file (bookend extents in file.c) |
913 | * | |
5d4f98a2 | 914 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
915 | * |
916 | * - Objectid of the subvolume root | |
d8d5f3e1 | 917 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
918 | * - original offset in the file |
919 | * - how many bookend extents | |
d8d5f3e1 | 920 | * |
5d4f98a2 YZ |
921 | * The key offset for the implicit back refs is hash of the first |
922 | * three fields. | |
d8d5f3e1 | 923 | * |
5d4f98a2 | 924 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 925 | * |
5d4f98a2 | 926 | * - number of pointers in the tree leaf |
d8d5f3e1 | 927 | * |
5d4f98a2 YZ |
928 | * The key offset for the implicit back refs is the first byte of |
929 | * the tree leaf | |
d8d5f3e1 | 930 | * |
5d4f98a2 YZ |
931 | * When a file extent is allocated, The implicit back refs is used. |
932 | * the fields are filled in: | |
d8d5f3e1 | 933 | * |
5d4f98a2 | 934 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 935 | * |
5d4f98a2 YZ |
936 | * When a file extent is removed file truncation, we find the |
937 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 938 | * |
5d4f98a2 | 939 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 940 | * |
5d4f98a2 | 941 | * Btree extents can be referenced by: |
d8d5f3e1 | 942 | * |
5d4f98a2 | 943 | * - Different subvolumes |
d8d5f3e1 | 944 | * |
5d4f98a2 YZ |
945 | * Both the implicit back refs and the full back refs for tree blocks |
946 | * only consist of key. The key offset for the implicit back refs is | |
947 | * objectid of block's owner tree. The key offset for the full back refs | |
948 | * is the first byte of parent block. | |
d8d5f3e1 | 949 | * |
5d4f98a2 YZ |
950 | * When implicit back refs is used, information about the lowest key and |
951 | * level of the tree block are required. These information are stored in | |
952 | * tree block info structure. | |
d8d5f3e1 | 953 | */ |
31840ae1 | 954 | |
5d4f98a2 YZ |
955 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
956 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
957 | struct btrfs_root *root, | |
958 | struct btrfs_path *path, | |
959 | u64 owner, u32 extra_size) | |
7bb86316 | 960 | { |
5d4f98a2 YZ |
961 | struct btrfs_extent_item *item; |
962 | struct btrfs_extent_item_v0 *ei0; | |
963 | struct btrfs_extent_ref_v0 *ref0; | |
964 | struct btrfs_tree_block_info *bi; | |
965 | struct extent_buffer *leaf; | |
7bb86316 | 966 | struct btrfs_key key; |
5d4f98a2 YZ |
967 | struct btrfs_key found_key; |
968 | u32 new_size = sizeof(*item); | |
969 | u64 refs; | |
970 | int ret; | |
971 | ||
972 | leaf = path->nodes[0]; | |
973 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
974 | ||
975 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
976 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
977 | struct btrfs_extent_item_v0); | |
978 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
979 | ||
980 | if (owner == (u64)-1) { | |
981 | while (1) { | |
982 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
983 | ret = btrfs_next_leaf(root, path); | |
984 | if (ret < 0) | |
985 | return ret; | |
79787eaa | 986 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
987 | leaf = path->nodes[0]; |
988 | } | |
989 | btrfs_item_key_to_cpu(leaf, &found_key, | |
990 | path->slots[0]); | |
991 | BUG_ON(key.objectid != found_key.objectid); | |
992 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
993 | path->slots[0]++; | |
994 | continue; | |
995 | } | |
996 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
997 | struct btrfs_extent_ref_v0); | |
998 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
999 | break; | |
1000 | } | |
1001 | } | |
b3b4aa74 | 1002 | btrfs_release_path(path); |
5d4f98a2 YZ |
1003 | |
1004 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1005 | new_size += sizeof(*bi); | |
1006 | ||
1007 | new_size -= sizeof(*ei0); | |
1008 | ret = btrfs_search_slot(trans, root, &key, path, | |
1009 | new_size + extra_size, 1); | |
1010 | if (ret < 0) | |
1011 | return ret; | |
79787eaa | 1012 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 1013 | |
143bede5 | 1014 | btrfs_extend_item(trans, root, path, new_size); |
5d4f98a2 YZ |
1015 | |
1016 | leaf = path->nodes[0]; | |
1017 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1018 | btrfs_set_extent_refs(leaf, item, refs); | |
1019 | /* FIXME: get real generation */ | |
1020 | btrfs_set_extent_generation(leaf, item, 0); | |
1021 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1022 | btrfs_set_extent_flags(leaf, item, | |
1023 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1024 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1025 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1026 | /* FIXME: get first key of the block */ | |
1027 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
1028 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
1029 | } else { | |
1030 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1031 | } | |
1032 | btrfs_mark_buffer_dirty(leaf); | |
1033 | return 0; | |
1034 | } | |
1035 | #endif | |
1036 | ||
1037 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
1038 | { | |
1039 | u32 high_crc = ~(u32)0; | |
1040 | u32 low_crc = ~(u32)0; | |
1041 | __le64 lenum; | |
1042 | ||
1043 | lenum = cpu_to_le64(root_objectid); | |
163e783e | 1044 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1045 | lenum = cpu_to_le64(owner); |
163e783e | 1046 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1047 | lenum = cpu_to_le64(offset); |
163e783e | 1048 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1049 | |
1050 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1051 | } | |
1052 | ||
1053 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1054 | struct btrfs_extent_data_ref *ref) | |
1055 | { | |
1056 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1057 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1058 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1059 | } | |
1060 | ||
1061 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1062 | struct btrfs_extent_data_ref *ref, | |
1063 | u64 root_objectid, u64 owner, u64 offset) | |
1064 | { | |
1065 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1066 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1067 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1068 | return 0; | |
1069 | return 1; | |
1070 | } | |
1071 | ||
1072 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
1073 | struct btrfs_root *root, | |
1074 | struct btrfs_path *path, | |
1075 | u64 bytenr, u64 parent, | |
1076 | u64 root_objectid, | |
1077 | u64 owner, u64 offset) | |
1078 | { | |
1079 | struct btrfs_key key; | |
1080 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1081 | struct extent_buffer *leaf; |
5d4f98a2 | 1082 | u32 nritems; |
74493f7a | 1083 | int ret; |
5d4f98a2 YZ |
1084 | int recow; |
1085 | int err = -ENOENT; | |
74493f7a | 1086 | |
31840ae1 | 1087 | key.objectid = bytenr; |
5d4f98a2 YZ |
1088 | if (parent) { |
1089 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1090 | key.offset = parent; | |
1091 | } else { | |
1092 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1093 | key.offset = hash_extent_data_ref(root_objectid, | |
1094 | owner, offset); | |
1095 | } | |
1096 | again: | |
1097 | recow = 0; | |
1098 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1099 | if (ret < 0) { | |
1100 | err = ret; | |
1101 | goto fail; | |
1102 | } | |
31840ae1 | 1103 | |
5d4f98a2 YZ |
1104 | if (parent) { |
1105 | if (!ret) | |
1106 | return 0; | |
1107 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1108 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1109 | btrfs_release_path(path); |
5d4f98a2 YZ |
1110 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1111 | if (ret < 0) { | |
1112 | err = ret; | |
1113 | goto fail; | |
1114 | } | |
1115 | if (!ret) | |
1116 | return 0; | |
1117 | #endif | |
1118 | goto fail; | |
31840ae1 ZY |
1119 | } |
1120 | ||
1121 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1122 | nritems = btrfs_header_nritems(leaf); |
1123 | while (1) { | |
1124 | if (path->slots[0] >= nritems) { | |
1125 | ret = btrfs_next_leaf(root, path); | |
1126 | if (ret < 0) | |
1127 | err = ret; | |
1128 | if (ret) | |
1129 | goto fail; | |
1130 | ||
1131 | leaf = path->nodes[0]; | |
1132 | nritems = btrfs_header_nritems(leaf); | |
1133 | recow = 1; | |
1134 | } | |
1135 | ||
1136 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1137 | if (key.objectid != bytenr || | |
1138 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1139 | goto fail; | |
1140 | ||
1141 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1142 | struct btrfs_extent_data_ref); | |
1143 | ||
1144 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1145 | owner, offset)) { | |
1146 | if (recow) { | |
b3b4aa74 | 1147 | btrfs_release_path(path); |
5d4f98a2 YZ |
1148 | goto again; |
1149 | } | |
1150 | err = 0; | |
1151 | break; | |
1152 | } | |
1153 | path->slots[0]++; | |
31840ae1 | 1154 | } |
5d4f98a2 YZ |
1155 | fail: |
1156 | return err; | |
31840ae1 ZY |
1157 | } |
1158 | ||
5d4f98a2 YZ |
1159 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
1160 | struct btrfs_root *root, | |
1161 | struct btrfs_path *path, | |
1162 | u64 bytenr, u64 parent, | |
1163 | u64 root_objectid, u64 owner, | |
1164 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
1165 | { |
1166 | struct btrfs_key key; | |
1167 | struct extent_buffer *leaf; | |
5d4f98a2 | 1168 | u32 size; |
31840ae1 ZY |
1169 | u32 num_refs; |
1170 | int ret; | |
74493f7a | 1171 | |
74493f7a | 1172 | key.objectid = bytenr; |
5d4f98a2 YZ |
1173 | if (parent) { |
1174 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1175 | key.offset = parent; | |
1176 | size = sizeof(struct btrfs_shared_data_ref); | |
1177 | } else { | |
1178 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1179 | key.offset = hash_extent_data_ref(root_objectid, | |
1180 | owner, offset); | |
1181 | size = sizeof(struct btrfs_extent_data_ref); | |
1182 | } | |
74493f7a | 1183 | |
5d4f98a2 YZ |
1184 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1185 | if (ret && ret != -EEXIST) | |
1186 | goto fail; | |
1187 | ||
1188 | leaf = path->nodes[0]; | |
1189 | if (parent) { | |
1190 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1191 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1192 | struct btrfs_shared_data_ref); |
1193 | if (ret == 0) { | |
1194 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1195 | } else { | |
1196 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1197 | num_refs += refs_to_add; | |
1198 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1199 | } |
5d4f98a2 YZ |
1200 | } else { |
1201 | struct btrfs_extent_data_ref *ref; | |
1202 | while (ret == -EEXIST) { | |
1203 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1204 | struct btrfs_extent_data_ref); | |
1205 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1206 | owner, offset)) | |
1207 | break; | |
b3b4aa74 | 1208 | btrfs_release_path(path); |
5d4f98a2 YZ |
1209 | key.offset++; |
1210 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1211 | size); | |
1212 | if (ret && ret != -EEXIST) | |
1213 | goto fail; | |
31840ae1 | 1214 | |
5d4f98a2 YZ |
1215 | leaf = path->nodes[0]; |
1216 | } | |
1217 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1218 | struct btrfs_extent_data_ref); | |
1219 | if (ret == 0) { | |
1220 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1221 | root_objectid); | |
1222 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1223 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1224 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1225 | } else { | |
1226 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1227 | num_refs += refs_to_add; | |
1228 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1229 | } |
31840ae1 | 1230 | } |
5d4f98a2 YZ |
1231 | btrfs_mark_buffer_dirty(leaf); |
1232 | ret = 0; | |
1233 | fail: | |
b3b4aa74 | 1234 | btrfs_release_path(path); |
7bb86316 | 1235 | return ret; |
74493f7a CM |
1236 | } |
1237 | ||
5d4f98a2 YZ |
1238 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
1239 | struct btrfs_root *root, | |
1240 | struct btrfs_path *path, | |
1241 | int refs_to_drop) | |
31840ae1 | 1242 | { |
5d4f98a2 YZ |
1243 | struct btrfs_key key; |
1244 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1245 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1246 | struct extent_buffer *leaf; |
5d4f98a2 | 1247 | u32 num_refs = 0; |
31840ae1 ZY |
1248 | int ret = 0; |
1249 | ||
1250 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1251 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1252 | ||
1253 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1254 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1255 | struct btrfs_extent_data_ref); | |
1256 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1257 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1258 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1259 | struct btrfs_shared_data_ref); | |
1260 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1261 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1262 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1263 | struct btrfs_extent_ref_v0 *ref0; | |
1264 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1265 | struct btrfs_extent_ref_v0); | |
1266 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1267 | #endif | |
1268 | } else { | |
1269 | BUG(); | |
1270 | } | |
1271 | ||
56bec294 CM |
1272 | BUG_ON(num_refs < refs_to_drop); |
1273 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1274 | |
31840ae1 ZY |
1275 | if (num_refs == 0) { |
1276 | ret = btrfs_del_item(trans, root, path); | |
1277 | } else { | |
5d4f98a2 YZ |
1278 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1279 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1280 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1281 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1282 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1283 | else { | |
1284 | struct btrfs_extent_ref_v0 *ref0; | |
1285 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1286 | struct btrfs_extent_ref_v0); | |
1287 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1288 | } | |
1289 | #endif | |
31840ae1 ZY |
1290 | btrfs_mark_buffer_dirty(leaf); |
1291 | } | |
31840ae1 ZY |
1292 | return ret; |
1293 | } | |
1294 | ||
5d4f98a2 YZ |
1295 | static noinline u32 extent_data_ref_count(struct btrfs_root *root, |
1296 | struct btrfs_path *path, | |
1297 | struct btrfs_extent_inline_ref *iref) | |
15916de8 | 1298 | { |
5d4f98a2 YZ |
1299 | struct btrfs_key key; |
1300 | struct extent_buffer *leaf; | |
1301 | struct btrfs_extent_data_ref *ref1; | |
1302 | struct btrfs_shared_data_ref *ref2; | |
1303 | u32 num_refs = 0; | |
1304 | ||
1305 | leaf = path->nodes[0]; | |
1306 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1307 | if (iref) { | |
1308 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1309 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1310 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1311 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1312 | } else { | |
1313 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1314 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1315 | } | |
1316 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1317 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1318 | struct btrfs_extent_data_ref); | |
1319 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1320 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1321 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1322 | struct btrfs_shared_data_ref); | |
1323 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1324 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1325 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1326 | struct btrfs_extent_ref_v0 *ref0; | |
1327 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1328 | struct btrfs_extent_ref_v0); | |
1329 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1330 | #endif |
5d4f98a2 YZ |
1331 | } else { |
1332 | WARN_ON(1); | |
1333 | } | |
1334 | return num_refs; | |
1335 | } | |
15916de8 | 1336 | |
5d4f98a2 YZ |
1337 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1338 | struct btrfs_root *root, | |
1339 | struct btrfs_path *path, | |
1340 | u64 bytenr, u64 parent, | |
1341 | u64 root_objectid) | |
1f3c79a2 | 1342 | { |
5d4f98a2 | 1343 | struct btrfs_key key; |
1f3c79a2 | 1344 | int ret; |
1f3c79a2 | 1345 | |
5d4f98a2 YZ |
1346 | key.objectid = bytenr; |
1347 | if (parent) { | |
1348 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1349 | key.offset = parent; | |
1350 | } else { | |
1351 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1352 | key.offset = root_objectid; | |
1f3c79a2 LH |
1353 | } |
1354 | ||
5d4f98a2 YZ |
1355 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1356 | if (ret > 0) | |
1357 | ret = -ENOENT; | |
1358 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1359 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1360 | btrfs_release_path(path); |
5d4f98a2 YZ |
1361 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1362 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1363 | if (ret > 0) | |
1364 | ret = -ENOENT; | |
1365 | } | |
1f3c79a2 | 1366 | #endif |
5d4f98a2 | 1367 | return ret; |
1f3c79a2 LH |
1368 | } |
1369 | ||
5d4f98a2 YZ |
1370 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1371 | struct btrfs_root *root, | |
1372 | struct btrfs_path *path, | |
1373 | u64 bytenr, u64 parent, | |
1374 | u64 root_objectid) | |
31840ae1 | 1375 | { |
5d4f98a2 | 1376 | struct btrfs_key key; |
31840ae1 | 1377 | int ret; |
31840ae1 | 1378 | |
5d4f98a2 YZ |
1379 | key.objectid = bytenr; |
1380 | if (parent) { | |
1381 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1382 | key.offset = parent; | |
1383 | } else { | |
1384 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1385 | key.offset = root_objectid; | |
1386 | } | |
1387 | ||
1388 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
b3b4aa74 | 1389 | btrfs_release_path(path); |
31840ae1 ZY |
1390 | return ret; |
1391 | } | |
1392 | ||
5d4f98a2 | 1393 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1394 | { |
5d4f98a2 YZ |
1395 | int type; |
1396 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1397 | if (parent > 0) | |
1398 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1399 | else | |
1400 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1401 | } else { | |
1402 | if (parent > 0) | |
1403 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1404 | else | |
1405 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1406 | } | |
1407 | return type; | |
31840ae1 | 1408 | } |
56bec294 | 1409 | |
2c47e605 YZ |
1410 | static int find_next_key(struct btrfs_path *path, int level, |
1411 | struct btrfs_key *key) | |
56bec294 | 1412 | |
02217ed2 | 1413 | { |
2c47e605 | 1414 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1415 | if (!path->nodes[level]) |
1416 | break; | |
5d4f98a2 YZ |
1417 | if (path->slots[level] + 1 >= |
1418 | btrfs_header_nritems(path->nodes[level])) | |
1419 | continue; | |
1420 | if (level == 0) | |
1421 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1422 | path->slots[level] + 1); | |
1423 | else | |
1424 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1425 | path->slots[level] + 1); | |
1426 | return 0; | |
1427 | } | |
1428 | return 1; | |
1429 | } | |
037e6390 | 1430 | |
5d4f98a2 YZ |
1431 | /* |
1432 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1433 | * to the address of inline back ref, and 0 is returned. | |
1434 | * | |
1435 | * if back ref isn't found, *ref_ret is set to the address where it | |
1436 | * should be inserted, and -ENOENT is returned. | |
1437 | * | |
1438 | * if insert is true and there are too many inline back refs, the path | |
1439 | * points to the extent item, and -EAGAIN is returned. | |
1440 | * | |
1441 | * NOTE: inline back refs are ordered in the same way that back ref | |
1442 | * items in the tree are ordered. | |
1443 | */ | |
1444 | static noinline_for_stack | |
1445 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1446 | struct btrfs_root *root, | |
1447 | struct btrfs_path *path, | |
1448 | struct btrfs_extent_inline_ref **ref_ret, | |
1449 | u64 bytenr, u64 num_bytes, | |
1450 | u64 parent, u64 root_objectid, | |
1451 | u64 owner, u64 offset, int insert) | |
1452 | { | |
1453 | struct btrfs_key key; | |
1454 | struct extent_buffer *leaf; | |
1455 | struct btrfs_extent_item *ei; | |
1456 | struct btrfs_extent_inline_ref *iref; | |
1457 | u64 flags; | |
1458 | u64 item_size; | |
1459 | unsigned long ptr; | |
1460 | unsigned long end; | |
1461 | int extra_size; | |
1462 | int type; | |
1463 | int want; | |
1464 | int ret; | |
1465 | int err = 0; | |
26b8003f | 1466 | |
db94535d | 1467 | key.objectid = bytenr; |
31840ae1 | 1468 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1469 | key.offset = num_bytes; |
31840ae1 | 1470 | |
5d4f98a2 YZ |
1471 | want = extent_ref_type(parent, owner); |
1472 | if (insert) { | |
1473 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1474 | path->keep_locks = 1; |
5d4f98a2 YZ |
1475 | } else |
1476 | extra_size = -1; | |
1477 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); | |
b9473439 | 1478 | if (ret < 0) { |
5d4f98a2 YZ |
1479 | err = ret; |
1480 | goto out; | |
1481 | } | |
79787eaa JM |
1482 | if (ret && !insert) { |
1483 | err = -ENOENT; | |
1484 | goto out; | |
1485 | } | |
1486 | BUG_ON(ret); /* Corruption */ | |
5d4f98a2 YZ |
1487 | |
1488 | leaf = path->nodes[0]; | |
1489 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1490 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1491 | if (item_size < sizeof(*ei)) { | |
1492 | if (!insert) { | |
1493 | err = -ENOENT; | |
1494 | goto out; | |
1495 | } | |
1496 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1497 | extra_size); | |
1498 | if (ret < 0) { | |
1499 | err = ret; | |
1500 | goto out; | |
1501 | } | |
1502 | leaf = path->nodes[0]; | |
1503 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1504 | } | |
1505 | #endif | |
1506 | BUG_ON(item_size < sizeof(*ei)); | |
1507 | ||
5d4f98a2 YZ |
1508 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1509 | flags = btrfs_extent_flags(leaf, ei); | |
1510 | ||
1511 | ptr = (unsigned long)(ei + 1); | |
1512 | end = (unsigned long)ei + item_size; | |
1513 | ||
1514 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
1515 | ptr += sizeof(struct btrfs_tree_block_info); | |
1516 | BUG_ON(ptr > end); | |
1517 | } else { | |
1518 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); | |
1519 | } | |
1520 | ||
1521 | err = -ENOENT; | |
1522 | while (1) { | |
1523 | if (ptr >= end) { | |
1524 | WARN_ON(ptr > end); | |
1525 | break; | |
1526 | } | |
1527 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1528 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1529 | if (want < type) | |
1530 | break; | |
1531 | if (want > type) { | |
1532 | ptr += btrfs_extent_inline_ref_size(type); | |
1533 | continue; | |
1534 | } | |
1535 | ||
1536 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1537 | struct btrfs_extent_data_ref *dref; | |
1538 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1539 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1540 | owner, offset)) { | |
1541 | err = 0; | |
1542 | break; | |
1543 | } | |
1544 | if (hash_extent_data_ref_item(leaf, dref) < | |
1545 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1546 | break; | |
1547 | } else { | |
1548 | u64 ref_offset; | |
1549 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1550 | if (parent > 0) { | |
1551 | if (parent == ref_offset) { | |
1552 | err = 0; | |
1553 | break; | |
1554 | } | |
1555 | if (ref_offset < parent) | |
1556 | break; | |
1557 | } else { | |
1558 | if (root_objectid == ref_offset) { | |
1559 | err = 0; | |
1560 | break; | |
1561 | } | |
1562 | if (ref_offset < root_objectid) | |
1563 | break; | |
1564 | } | |
1565 | } | |
1566 | ptr += btrfs_extent_inline_ref_size(type); | |
1567 | } | |
1568 | if (err == -ENOENT && insert) { | |
1569 | if (item_size + extra_size >= | |
1570 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1571 | err = -EAGAIN; | |
1572 | goto out; | |
1573 | } | |
1574 | /* | |
1575 | * To add new inline back ref, we have to make sure | |
1576 | * there is no corresponding back ref item. | |
1577 | * For simplicity, we just do not add new inline back | |
1578 | * ref if there is any kind of item for this block | |
1579 | */ | |
2c47e605 YZ |
1580 | if (find_next_key(path, 0, &key) == 0 && |
1581 | key.objectid == bytenr && | |
85d4198e | 1582 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1583 | err = -EAGAIN; |
1584 | goto out; | |
1585 | } | |
1586 | } | |
1587 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1588 | out: | |
85d4198e | 1589 | if (insert) { |
5d4f98a2 YZ |
1590 | path->keep_locks = 0; |
1591 | btrfs_unlock_up_safe(path, 1); | |
1592 | } | |
1593 | return err; | |
1594 | } | |
1595 | ||
1596 | /* | |
1597 | * helper to add new inline back ref | |
1598 | */ | |
1599 | static noinline_for_stack | |
143bede5 JM |
1600 | void setup_inline_extent_backref(struct btrfs_trans_handle *trans, |
1601 | struct btrfs_root *root, | |
1602 | struct btrfs_path *path, | |
1603 | struct btrfs_extent_inline_ref *iref, | |
1604 | u64 parent, u64 root_objectid, | |
1605 | u64 owner, u64 offset, int refs_to_add, | |
1606 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1607 | { |
1608 | struct extent_buffer *leaf; | |
1609 | struct btrfs_extent_item *ei; | |
1610 | unsigned long ptr; | |
1611 | unsigned long end; | |
1612 | unsigned long item_offset; | |
1613 | u64 refs; | |
1614 | int size; | |
1615 | int type; | |
5d4f98a2 YZ |
1616 | |
1617 | leaf = path->nodes[0]; | |
1618 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1619 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1620 | ||
1621 | type = extent_ref_type(parent, owner); | |
1622 | size = btrfs_extent_inline_ref_size(type); | |
1623 | ||
143bede5 | 1624 | btrfs_extend_item(trans, root, path, size); |
5d4f98a2 YZ |
1625 | |
1626 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1627 | refs = btrfs_extent_refs(leaf, ei); | |
1628 | refs += refs_to_add; | |
1629 | btrfs_set_extent_refs(leaf, ei, refs); | |
1630 | if (extent_op) | |
1631 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1632 | ||
1633 | ptr = (unsigned long)ei + item_offset; | |
1634 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1635 | if (ptr < end - size) | |
1636 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1637 | end - size - ptr); | |
1638 | ||
1639 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1640 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1641 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1642 | struct btrfs_extent_data_ref *dref; | |
1643 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1644 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1645 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1646 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1647 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1648 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1649 | struct btrfs_shared_data_ref *sref; | |
1650 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1651 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1652 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1653 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1654 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1655 | } else { | |
1656 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1657 | } | |
1658 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1659 | } |
1660 | ||
1661 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1662 | struct btrfs_root *root, | |
1663 | struct btrfs_path *path, | |
1664 | struct btrfs_extent_inline_ref **ref_ret, | |
1665 | u64 bytenr, u64 num_bytes, u64 parent, | |
1666 | u64 root_objectid, u64 owner, u64 offset) | |
1667 | { | |
1668 | int ret; | |
1669 | ||
1670 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1671 | bytenr, num_bytes, parent, | |
1672 | root_objectid, owner, offset, 0); | |
1673 | if (ret != -ENOENT) | |
54aa1f4d | 1674 | return ret; |
5d4f98a2 | 1675 | |
b3b4aa74 | 1676 | btrfs_release_path(path); |
5d4f98a2 YZ |
1677 | *ref_ret = NULL; |
1678 | ||
1679 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1680 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1681 | root_objectid); | |
1682 | } else { | |
1683 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1684 | root_objectid, owner, offset); | |
b9473439 | 1685 | } |
5d4f98a2 YZ |
1686 | return ret; |
1687 | } | |
31840ae1 | 1688 | |
5d4f98a2 YZ |
1689 | /* |
1690 | * helper to update/remove inline back ref | |
1691 | */ | |
1692 | static noinline_for_stack | |
143bede5 JM |
1693 | void update_inline_extent_backref(struct btrfs_trans_handle *trans, |
1694 | struct btrfs_root *root, | |
1695 | struct btrfs_path *path, | |
1696 | struct btrfs_extent_inline_ref *iref, | |
1697 | int refs_to_mod, | |
1698 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1699 | { |
1700 | struct extent_buffer *leaf; | |
1701 | struct btrfs_extent_item *ei; | |
1702 | struct btrfs_extent_data_ref *dref = NULL; | |
1703 | struct btrfs_shared_data_ref *sref = NULL; | |
1704 | unsigned long ptr; | |
1705 | unsigned long end; | |
1706 | u32 item_size; | |
1707 | int size; | |
1708 | int type; | |
5d4f98a2 YZ |
1709 | u64 refs; |
1710 | ||
1711 | leaf = path->nodes[0]; | |
1712 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1713 | refs = btrfs_extent_refs(leaf, ei); | |
1714 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1715 | refs += refs_to_mod; | |
1716 | btrfs_set_extent_refs(leaf, ei, refs); | |
1717 | if (extent_op) | |
1718 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1719 | ||
1720 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1721 | ||
1722 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1723 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1724 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1725 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1726 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1727 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1728 | } else { | |
1729 | refs = 1; | |
1730 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1731 | } |
31840ae1 | 1732 | |
5d4f98a2 YZ |
1733 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1734 | refs += refs_to_mod; | |
1735 | ||
1736 | if (refs > 0) { | |
1737 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1738 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1739 | else | |
1740 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1741 | } else { | |
1742 | size = btrfs_extent_inline_ref_size(type); | |
1743 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1744 | ptr = (unsigned long)iref; | |
1745 | end = (unsigned long)ei + item_size; | |
1746 | if (ptr + size < end) | |
1747 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1748 | end - ptr - size); | |
1749 | item_size -= size; | |
143bede5 | 1750 | btrfs_truncate_item(trans, root, path, item_size, 1); |
5d4f98a2 YZ |
1751 | } |
1752 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1753 | } |
1754 | ||
1755 | static noinline_for_stack | |
1756 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1757 | struct btrfs_root *root, | |
1758 | struct btrfs_path *path, | |
1759 | u64 bytenr, u64 num_bytes, u64 parent, | |
1760 | u64 root_objectid, u64 owner, | |
1761 | u64 offset, int refs_to_add, | |
1762 | struct btrfs_delayed_extent_op *extent_op) | |
1763 | { | |
1764 | struct btrfs_extent_inline_ref *iref; | |
1765 | int ret; | |
1766 | ||
1767 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1768 | bytenr, num_bytes, parent, | |
1769 | root_objectid, owner, offset, 1); | |
1770 | if (ret == 0) { | |
1771 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
143bede5 JM |
1772 | update_inline_extent_backref(trans, root, path, iref, |
1773 | refs_to_add, extent_op); | |
5d4f98a2 | 1774 | } else if (ret == -ENOENT) { |
143bede5 JM |
1775 | setup_inline_extent_backref(trans, root, path, iref, parent, |
1776 | root_objectid, owner, offset, | |
1777 | refs_to_add, extent_op); | |
1778 | ret = 0; | |
771ed689 | 1779 | } |
5d4f98a2 YZ |
1780 | return ret; |
1781 | } | |
31840ae1 | 1782 | |
5d4f98a2 YZ |
1783 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1784 | struct btrfs_root *root, | |
1785 | struct btrfs_path *path, | |
1786 | u64 bytenr, u64 parent, u64 root_objectid, | |
1787 | u64 owner, u64 offset, int refs_to_add) | |
1788 | { | |
1789 | int ret; | |
1790 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1791 | BUG_ON(refs_to_add != 1); | |
1792 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1793 | parent, root_objectid); | |
1794 | } else { | |
1795 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1796 | parent, root_objectid, | |
1797 | owner, offset, refs_to_add); | |
1798 | } | |
1799 | return ret; | |
1800 | } | |
56bec294 | 1801 | |
5d4f98a2 YZ |
1802 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1803 | struct btrfs_root *root, | |
1804 | struct btrfs_path *path, | |
1805 | struct btrfs_extent_inline_ref *iref, | |
1806 | int refs_to_drop, int is_data) | |
1807 | { | |
143bede5 | 1808 | int ret = 0; |
b9473439 | 1809 | |
5d4f98a2 YZ |
1810 | BUG_ON(!is_data && refs_to_drop != 1); |
1811 | if (iref) { | |
143bede5 JM |
1812 | update_inline_extent_backref(trans, root, path, iref, |
1813 | -refs_to_drop, NULL); | |
5d4f98a2 YZ |
1814 | } else if (is_data) { |
1815 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop); | |
1816 | } else { | |
1817 | ret = btrfs_del_item(trans, root, path); | |
1818 | } | |
1819 | return ret; | |
1820 | } | |
1821 | ||
5378e607 | 1822 | static int btrfs_issue_discard(struct block_device *bdev, |
5d4f98a2 YZ |
1823 | u64 start, u64 len) |
1824 | { | |
5378e607 | 1825 | return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0); |
5d4f98a2 | 1826 | } |
5d4f98a2 YZ |
1827 | |
1828 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 1829 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1830 | { |
5d4f98a2 | 1831 | int ret; |
5378e607 | 1832 | u64 discarded_bytes = 0; |
a1d3c478 | 1833 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1834 | |
e244a0ae | 1835 | |
5d4f98a2 | 1836 | /* Tell the block device(s) that the sectors can be discarded */ |
5378e607 | 1837 | ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD, |
a1d3c478 | 1838 | bytenr, &num_bytes, &bbio, 0); |
79787eaa | 1839 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 1840 | if (!ret) { |
a1d3c478 | 1841 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
1842 | int i; |
1843 | ||
5d4f98a2 | 1844 | |
a1d3c478 | 1845 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d5e2003c JB |
1846 | if (!stripe->dev->can_discard) |
1847 | continue; | |
1848 | ||
5378e607 LD |
1849 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1850 | stripe->physical, | |
1851 | stripe->length); | |
1852 | if (!ret) | |
1853 | discarded_bytes += stripe->length; | |
1854 | else if (ret != -EOPNOTSUPP) | |
79787eaa | 1855 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
1856 | |
1857 | /* | |
1858 | * Just in case we get back EOPNOTSUPP for some reason, | |
1859 | * just ignore the return value so we don't screw up | |
1860 | * people calling discard_extent. | |
1861 | */ | |
1862 | ret = 0; | |
5d4f98a2 | 1863 | } |
a1d3c478 | 1864 | kfree(bbio); |
5d4f98a2 | 1865 | } |
5378e607 LD |
1866 | |
1867 | if (actual_bytes) | |
1868 | *actual_bytes = discarded_bytes; | |
1869 | ||
5d4f98a2 YZ |
1870 | |
1871 | return ret; | |
5d4f98a2 YZ |
1872 | } |
1873 | ||
79787eaa | 1874 | /* Can return -ENOMEM */ |
5d4f98a2 YZ |
1875 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
1876 | struct btrfs_root *root, | |
1877 | u64 bytenr, u64 num_bytes, u64 parent, | |
66d7e7f0 | 1878 | u64 root_objectid, u64 owner, u64 offset, int for_cow) |
5d4f98a2 YZ |
1879 | { |
1880 | int ret; | |
66d7e7f0 AJ |
1881 | struct btrfs_fs_info *fs_info = root->fs_info; |
1882 | ||
5d4f98a2 YZ |
1883 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
1884 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
1885 | ||
1886 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
66d7e7f0 AJ |
1887 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
1888 | num_bytes, | |
5d4f98a2 | 1889 | parent, root_objectid, (int)owner, |
66d7e7f0 | 1890 | BTRFS_ADD_DELAYED_REF, NULL, for_cow); |
5d4f98a2 | 1891 | } else { |
66d7e7f0 AJ |
1892 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
1893 | num_bytes, | |
5d4f98a2 | 1894 | parent, root_objectid, owner, offset, |
66d7e7f0 | 1895 | BTRFS_ADD_DELAYED_REF, NULL, for_cow); |
5d4f98a2 YZ |
1896 | } |
1897 | return ret; | |
1898 | } | |
1899 | ||
1900 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1901 | struct btrfs_root *root, | |
1902 | u64 bytenr, u64 num_bytes, | |
1903 | u64 parent, u64 root_objectid, | |
1904 | u64 owner, u64 offset, int refs_to_add, | |
1905 | struct btrfs_delayed_extent_op *extent_op) | |
1906 | { | |
1907 | struct btrfs_path *path; | |
1908 | struct extent_buffer *leaf; | |
1909 | struct btrfs_extent_item *item; | |
1910 | u64 refs; | |
1911 | int ret; | |
1912 | int err = 0; | |
1913 | ||
1914 | path = btrfs_alloc_path(); | |
1915 | if (!path) | |
1916 | return -ENOMEM; | |
1917 | ||
1918 | path->reada = 1; | |
1919 | path->leave_spinning = 1; | |
1920 | /* this will setup the path even if it fails to insert the back ref */ | |
1921 | ret = insert_inline_extent_backref(trans, root->fs_info->extent_root, | |
1922 | path, bytenr, num_bytes, parent, | |
1923 | root_objectid, owner, offset, | |
1924 | refs_to_add, extent_op); | |
1925 | if (ret == 0) | |
1926 | goto out; | |
1927 | ||
1928 | if (ret != -EAGAIN) { | |
1929 | err = ret; | |
1930 | goto out; | |
1931 | } | |
1932 | ||
1933 | leaf = path->nodes[0]; | |
1934 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1935 | refs = btrfs_extent_refs(leaf, item); | |
1936 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1937 | if (extent_op) | |
1938 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1939 | |
5d4f98a2 | 1940 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 1941 | btrfs_release_path(path); |
56bec294 CM |
1942 | |
1943 | path->reada = 1; | |
b9473439 CM |
1944 | path->leave_spinning = 1; |
1945 | ||
56bec294 CM |
1946 | /* now insert the actual backref */ |
1947 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
1948 | path, bytenr, parent, root_objectid, |
1949 | owner, offset, refs_to_add); | |
79787eaa JM |
1950 | if (ret) |
1951 | btrfs_abort_transaction(trans, root, ret); | |
5d4f98a2 | 1952 | out: |
56bec294 | 1953 | btrfs_free_path(path); |
5d4f98a2 | 1954 | return err; |
56bec294 CM |
1955 | } |
1956 | ||
5d4f98a2 YZ |
1957 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
1958 | struct btrfs_root *root, | |
1959 | struct btrfs_delayed_ref_node *node, | |
1960 | struct btrfs_delayed_extent_op *extent_op, | |
1961 | int insert_reserved) | |
56bec294 | 1962 | { |
5d4f98a2 YZ |
1963 | int ret = 0; |
1964 | struct btrfs_delayed_data_ref *ref; | |
1965 | struct btrfs_key ins; | |
1966 | u64 parent = 0; | |
1967 | u64 ref_root = 0; | |
1968 | u64 flags = 0; | |
1969 | ||
1970 | ins.objectid = node->bytenr; | |
1971 | ins.offset = node->num_bytes; | |
1972 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1973 | ||
1974 | ref = btrfs_delayed_node_to_data_ref(node); | |
1975 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) | |
1976 | parent = ref->parent; | |
1977 | else | |
1978 | ref_root = ref->root; | |
1979 | ||
1980 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
1981 | if (extent_op) { | |
1982 | BUG_ON(extent_op->update_key); | |
1983 | flags |= extent_op->flags_to_set; | |
1984 | } | |
1985 | ret = alloc_reserved_file_extent(trans, root, | |
1986 | parent, ref_root, flags, | |
1987 | ref->objectid, ref->offset, | |
1988 | &ins, node->ref_mod); | |
5d4f98a2 YZ |
1989 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
1990 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
1991 | node->num_bytes, parent, | |
1992 | ref_root, ref->objectid, | |
1993 | ref->offset, node->ref_mod, | |
1994 | extent_op); | |
1995 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
1996 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
1997 | node->num_bytes, parent, | |
1998 | ref_root, ref->objectid, | |
1999 | ref->offset, node->ref_mod, | |
2000 | extent_op); | |
2001 | } else { | |
2002 | BUG(); | |
2003 | } | |
2004 | return ret; | |
2005 | } | |
2006 | ||
2007 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2008 | struct extent_buffer *leaf, | |
2009 | struct btrfs_extent_item *ei) | |
2010 | { | |
2011 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2012 | if (extent_op->update_flags) { | |
2013 | flags |= extent_op->flags_to_set; | |
2014 | btrfs_set_extent_flags(leaf, ei, flags); | |
2015 | } | |
2016 | ||
2017 | if (extent_op->update_key) { | |
2018 | struct btrfs_tree_block_info *bi; | |
2019 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2020 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2021 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2022 | } | |
2023 | } | |
2024 | ||
2025 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2026 | struct btrfs_root *root, | |
2027 | struct btrfs_delayed_ref_node *node, | |
2028 | struct btrfs_delayed_extent_op *extent_op) | |
2029 | { | |
2030 | struct btrfs_key key; | |
2031 | struct btrfs_path *path; | |
2032 | struct btrfs_extent_item *ei; | |
2033 | struct extent_buffer *leaf; | |
2034 | u32 item_size; | |
56bec294 | 2035 | int ret; |
5d4f98a2 YZ |
2036 | int err = 0; |
2037 | ||
79787eaa JM |
2038 | if (trans->aborted) |
2039 | return 0; | |
2040 | ||
5d4f98a2 YZ |
2041 | path = btrfs_alloc_path(); |
2042 | if (!path) | |
2043 | return -ENOMEM; | |
2044 | ||
2045 | key.objectid = node->bytenr; | |
2046 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2047 | key.offset = node->num_bytes; | |
2048 | ||
2049 | path->reada = 1; | |
2050 | path->leave_spinning = 1; | |
2051 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
2052 | path, 0, 1); | |
2053 | if (ret < 0) { | |
2054 | err = ret; | |
2055 | goto out; | |
2056 | } | |
2057 | if (ret > 0) { | |
2058 | err = -EIO; | |
2059 | goto out; | |
2060 | } | |
2061 | ||
2062 | leaf = path->nodes[0]; | |
2063 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2064 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2065 | if (item_size < sizeof(*ei)) { | |
2066 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
2067 | path, (u64)-1, 0); | |
2068 | if (ret < 0) { | |
2069 | err = ret; | |
2070 | goto out; | |
2071 | } | |
2072 | leaf = path->nodes[0]; | |
2073 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2074 | } | |
2075 | #endif | |
2076 | BUG_ON(item_size < sizeof(*ei)); | |
2077 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2078 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2079 | |
5d4f98a2 YZ |
2080 | btrfs_mark_buffer_dirty(leaf); |
2081 | out: | |
2082 | btrfs_free_path(path); | |
2083 | return err; | |
56bec294 CM |
2084 | } |
2085 | ||
5d4f98a2 YZ |
2086 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2087 | struct btrfs_root *root, | |
2088 | struct btrfs_delayed_ref_node *node, | |
2089 | struct btrfs_delayed_extent_op *extent_op, | |
2090 | int insert_reserved) | |
56bec294 CM |
2091 | { |
2092 | int ret = 0; | |
5d4f98a2 YZ |
2093 | struct btrfs_delayed_tree_ref *ref; |
2094 | struct btrfs_key ins; | |
2095 | u64 parent = 0; | |
2096 | u64 ref_root = 0; | |
56bec294 | 2097 | |
5d4f98a2 YZ |
2098 | ins.objectid = node->bytenr; |
2099 | ins.offset = node->num_bytes; | |
2100 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
56bec294 | 2101 | |
5d4f98a2 YZ |
2102 | ref = btrfs_delayed_node_to_tree_ref(node); |
2103 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2104 | parent = ref->parent; | |
2105 | else | |
2106 | ref_root = ref->root; | |
2107 | ||
2108 | BUG_ON(node->ref_mod != 1); | |
2109 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
2110 | BUG_ON(!extent_op || !extent_op->update_flags || | |
2111 | !extent_op->update_key); | |
2112 | ret = alloc_reserved_tree_block(trans, root, | |
2113 | parent, ref_root, | |
2114 | extent_op->flags_to_set, | |
2115 | &extent_op->key, | |
2116 | ref->level, &ins); | |
5d4f98a2 YZ |
2117 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2118 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
2119 | node->num_bytes, parent, ref_root, | |
2120 | ref->level, 0, 1, extent_op); | |
2121 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
2122 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
2123 | node->num_bytes, parent, ref_root, | |
2124 | ref->level, 0, 1, extent_op); | |
2125 | } else { | |
2126 | BUG(); | |
2127 | } | |
56bec294 CM |
2128 | return ret; |
2129 | } | |
2130 | ||
2131 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 YZ |
2132 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2133 | struct btrfs_root *root, | |
2134 | struct btrfs_delayed_ref_node *node, | |
2135 | struct btrfs_delayed_extent_op *extent_op, | |
2136 | int insert_reserved) | |
56bec294 | 2137 | { |
79787eaa JM |
2138 | int ret = 0; |
2139 | ||
2140 | if (trans->aborted) | |
2141 | return 0; | |
2142 | ||
5d4f98a2 | 2143 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2144 | struct btrfs_delayed_ref_head *head; |
2145 | /* | |
2146 | * we've hit the end of the chain and we were supposed | |
2147 | * to insert this extent into the tree. But, it got | |
2148 | * deleted before we ever needed to insert it, so all | |
2149 | * we have to do is clean up the accounting | |
2150 | */ | |
5d4f98a2 YZ |
2151 | BUG_ON(extent_op); |
2152 | head = btrfs_delayed_node_to_head(node); | |
56bec294 | 2153 | if (insert_reserved) { |
f0486c68 YZ |
2154 | btrfs_pin_extent(root, node->bytenr, |
2155 | node->num_bytes, 1); | |
5d4f98a2 YZ |
2156 | if (head->is_data) { |
2157 | ret = btrfs_del_csums(trans, root, | |
2158 | node->bytenr, | |
2159 | node->num_bytes); | |
5d4f98a2 | 2160 | } |
56bec294 | 2161 | } |
56bec294 | 2162 | mutex_unlock(&head->mutex); |
79787eaa | 2163 | return ret; |
56bec294 CM |
2164 | } |
2165 | ||
5d4f98a2 YZ |
2166 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2167 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2168 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2169 | insert_reserved); | |
2170 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2171 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2172 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2173 | insert_reserved); | |
2174 | else | |
2175 | BUG(); | |
2176 | return ret; | |
56bec294 CM |
2177 | } |
2178 | ||
2179 | static noinline struct btrfs_delayed_ref_node * | |
2180 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
2181 | { | |
2182 | struct rb_node *node; | |
2183 | struct btrfs_delayed_ref_node *ref; | |
2184 | int action = BTRFS_ADD_DELAYED_REF; | |
2185 | again: | |
2186 | /* | |
2187 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2188 | * this prevents ref count from going down to zero when | |
2189 | * there still are pending delayed ref. | |
2190 | */ | |
2191 | node = rb_prev(&head->node.rb_node); | |
2192 | while (1) { | |
2193 | if (!node) | |
2194 | break; | |
2195 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2196 | rb_node); | |
2197 | if (ref->bytenr != head->node.bytenr) | |
2198 | break; | |
5d4f98a2 | 2199 | if (ref->action == action) |
56bec294 CM |
2200 | return ref; |
2201 | node = rb_prev(node); | |
2202 | } | |
2203 | if (action == BTRFS_ADD_DELAYED_REF) { | |
2204 | action = BTRFS_DROP_DELAYED_REF; | |
2205 | goto again; | |
2206 | } | |
2207 | return NULL; | |
2208 | } | |
2209 | ||
79787eaa JM |
2210 | /* |
2211 | * Returns 0 on success or if called with an already aborted transaction. | |
2212 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2213 | */ | |
c3e69d58 CM |
2214 | static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, |
2215 | struct btrfs_root *root, | |
2216 | struct list_head *cluster) | |
56bec294 | 2217 | { |
56bec294 CM |
2218 | struct btrfs_delayed_ref_root *delayed_refs; |
2219 | struct btrfs_delayed_ref_node *ref; | |
2220 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2221 | struct btrfs_delayed_extent_op *extent_op; |
097b8a7c | 2222 | struct btrfs_fs_info *fs_info = root->fs_info; |
56bec294 | 2223 | int ret; |
c3e69d58 | 2224 | int count = 0; |
56bec294 | 2225 | int must_insert_reserved = 0; |
56bec294 CM |
2226 | |
2227 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2228 | while (1) { |
2229 | if (!locked_ref) { | |
c3e69d58 CM |
2230 | /* pick a new head ref from the cluster list */ |
2231 | if (list_empty(cluster)) | |
56bec294 | 2232 | break; |
56bec294 | 2233 | |
c3e69d58 CM |
2234 | locked_ref = list_entry(cluster->next, |
2235 | struct btrfs_delayed_ref_head, cluster); | |
2236 | ||
2237 | /* grab the lock that says we are going to process | |
2238 | * all the refs for this head */ | |
2239 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
2240 | ||
2241 | /* | |
2242 | * we may have dropped the spin lock to get the head | |
2243 | * mutex lock, and that might have given someone else | |
2244 | * time to free the head. If that's true, it has been | |
2245 | * removed from our list and we can move on. | |
2246 | */ | |
2247 | if (ret == -EAGAIN) { | |
2248 | locked_ref = NULL; | |
2249 | count++; | |
2250 | continue; | |
56bec294 CM |
2251 | } |
2252 | } | |
a28ec197 | 2253 | |
d1270cd9 AJ |
2254 | /* |
2255 | * locked_ref is the head node, so we have to go one | |
2256 | * node back for any delayed ref updates | |
2257 | */ | |
2258 | ref = select_delayed_ref(locked_ref); | |
2259 | ||
2260 | if (ref && ref->seq && | |
097b8a7c | 2261 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d1270cd9 AJ |
2262 | /* |
2263 | * there are still refs with lower seq numbers in the | |
2264 | * process of being added. Don't run this ref yet. | |
2265 | */ | |
2266 | list_del_init(&locked_ref->cluster); | |
2267 | mutex_unlock(&locked_ref->mutex); | |
2268 | locked_ref = NULL; | |
2269 | delayed_refs->num_heads_ready++; | |
2270 | spin_unlock(&delayed_refs->lock); | |
2271 | cond_resched(); | |
2272 | spin_lock(&delayed_refs->lock); | |
2273 | continue; | |
2274 | } | |
2275 | ||
56bec294 CM |
2276 | /* |
2277 | * record the must insert reserved flag before we | |
2278 | * drop the spin lock. | |
2279 | */ | |
2280 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2281 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2282 | |
5d4f98a2 YZ |
2283 | extent_op = locked_ref->extent_op; |
2284 | locked_ref->extent_op = NULL; | |
2285 | ||
56bec294 CM |
2286 | if (!ref) { |
2287 | /* All delayed refs have been processed, Go ahead | |
2288 | * and send the head node to run_one_delayed_ref, | |
2289 | * so that any accounting fixes can happen | |
2290 | */ | |
2291 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2292 | |
2293 | if (extent_op && must_insert_reserved) { | |
2294 | kfree(extent_op); | |
2295 | extent_op = NULL; | |
2296 | } | |
2297 | ||
2298 | if (extent_op) { | |
2299 | spin_unlock(&delayed_refs->lock); | |
2300 | ||
2301 | ret = run_delayed_extent_op(trans, root, | |
2302 | ref, extent_op); | |
5d4f98a2 YZ |
2303 | kfree(extent_op); |
2304 | ||
79787eaa JM |
2305 | if (ret) { |
2306 | printk(KERN_DEBUG "btrfs: run_delayed_extent_op returned %d\n", ret); | |
253beebd | 2307 | spin_lock(&delayed_refs->lock); |
79787eaa JM |
2308 | return ret; |
2309 | } | |
2310 | ||
203bf287 | 2311 | goto next; |
5d4f98a2 YZ |
2312 | } |
2313 | ||
c3e69d58 | 2314 | list_del_init(&locked_ref->cluster); |
56bec294 CM |
2315 | locked_ref = NULL; |
2316 | } | |
02217ed2 | 2317 | |
56bec294 CM |
2318 | ref->in_tree = 0; |
2319 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
2320 | delayed_refs->num_entries--; | |
a168650c JS |
2321 | /* |
2322 | * we modified num_entries, but as we're currently running | |
2323 | * delayed refs, skip | |
2324 | * wake_up(&delayed_refs->seq_wait); | |
2325 | * here. | |
2326 | */ | |
56bec294 | 2327 | spin_unlock(&delayed_refs->lock); |
925baedd | 2328 | |
5d4f98a2 | 2329 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 | 2330 | must_insert_reserved); |
eb099670 | 2331 | |
5d4f98a2 YZ |
2332 | btrfs_put_delayed_ref(ref); |
2333 | kfree(extent_op); | |
c3e69d58 | 2334 | count++; |
79787eaa JM |
2335 | |
2336 | if (ret) { | |
2337 | printk(KERN_DEBUG "btrfs: run_one_delayed_ref returned %d\n", ret); | |
253beebd | 2338 | spin_lock(&delayed_refs->lock); |
79787eaa JM |
2339 | return ret; |
2340 | } | |
2341 | ||
203bf287 | 2342 | next: |
097b8a7c | 2343 | do_chunk_alloc(trans, fs_info->extent_root, |
203bf287 CM |
2344 | 2 * 1024 * 1024, |
2345 | btrfs_get_alloc_profile(root, 0), | |
2346 | CHUNK_ALLOC_NO_FORCE); | |
c3e69d58 CM |
2347 | cond_resched(); |
2348 | spin_lock(&delayed_refs->lock); | |
2349 | } | |
2350 | return count; | |
2351 | } | |
2352 | ||
097b8a7c JS |
2353 | static void wait_for_more_refs(struct btrfs_fs_info *fs_info, |
2354 | struct btrfs_delayed_ref_root *delayed_refs, | |
8ca78f3e JS |
2355 | unsigned long num_refs, |
2356 | struct list_head *first_seq) | |
a168650c | 2357 | { |
a168650c JS |
2358 | spin_unlock(&delayed_refs->lock); |
2359 | pr_debug("waiting for more refs (num %ld, first %p)\n", | |
2360 | num_refs, first_seq); | |
097b8a7c | 2361 | wait_event(fs_info->tree_mod_seq_wait, |
a168650c | 2362 | num_refs != delayed_refs->num_entries || |
097b8a7c | 2363 | fs_info->tree_mod_seq_list.next != first_seq); |
a168650c | 2364 | pr_debug("done waiting for more refs (num %ld, first %p)\n", |
097b8a7c | 2365 | delayed_refs->num_entries, fs_info->tree_mod_seq_list.next); |
a168650c JS |
2366 | spin_lock(&delayed_refs->lock); |
2367 | } | |
2368 | ||
709c0486 AJ |
2369 | #ifdef SCRAMBLE_DELAYED_REFS |
2370 | /* | |
2371 | * Normally delayed refs get processed in ascending bytenr order. This | |
2372 | * correlates in most cases to the order added. To expose dependencies on this | |
2373 | * order, we start to process the tree in the middle instead of the beginning | |
2374 | */ | |
2375 | static u64 find_middle(struct rb_root *root) | |
2376 | { | |
2377 | struct rb_node *n = root->rb_node; | |
2378 | struct btrfs_delayed_ref_node *entry; | |
2379 | int alt = 1; | |
2380 | u64 middle; | |
2381 | u64 first = 0, last = 0; | |
2382 | ||
2383 | n = rb_first(root); | |
2384 | if (n) { | |
2385 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2386 | first = entry->bytenr; | |
2387 | } | |
2388 | n = rb_last(root); | |
2389 | if (n) { | |
2390 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2391 | last = entry->bytenr; | |
2392 | } | |
2393 | n = root->rb_node; | |
2394 | ||
2395 | while (n) { | |
2396 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2397 | WARN_ON(!entry->in_tree); | |
2398 | ||
2399 | middle = entry->bytenr; | |
2400 | ||
2401 | if (alt) | |
2402 | n = n->rb_left; | |
2403 | else | |
2404 | n = n->rb_right; | |
2405 | ||
2406 | alt = 1 - alt; | |
2407 | } | |
2408 | return middle; | |
2409 | } | |
2410 | #endif | |
2411 | ||
bed92eae AJ |
2412 | int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans, |
2413 | struct btrfs_fs_info *fs_info) | |
2414 | { | |
2415 | struct qgroup_update *qgroup_update; | |
2416 | int ret = 0; | |
2417 | ||
2418 | if (list_empty(&trans->qgroup_ref_list) != | |
2419 | !trans->delayed_ref_elem.seq) { | |
2420 | /* list without seq or seq without list */ | |
2421 | printk(KERN_ERR "btrfs: qgroup accounting update error, list is%s empty, seq is %llu\n", | |
2422 | list_empty(&trans->qgroup_ref_list) ? "" : " not", | |
2423 | trans->delayed_ref_elem.seq); | |
2424 | BUG(); | |
2425 | } | |
2426 | ||
2427 | if (!trans->delayed_ref_elem.seq) | |
2428 | return 0; | |
2429 | ||
2430 | while (!list_empty(&trans->qgroup_ref_list)) { | |
2431 | qgroup_update = list_first_entry(&trans->qgroup_ref_list, | |
2432 | struct qgroup_update, list); | |
2433 | list_del(&qgroup_update->list); | |
2434 | if (!ret) | |
2435 | ret = btrfs_qgroup_account_ref( | |
2436 | trans, fs_info, qgroup_update->node, | |
2437 | qgroup_update->extent_op); | |
2438 | kfree(qgroup_update); | |
2439 | } | |
2440 | ||
2441 | btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem); | |
2442 | ||
2443 | return ret; | |
2444 | } | |
2445 | ||
c3e69d58 CM |
2446 | /* |
2447 | * this starts processing the delayed reference count updates and | |
2448 | * extent insertions we have queued up so far. count can be | |
2449 | * 0, which means to process everything in the tree at the start | |
2450 | * of the run (but not newly added entries), or it can be some target | |
2451 | * number you'd like to process. | |
79787eaa JM |
2452 | * |
2453 | * Returns 0 on success or if called with an aborted transaction | |
2454 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2455 | */ |
2456 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2457 | struct btrfs_root *root, unsigned long count) | |
2458 | { | |
2459 | struct rb_node *node; | |
2460 | struct btrfs_delayed_ref_root *delayed_refs; | |
2461 | struct btrfs_delayed_ref_node *ref; | |
2462 | struct list_head cluster; | |
8ca78f3e | 2463 | struct list_head *first_seq = NULL; |
c3e69d58 | 2464 | int ret; |
a168650c | 2465 | u64 delayed_start; |
c3e69d58 CM |
2466 | int run_all = count == (unsigned long)-1; |
2467 | int run_most = 0; | |
a168650c JS |
2468 | unsigned long num_refs = 0; |
2469 | int consider_waiting; | |
c3e69d58 | 2470 | |
79787eaa JM |
2471 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2472 | if (trans->aborted) | |
2473 | return 0; | |
2474 | ||
c3e69d58 CM |
2475 | if (root == root->fs_info->extent_root) |
2476 | root = root->fs_info->tree_root; | |
2477 | ||
203bf287 CM |
2478 | do_chunk_alloc(trans, root->fs_info->extent_root, |
2479 | 2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0), | |
2480 | CHUNK_ALLOC_NO_FORCE); | |
2481 | ||
edf39272 JS |
2482 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); |
2483 | ||
c3e69d58 CM |
2484 | delayed_refs = &trans->transaction->delayed_refs; |
2485 | INIT_LIST_HEAD(&cluster); | |
2486 | again: | |
a168650c | 2487 | consider_waiting = 0; |
c3e69d58 | 2488 | spin_lock(&delayed_refs->lock); |
097b8a7c | 2489 | |
709c0486 AJ |
2490 | #ifdef SCRAMBLE_DELAYED_REFS |
2491 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2492 | #endif | |
2493 | ||
c3e69d58 CM |
2494 | if (count == 0) { |
2495 | count = delayed_refs->num_entries * 2; | |
2496 | run_most = 1; | |
2497 | } | |
2498 | while (1) { | |
2499 | if (!(run_all || run_most) && | |
2500 | delayed_refs->num_heads_ready < 64) | |
2501 | break; | |
eb099670 | 2502 | |
56bec294 | 2503 | /* |
c3e69d58 CM |
2504 | * go find something we can process in the rbtree. We start at |
2505 | * the beginning of the tree, and then build a cluster | |
2506 | * of refs to process starting at the first one we are able to | |
2507 | * lock | |
56bec294 | 2508 | */ |
a168650c | 2509 | delayed_start = delayed_refs->run_delayed_start; |
c3e69d58 CM |
2510 | ret = btrfs_find_ref_cluster(trans, &cluster, |
2511 | delayed_refs->run_delayed_start); | |
2512 | if (ret) | |
56bec294 CM |
2513 | break; |
2514 | ||
a168650c JS |
2515 | if (delayed_start >= delayed_refs->run_delayed_start) { |
2516 | if (consider_waiting == 0) { | |
2517 | /* | |
2518 | * btrfs_find_ref_cluster looped. let's do one | |
2519 | * more cycle. if we don't run any delayed ref | |
2520 | * during that cycle (because we can't because | |
2521 | * all of them are blocked) and if the number of | |
2522 | * refs doesn't change, we avoid busy waiting. | |
2523 | */ | |
2524 | consider_waiting = 1; | |
2525 | num_refs = delayed_refs->num_entries; | |
8ca78f3e | 2526 | first_seq = root->fs_info->tree_mod_seq_list.next; |
a168650c | 2527 | } else { |
097b8a7c | 2528 | wait_for_more_refs(root->fs_info, delayed_refs, |
8ca78f3e | 2529 | num_refs, first_seq); |
a168650c JS |
2530 | /* |
2531 | * after waiting, things have changed. we | |
2532 | * dropped the lock and someone else might have | |
2533 | * run some refs, built new clusters and so on. | |
2534 | * therefore, we restart staleness detection. | |
2535 | */ | |
2536 | consider_waiting = 0; | |
2537 | } | |
2538 | } | |
2539 | ||
c3e69d58 | 2540 | ret = run_clustered_refs(trans, root, &cluster); |
79787eaa JM |
2541 | if (ret < 0) { |
2542 | spin_unlock(&delayed_refs->lock); | |
2543 | btrfs_abort_transaction(trans, root, ret); | |
2544 | return ret; | |
2545 | } | |
c3e69d58 CM |
2546 | |
2547 | count -= min_t(unsigned long, ret, count); | |
2548 | ||
2549 | if (count == 0) | |
2550 | break; | |
a168650c JS |
2551 | |
2552 | if (ret || delayed_refs->run_delayed_start == 0) { | |
2553 | /* refs were run, let's reset staleness detection */ | |
2554 | consider_waiting = 0; | |
2555 | } | |
eb099670 | 2556 | } |
c3e69d58 | 2557 | |
56bec294 | 2558 | if (run_all) { |
56bec294 | 2559 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 2560 | if (!node) |
56bec294 | 2561 | goto out; |
c3e69d58 | 2562 | count = (unsigned long)-1; |
e9d0b13b | 2563 | |
56bec294 CM |
2564 | while (node) { |
2565 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2566 | rb_node); | |
2567 | if (btrfs_delayed_ref_is_head(ref)) { | |
2568 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 2569 | |
56bec294 CM |
2570 | head = btrfs_delayed_node_to_head(ref); |
2571 | atomic_inc(&ref->refs); | |
2572 | ||
2573 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2574 | /* |
2575 | * Mutex was contended, block until it's | |
2576 | * released and try again | |
2577 | */ | |
56bec294 CM |
2578 | mutex_lock(&head->mutex); |
2579 | mutex_unlock(&head->mutex); | |
2580 | ||
2581 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2582 | cond_resched(); |
56bec294 CM |
2583 | goto again; |
2584 | } | |
2585 | node = rb_next(node); | |
2586 | } | |
2587 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
2588 | schedule_timeout(1); |
2589 | goto again; | |
5f39d397 | 2590 | } |
54aa1f4d | 2591 | out: |
c3e69d58 | 2592 | spin_unlock(&delayed_refs->lock); |
edf39272 | 2593 | assert_qgroups_uptodate(trans); |
a28ec197 CM |
2594 | return 0; |
2595 | } | |
2596 | ||
5d4f98a2 YZ |
2597 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2598 | struct btrfs_root *root, | |
2599 | u64 bytenr, u64 num_bytes, u64 flags, | |
2600 | int is_data) | |
2601 | { | |
2602 | struct btrfs_delayed_extent_op *extent_op; | |
2603 | int ret; | |
2604 | ||
2605 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
2606 | if (!extent_op) | |
2607 | return -ENOMEM; | |
2608 | ||
2609 | extent_op->flags_to_set = flags; | |
2610 | extent_op->update_flags = 1; | |
2611 | extent_op->update_key = 0; | |
2612 | extent_op->is_data = is_data ? 1 : 0; | |
2613 | ||
66d7e7f0 AJ |
2614 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, |
2615 | num_bytes, extent_op); | |
5d4f98a2 YZ |
2616 | if (ret) |
2617 | kfree(extent_op); | |
2618 | return ret; | |
2619 | } | |
2620 | ||
2621 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2622 | struct btrfs_root *root, | |
2623 | struct btrfs_path *path, | |
2624 | u64 objectid, u64 offset, u64 bytenr) | |
2625 | { | |
2626 | struct btrfs_delayed_ref_head *head; | |
2627 | struct btrfs_delayed_ref_node *ref; | |
2628 | struct btrfs_delayed_data_ref *data_ref; | |
2629 | struct btrfs_delayed_ref_root *delayed_refs; | |
2630 | struct rb_node *node; | |
2631 | int ret = 0; | |
2632 | ||
2633 | ret = -ENOENT; | |
2634 | delayed_refs = &trans->transaction->delayed_refs; | |
2635 | spin_lock(&delayed_refs->lock); | |
2636 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2637 | if (!head) | |
2638 | goto out; | |
2639 | ||
2640 | if (!mutex_trylock(&head->mutex)) { | |
2641 | atomic_inc(&head->node.refs); | |
2642 | spin_unlock(&delayed_refs->lock); | |
2643 | ||
b3b4aa74 | 2644 | btrfs_release_path(path); |
5d4f98a2 | 2645 | |
8cc33e5c DS |
2646 | /* |
2647 | * Mutex was contended, block until it's released and let | |
2648 | * caller try again | |
2649 | */ | |
5d4f98a2 YZ |
2650 | mutex_lock(&head->mutex); |
2651 | mutex_unlock(&head->mutex); | |
2652 | btrfs_put_delayed_ref(&head->node); | |
2653 | return -EAGAIN; | |
2654 | } | |
2655 | ||
2656 | node = rb_prev(&head->node.rb_node); | |
2657 | if (!node) | |
2658 | goto out_unlock; | |
2659 | ||
2660 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2661 | ||
2662 | if (ref->bytenr != bytenr) | |
2663 | goto out_unlock; | |
2664 | ||
2665 | ret = 1; | |
2666 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2667 | goto out_unlock; | |
2668 | ||
2669 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2670 | ||
2671 | node = rb_prev(node); | |
2672 | if (node) { | |
df57dbe6 LB |
2673 | int seq = ref->seq; |
2674 | ||
5d4f98a2 | 2675 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); |
df57dbe6 | 2676 | if (ref->bytenr == bytenr && ref->seq == seq) |
5d4f98a2 YZ |
2677 | goto out_unlock; |
2678 | } | |
2679 | ||
2680 | if (data_ref->root != root->root_key.objectid || | |
2681 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2682 | goto out_unlock; | |
2683 | ||
2684 | ret = 0; | |
2685 | out_unlock: | |
2686 | mutex_unlock(&head->mutex); | |
2687 | out: | |
2688 | spin_unlock(&delayed_refs->lock); | |
2689 | return ret; | |
2690 | } | |
2691 | ||
2692 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2693 | struct btrfs_root *root, | |
2694 | struct btrfs_path *path, | |
2695 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2696 | { |
2697 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2698 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2699 | struct btrfs_extent_data_ref *ref; |
2700 | struct btrfs_extent_inline_ref *iref; | |
2701 | struct btrfs_extent_item *ei; | |
f321e491 | 2702 | struct btrfs_key key; |
5d4f98a2 | 2703 | u32 item_size; |
be20aa9d | 2704 | int ret; |
925baedd | 2705 | |
be20aa9d | 2706 | key.objectid = bytenr; |
31840ae1 | 2707 | key.offset = (u64)-1; |
f321e491 | 2708 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2709 | |
be20aa9d CM |
2710 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2711 | if (ret < 0) | |
2712 | goto out; | |
79787eaa | 2713 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2714 | |
2715 | ret = -ENOENT; | |
2716 | if (path->slots[0] == 0) | |
31840ae1 | 2717 | goto out; |
be20aa9d | 2718 | |
31840ae1 | 2719 | path->slots[0]--; |
f321e491 | 2720 | leaf = path->nodes[0]; |
5d4f98a2 | 2721 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2722 | |
5d4f98a2 | 2723 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2724 | goto out; |
f321e491 | 2725 | |
5d4f98a2 YZ |
2726 | ret = 1; |
2727 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2728 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2729 | if (item_size < sizeof(*ei)) { | |
2730 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2731 | goto out; | |
2732 | } | |
2733 | #endif | |
2734 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2735 | |
5d4f98a2 YZ |
2736 | if (item_size != sizeof(*ei) + |
2737 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2738 | goto out; | |
be20aa9d | 2739 | |
5d4f98a2 YZ |
2740 | if (btrfs_extent_generation(leaf, ei) <= |
2741 | btrfs_root_last_snapshot(&root->root_item)) | |
2742 | goto out; | |
2743 | ||
2744 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2745 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2746 | BTRFS_EXTENT_DATA_REF_KEY) | |
2747 | goto out; | |
2748 | ||
2749 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2750 | if (btrfs_extent_refs(leaf, ei) != | |
2751 | btrfs_extent_data_ref_count(leaf, ref) || | |
2752 | btrfs_extent_data_ref_root(leaf, ref) != | |
2753 | root->root_key.objectid || | |
2754 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2755 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2756 | goto out; | |
2757 | ||
2758 | ret = 0; | |
2759 | out: | |
2760 | return ret; | |
2761 | } | |
2762 | ||
2763 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2764 | struct btrfs_root *root, | |
2765 | u64 objectid, u64 offset, u64 bytenr) | |
2766 | { | |
2767 | struct btrfs_path *path; | |
2768 | int ret; | |
2769 | int ret2; | |
2770 | ||
2771 | path = btrfs_alloc_path(); | |
2772 | if (!path) | |
2773 | return -ENOENT; | |
2774 | ||
2775 | do { | |
2776 | ret = check_committed_ref(trans, root, path, objectid, | |
2777 | offset, bytenr); | |
2778 | if (ret && ret != -ENOENT) | |
f321e491 | 2779 | goto out; |
80ff3856 | 2780 | |
5d4f98a2 YZ |
2781 | ret2 = check_delayed_ref(trans, root, path, objectid, |
2782 | offset, bytenr); | |
2783 | } while (ret2 == -EAGAIN); | |
2784 | ||
2785 | if (ret2 && ret2 != -ENOENT) { | |
2786 | ret = ret2; | |
2787 | goto out; | |
f321e491 | 2788 | } |
5d4f98a2 YZ |
2789 | |
2790 | if (ret != -ENOENT || ret2 != -ENOENT) | |
2791 | ret = 0; | |
be20aa9d | 2792 | out: |
80ff3856 | 2793 | btrfs_free_path(path); |
f0486c68 YZ |
2794 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2795 | WARN_ON(ret > 0); | |
f321e491 | 2796 | return ret; |
be20aa9d | 2797 | } |
c5739bba | 2798 | |
5d4f98a2 | 2799 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2800 | struct btrfs_root *root, |
5d4f98a2 | 2801 | struct extent_buffer *buf, |
66d7e7f0 | 2802 | int full_backref, int inc, int for_cow) |
31840ae1 ZY |
2803 | { |
2804 | u64 bytenr; | |
5d4f98a2 YZ |
2805 | u64 num_bytes; |
2806 | u64 parent; | |
31840ae1 | 2807 | u64 ref_root; |
31840ae1 | 2808 | u32 nritems; |
31840ae1 ZY |
2809 | struct btrfs_key key; |
2810 | struct btrfs_file_extent_item *fi; | |
2811 | int i; | |
2812 | int level; | |
2813 | int ret = 0; | |
31840ae1 | 2814 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
66d7e7f0 | 2815 | u64, u64, u64, u64, u64, u64, int); |
31840ae1 ZY |
2816 | |
2817 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
2818 | nritems = btrfs_header_nritems(buf); |
2819 | level = btrfs_header_level(buf); | |
2820 | ||
5d4f98a2 YZ |
2821 | if (!root->ref_cows && level == 0) |
2822 | return 0; | |
31840ae1 | 2823 | |
5d4f98a2 YZ |
2824 | if (inc) |
2825 | process_func = btrfs_inc_extent_ref; | |
2826 | else | |
2827 | process_func = btrfs_free_extent; | |
31840ae1 | 2828 | |
5d4f98a2 YZ |
2829 | if (full_backref) |
2830 | parent = buf->start; | |
2831 | else | |
2832 | parent = 0; | |
2833 | ||
2834 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2835 | if (level == 0) { |
5d4f98a2 | 2836 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
2837 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
2838 | continue; | |
5d4f98a2 | 2839 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2840 | struct btrfs_file_extent_item); |
2841 | if (btrfs_file_extent_type(buf, fi) == | |
2842 | BTRFS_FILE_EXTENT_INLINE) | |
2843 | continue; | |
2844 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2845 | if (bytenr == 0) | |
2846 | continue; | |
5d4f98a2 YZ |
2847 | |
2848 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2849 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2850 | ret = process_func(trans, root, bytenr, num_bytes, | |
2851 | parent, ref_root, key.objectid, | |
66d7e7f0 | 2852 | key.offset, for_cow); |
31840ae1 ZY |
2853 | if (ret) |
2854 | goto fail; | |
2855 | } else { | |
5d4f98a2 YZ |
2856 | bytenr = btrfs_node_blockptr(buf, i); |
2857 | num_bytes = btrfs_level_size(root, level - 1); | |
2858 | ret = process_func(trans, root, bytenr, num_bytes, | |
66d7e7f0 AJ |
2859 | parent, ref_root, level - 1, 0, |
2860 | for_cow); | |
31840ae1 ZY |
2861 | if (ret) |
2862 | goto fail; | |
2863 | } | |
2864 | } | |
2865 | return 0; | |
2866 | fail: | |
5d4f98a2 YZ |
2867 | return ret; |
2868 | } | |
2869 | ||
2870 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 2871 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 2872 | { |
66d7e7f0 | 2873 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow); |
5d4f98a2 YZ |
2874 | } |
2875 | ||
2876 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 2877 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 2878 | { |
66d7e7f0 | 2879 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow); |
31840ae1 ZY |
2880 | } |
2881 | ||
9078a3e1 CM |
2882 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2883 | struct btrfs_root *root, | |
2884 | struct btrfs_path *path, | |
2885 | struct btrfs_block_group_cache *cache) | |
2886 | { | |
2887 | int ret; | |
9078a3e1 | 2888 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
2889 | unsigned long bi; |
2890 | struct extent_buffer *leaf; | |
9078a3e1 | 2891 | |
9078a3e1 | 2892 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
2893 | if (ret < 0) |
2894 | goto fail; | |
79787eaa | 2895 | BUG_ON(ret); /* Corruption */ |
5f39d397 CM |
2896 | |
2897 | leaf = path->nodes[0]; | |
2898 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2899 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
2900 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 2901 | btrfs_release_path(path); |
54aa1f4d | 2902 | fail: |
79787eaa JM |
2903 | if (ret) { |
2904 | btrfs_abort_transaction(trans, root, ret); | |
9078a3e1 | 2905 | return ret; |
79787eaa | 2906 | } |
9078a3e1 CM |
2907 | return 0; |
2908 | ||
2909 | } | |
2910 | ||
4a8c9a62 YZ |
2911 | static struct btrfs_block_group_cache * |
2912 | next_block_group(struct btrfs_root *root, | |
2913 | struct btrfs_block_group_cache *cache) | |
2914 | { | |
2915 | struct rb_node *node; | |
2916 | spin_lock(&root->fs_info->block_group_cache_lock); | |
2917 | node = rb_next(&cache->cache_node); | |
2918 | btrfs_put_block_group(cache); | |
2919 | if (node) { | |
2920 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
2921 | cache_node); | |
11dfe35a | 2922 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
2923 | } else |
2924 | cache = NULL; | |
2925 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
2926 | return cache; | |
2927 | } | |
2928 | ||
0af3d00b JB |
2929 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
2930 | struct btrfs_trans_handle *trans, | |
2931 | struct btrfs_path *path) | |
2932 | { | |
2933 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
2934 | struct inode *inode = NULL; | |
2935 | u64 alloc_hint = 0; | |
2b20982e | 2936 | int dcs = BTRFS_DC_ERROR; |
0af3d00b JB |
2937 | int num_pages = 0; |
2938 | int retries = 0; | |
2939 | int ret = 0; | |
2940 | ||
2941 | /* | |
2942 | * If this block group is smaller than 100 megs don't bother caching the | |
2943 | * block group. | |
2944 | */ | |
2945 | if (block_group->key.offset < (100 * 1024 * 1024)) { | |
2946 | spin_lock(&block_group->lock); | |
2947 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
2948 | spin_unlock(&block_group->lock); | |
2949 | return 0; | |
2950 | } | |
2951 | ||
2952 | again: | |
2953 | inode = lookup_free_space_inode(root, block_group, path); | |
2954 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
2955 | ret = PTR_ERR(inode); | |
b3b4aa74 | 2956 | btrfs_release_path(path); |
0af3d00b JB |
2957 | goto out; |
2958 | } | |
2959 | ||
2960 | if (IS_ERR(inode)) { | |
2961 | BUG_ON(retries); | |
2962 | retries++; | |
2963 | ||
2964 | if (block_group->ro) | |
2965 | goto out_free; | |
2966 | ||
2967 | ret = create_free_space_inode(root, trans, block_group, path); | |
2968 | if (ret) | |
2969 | goto out_free; | |
2970 | goto again; | |
2971 | } | |
2972 | ||
5b0e95bf JB |
2973 | /* We've already setup this transaction, go ahead and exit */ |
2974 | if (block_group->cache_generation == trans->transid && | |
2975 | i_size_read(inode)) { | |
2976 | dcs = BTRFS_DC_SETUP; | |
2977 | goto out_put; | |
2978 | } | |
2979 | ||
0af3d00b JB |
2980 | /* |
2981 | * We want to set the generation to 0, that way if anything goes wrong | |
2982 | * from here on out we know not to trust this cache when we load up next | |
2983 | * time. | |
2984 | */ | |
2985 | BTRFS_I(inode)->generation = 0; | |
2986 | ret = btrfs_update_inode(trans, root, inode); | |
2987 | WARN_ON(ret); | |
2988 | ||
2989 | if (i_size_read(inode) > 0) { | |
2990 | ret = btrfs_truncate_free_space_cache(root, trans, path, | |
2991 | inode); | |
2992 | if (ret) | |
2993 | goto out_put; | |
2994 | } | |
2995 | ||
2996 | spin_lock(&block_group->lock); | |
cf7c1ef6 LB |
2997 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
2998 | !btrfs_test_opt(root, SPACE_CACHE)) { | |
2999 | /* | |
3000 | * don't bother trying to write stuff out _if_ | |
3001 | * a) we're not cached, | |
3002 | * b) we're with nospace_cache mount option. | |
3003 | */ | |
2b20982e | 3004 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3005 | spin_unlock(&block_group->lock); |
3006 | goto out_put; | |
3007 | } | |
3008 | spin_unlock(&block_group->lock); | |
3009 | ||
3010 | num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024); | |
3011 | if (!num_pages) | |
3012 | num_pages = 1; | |
3013 | ||
3014 | /* | |
3015 | * Just to make absolutely sure we have enough space, we're going to | |
3016 | * preallocate 12 pages worth of space for each block group. In | |
3017 | * practice we ought to use at most 8, but we need extra space so we can | |
3018 | * add our header and have a terminator between the extents and the | |
3019 | * bitmaps. | |
3020 | */ | |
3021 | num_pages *= 16; | |
3022 | num_pages *= PAGE_CACHE_SIZE; | |
3023 | ||
3024 | ret = btrfs_check_data_free_space(inode, num_pages); | |
3025 | if (ret) | |
3026 | goto out_put; | |
3027 | ||
3028 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3029 | num_pages, num_pages, | |
3030 | &alloc_hint); | |
2b20982e JB |
3031 | if (!ret) |
3032 | dcs = BTRFS_DC_SETUP; | |
0af3d00b | 3033 | btrfs_free_reserved_data_space(inode, num_pages); |
c09544e0 | 3034 | |
0af3d00b JB |
3035 | out_put: |
3036 | iput(inode); | |
3037 | out_free: | |
b3b4aa74 | 3038 | btrfs_release_path(path); |
0af3d00b JB |
3039 | out: |
3040 | spin_lock(&block_group->lock); | |
e65cbb94 | 3041 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3042 | block_group->cache_generation = trans->transid; |
2b20982e | 3043 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3044 | spin_unlock(&block_group->lock); |
3045 | ||
3046 | return ret; | |
3047 | } | |
3048 | ||
96b5179d CM |
3049 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
3050 | struct btrfs_root *root) | |
9078a3e1 | 3051 | { |
4a8c9a62 | 3052 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 3053 | int err = 0; |
9078a3e1 | 3054 | struct btrfs_path *path; |
96b5179d | 3055 | u64 last = 0; |
9078a3e1 CM |
3056 | |
3057 | path = btrfs_alloc_path(); | |
3058 | if (!path) | |
3059 | return -ENOMEM; | |
3060 | ||
0af3d00b JB |
3061 | again: |
3062 | while (1) { | |
3063 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
3064 | while (cache) { | |
3065 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3066 | break; | |
3067 | cache = next_block_group(root, cache); | |
3068 | } | |
3069 | if (!cache) { | |
3070 | if (last == 0) | |
3071 | break; | |
3072 | last = 0; | |
3073 | continue; | |
3074 | } | |
3075 | err = cache_save_setup(cache, trans, path); | |
3076 | last = cache->key.objectid + cache->key.offset; | |
3077 | btrfs_put_block_group(cache); | |
3078 | } | |
3079 | ||
d397712b | 3080 | while (1) { |
4a8c9a62 YZ |
3081 | if (last == 0) { |
3082 | err = btrfs_run_delayed_refs(trans, root, | |
3083 | (unsigned long)-1); | |
79787eaa JM |
3084 | if (err) /* File system offline */ |
3085 | goto out; | |
0f9dd46c | 3086 | } |
54aa1f4d | 3087 | |
4a8c9a62 YZ |
3088 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
3089 | while (cache) { | |
0af3d00b JB |
3090 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) { |
3091 | btrfs_put_block_group(cache); | |
3092 | goto again; | |
3093 | } | |
3094 | ||
4a8c9a62 YZ |
3095 | if (cache->dirty) |
3096 | break; | |
3097 | cache = next_block_group(root, cache); | |
3098 | } | |
3099 | if (!cache) { | |
3100 | if (last == 0) | |
3101 | break; | |
3102 | last = 0; | |
3103 | continue; | |
3104 | } | |
0f9dd46c | 3105 | |
0cb59c99 JB |
3106 | if (cache->disk_cache_state == BTRFS_DC_SETUP) |
3107 | cache->disk_cache_state = BTRFS_DC_NEED_WRITE; | |
e8569813 | 3108 | cache->dirty = 0; |
4a8c9a62 | 3109 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 3110 | |
4a8c9a62 | 3111 | err = write_one_cache_group(trans, root, path, cache); |
79787eaa JM |
3112 | if (err) /* File system offline */ |
3113 | goto out; | |
3114 | ||
4a8c9a62 | 3115 | btrfs_put_block_group(cache); |
9078a3e1 | 3116 | } |
4a8c9a62 | 3117 | |
0cb59c99 JB |
3118 | while (1) { |
3119 | /* | |
3120 | * I don't think this is needed since we're just marking our | |
3121 | * preallocated extent as written, but just in case it can't | |
3122 | * hurt. | |
3123 | */ | |
3124 | if (last == 0) { | |
3125 | err = btrfs_run_delayed_refs(trans, root, | |
3126 | (unsigned long)-1); | |
79787eaa JM |
3127 | if (err) /* File system offline */ |
3128 | goto out; | |
0cb59c99 JB |
3129 | } |
3130 | ||
3131 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
3132 | while (cache) { | |
3133 | /* | |
3134 | * Really this shouldn't happen, but it could if we | |
3135 | * couldn't write the entire preallocated extent and | |
3136 | * splitting the extent resulted in a new block. | |
3137 | */ | |
3138 | if (cache->dirty) { | |
3139 | btrfs_put_block_group(cache); | |
3140 | goto again; | |
3141 | } | |
3142 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
3143 | break; | |
3144 | cache = next_block_group(root, cache); | |
3145 | } | |
3146 | if (!cache) { | |
3147 | if (last == 0) | |
3148 | break; | |
3149 | last = 0; | |
3150 | continue; | |
3151 | } | |
3152 | ||
79787eaa | 3153 | err = btrfs_write_out_cache(root, trans, cache, path); |
0cb59c99 JB |
3154 | |
3155 | /* | |
3156 | * If we didn't have an error then the cache state is still | |
3157 | * NEED_WRITE, so we can set it to WRITTEN. | |
3158 | */ | |
79787eaa | 3159 | if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE) |
0cb59c99 JB |
3160 | cache->disk_cache_state = BTRFS_DC_WRITTEN; |
3161 | last = cache->key.objectid + cache->key.offset; | |
3162 | btrfs_put_block_group(cache); | |
3163 | } | |
79787eaa | 3164 | out: |
0cb59c99 | 3165 | |
9078a3e1 | 3166 | btrfs_free_path(path); |
79787eaa | 3167 | return err; |
9078a3e1 CM |
3168 | } |
3169 | ||
d2fb3437 YZ |
3170 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
3171 | { | |
3172 | struct btrfs_block_group_cache *block_group; | |
3173 | int readonly = 0; | |
3174 | ||
3175 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
3176 | if (!block_group || block_group->ro) | |
3177 | readonly = 1; | |
3178 | if (block_group) | |
fa9c0d79 | 3179 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3180 | return readonly; |
3181 | } | |
3182 | ||
593060d7 CM |
3183 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3184 | u64 total_bytes, u64 bytes_used, | |
3185 | struct btrfs_space_info **space_info) | |
3186 | { | |
3187 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3188 | int i; |
3189 | int factor; | |
3190 | ||
3191 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3192 | BTRFS_BLOCK_GROUP_RAID10)) | |
3193 | factor = 2; | |
3194 | else | |
3195 | factor = 1; | |
593060d7 CM |
3196 | |
3197 | found = __find_space_info(info, flags); | |
3198 | if (found) { | |
25179201 | 3199 | spin_lock(&found->lock); |
593060d7 | 3200 | found->total_bytes += total_bytes; |
89a55897 | 3201 | found->disk_total += total_bytes * factor; |
593060d7 | 3202 | found->bytes_used += bytes_used; |
b742bb82 | 3203 | found->disk_used += bytes_used * factor; |
8f18cf13 | 3204 | found->full = 0; |
25179201 | 3205 | spin_unlock(&found->lock); |
593060d7 CM |
3206 | *space_info = found; |
3207 | return 0; | |
3208 | } | |
c146afad | 3209 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3210 | if (!found) |
3211 | return -ENOMEM; | |
3212 | ||
b742bb82 YZ |
3213 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
3214 | INIT_LIST_HEAD(&found->block_groups[i]); | |
80eb234a | 3215 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3216 | spin_lock_init(&found->lock); |
52ba6929 | 3217 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
593060d7 | 3218 | found->total_bytes = total_bytes; |
89a55897 | 3219 | found->disk_total = total_bytes * factor; |
593060d7 | 3220 | found->bytes_used = bytes_used; |
b742bb82 | 3221 | found->disk_used = bytes_used * factor; |
593060d7 | 3222 | found->bytes_pinned = 0; |
e8569813 | 3223 | found->bytes_reserved = 0; |
c146afad | 3224 | found->bytes_readonly = 0; |
f0486c68 | 3225 | found->bytes_may_use = 0; |
593060d7 | 3226 | found->full = 0; |
0e4f8f88 | 3227 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3228 | found->chunk_alloc = 0; |
fdb5effd JB |
3229 | found->flush = 0; |
3230 | init_waitqueue_head(&found->wait); | |
593060d7 | 3231 | *space_info = found; |
4184ea7f | 3232 | list_add_rcu(&found->list, &info->space_info); |
b4d7c3c9 LZ |
3233 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3234 | info->data_sinfo = found; | |
593060d7 CM |
3235 | return 0; |
3236 | } | |
3237 | ||
8790d502 CM |
3238 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3239 | { | |
899c81ea ID |
3240 | u64 extra_flags = chunk_to_extended(flags) & |
3241 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 ID |
3242 | |
3243 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3244 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3245 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3246 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3247 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3248 | fs_info->avail_system_alloc_bits |= extra_flags; | |
8790d502 | 3249 | } |
593060d7 | 3250 | |
fc67c450 ID |
3251 | /* |
3252 | * returns target flags in extended format or 0 if restripe for this | |
3253 | * chunk_type is not in progress | |
c6664b42 ID |
3254 | * |
3255 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
3256 | */ |
3257 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
3258 | { | |
3259 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3260 | u64 target = 0; | |
3261 | ||
fc67c450 ID |
3262 | if (!bctl) |
3263 | return 0; | |
3264 | ||
3265 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
3266 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3267 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3268 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
3269 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3270 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3271 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
3272 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3273 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3274 | } | |
3275 | ||
3276 | return target; | |
3277 | } | |
3278 | ||
a46d11a8 ID |
3279 | /* |
3280 | * @flags: available profiles in extended format (see ctree.h) | |
3281 | * | |
e4d8ec0f ID |
3282 | * Returns reduced profile in chunk format. If profile changing is in |
3283 | * progress (either running or paused) picks the target profile (if it's | |
3284 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 3285 | */ |
2b82032c | 3286 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 3287 | { |
cd02dca5 CM |
3288 | /* |
3289 | * we add in the count of missing devices because we want | |
3290 | * to make sure that any RAID levels on a degraded FS | |
3291 | * continue to be honored. | |
3292 | */ | |
3293 | u64 num_devices = root->fs_info->fs_devices->rw_devices + | |
3294 | root->fs_info->fs_devices->missing_devices; | |
fc67c450 | 3295 | u64 target; |
a061fc8d | 3296 | |
fc67c450 ID |
3297 | /* |
3298 | * see if restripe for this chunk_type is in progress, if so | |
3299 | * try to reduce to the target profile | |
3300 | */ | |
e4d8ec0f | 3301 | spin_lock(&root->fs_info->balance_lock); |
fc67c450 ID |
3302 | target = get_restripe_target(root->fs_info, flags); |
3303 | if (target) { | |
3304 | /* pick target profile only if it's already available */ | |
3305 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
e4d8ec0f | 3306 | spin_unlock(&root->fs_info->balance_lock); |
fc67c450 | 3307 | return extended_to_chunk(target); |
e4d8ec0f ID |
3308 | } |
3309 | } | |
3310 | spin_unlock(&root->fs_info->balance_lock); | |
3311 | ||
a061fc8d CM |
3312 | if (num_devices == 1) |
3313 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
3314 | if (num_devices < 4) | |
3315 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
3316 | ||
ec44a35c CM |
3317 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
3318 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 3319 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 3320 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 3321 | } |
ec44a35c CM |
3322 | |
3323 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 3324 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 3325 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 3326 | } |
ec44a35c CM |
3327 | |
3328 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
3329 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
3330 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
a46d11a8 | 3331 | (flags & BTRFS_BLOCK_GROUP_DUP))) { |
ec44a35c | 3332 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; |
a46d11a8 ID |
3333 | } |
3334 | ||
899c81ea | 3335 | return extended_to_chunk(flags); |
ec44a35c CM |
3336 | } |
3337 | ||
b742bb82 | 3338 | static u64 get_alloc_profile(struct btrfs_root *root, u64 flags) |
6a63209f | 3339 | { |
b742bb82 | 3340 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
6fef8df1 | 3341 | flags |= root->fs_info->avail_data_alloc_bits; |
b742bb82 | 3342 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
6fef8df1 | 3343 | flags |= root->fs_info->avail_system_alloc_bits; |
b742bb82 | 3344 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
6fef8df1 ID |
3345 | flags |= root->fs_info->avail_metadata_alloc_bits; |
3346 | ||
b742bb82 | 3347 | return btrfs_reduce_alloc_profile(root, flags); |
6a63209f JB |
3348 | } |
3349 | ||
6d07bcec | 3350 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 3351 | { |
b742bb82 | 3352 | u64 flags; |
9ed74f2d | 3353 | |
b742bb82 YZ |
3354 | if (data) |
3355 | flags = BTRFS_BLOCK_GROUP_DATA; | |
3356 | else if (root == root->fs_info->chunk_root) | |
3357 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 3358 | else |
b742bb82 | 3359 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3360 | |
b742bb82 | 3361 | return get_alloc_profile(root, flags); |
6a63209f | 3362 | } |
9ed74f2d | 3363 | |
6a63209f | 3364 | /* |
6a63209f JB |
3365 | * This will check the space that the inode allocates from to make sure we have |
3366 | * enough space for bytes. | |
6a63209f | 3367 | */ |
0ca1f7ce | 3368 | int btrfs_check_data_free_space(struct inode *inode, u64 bytes) |
6a63209f | 3369 | { |
6a63209f | 3370 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 3371 | struct btrfs_root *root = BTRFS_I(inode)->root; |
b4d7c3c9 | 3372 | struct btrfs_fs_info *fs_info = root->fs_info; |
ab6e2410 | 3373 | u64 used; |
0af3d00b | 3374 | int ret = 0, committed = 0, alloc_chunk = 1; |
6a63209f | 3375 | |
6a63209f JB |
3376 | /* make sure bytes are sectorsize aligned */ |
3377 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
6a63209f | 3378 | |
82d5902d LZ |
3379 | if (root == root->fs_info->tree_root || |
3380 | BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) { | |
0af3d00b JB |
3381 | alloc_chunk = 0; |
3382 | committed = 1; | |
3383 | } | |
3384 | ||
b4d7c3c9 | 3385 | data_sinfo = fs_info->data_sinfo; |
33b4d47f CM |
3386 | if (!data_sinfo) |
3387 | goto alloc; | |
9ed74f2d | 3388 | |
6a63209f JB |
3389 | again: |
3390 | /* make sure we have enough space to handle the data first */ | |
3391 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
3392 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
3393 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
3394 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
3395 | |
3396 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 3397 | struct btrfs_trans_handle *trans; |
9ed74f2d | 3398 | |
6a63209f JB |
3399 | /* |
3400 | * if we don't have enough free bytes in this space then we need | |
3401 | * to alloc a new chunk. | |
3402 | */ | |
0af3d00b | 3403 | if (!data_sinfo->full && alloc_chunk) { |
6a63209f | 3404 | u64 alloc_target; |
9ed74f2d | 3405 | |
0e4f8f88 | 3406 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 3407 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 3408 | alloc: |
6a63209f | 3409 | alloc_target = btrfs_get_alloc_profile(root, 1); |
7a7eaa40 | 3410 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3411 | if (IS_ERR(trans)) |
3412 | return PTR_ERR(trans); | |
9ed74f2d | 3413 | |
6a63209f JB |
3414 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
3415 | bytes + 2 * 1024 * 1024, | |
0e4f8f88 CM |
3416 | alloc_target, |
3417 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 3418 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
3419 | if (ret < 0) { |
3420 | if (ret != -ENOSPC) | |
3421 | return ret; | |
3422 | else | |
3423 | goto commit_trans; | |
3424 | } | |
9ed74f2d | 3425 | |
b4d7c3c9 LZ |
3426 | if (!data_sinfo) |
3427 | data_sinfo = fs_info->data_sinfo; | |
3428 | ||
6a63209f JB |
3429 | goto again; |
3430 | } | |
f2bb8f5c JB |
3431 | |
3432 | /* | |
3433 | * If we have less pinned bytes than we want to allocate then | |
3434 | * don't bother committing the transaction, it won't help us. | |
3435 | */ | |
3436 | if (data_sinfo->bytes_pinned < bytes) | |
3437 | committed = 1; | |
6a63209f | 3438 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3439 | |
4e06bdd6 | 3440 | /* commit the current transaction and try again */ |
d52a5b5f | 3441 | commit_trans: |
a4abeea4 JB |
3442 | if (!committed && |
3443 | !atomic_read(&root->fs_info->open_ioctl_trans)) { | |
4e06bdd6 | 3444 | committed = 1; |
7a7eaa40 | 3445 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3446 | if (IS_ERR(trans)) |
3447 | return PTR_ERR(trans); | |
4e06bdd6 JB |
3448 | ret = btrfs_commit_transaction(trans, root); |
3449 | if (ret) | |
3450 | return ret; | |
3451 | goto again; | |
3452 | } | |
9ed74f2d | 3453 | |
6a63209f JB |
3454 | return -ENOSPC; |
3455 | } | |
3456 | data_sinfo->bytes_may_use += bytes; | |
8c2a3ca2 | 3457 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3458 | data_sinfo->flags, bytes, 1); |
6a63209f | 3459 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3460 | |
9ed74f2d | 3461 | return 0; |
9ed74f2d | 3462 | } |
6a63209f | 3463 | |
6a63209f | 3464 | /* |
fb25e914 | 3465 | * Called if we need to clear a data reservation for this inode. |
6a63209f | 3466 | */ |
0ca1f7ce | 3467 | void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes) |
e3ccfa98 | 3468 | { |
0ca1f7ce | 3469 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6a63209f | 3470 | struct btrfs_space_info *data_sinfo; |
e3ccfa98 | 3471 | |
6a63209f JB |
3472 | /* make sure bytes are sectorsize aligned */ |
3473 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
e3ccfa98 | 3474 | |
b4d7c3c9 | 3475 | data_sinfo = root->fs_info->data_sinfo; |
6a63209f JB |
3476 | spin_lock(&data_sinfo->lock); |
3477 | data_sinfo->bytes_may_use -= bytes; | |
8c2a3ca2 | 3478 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3479 | data_sinfo->flags, bytes, 0); |
6a63209f | 3480 | spin_unlock(&data_sinfo->lock); |
e3ccfa98 JB |
3481 | } |
3482 | ||
97e728d4 | 3483 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 3484 | { |
97e728d4 JB |
3485 | struct list_head *head = &info->space_info; |
3486 | struct btrfs_space_info *found; | |
e3ccfa98 | 3487 | |
97e728d4 JB |
3488 | rcu_read_lock(); |
3489 | list_for_each_entry_rcu(found, head, list) { | |
3490 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 3491 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 3492 | } |
97e728d4 | 3493 | rcu_read_unlock(); |
e3ccfa98 JB |
3494 | } |
3495 | ||
e5bc2458 | 3496 | static int should_alloc_chunk(struct btrfs_root *root, |
0e4f8f88 CM |
3497 | struct btrfs_space_info *sinfo, u64 alloc_bytes, |
3498 | int force) | |
32c00aff | 3499 | { |
fb25e914 | 3500 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
424499db | 3501 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 3502 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 3503 | u64 thresh; |
e3ccfa98 | 3504 | |
0e4f8f88 CM |
3505 | if (force == CHUNK_ALLOC_FORCE) |
3506 | return 1; | |
3507 | ||
fb25e914 JB |
3508 | /* |
3509 | * We need to take into account the global rsv because for all intents | |
3510 | * and purposes it's used space. Don't worry about locking the | |
3511 | * global_rsv, it doesn't change except when the transaction commits. | |
3512 | */ | |
3513 | num_allocated += global_rsv->size; | |
3514 | ||
0e4f8f88 CM |
3515 | /* |
3516 | * in limited mode, we want to have some free space up to | |
3517 | * about 1% of the FS size. | |
3518 | */ | |
3519 | if (force == CHUNK_ALLOC_LIMITED) { | |
6c41761f | 3520 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 CM |
3521 | thresh = max_t(u64, 64 * 1024 * 1024, |
3522 | div_factor_fine(thresh, 1)); | |
3523 | ||
3524 | if (num_bytes - num_allocated < thresh) | |
3525 | return 1; | |
3526 | } | |
6c41761f | 3527 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 | 3528 | |
cf1d72c9 CM |
3529 | /* 256MB or 2% of the FS */ |
3530 | thresh = max_t(u64, 256 * 1024 * 1024, div_factor_fine(thresh, 2)); | |
96bdc7dc CM |
3531 | /* system chunks need a much small threshold */ |
3532 | if (sinfo->flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3533 | thresh = 32 * 1024 * 1024; | |
e5bc2458 | 3534 | |
cf1d72c9 | 3535 | if (num_bytes > thresh && sinfo->bytes_used < div_factor(num_bytes, 8)) |
14ed0ca6 | 3536 | return 0; |
424499db | 3537 | return 1; |
32c00aff JB |
3538 | } |
3539 | ||
15d1ff81 LB |
3540 | static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type) |
3541 | { | |
3542 | u64 num_dev; | |
3543 | ||
3544 | if (type & BTRFS_BLOCK_GROUP_RAID10 || | |
3545 | type & BTRFS_BLOCK_GROUP_RAID0) | |
3546 | num_dev = root->fs_info->fs_devices->rw_devices; | |
3547 | else if (type & BTRFS_BLOCK_GROUP_RAID1) | |
3548 | num_dev = 2; | |
3549 | else | |
3550 | num_dev = 1; /* DUP or single */ | |
3551 | ||
3552 | /* metadata for updaing devices and chunk tree */ | |
3553 | return btrfs_calc_trans_metadata_size(root, num_dev + 1); | |
3554 | } | |
3555 | ||
3556 | static void check_system_chunk(struct btrfs_trans_handle *trans, | |
3557 | struct btrfs_root *root, u64 type) | |
3558 | { | |
3559 | struct btrfs_space_info *info; | |
3560 | u64 left; | |
3561 | u64 thresh; | |
3562 | ||
3563 | info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
3564 | spin_lock(&info->lock); | |
3565 | left = info->total_bytes - info->bytes_used - info->bytes_pinned - | |
3566 | info->bytes_reserved - info->bytes_readonly; | |
3567 | spin_unlock(&info->lock); | |
3568 | ||
3569 | thresh = get_system_chunk_thresh(root, type); | |
3570 | if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
3571 | printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n", | |
3572 | left, thresh, type); | |
3573 | dump_space_info(info, 0, 0); | |
3574 | } | |
3575 | ||
3576 | if (left < thresh) { | |
3577 | u64 flags; | |
3578 | ||
3579 | flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0); | |
3580 | btrfs_alloc_chunk(trans, root, flags); | |
3581 | } | |
3582 | } | |
3583 | ||
6324fbf3 CM |
3584 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
3585 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 3586 | u64 flags, int force) |
9ed74f2d | 3587 | { |
6324fbf3 | 3588 | struct btrfs_space_info *space_info; |
97e728d4 | 3589 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 3590 | int wait_for_alloc = 0; |
9ed74f2d | 3591 | int ret = 0; |
9ed74f2d | 3592 | |
6324fbf3 | 3593 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
3594 | if (!space_info) { |
3595 | ret = update_space_info(extent_root->fs_info, flags, | |
3596 | 0, 0, &space_info); | |
79787eaa | 3597 | BUG_ON(ret); /* -ENOMEM */ |
9ed74f2d | 3598 | } |
79787eaa | 3599 | BUG_ON(!space_info); /* Logic error */ |
9ed74f2d | 3600 | |
6d74119f | 3601 | again: |
25179201 | 3602 | spin_lock(&space_info->lock); |
9e622d6b | 3603 | if (force < space_info->force_alloc) |
0e4f8f88 | 3604 | force = space_info->force_alloc; |
25179201 JB |
3605 | if (space_info->full) { |
3606 | spin_unlock(&space_info->lock); | |
6d74119f | 3607 | return 0; |
9ed74f2d JB |
3608 | } |
3609 | ||
0e4f8f88 | 3610 | if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) { |
25179201 | 3611 | spin_unlock(&space_info->lock); |
6d74119f JB |
3612 | return 0; |
3613 | } else if (space_info->chunk_alloc) { | |
3614 | wait_for_alloc = 1; | |
3615 | } else { | |
3616 | space_info->chunk_alloc = 1; | |
9ed74f2d | 3617 | } |
0e4f8f88 | 3618 | |
25179201 | 3619 | spin_unlock(&space_info->lock); |
9ed74f2d | 3620 | |
6d74119f JB |
3621 | mutex_lock(&fs_info->chunk_mutex); |
3622 | ||
3623 | /* | |
3624 | * The chunk_mutex is held throughout the entirety of a chunk | |
3625 | * allocation, so once we've acquired the chunk_mutex we know that the | |
3626 | * other guy is done and we need to recheck and see if we should | |
3627 | * allocate. | |
3628 | */ | |
3629 | if (wait_for_alloc) { | |
3630 | mutex_unlock(&fs_info->chunk_mutex); | |
3631 | wait_for_alloc = 0; | |
3632 | goto again; | |
3633 | } | |
3634 | ||
67377734 JB |
3635 | /* |
3636 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3637 | * allocating mixed chunks instead of individual chunks. | |
3638 | */ | |
3639 | if (btrfs_mixed_space_info(space_info)) | |
3640 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3641 | ||
97e728d4 JB |
3642 | /* |
3643 | * if we're doing a data chunk, go ahead and make sure that | |
3644 | * we keep a reasonable number of metadata chunks allocated in the | |
3645 | * FS as well. | |
3646 | */ | |
9ed74f2d | 3647 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
3648 | fs_info->data_chunk_allocations++; |
3649 | if (!(fs_info->data_chunk_allocations % | |
3650 | fs_info->metadata_ratio)) | |
3651 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
3652 | } |
3653 | ||
15d1ff81 LB |
3654 | /* |
3655 | * Check if we have enough space in SYSTEM chunk because we may need | |
3656 | * to update devices. | |
3657 | */ | |
3658 | check_system_chunk(trans, extent_root, flags); | |
3659 | ||
2b82032c | 3660 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
92b8e897 MF |
3661 | if (ret < 0 && ret != -ENOSPC) |
3662 | goto out; | |
3663 | ||
9ed74f2d | 3664 | spin_lock(&space_info->lock); |
9ed74f2d | 3665 | if (ret) |
6324fbf3 | 3666 | space_info->full = 1; |
424499db YZ |
3667 | else |
3668 | ret = 1; | |
6d74119f | 3669 | |
0e4f8f88 | 3670 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3671 | space_info->chunk_alloc = 0; |
9ed74f2d | 3672 | spin_unlock(&space_info->lock); |
92b8e897 | 3673 | out: |
a25c75d5 | 3674 | mutex_unlock(&fs_info->chunk_mutex); |
0f9dd46c | 3675 | return ret; |
6324fbf3 | 3676 | } |
9ed74f2d | 3677 | |
9ed74f2d | 3678 | /* |
5da9d01b | 3679 | * shrink metadata reservation for delalloc |
9ed74f2d | 3680 | */ |
f4c738c2 JB |
3681 | static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig, |
3682 | bool wait_ordered) | |
5da9d01b | 3683 | { |
0ca1f7ce | 3684 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 3685 | struct btrfs_space_info *space_info; |
663350ac | 3686 | struct btrfs_trans_handle *trans; |
f4c738c2 | 3687 | u64 delalloc_bytes; |
5da9d01b | 3688 | u64 max_reclaim; |
b1953bce | 3689 | long time_left; |
877da174 | 3690 | unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT; |
b1953bce | 3691 | int loops = 0; |
5da9d01b | 3692 | |
663350ac | 3693 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0ca1f7ce | 3694 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 3695 | space_info = block_rsv->space_info; |
bf9022e0 CM |
3696 | |
3697 | smp_mb(); | |
f4c738c2 JB |
3698 | delalloc_bytes = root->fs_info->delalloc_bytes; |
3699 | if (delalloc_bytes == 0) { | |
fdb5effd | 3700 | if (trans) |
f4c738c2 | 3701 | return; |
fdb5effd | 3702 | btrfs_wait_ordered_extents(root, 0, 0); |
f4c738c2 | 3703 | return; |
fdb5effd JB |
3704 | } |
3705 | ||
f4c738c2 JB |
3706 | while (delalloc_bytes && loops < 3) { |
3707 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
3708 | nr_pages = max_reclaim >> PAGE_CACHE_SHIFT; | |
0e175a18 | 3709 | writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages, |
f4c738c2 | 3710 | WB_REASON_FS_FREE_SPACE); |
5da9d01b | 3711 | |
0019f10d | 3712 | spin_lock(&space_info->lock); |
f4c738c2 JB |
3713 | if (space_info->bytes_used + space_info->bytes_reserved + |
3714 | space_info->bytes_pinned + space_info->bytes_readonly + | |
3715 | space_info->bytes_may_use + orig <= | |
3716 | space_info->total_bytes) { | |
3717 | spin_unlock(&space_info->lock); | |
3718 | break; | |
3719 | } | |
0019f10d | 3720 | spin_unlock(&space_info->lock); |
5da9d01b | 3721 | |
36e39c40 | 3722 | loops++; |
f104d044 JB |
3723 | if (wait_ordered && !trans) { |
3724 | btrfs_wait_ordered_extents(root, 0, 0); | |
3725 | } else { | |
f4c738c2 | 3726 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
3727 | if (time_left) |
3728 | break; | |
3729 | } | |
f4c738c2 JB |
3730 | smp_mb(); |
3731 | delalloc_bytes = root->fs_info->delalloc_bytes; | |
5da9d01b | 3732 | } |
5da9d01b YZ |
3733 | } |
3734 | ||
663350ac JB |
3735 | /** |
3736 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
3737 | * @root - the root we're allocating for | |
3738 | * @bytes - the number of bytes we want to reserve | |
3739 | * @force - force the commit | |
8bb8ab2e | 3740 | * |
663350ac JB |
3741 | * This will check to make sure that committing the transaction will actually |
3742 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
3743 | * will return -ENOSPC. | |
8bb8ab2e | 3744 | */ |
663350ac JB |
3745 | static int may_commit_transaction(struct btrfs_root *root, |
3746 | struct btrfs_space_info *space_info, | |
3747 | u64 bytes, int force) | |
3748 | { | |
3749 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; | |
3750 | struct btrfs_trans_handle *trans; | |
3751 | ||
3752 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
3753 | if (trans) | |
3754 | return -EAGAIN; | |
3755 | ||
3756 | if (force) | |
3757 | goto commit; | |
3758 | ||
3759 | /* See if there is enough pinned space to make this reservation */ | |
3760 | spin_lock(&space_info->lock); | |
3761 | if (space_info->bytes_pinned >= bytes) { | |
3762 | spin_unlock(&space_info->lock); | |
3763 | goto commit; | |
3764 | } | |
3765 | spin_unlock(&space_info->lock); | |
3766 | ||
3767 | /* | |
3768 | * See if there is some space in the delayed insertion reservation for | |
3769 | * this reservation. | |
3770 | */ | |
3771 | if (space_info != delayed_rsv->space_info) | |
3772 | return -ENOSPC; | |
3773 | ||
d9b0218f | 3774 | spin_lock(&space_info->lock); |
663350ac | 3775 | spin_lock(&delayed_rsv->lock); |
d9b0218f | 3776 | if (space_info->bytes_pinned + delayed_rsv->size < bytes) { |
663350ac | 3777 | spin_unlock(&delayed_rsv->lock); |
d9b0218f | 3778 | spin_unlock(&space_info->lock); |
663350ac JB |
3779 | return -ENOSPC; |
3780 | } | |
3781 | spin_unlock(&delayed_rsv->lock); | |
d9b0218f | 3782 | spin_unlock(&space_info->lock); |
663350ac JB |
3783 | |
3784 | commit: | |
3785 | trans = btrfs_join_transaction(root); | |
3786 | if (IS_ERR(trans)) | |
3787 | return -ENOSPC; | |
3788 | ||
3789 | return btrfs_commit_transaction(trans, root); | |
3790 | } | |
3791 | ||
96c3f433 JB |
3792 | enum flush_state { |
3793 | FLUSH_DELALLOC = 1, | |
3794 | FLUSH_DELALLOC_WAIT = 2, | |
3795 | FLUSH_DELAYED_ITEMS_NR = 3, | |
3796 | FLUSH_DELAYED_ITEMS = 4, | |
3797 | COMMIT_TRANS = 5, | |
3798 | }; | |
3799 | ||
3800 | static int flush_space(struct btrfs_root *root, | |
3801 | struct btrfs_space_info *space_info, u64 num_bytes, | |
3802 | u64 orig_bytes, int state) | |
3803 | { | |
3804 | struct btrfs_trans_handle *trans; | |
3805 | int nr; | |
f4c738c2 | 3806 | int ret = 0; |
96c3f433 JB |
3807 | |
3808 | switch (state) { | |
3809 | case FLUSH_DELALLOC: | |
3810 | case FLUSH_DELALLOC_WAIT: | |
f4c738c2 JB |
3811 | shrink_delalloc(root, num_bytes, orig_bytes, |
3812 | state == FLUSH_DELALLOC_WAIT); | |
96c3f433 JB |
3813 | break; |
3814 | case FLUSH_DELAYED_ITEMS_NR: | |
3815 | case FLUSH_DELAYED_ITEMS: | |
3816 | if (state == FLUSH_DELAYED_ITEMS_NR) { | |
3817 | u64 bytes = btrfs_calc_trans_metadata_size(root, 1); | |
3818 | ||
3819 | nr = (int)div64_u64(num_bytes, bytes); | |
3820 | if (!nr) | |
3821 | nr = 1; | |
3822 | nr *= 2; | |
3823 | } else { | |
3824 | nr = -1; | |
3825 | } | |
3826 | trans = btrfs_join_transaction(root); | |
3827 | if (IS_ERR(trans)) { | |
3828 | ret = PTR_ERR(trans); | |
3829 | break; | |
3830 | } | |
3831 | ret = btrfs_run_delayed_items_nr(trans, root, nr); | |
3832 | btrfs_end_transaction(trans, root); | |
3833 | break; | |
3834 | case COMMIT_TRANS: | |
3835 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); | |
3836 | break; | |
3837 | default: | |
3838 | ret = -ENOSPC; | |
3839 | break; | |
3840 | } | |
3841 | ||
3842 | return ret; | |
3843 | } | |
4a92b1b8 JB |
3844 | /** |
3845 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
3846 | * @root - the root we're allocating for | |
3847 | * @block_rsv - the block_rsv we're allocating for | |
3848 | * @orig_bytes - the number of bytes we want | |
3849 | * @flush - wether or not we can flush to make our reservation | |
8bb8ab2e | 3850 | * |
4a92b1b8 JB |
3851 | * This will reserve orgi_bytes number of bytes from the space info associated |
3852 | * with the block_rsv. If there is not enough space it will make an attempt to | |
3853 | * flush out space to make room. It will do this by flushing delalloc if | |
3854 | * possible or committing the transaction. If flush is 0 then no attempts to | |
3855 | * regain reservations will be made and this will fail if there is not enough | |
3856 | * space already. | |
8bb8ab2e | 3857 | */ |
4a92b1b8 | 3858 | static int reserve_metadata_bytes(struct btrfs_root *root, |
8bb8ab2e JB |
3859 | struct btrfs_block_rsv *block_rsv, |
3860 | u64 orig_bytes, int flush) | |
9ed74f2d | 3861 | { |
f0486c68 | 3862 | struct btrfs_space_info *space_info = block_rsv->space_info; |
2bf64758 | 3863 | u64 used; |
8bb8ab2e | 3864 | u64 num_bytes = orig_bytes; |
96c3f433 | 3865 | int flush_state = FLUSH_DELALLOC; |
8bb8ab2e | 3866 | int ret = 0; |
fdb5effd | 3867 | bool flushing = false; |
96c3f433 | 3868 | bool committed = false; |
9ed74f2d | 3869 | |
8bb8ab2e | 3870 | again: |
fdb5effd | 3871 | ret = 0; |
8bb8ab2e | 3872 | spin_lock(&space_info->lock); |
fdb5effd JB |
3873 | /* |
3874 | * We only want to wait if somebody other than us is flushing and we are | |
3875 | * actually alloed to flush. | |
3876 | */ | |
3877 | while (flush && !flushing && space_info->flush) { | |
3878 | spin_unlock(&space_info->lock); | |
3879 | /* | |
3880 | * If we have a trans handle we can't wait because the flusher | |
3881 | * may have to commit the transaction, which would mean we would | |
3882 | * deadlock since we are waiting for the flusher to finish, but | |
3883 | * hold the current transaction open. | |
3884 | */ | |
663350ac | 3885 | if (current->journal_info) |
fdb5effd | 3886 | return -EAGAIN; |
b9688bb8 AJ |
3887 | ret = wait_event_killable(space_info->wait, !space_info->flush); |
3888 | /* Must have been killed, return */ | |
3889 | if (ret) | |
fdb5effd JB |
3890 | return -EINTR; |
3891 | ||
3892 | spin_lock(&space_info->lock); | |
3893 | } | |
3894 | ||
3895 | ret = -ENOSPC; | |
2bf64758 JB |
3896 | used = space_info->bytes_used + space_info->bytes_reserved + |
3897 | space_info->bytes_pinned + space_info->bytes_readonly + | |
3898 | space_info->bytes_may_use; | |
9ed74f2d | 3899 | |
8bb8ab2e JB |
3900 | /* |
3901 | * The idea here is that we've not already over-reserved the block group | |
3902 | * then we can go ahead and save our reservation first and then start | |
3903 | * flushing if we need to. Otherwise if we've already overcommitted | |
3904 | * lets start flushing stuff first and then come back and try to make | |
3905 | * our reservation. | |
3906 | */ | |
2bf64758 JB |
3907 | if (used <= space_info->total_bytes) { |
3908 | if (used + orig_bytes <= space_info->total_bytes) { | |
fb25e914 | 3909 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 3910 | trace_btrfs_space_reservation(root->fs_info, |
2bcc0328 | 3911 | "space_info", space_info->flags, orig_bytes, 1); |
8bb8ab2e JB |
3912 | ret = 0; |
3913 | } else { | |
3914 | /* | |
3915 | * Ok set num_bytes to orig_bytes since we aren't | |
3916 | * overocmmitted, this way we only try and reclaim what | |
3917 | * we need. | |
3918 | */ | |
3919 | num_bytes = orig_bytes; | |
3920 | } | |
3921 | } else { | |
3922 | /* | |
3923 | * Ok we're over committed, set num_bytes to the overcommitted | |
3924 | * amount plus the amount of bytes that we need for this | |
3925 | * reservation. | |
3926 | */ | |
2bf64758 | 3927 | num_bytes = used - space_info->total_bytes + |
96c3f433 | 3928 | (orig_bytes * 2); |
8bb8ab2e | 3929 | } |
9ed74f2d | 3930 | |
36ba022a | 3931 | if (ret) { |
2bf64758 JB |
3932 | u64 profile = btrfs_get_alloc_profile(root, 0); |
3933 | u64 avail; | |
3934 | ||
7e355b83 JB |
3935 | /* |
3936 | * If we have a lot of space that's pinned, don't bother doing | |
3937 | * the overcommit dance yet and just commit the transaction. | |
3938 | */ | |
3939 | avail = (space_info->total_bytes - space_info->bytes_used) * 8; | |
3940 | do_div(avail, 10); | |
663350ac | 3941 | if (space_info->bytes_pinned >= avail && flush && !committed) { |
7e355b83 JB |
3942 | space_info->flush = 1; |
3943 | flushing = true; | |
3944 | spin_unlock(&space_info->lock); | |
663350ac JB |
3945 | ret = may_commit_transaction(root, space_info, |
3946 | orig_bytes, 1); | |
3947 | if (ret) | |
3948 | goto out; | |
3949 | committed = true; | |
3950 | goto again; | |
7e355b83 JB |
3951 | } |
3952 | ||
2bf64758 JB |
3953 | spin_lock(&root->fs_info->free_chunk_lock); |
3954 | avail = root->fs_info->free_chunk_space; | |
3955 | ||
3956 | /* | |
3957 | * If we have dup, raid1 or raid10 then only half of the free | |
3958 | * space is actually useable. | |
3959 | */ | |
3960 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
3961 | BTRFS_BLOCK_GROUP_RAID1 | | |
3962 | BTRFS_BLOCK_GROUP_RAID10)) | |
3963 | avail >>= 1; | |
3964 | ||
3965 | /* | |
3966 | * If we aren't flushing don't let us overcommit too much, say | |
3967 | * 1/8th of the space. If we can flush, let it overcommit up to | |
3968 | * 1/2 of the space. | |
3969 | */ | |
3970 | if (flush) | |
3971 | avail >>= 3; | |
3972 | else | |
3973 | avail >>= 1; | |
3974 | spin_unlock(&root->fs_info->free_chunk_lock); | |
3975 | ||
9a82ca65 | 3976 | if (used + num_bytes < space_info->total_bytes + avail) { |
2bf64758 | 3977 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 3978 | trace_btrfs_space_reservation(root->fs_info, |
2bcc0328 | 3979 | "space_info", space_info->flags, orig_bytes, 1); |
2bf64758 JB |
3980 | ret = 0; |
3981 | } | |
3982 | } | |
3983 | ||
8bb8ab2e JB |
3984 | /* |
3985 | * Couldn't make our reservation, save our place so while we're trying | |
3986 | * to reclaim space we can actually use it instead of somebody else | |
3987 | * stealing it from us. | |
3988 | */ | |
fdb5effd JB |
3989 | if (ret && flush) { |
3990 | flushing = true; | |
3991 | space_info->flush = 1; | |
8bb8ab2e | 3992 | } |
9ed74f2d | 3993 | |
f0486c68 | 3994 | spin_unlock(&space_info->lock); |
9ed74f2d | 3995 | |
fdb5effd | 3996 | if (!ret || !flush) |
8bb8ab2e | 3997 | goto out; |
f0486c68 | 3998 | |
96c3f433 JB |
3999 | ret = flush_space(root, space_info, num_bytes, orig_bytes, |
4000 | flush_state); | |
4001 | flush_state++; | |
4002 | if (!ret) | |
8bb8ab2e | 4003 | goto again; |
96c3f433 | 4004 | else if (flush_state <= COMMIT_TRANS) |
8bb8ab2e JB |
4005 | goto again; |
4006 | ||
4007 | out: | |
fdb5effd | 4008 | if (flushing) { |
8bb8ab2e | 4009 | spin_lock(&space_info->lock); |
fdb5effd JB |
4010 | space_info->flush = 0; |
4011 | wake_up_all(&space_info->wait); | |
8bb8ab2e | 4012 | spin_unlock(&space_info->lock); |
f0486c68 | 4013 | } |
f0486c68 YZ |
4014 | return ret; |
4015 | } | |
4016 | ||
79787eaa JM |
4017 | static struct btrfs_block_rsv *get_block_rsv( |
4018 | const struct btrfs_trans_handle *trans, | |
4019 | const struct btrfs_root *root) | |
f0486c68 | 4020 | { |
4c13d758 JB |
4021 | struct btrfs_block_rsv *block_rsv = NULL; |
4022 | ||
0e721106 JB |
4023 | if (root->ref_cows) |
4024 | block_rsv = trans->block_rsv; | |
4025 | ||
4026 | if (root == root->fs_info->csum_root && trans->adding_csums) | |
f0486c68 | 4027 | block_rsv = trans->block_rsv; |
4c13d758 JB |
4028 | |
4029 | if (!block_rsv) | |
f0486c68 YZ |
4030 | block_rsv = root->block_rsv; |
4031 | ||
4032 | if (!block_rsv) | |
4033 | block_rsv = &root->fs_info->empty_block_rsv; | |
4034 | ||
4035 | return block_rsv; | |
4036 | } | |
4037 | ||
4038 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
4039 | u64 num_bytes) | |
4040 | { | |
4041 | int ret = -ENOSPC; | |
4042 | spin_lock(&block_rsv->lock); | |
4043 | if (block_rsv->reserved >= num_bytes) { | |
4044 | block_rsv->reserved -= num_bytes; | |
4045 | if (block_rsv->reserved < block_rsv->size) | |
4046 | block_rsv->full = 0; | |
4047 | ret = 0; | |
4048 | } | |
4049 | spin_unlock(&block_rsv->lock); | |
4050 | return ret; | |
4051 | } | |
4052 | ||
4053 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
4054 | u64 num_bytes, int update_size) | |
4055 | { | |
4056 | spin_lock(&block_rsv->lock); | |
4057 | block_rsv->reserved += num_bytes; | |
4058 | if (update_size) | |
4059 | block_rsv->size += num_bytes; | |
4060 | else if (block_rsv->reserved >= block_rsv->size) | |
4061 | block_rsv->full = 1; | |
4062 | spin_unlock(&block_rsv->lock); | |
4063 | } | |
4064 | ||
8c2a3ca2 JB |
4065 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
4066 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 4067 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
4068 | { |
4069 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
4070 | ||
4071 | spin_lock(&block_rsv->lock); | |
4072 | if (num_bytes == (u64)-1) | |
4073 | num_bytes = block_rsv->size; | |
4074 | block_rsv->size -= num_bytes; | |
4075 | if (block_rsv->reserved >= block_rsv->size) { | |
4076 | num_bytes = block_rsv->reserved - block_rsv->size; | |
4077 | block_rsv->reserved = block_rsv->size; | |
4078 | block_rsv->full = 1; | |
4079 | } else { | |
4080 | num_bytes = 0; | |
4081 | } | |
4082 | spin_unlock(&block_rsv->lock); | |
4083 | ||
4084 | if (num_bytes > 0) { | |
4085 | if (dest) { | |
e9e22899 JB |
4086 | spin_lock(&dest->lock); |
4087 | if (!dest->full) { | |
4088 | u64 bytes_to_add; | |
4089 | ||
4090 | bytes_to_add = dest->size - dest->reserved; | |
4091 | bytes_to_add = min(num_bytes, bytes_to_add); | |
4092 | dest->reserved += bytes_to_add; | |
4093 | if (dest->reserved >= dest->size) | |
4094 | dest->full = 1; | |
4095 | num_bytes -= bytes_to_add; | |
4096 | } | |
4097 | spin_unlock(&dest->lock); | |
4098 | } | |
4099 | if (num_bytes) { | |
f0486c68 | 4100 | spin_lock(&space_info->lock); |
fb25e914 | 4101 | space_info->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4102 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4103 | space_info->flags, num_bytes, 0); |
36e39c40 | 4104 | space_info->reservation_progress++; |
f0486c68 | 4105 | spin_unlock(&space_info->lock); |
4e06bdd6 | 4106 | } |
9ed74f2d | 4107 | } |
f0486c68 | 4108 | } |
4e06bdd6 | 4109 | |
f0486c68 YZ |
4110 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
4111 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
4112 | { | |
4113 | int ret; | |
9ed74f2d | 4114 | |
f0486c68 YZ |
4115 | ret = block_rsv_use_bytes(src, num_bytes); |
4116 | if (ret) | |
4117 | return ret; | |
9ed74f2d | 4118 | |
f0486c68 | 4119 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
4120 | return 0; |
4121 | } | |
4122 | ||
f0486c68 | 4123 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv) |
9ed74f2d | 4124 | { |
f0486c68 YZ |
4125 | memset(rsv, 0, sizeof(*rsv)); |
4126 | spin_lock_init(&rsv->lock); | |
f0486c68 YZ |
4127 | } |
4128 | ||
4129 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root) | |
4130 | { | |
4131 | struct btrfs_block_rsv *block_rsv; | |
4132 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 4133 | |
f0486c68 YZ |
4134 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
4135 | if (!block_rsv) | |
4136 | return NULL; | |
9ed74f2d | 4137 | |
f0486c68 | 4138 | btrfs_init_block_rsv(block_rsv); |
f0486c68 YZ |
4139 | block_rsv->space_info = __find_space_info(fs_info, |
4140 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
4141 | return block_rsv; |
4142 | } | |
9ed74f2d | 4143 | |
f0486c68 YZ |
4144 | void btrfs_free_block_rsv(struct btrfs_root *root, |
4145 | struct btrfs_block_rsv *rsv) | |
4146 | { | |
dabdb640 JB |
4147 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
4148 | kfree(rsv); | |
9ed74f2d JB |
4149 | } |
4150 | ||
61b520a9 MX |
4151 | static inline int __block_rsv_add(struct btrfs_root *root, |
4152 | struct btrfs_block_rsv *block_rsv, | |
4153 | u64 num_bytes, int flush) | |
9ed74f2d | 4154 | { |
f0486c68 | 4155 | int ret; |
9ed74f2d | 4156 | |
f0486c68 YZ |
4157 | if (num_bytes == 0) |
4158 | return 0; | |
8bb8ab2e | 4159 | |
61b520a9 | 4160 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
4161 | if (!ret) { |
4162 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
4163 | return 0; | |
4164 | } | |
9ed74f2d | 4165 | |
f0486c68 | 4166 | return ret; |
f0486c68 | 4167 | } |
9ed74f2d | 4168 | |
61b520a9 MX |
4169 | int btrfs_block_rsv_add(struct btrfs_root *root, |
4170 | struct btrfs_block_rsv *block_rsv, | |
4171 | u64 num_bytes) | |
4172 | { | |
4173 | return __block_rsv_add(root, block_rsv, num_bytes, 1); | |
4174 | } | |
4175 | ||
c06a0e12 JB |
4176 | int btrfs_block_rsv_add_noflush(struct btrfs_root *root, |
4177 | struct btrfs_block_rsv *block_rsv, | |
4178 | u64 num_bytes) | |
f0486c68 | 4179 | { |
61b520a9 | 4180 | return __block_rsv_add(root, block_rsv, num_bytes, 0); |
f0486c68 | 4181 | } |
9ed74f2d | 4182 | |
4a92b1b8 | 4183 | int btrfs_block_rsv_check(struct btrfs_root *root, |
36ba022a | 4184 | struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
4185 | { |
4186 | u64 num_bytes = 0; | |
f0486c68 | 4187 | int ret = -ENOSPC; |
9ed74f2d | 4188 | |
f0486c68 YZ |
4189 | if (!block_rsv) |
4190 | return 0; | |
9ed74f2d | 4191 | |
f0486c68 | 4192 | spin_lock(&block_rsv->lock); |
36ba022a JB |
4193 | num_bytes = div_factor(block_rsv->size, min_factor); |
4194 | if (block_rsv->reserved >= num_bytes) | |
4195 | ret = 0; | |
4196 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 4197 | |
36ba022a JB |
4198 | return ret; |
4199 | } | |
4200 | ||
aa38a711 MX |
4201 | static inline int __btrfs_block_rsv_refill(struct btrfs_root *root, |
4202 | struct btrfs_block_rsv *block_rsv, | |
4203 | u64 min_reserved, int flush) | |
36ba022a JB |
4204 | { |
4205 | u64 num_bytes = 0; | |
4206 | int ret = -ENOSPC; | |
4207 | ||
4208 | if (!block_rsv) | |
4209 | return 0; | |
4210 | ||
4211 | spin_lock(&block_rsv->lock); | |
4212 | num_bytes = min_reserved; | |
13553e52 | 4213 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 4214 | ret = 0; |
13553e52 | 4215 | else |
f0486c68 | 4216 | num_bytes -= block_rsv->reserved; |
f0486c68 | 4217 | spin_unlock(&block_rsv->lock); |
13553e52 | 4218 | |
f0486c68 YZ |
4219 | if (!ret) |
4220 | return 0; | |
4221 | ||
aa38a711 | 4222 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
4223 | if (!ret) { |
4224 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 4225 | return 0; |
6a63209f | 4226 | } |
9ed74f2d | 4227 | |
13553e52 | 4228 | return ret; |
f0486c68 YZ |
4229 | } |
4230 | ||
aa38a711 MX |
4231 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
4232 | struct btrfs_block_rsv *block_rsv, | |
4233 | u64 min_reserved) | |
4234 | { | |
4235 | return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1); | |
4236 | } | |
4237 | ||
4238 | int btrfs_block_rsv_refill_noflush(struct btrfs_root *root, | |
4239 | struct btrfs_block_rsv *block_rsv, | |
4240 | u64 min_reserved) | |
4241 | { | |
4242 | return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0); | |
4243 | } | |
4244 | ||
f0486c68 YZ |
4245 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, |
4246 | struct btrfs_block_rsv *dst_rsv, | |
4247 | u64 num_bytes) | |
4248 | { | |
4249 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
4250 | } | |
4251 | ||
4252 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
4253 | struct btrfs_block_rsv *block_rsv, | |
4254 | u64 num_bytes) | |
4255 | { | |
4256 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
4257 | if (global_rsv->full || global_rsv == block_rsv || | |
4258 | block_rsv->space_info != global_rsv->space_info) | |
4259 | global_rsv = NULL; | |
8c2a3ca2 JB |
4260 | block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv, |
4261 | num_bytes); | |
6a63209f JB |
4262 | } |
4263 | ||
4264 | /* | |
8929ecfa YZ |
4265 | * helper to calculate size of global block reservation. |
4266 | * the desired value is sum of space used by extent tree, | |
4267 | * checksum tree and root tree | |
6a63209f | 4268 | */ |
8929ecfa | 4269 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 4270 | { |
8929ecfa YZ |
4271 | struct btrfs_space_info *sinfo; |
4272 | u64 num_bytes; | |
4273 | u64 meta_used; | |
4274 | u64 data_used; | |
6c41761f | 4275 | int csum_size = btrfs_super_csum_size(fs_info->super_copy); |
6a63209f | 4276 | |
8929ecfa YZ |
4277 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
4278 | spin_lock(&sinfo->lock); | |
4279 | data_used = sinfo->bytes_used; | |
4280 | spin_unlock(&sinfo->lock); | |
33b4d47f | 4281 | |
8929ecfa YZ |
4282 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
4283 | spin_lock(&sinfo->lock); | |
6d48755d JB |
4284 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
4285 | data_used = 0; | |
8929ecfa YZ |
4286 | meta_used = sinfo->bytes_used; |
4287 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 4288 | |
8929ecfa YZ |
4289 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
4290 | csum_size * 2; | |
4291 | num_bytes += div64_u64(data_used + meta_used, 50); | |
4e06bdd6 | 4292 | |
8929ecfa | 4293 | if (num_bytes * 3 > meta_used) |
8e62c2de | 4294 | num_bytes = div64_u64(meta_used, 3); |
ab6e2410 | 4295 | |
8929ecfa YZ |
4296 | return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10); |
4297 | } | |
6a63209f | 4298 | |
8929ecfa YZ |
4299 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
4300 | { | |
4301 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
4302 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
4303 | u64 num_bytes; | |
6a63209f | 4304 | |
8929ecfa | 4305 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 4306 | |
8929ecfa | 4307 | spin_lock(&sinfo->lock); |
1f699d38 | 4308 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 4309 | |
8929ecfa | 4310 | block_rsv->size = num_bytes; |
4e06bdd6 | 4311 | |
8929ecfa | 4312 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
6d48755d JB |
4313 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
4314 | sinfo->bytes_may_use; | |
8929ecfa YZ |
4315 | |
4316 | if (sinfo->total_bytes > num_bytes) { | |
4317 | num_bytes = sinfo->total_bytes - num_bytes; | |
4318 | block_rsv->reserved += num_bytes; | |
fb25e914 | 4319 | sinfo->bytes_may_use += num_bytes; |
8c2a3ca2 | 4320 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4321 | sinfo->flags, num_bytes, 1); |
6a63209f | 4322 | } |
6a63209f | 4323 | |
8929ecfa YZ |
4324 | if (block_rsv->reserved >= block_rsv->size) { |
4325 | num_bytes = block_rsv->reserved - block_rsv->size; | |
fb25e914 | 4326 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4327 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4328 | sinfo->flags, num_bytes, 0); |
36e39c40 | 4329 | sinfo->reservation_progress++; |
8929ecfa YZ |
4330 | block_rsv->reserved = block_rsv->size; |
4331 | block_rsv->full = 1; | |
4332 | } | |
182608c8 | 4333 | |
8929ecfa | 4334 | spin_unlock(&block_rsv->lock); |
1f699d38 | 4335 | spin_unlock(&sinfo->lock); |
6a63209f JB |
4336 | } |
4337 | ||
f0486c68 | 4338 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4339 | { |
f0486c68 | 4340 | struct btrfs_space_info *space_info; |
6a63209f | 4341 | |
f0486c68 YZ |
4342 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
4343 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 4344 | |
f0486c68 | 4345 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 4346 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 4347 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
4348 | fs_info->trans_block_rsv.space_info = space_info; |
4349 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 4350 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 4351 | |
8929ecfa YZ |
4352 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
4353 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
4354 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
4355 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 4356 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 4357 | |
8929ecfa | 4358 | update_global_block_rsv(fs_info); |
6a63209f JB |
4359 | } |
4360 | ||
8929ecfa | 4361 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4362 | { |
8c2a3ca2 JB |
4363 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
4364 | (u64)-1); | |
8929ecfa YZ |
4365 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
4366 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
4367 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
4368 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
4369 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
4370 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
4371 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
4372 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
4373 | } |
4374 | ||
a22285a6 YZ |
4375 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
4376 | struct btrfs_root *root) | |
6a63209f | 4377 | { |
0e721106 JB |
4378 | if (!trans->block_rsv) |
4379 | return; | |
4380 | ||
a22285a6 YZ |
4381 | if (!trans->bytes_reserved) |
4382 | return; | |
6a63209f | 4383 | |
e77266e4 | 4384 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 4385 | trans->transid, trans->bytes_reserved, 0); |
b24e03db | 4386 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); |
a22285a6 YZ |
4387 | trans->bytes_reserved = 0; |
4388 | } | |
6a63209f | 4389 | |
79787eaa | 4390 | /* Can only return 0 or -ENOSPC */ |
d68fc57b YZ |
4391 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
4392 | struct inode *inode) | |
4393 | { | |
4394 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4395 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4396 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
4397 | ||
4398 | /* | |
fcb80c2a JB |
4399 | * We need to hold space in order to delete our orphan item once we've |
4400 | * added it, so this takes the reservation so we can release it later | |
4401 | * when we are truly done with the orphan item. | |
d68fc57b | 4402 | */ |
ff5714cc | 4403 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4404 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4405 | btrfs_ino(inode), num_bytes, 1); | |
d68fc57b | 4406 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
6a63209f JB |
4407 | } |
4408 | ||
d68fc57b | 4409 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 4410 | { |
d68fc57b | 4411 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 4412 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4413 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4414 | btrfs_ino(inode), num_bytes, 0); | |
d68fc57b YZ |
4415 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
4416 | } | |
97e728d4 | 4417 | |
a22285a6 YZ |
4418 | int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans, |
4419 | struct btrfs_pending_snapshot *pending) | |
4420 | { | |
4421 | struct btrfs_root *root = pending->root; | |
4422 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4423 | struct btrfs_block_rsv *dst_rsv = &pending->block_rsv; | |
4424 | /* | |
4425 | * two for root back/forward refs, two for directory entries | |
4426 | * and one for root of the snapshot. | |
4427 | */ | |
16cdcec7 | 4428 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5); |
a22285a6 YZ |
4429 | dst_rsv->space_info = src_rsv->space_info; |
4430 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
97e728d4 JB |
4431 | } |
4432 | ||
7709cde3 JB |
4433 | /** |
4434 | * drop_outstanding_extent - drop an outstanding extent | |
4435 | * @inode: the inode we're dropping the extent for | |
4436 | * | |
4437 | * This is called when we are freeing up an outstanding extent, either called | |
4438 | * after an error or after an extent is written. This will return the number of | |
4439 | * reserved extents that need to be freed. This must be called with | |
4440 | * BTRFS_I(inode)->lock held. | |
4441 | */ | |
9e0baf60 JB |
4442 | static unsigned drop_outstanding_extent(struct inode *inode) |
4443 | { | |
7fd2ae21 | 4444 | unsigned drop_inode_space = 0; |
9e0baf60 JB |
4445 | unsigned dropped_extents = 0; |
4446 | ||
9e0baf60 JB |
4447 | BUG_ON(!BTRFS_I(inode)->outstanding_extents); |
4448 | BTRFS_I(inode)->outstanding_extents--; | |
4449 | ||
7fd2ae21 | 4450 | if (BTRFS_I(inode)->outstanding_extents == 0 && |
72ac3c0d JB |
4451 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4452 | &BTRFS_I(inode)->runtime_flags)) | |
7fd2ae21 | 4453 | drop_inode_space = 1; |
7fd2ae21 | 4454 | |
9e0baf60 JB |
4455 | /* |
4456 | * If we have more or the same amount of outsanding extents than we have | |
4457 | * reserved then we need to leave the reserved extents count alone. | |
4458 | */ | |
4459 | if (BTRFS_I(inode)->outstanding_extents >= | |
4460 | BTRFS_I(inode)->reserved_extents) | |
7fd2ae21 | 4461 | return drop_inode_space; |
9e0baf60 JB |
4462 | |
4463 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
4464 | BTRFS_I(inode)->outstanding_extents; | |
4465 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
7fd2ae21 | 4466 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
4467 | } |
4468 | ||
7709cde3 JB |
4469 | /** |
4470 | * calc_csum_metadata_size - return the amount of metada space that must be | |
4471 | * reserved/free'd for the given bytes. | |
4472 | * @inode: the inode we're manipulating | |
4473 | * @num_bytes: the number of bytes in question | |
4474 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
4475 | * | |
4476 | * This adjusts the number of csum_bytes in the inode and then returns the | |
4477 | * correct amount of metadata that must either be reserved or freed. We | |
4478 | * calculate how many checksums we can fit into one leaf and then divide the | |
4479 | * number of bytes that will need to be checksumed by this value to figure out | |
4480 | * how many checksums will be required. If we are adding bytes then the number | |
4481 | * may go up and we will return the number of additional bytes that must be | |
4482 | * reserved. If it is going down we will return the number of bytes that must | |
4483 | * be freed. | |
4484 | * | |
4485 | * This must be called with BTRFS_I(inode)->lock held. | |
4486 | */ | |
4487 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, | |
4488 | int reserve) | |
6324fbf3 | 4489 | { |
7709cde3 JB |
4490 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4491 | u64 csum_size; | |
4492 | int num_csums_per_leaf; | |
4493 | int num_csums; | |
4494 | int old_csums; | |
4495 | ||
4496 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && | |
4497 | BTRFS_I(inode)->csum_bytes == 0) | |
4498 | return 0; | |
4499 | ||
4500 | old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4501 | if (reserve) | |
4502 | BTRFS_I(inode)->csum_bytes += num_bytes; | |
4503 | else | |
4504 | BTRFS_I(inode)->csum_bytes -= num_bytes; | |
4505 | csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item); | |
4506 | num_csums_per_leaf = (int)div64_u64(csum_size, | |
4507 | sizeof(struct btrfs_csum_item) + | |
4508 | sizeof(struct btrfs_disk_key)); | |
4509 | num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4510 | num_csums = num_csums + num_csums_per_leaf - 1; | |
4511 | num_csums = num_csums / num_csums_per_leaf; | |
4512 | ||
4513 | old_csums = old_csums + num_csums_per_leaf - 1; | |
4514 | old_csums = old_csums / num_csums_per_leaf; | |
4515 | ||
4516 | /* No change, no need to reserve more */ | |
4517 | if (old_csums == num_csums) | |
4518 | return 0; | |
4519 | ||
4520 | if (reserve) | |
4521 | return btrfs_calc_trans_metadata_size(root, | |
4522 | num_csums - old_csums); | |
4523 | ||
4524 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); | |
0ca1f7ce | 4525 | } |
c146afad | 4526 | |
0ca1f7ce YZ |
4527 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
4528 | { | |
4529 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4530 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 | 4531 | u64 to_reserve = 0; |
660d3f6c | 4532 | u64 csum_bytes; |
9e0baf60 | 4533 | unsigned nr_extents = 0; |
660d3f6c | 4534 | int extra_reserve = 0; |
c09544e0 | 4535 | int flush = 1; |
0ca1f7ce | 4536 | int ret; |
6324fbf3 | 4537 | |
660d3f6c | 4538 | /* Need to be holding the i_mutex here if we aren't free space cache */ |
83eea1f1 | 4539 | if (btrfs_is_free_space_inode(inode)) |
c09544e0 JB |
4540 | flush = 0; |
4541 | ||
4542 | if (flush && btrfs_transaction_in_commit(root->fs_info)) | |
0ca1f7ce | 4543 | schedule_timeout(1); |
ec44a35c | 4544 | |
f248679e | 4545 | mutex_lock(&BTRFS_I(inode)->delalloc_mutex); |
0ca1f7ce | 4546 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 4547 | |
9e0baf60 JB |
4548 | spin_lock(&BTRFS_I(inode)->lock); |
4549 | BTRFS_I(inode)->outstanding_extents++; | |
4550 | ||
4551 | if (BTRFS_I(inode)->outstanding_extents > | |
660d3f6c | 4552 | BTRFS_I(inode)->reserved_extents) |
9e0baf60 JB |
4553 | nr_extents = BTRFS_I(inode)->outstanding_extents - |
4554 | BTRFS_I(inode)->reserved_extents; | |
57a45ced | 4555 | |
7fd2ae21 JB |
4556 | /* |
4557 | * Add an item to reserve for updating the inode when we complete the | |
4558 | * delalloc io. | |
4559 | */ | |
72ac3c0d JB |
4560 | if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4561 | &BTRFS_I(inode)->runtime_flags)) { | |
7fd2ae21 | 4562 | nr_extents++; |
660d3f6c | 4563 | extra_reserve = 1; |
593060d7 | 4564 | } |
7fd2ae21 JB |
4565 | |
4566 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); | |
7709cde3 | 4567 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
660d3f6c | 4568 | csum_bytes = BTRFS_I(inode)->csum_bytes; |
9e0baf60 | 4569 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 4570 | |
c5567237 AJ |
4571 | if (root->fs_info->quota_enabled) { |
4572 | ret = btrfs_qgroup_reserve(root, num_bytes + | |
4573 | nr_extents * root->leafsize); | |
4574 | if (ret) | |
4575 | return ret; | |
4576 | } | |
4577 | ||
36ba022a | 4578 | ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush); |
9e0baf60 | 4579 | if (ret) { |
7ed49f18 | 4580 | u64 to_free = 0; |
9e0baf60 | 4581 | unsigned dropped; |
7ed49f18 | 4582 | |
7709cde3 | 4583 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 4584 | dropped = drop_outstanding_extent(inode); |
9e0baf60 | 4585 | /* |
660d3f6c JB |
4586 | * If the inodes csum_bytes is the same as the original |
4587 | * csum_bytes then we know we haven't raced with any free()ers | |
4588 | * so we can just reduce our inodes csum bytes and carry on. | |
4589 | * Otherwise we have to do the normal free thing to account for | |
4590 | * the case that the free side didn't free up its reserve | |
4591 | * because of this outstanding reservation. | |
9e0baf60 | 4592 | */ |
660d3f6c JB |
4593 | if (BTRFS_I(inode)->csum_bytes == csum_bytes) |
4594 | calc_csum_metadata_size(inode, num_bytes, 0); | |
4595 | else | |
4596 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
4597 | spin_unlock(&BTRFS_I(inode)->lock); | |
4598 | if (dropped) | |
4599 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
4600 | ||
8c2a3ca2 | 4601 | if (to_free) { |
7ed49f18 | 4602 | btrfs_block_rsv_release(root, block_rsv, to_free); |
8c2a3ca2 JB |
4603 | trace_btrfs_space_reservation(root->fs_info, |
4604 | "delalloc", | |
4605 | btrfs_ino(inode), | |
4606 | to_free, 0); | |
4607 | } | |
f248679e | 4608 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); |
0ca1f7ce | 4609 | return ret; |
9e0baf60 | 4610 | } |
25179201 | 4611 | |
660d3f6c JB |
4612 | spin_lock(&BTRFS_I(inode)->lock); |
4613 | if (extra_reserve) { | |
72ac3c0d JB |
4614 | set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4615 | &BTRFS_I(inode)->runtime_flags); | |
660d3f6c JB |
4616 | nr_extents--; |
4617 | } | |
4618 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
4619 | spin_unlock(&BTRFS_I(inode)->lock); | |
f248679e | 4620 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); |
660d3f6c | 4621 | |
8c2a3ca2 JB |
4622 | if (to_reserve) |
4623 | trace_btrfs_space_reservation(root->fs_info,"delalloc", | |
4624 | btrfs_ino(inode), to_reserve, 1); | |
0ca1f7ce YZ |
4625 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
4626 | ||
0ca1f7ce YZ |
4627 | return 0; |
4628 | } | |
4629 | ||
7709cde3 JB |
4630 | /** |
4631 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
4632 | * @inode: the inode to release the reservation for | |
4633 | * @num_bytes: the number of bytes we're releasing | |
4634 | * | |
4635 | * This will release the metadata reservation for an inode. This can be called | |
4636 | * once we complete IO for a given set of bytes to release their metadata | |
4637 | * reservations. | |
4638 | */ | |
0ca1f7ce YZ |
4639 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) |
4640 | { | |
4641 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
4642 | u64 to_free = 0; |
4643 | unsigned dropped; | |
0ca1f7ce YZ |
4644 | |
4645 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
7709cde3 | 4646 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 4647 | dropped = drop_outstanding_extent(inode); |
97e728d4 | 4648 | |
7709cde3 JB |
4649 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); |
4650 | spin_unlock(&BTRFS_I(inode)->lock); | |
9e0baf60 JB |
4651 | if (dropped > 0) |
4652 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce | 4653 | |
8c2a3ca2 JB |
4654 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
4655 | btrfs_ino(inode), to_free, 0); | |
c5567237 AJ |
4656 | if (root->fs_info->quota_enabled) { |
4657 | btrfs_qgroup_free(root, num_bytes + | |
4658 | dropped * root->leafsize); | |
4659 | } | |
4660 | ||
0ca1f7ce YZ |
4661 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, |
4662 | to_free); | |
4663 | } | |
4664 | ||
7709cde3 JB |
4665 | /** |
4666 | * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc | |
4667 | * @inode: inode we're writing to | |
4668 | * @num_bytes: the number of bytes we want to allocate | |
4669 | * | |
4670 | * This will do the following things | |
4671 | * | |
4672 | * o reserve space in the data space info for num_bytes | |
4673 | * o reserve space in the metadata space info based on number of outstanding | |
4674 | * extents and how much csums will be needed | |
4675 | * o add to the inodes ->delalloc_bytes | |
4676 | * o add it to the fs_info's delalloc inodes list. | |
4677 | * | |
4678 | * This will return 0 for success and -ENOSPC if there is no space left. | |
4679 | */ | |
0ca1f7ce YZ |
4680 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes) |
4681 | { | |
4682 | int ret; | |
4683 | ||
4684 | ret = btrfs_check_data_free_space(inode, num_bytes); | |
d397712b | 4685 | if (ret) |
0ca1f7ce YZ |
4686 | return ret; |
4687 | ||
4688 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes); | |
4689 | if (ret) { | |
4690 | btrfs_free_reserved_data_space(inode, num_bytes); | |
4691 | return ret; | |
4692 | } | |
4693 | ||
4694 | return 0; | |
4695 | } | |
4696 | ||
7709cde3 JB |
4697 | /** |
4698 | * btrfs_delalloc_release_space - release data and metadata space for delalloc | |
4699 | * @inode: inode we're releasing space for | |
4700 | * @num_bytes: the number of bytes we want to free up | |
4701 | * | |
4702 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
4703 | * called in the case that we don't need the metadata AND data reservations | |
4704 | * anymore. So if there is an error or we insert an inline extent. | |
4705 | * | |
4706 | * This function will release the metadata space that was not used and will | |
4707 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
4708 | * list if there are no delalloc bytes left. | |
4709 | */ | |
0ca1f7ce YZ |
4710 | void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes) |
4711 | { | |
4712 | btrfs_delalloc_release_metadata(inode, num_bytes); | |
4713 | btrfs_free_reserved_data_space(inode, num_bytes); | |
6324fbf3 CM |
4714 | } |
4715 | ||
9078a3e1 CM |
4716 | static int update_block_group(struct btrfs_trans_handle *trans, |
4717 | struct btrfs_root *root, | |
f0486c68 | 4718 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 4719 | { |
0af3d00b | 4720 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 4721 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 4722 | u64 total = num_bytes; |
9078a3e1 | 4723 | u64 old_val; |
db94535d | 4724 | u64 byte_in_group; |
0af3d00b | 4725 | int factor; |
3e1ad54f | 4726 | |
5d4f98a2 YZ |
4727 | /* block accounting for super block */ |
4728 | spin_lock(&info->delalloc_lock); | |
6c41761f | 4729 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
4730 | if (alloc) |
4731 | old_val += num_bytes; | |
4732 | else | |
4733 | old_val -= num_bytes; | |
6c41761f | 4734 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
5d4f98a2 YZ |
4735 | spin_unlock(&info->delalloc_lock); |
4736 | ||
d397712b | 4737 | while (total) { |
db94535d | 4738 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 4739 | if (!cache) |
79787eaa | 4740 | return -ENOENT; |
b742bb82 YZ |
4741 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
4742 | BTRFS_BLOCK_GROUP_RAID1 | | |
4743 | BTRFS_BLOCK_GROUP_RAID10)) | |
4744 | factor = 2; | |
4745 | else | |
4746 | factor = 1; | |
9d66e233 JB |
4747 | /* |
4748 | * If this block group has free space cache written out, we | |
4749 | * need to make sure to load it if we are removing space. This | |
4750 | * is because we need the unpinning stage to actually add the | |
4751 | * space back to the block group, otherwise we will leak space. | |
4752 | */ | |
4753 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
b8399dee | 4754 | cache_block_group(cache, trans, NULL, 1); |
0af3d00b | 4755 | |
db94535d CM |
4756 | byte_in_group = bytenr - cache->key.objectid; |
4757 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 4758 | |
25179201 | 4759 | spin_lock(&cache->space_info->lock); |
c286ac48 | 4760 | spin_lock(&cache->lock); |
0af3d00b | 4761 | |
73bc1876 | 4762 | if (btrfs_test_opt(root, SPACE_CACHE) && |
0af3d00b JB |
4763 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
4764 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
4765 | ||
0f9dd46c | 4766 | cache->dirty = 1; |
9078a3e1 | 4767 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 4768 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 4769 | if (alloc) { |
db94535d | 4770 | old_val += num_bytes; |
11833d66 YZ |
4771 | btrfs_set_block_group_used(&cache->item, old_val); |
4772 | cache->reserved -= num_bytes; | |
11833d66 | 4773 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
4774 | cache->space_info->bytes_used += num_bytes; |
4775 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 4776 | spin_unlock(&cache->lock); |
25179201 | 4777 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 4778 | } else { |
db94535d | 4779 | old_val -= num_bytes; |
c286ac48 | 4780 | btrfs_set_block_group_used(&cache->item, old_val); |
f0486c68 YZ |
4781 | cache->pinned += num_bytes; |
4782 | cache->space_info->bytes_pinned += num_bytes; | |
6324fbf3 | 4783 | cache->space_info->bytes_used -= num_bytes; |
b742bb82 | 4784 | cache->space_info->disk_used -= num_bytes * factor; |
c286ac48 | 4785 | spin_unlock(&cache->lock); |
25179201 | 4786 | spin_unlock(&cache->space_info->lock); |
1f3c79a2 | 4787 | |
f0486c68 YZ |
4788 | set_extent_dirty(info->pinned_extents, |
4789 | bytenr, bytenr + num_bytes - 1, | |
4790 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 4791 | } |
fa9c0d79 | 4792 | btrfs_put_block_group(cache); |
db94535d CM |
4793 | total -= num_bytes; |
4794 | bytenr += num_bytes; | |
9078a3e1 CM |
4795 | } |
4796 | return 0; | |
4797 | } | |
6324fbf3 | 4798 | |
a061fc8d CM |
4799 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
4800 | { | |
0f9dd46c | 4801 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 4802 | u64 bytenr; |
0f9dd46c JB |
4803 | |
4804 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); | |
4805 | if (!cache) | |
a061fc8d | 4806 | return 0; |
0f9dd46c | 4807 | |
d2fb3437 | 4808 | bytenr = cache->key.objectid; |
fa9c0d79 | 4809 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
4810 | |
4811 | return bytenr; | |
a061fc8d CM |
4812 | } |
4813 | ||
f0486c68 YZ |
4814 | static int pin_down_extent(struct btrfs_root *root, |
4815 | struct btrfs_block_group_cache *cache, | |
4816 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 4817 | { |
11833d66 YZ |
4818 | spin_lock(&cache->space_info->lock); |
4819 | spin_lock(&cache->lock); | |
4820 | cache->pinned += num_bytes; | |
4821 | cache->space_info->bytes_pinned += num_bytes; | |
4822 | if (reserved) { | |
4823 | cache->reserved -= num_bytes; | |
4824 | cache->space_info->bytes_reserved -= num_bytes; | |
4825 | } | |
4826 | spin_unlock(&cache->lock); | |
4827 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 4828 | |
f0486c68 YZ |
4829 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
4830 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
4831 | return 0; | |
4832 | } | |
68b38550 | 4833 | |
f0486c68 YZ |
4834 | /* |
4835 | * this function must be called within transaction | |
4836 | */ | |
4837 | int btrfs_pin_extent(struct btrfs_root *root, | |
4838 | u64 bytenr, u64 num_bytes, int reserved) | |
4839 | { | |
4840 | struct btrfs_block_group_cache *cache; | |
68b38550 | 4841 | |
f0486c68 | 4842 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
79787eaa | 4843 | BUG_ON(!cache); /* Logic error */ |
f0486c68 YZ |
4844 | |
4845 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
4846 | ||
4847 | btrfs_put_block_group(cache); | |
11833d66 YZ |
4848 | return 0; |
4849 | } | |
4850 | ||
f0486c68 | 4851 | /* |
e688b725 CM |
4852 | * this function must be called within transaction |
4853 | */ | |
4854 | int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans, | |
4855 | struct btrfs_root *root, | |
4856 | u64 bytenr, u64 num_bytes) | |
4857 | { | |
4858 | struct btrfs_block_group_cache *cache; | |
4859 | ||
4860 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
79787eaa | 4861 | BUG_ON(!cache); /* Logic error */ |
e688b725 CM |
4862 | |
4863 | /* | |
4864 | * pull in the free space cache (if any) so that our pin | |
4865 | * removes the free space from the cache. We have load_only set | |
4866 | * to one because the slow code to read in the free extents does check | |
4867 | * the pinned extents. | |
4868 | */ | |
4869 | cache_block_group(cache, trans, root, 1); | |
4870 | ||
4871 | pin_down_extent(root, cache, bytenr, num_bytes, 0); | |
4872 | ||
4873 | /* remove us from the free space cache (if we're there at all) */ | |
4874 | btrfs_remove_free_space(cache, bytenr, num_bytes); | |
4875 | btrfs_put_block_group(cache); | |
4876 | return 0; | |
4877 | } | |
4878 | ||
fb25e914 JB |
4879 | /** |
4880 | * btrfs_update_reserved_bytes - update the block_group and space info counters | |
4881 | * @cache: The cache we are manipulating | |
4882 | * @num_bytes: The number of bytes in question | |
4883 | * @reserve: One of the reservation enums | |
4884 | * | |
4885 | * This is called by the allocator when it reserves space, or by somebody who is | |
4886 | * freeing space that was never actually used on disk. For example if you | |
4887 | * reserve some space for a new leaf in transaction A and before transaction A | |
4888 | * commits you free that leaf, you call this with reserve set to 0 in order to | |
4889 | * clear the reservation. | |
4890 | * | |
4891 | * Metadata reservations should be called with RESERVE_ALLOC so we do the proper | |
4892 | * ENOSPC accounting. For data we handle the reservation through clearing the | |
4893 | * delalloc bits in the io_tree. We have to do this since we could end up | |
4894 | * allocating less disk space for the amount of data we have reserved in the | |
4895 | * case of compression. | |
4896 | * | |
4897 | * If this is a reservation and the block group has become read only we cannot | |
4898 | * make the reservation and return -EAGAIN, otherwise this function always | |
4899 | * succeeds. | |
f0486c68 | 4900 | */ |
fb25e914 JB |
4901 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
4902 | u64 num_bytes, int reserve) | |
11833d66 | 4903 | { |
fb25e914 | 4904 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 4905 | int ret = 0; |
79787eaa | 4906 | |
fb25e914 JB |
4907 | spin_lock(&space_info->lock); |
4908 | spin_lock(&cache->lock); | |
4909 | if (reserve != RESERVE_FREE) { | |
f0486c68 YZ |
4910 | if (cache->ro) { |
4911 | ret = -EAGAIN; | |
4912 | } else { | |
fb25e914 JB |
4913 | cache->reserved += num_bytes; |
4914 | space_info->bytes_reserved += num_bytes; | |
4915 | if (reserve == RESERVE_ALLOC) { | |
8c2a3ca2 | 4916 | trace_btrfs_space_reservation(cache->fs_info, |
2bcc0328 LB |
4917 | "space_info", space_info->flags, |
4918 | num_bytes, 0); | |
fb25e914 JB |
4919 | space_info->bytes_may_use -= num_bytes; |
4920 | } | |
f0486c68 | 4921 | } |
fb25e914 JB |
4922 | } else { |
4923 | if (cache->ro) | |
4924 | space_info->bytes_readonly += num_bytes; | |
4925 | cache->reserved -= num_bytes; | |
4926 | space_info->bytes_reserved -= num_bytes; | |
4927 | space_info->reservation_progress++; | |
324ae4df | 4928 | } |
fb25e914 JB |
4929 | spin_unlock(&cache->lock); |
4930 | spin_unlock(&space_info->lock); | |
f0486c68 | 4931 | return ret; |
324ae4df | 4932 | } |
9078a3e1 | 4933 | |
143bede5 | 4934 | void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
11833d66 | 4935 | struct btrfs_root *root) |
e8569813 | 4936 | { |
e8569813 | 4937 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
4938 | struct btrfs_caching_control *next; |
4939 | struct btrfs_caching_control *caching_ctl; | |
4940 | struct btrfs_block_group_cache *cache; | |
e8569813 | 4941 | |
11833d66 | 4942 | down_write(&fs_info->extent_commit_sem); |
25179201 | 4943 | |
11833d66 YZ |
4944 | list_for_each_entry_safe(caching_ctl, next, |
4945 | &fs_info->caching_block_groups, list) { | |
4946 | cache = caching_ctl->block_group; | |
4947 | if (block_group_cache_done(cache)) { | |
4948 | cache->last_byte_to_unpin = (u64)-1; | |
4949 | list_del_init(&caching_ctl->list); | |
4950 | put_caching_control(caching_ctl); | |
e8569813 | 4951 | } else { |
11833d66 | 4952 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 4953 | } |
e8569813 | 4954 | } |
11833d66 YZ |
4955 | |
4956 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
4957 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
4958 | else | |
4959 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
4960 | ||
4961 | up_write(&fs_info->extent_commit_sem); | |
8929ecfa YZ |
4962 | |
4963 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
4964 | } |
4965 | ||
11833d66 | 4966 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
ccd467d6 | 4967 | { |
11833d66 YZ |
4968 | struct btrfs_fs_info *fs_info = root->fs_info; |
4969 | struct btrfs_block_group_cache *cache = NULL; | |
4970 | u64 len; | |
ccd467d6 | 4971 | |
11833d66 YZ |
4972 | while (start <= end) { |
4973 | if (!cache || | |
4974 | start >= cache->key.objectid + cache->key.offset) { | |
4975 | if (cache) | |
4976 | btrfs_put_block_group(cache); | |
4977 | cache = btrfs_lookup_block_group(fs_info, start); | |
79787eaa | 4978 | BUG_ON(!cache); /* Logic error */ |
11833d66 YZ |
4979 | } |
4980 | ||
4981 | len = cache->key.objectid + cache->key.offset - start; | |
4982 | len = min(len, end + 1 - start); | |
4983 | ||
4984 | if (start < cache->last_byte_to_unpin) { | |
4985 | len = min(len, cache->last_byte_to_unpin - start); | |
4986 | btrfs_add_free_space(cache, start, len); | |
4987 | } | |
4988 | ||
f0486c68 YZ |
4989 | start += len; |
4990 | ||
11833d66 YZ |
4991 | spin_lock(&cache->space_info->lock); |
4992 | spin_lock(&cache->lock); | |
4993 | cache->pinned -= len; | |
4994 | cache->space_info->bytes_pinned -= len; | |
37be25bc | 4995 | if (cache->ro) |
f0486c68 | 4996 | cache->space_info->bytes_readonly += len; |
11833d66 YZ |
4997 | spin_unlock(&cache->lock); |
4998 | spin_unlock(&cache->space_info->lock); | |
ccd467d6 | 4999 | } |
11833d66 YZ |
5000 | |
5001 | if (cache) | |
5002 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
5003 | return 0; |
5004 | } | |
5005 | ||
5006 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 5007 | struct btrfs_root *root) |
a28ec197 | 5008 | { |
11833d66 YZ |
5009 | struct btrfs_fs_info *fs_info = root->fs_info; |
5010 | struct extent_io_tree *unpin; | |
1a5bc167 CM |
5011 | u64 start; |
5012 | u64 end; | |
a28ec197 | 5013 | int ret; |
a28ec197 | 5014 | |
79787eaa JM |
5015 | if (trans->aborted) |
5016 | return 0; | |
5017 | ||
11833d66 YZ |
5018 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
5019 | unpin = &fs_info->freed_extents[1]; | |
5020 | else | |
5021 | unpin = &fs_info->freed_extents[0]; | |
5022 | ||
d397712b | 5023 | while (1) { |
1a5bc167 CM |
5024 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
5025 | EXTENT_DIRTY); | |
5026 | if (ret) | |
a28ec197 | 5027 | break; |
1f3c79a2 | 5028 | |
5378e607 LD |
5029 | if (btrfs_test_opt(root, DISCARD)) |
5030 | ret = btrfs_discard_extent(root, start, | |
5031 | end + 1 - start, NULL); | |
1f3c79a2 | 5032 | |
1a5bc167 | 5033 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
11833d66 | 5034 | unpin_extent_range(root, start, end); |
b9473439 | 5035 | cond_resched(); |
a28ec197 | 5036 | } |
817d52f8 | 5037 | |
e20d96d6 CM |
5038 | return 0; |
5039 | } | |
5040 | ||
5d4f98a2 YZ |
5041 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
5042 | struct btrfs_root *root, | |
5043 | u64 bytenr, u64 num_bytes, u64 parent, | |
5044 | u64 root_objectid, u64 owner_objectid, | |
5045 | u64 owner_offset, int refs_to_drop, | |
5046 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 5047 | { |
e2fa7227 | 5048 | struct btrfs_key key; |
5d4f98a2 | 5049 | struct btrfs_path *path; |
1261ec42 CM |
5050 | struct btrfs_fs_info *info = root->fs_info; |
5051 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 5052 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
5053 | struct btrfs_extent_item *ei; |
5054 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 5055 | int ret; |
5d4f98a2 | 5056 | int is_data; |
952fccac CM |
5057 | int extent_slot = 0; |
5058 | int found_extent = 0; | |
5059 | int num_to_del = 1; | |
5d4f98a2 YZ |
5060 | u32 item_size; |
5061 | u64 refs; | |
037e6390 | 5062 | |
5caf2a00 | 5063 | path = btrfs_alloc_path(); |
54aa1f4d CM |
5064 | if (!path) |
5065 | return -ENOMEM; | |
5f26f772 | 5066 | |
3c12ac72 | 5067 | path->reada = 1; |
b9473439 | 5068 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5069 | |
5070 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
5071 | BUG_ON(!is_data && refs_to_drop != 1); | |
5072 | ||
5073 | ret = lookup_extent_backref(trans, extent_root, path, &iref, | |
5074 | bytenr, num_bytes, parent, | |
5075 | root_objectid, owner_objectid, | |
5076 | owner_offset); | |
7bb86316 | 5077 | if (ret == 0) { |
952fccac | 5078 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
5079 | while (extent_slot >= 0) { |
5080 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 5081 | extent_slot); |
5d4f98a2 | 5082 | if (key.objectid != bytenr) |
952fccac | 5083 | break; |
5d4f98a2 YZ |
5084 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
5085 | key.offset == num_bytes) { | |
952fccac CM |
5086 | found_extent = 1; |
5087 | break; | |
5088 | } | |
5089 | if (path->slots[0] - extent_slot > 5) | |
5090 | break; | |
5d4f98a2 | 5091 | extent_slot--; |
952fccac | 5092 | } |
5d4f98a2 YZ |
5093 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
5094 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
5095 | if (found_extent && item_size < sizeof(*ei)) | |
5096 | found_extent = 0; | |
5097 | #endif | |
31840ae1 | 5098 | if (!found_extent) { |
5d4f98a2 | 5099 | BUG_ON(iref); |
56bec294 | 5100 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
5101 | NULL, refs_to_drop, |
5102 | is_data); | |
79787eaa JM |
5103 | if (ret) |
5104 | goto abort; | |
b3b4aa74 | 5105 | btrfs_release_path(path); |
b9473439 | 5106 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5107 | |
5108 | key.objectid = bytenr; | |
5109 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5110 | key.offset = num_bytes; | |
5111 | ||
31840ae1 ZY |
5112 | ret = btrfs_search_slot(trans, extent_root, |
5113 | &key, path, -1, 1); | |
f3465ca4 JB |
5114 | if (ret) { |
5115 | printk(KERN_ERR "umm, got %d back from search" | |
d397712b CM |
5116 | ", was looking for %llu\n", ret, |
5117 | (unsigned long long)bytenr); | |
b783e62d JB |
5118 | if (ret > 0) |
5119 | btrfs_print_leaf(extent_root, | |
5120 | path->nodes[0]); | |
f3465ca4 | 5121 | } |
79787eaa JM |
5122 | if (ret < 0) |
5123 | goto abort; | |
31840ae1 ZY |
5124 | extent_slot = path->slots[0]; |
5125 | } | |
79787eaa | 5126 | } else if (ret == -ENOENT) { |
7bb86316 CM |
5127 | btrfs_print_leaf(extent_root, path->nodes[0]); |
5128 | WARN_ON(1); | |
d397712b | 5129 | printk(KERN_ERR "btrfs unable to find ref byte nr %llu " |
5d4f98a2 | 5130 | "parent %llu root %llu owner %llu offset %llu\n", |
d397712b | 5131 | (unsigned long long)bytenr, |
56bec294 | 5132 | (unsigned long long)parent, |
d397712b | 5133 | (unsigned long long)root_objectid, |
5d4f98a2 YZ |
5134 | (unsigned long long)owner_objectid, |
5135 | (unsigned long long)owner_offset); | |
79787eaa JM |
5136 | } else { |
5137 | goto abort; | |
7bb86316 | 5138 | } |
5f39d397 CM |
5139 | |
5140 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
5141 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
5142 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
5143 | if (item_size < sizeof(*ei)) { | |
5144 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
5145 | ret = convert_extent_item_v0(trans, extent_root, path, | |
5146 | owner_objectid, 0); | |
79787eaa JM |
5147 | if (ret < 0) |
5148 | goto abort; | |
5d4f98a2 | 5149 | |
b3b4aa74 | 5150 | btrfs_release_path(path); |
5d4f98a2 YZ |
5151 | path->leave_spinning = 1; |
5152 | ||
5153 | key.objectid = bytenr; | |
5154 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5155 | key.offset = num_bytes; | |
5156 | ||
5157 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
5158 | -1, 1); | |
5159 | if (ret) { | |
5160 | printk(KERN_ERR "umm, got %d back from search" | |
5161 | ", was looking for %llu\n", ret, | |
5162 | (unsigned long long)bytenr); | |
5163 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
5164 | } | |
79787eaa JM |
5165 | if (ret < 0) |
5166 | goto abort; | |
5d4f98a2 YZ |
5167 | extent_slot = path->slots[0]; |
5168 | leaf = path->nodes[0]; | |
5169 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
5170 | } | |
5171 | #endif | |
5172 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 5173 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 5174 | struct btrfs_extent_item); |
5d4f98a2 YZ |
5175 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { |
5176 | struct btrfs_tree_block_info *bi; | |
5177 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
5178 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
5179 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
5180 | } | |
56bec294 | 5181 | |
5d4f98a2 | 5182 | refs = btrfs_extent_refs(leaf, ei); |
56bec294 CM |
5183 | BUG_ON(refs < refs_to_drop); |
5184 | refs -= refs_to_drop; | |
5f39d397 | 5185 | |
5d4f98a2 YZ |
5186 | if (refs > 0) { |
5187 | if (extent_op) | |
5188 | __run_delayed_extent_op(extent_op, leaf, ei); | |
5189 | /* | |
5190 | * In the case of inline back ref, reference count will | |
5191 | * be updated by remove_extent_backref | |
952fccac | 5192 | */ |
5d4f98a2 YZ |
5193 | if (iref) { |
5194 | BUG_ON(!found_extent); | |
5195 | } else { | |
5196 | btrfs_set_extent_refs(leaf, ei, refs); | |
5197 | btrfs_mark_buffer_dirty(leaf); | |
5198 | } | |
5199 | if (found_extent) { | |
5200 | ret = remove_extent_backref(trans, extent_root, path, | |
5201 | iref, refs_to_drop, | |
5202 | is_data); | |
79787eaa JM |
5203 | if (ret) |
5204 | goto abort; | |
952fccac | 5205 | } |
5d4f98a2 | 5206 | } else { |
5d4f98a2 YZ |
5207 | if (found_extent) { |
5208 | BUG_ON(is_data && refs_to_drop != | |
5209 | extent_data_ref_count(root, path, iref)); | |
5210 | if (iref) { | |
5211 | BUG_ON(path->slots[0] != extent_slot); | |
5212 | } else { | |
5213 | BUG_ON(path->slots[0] != extent_slot + 1); | |
5214 | path->slots[0] = extent_slot; | |
5215 | num_to_del = 2; | |
5216 | } | |
78fae27e | 5217 | } |
b9473439 | 5218 | |
952fccac CM |
5219 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
5220 | num_to_del); | |
79787eaa JM |
5221 | if (ret) |
5222 | goto abort; | |
b3b4aa74 | 5223 | btrfs_release_path(path); |
21af804c | 5224 | |
5d4f98a2 | 5225 | if (is_data) { |
459931ec | 5226 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
79787eaa JM |
5227 | if (ret) |
5228 | goto abort; | |
459931ec CM |
5229 | } |
5230 | ||
f0486c68 | 5231 | ret = update_block_group(trans, root, bytenr, num_bytes, 0); |
79787eaa JM |
5232 | if (ret) |
5233 | goto abort; | |
a28ec197 | 5234 | } |
79787eaa | 5235 | out: |
5caf2a00 | 5236 | btrfs_free_path(path); |
a28ec197 | 5237 | return ret; |
79787eaa JM |
5238 | |
5239 | abort: | |
5240 | btrfs_abort_transaction(trans, extent_root, ret); | |
5241 | goto out; | |
a28ec197 CM |
5242 | } |
5243 | ||
1887be66 | 5244 | /* |
f0486c68 | 5245 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
5246 | * delayed ref for that extent as well. This searches the delayed ref tree for |
5247 | * a given extent, and if there are no other delayed refs to be processed, it | |
5248 | * removes it from the tree. | |
5249 | */ | |
5250 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
5251 | struct btrfs_root *root, u64 bytenr) | |
5252 | { | |
5253 | struct btrfs_delayed_ref_head *head; | |
5254 | struct btrfs_delayed_ref_root *delayed_refs; | |
5255 | struct btrfs_delayed_ref_node *ref; | |
5256 | struct rb_node *node; | |
f0486c68 | 5257 | int ret = 0; |
1887be66 CM |
5258 | |
5259 | delayed_refs = &trans->transaction->delayed_refs; | |
5260 | spin_lock(&delayed_refs->lock); | |
5261 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
5262 | if (!head) | |
5263 | goto out; | |
5264 | ||
5265 | node = rb_prev(&head->node.rb_node); | |
5266 | if (!node) | |
5267 | goto out; | |
5268 | ||
5269 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
5270 | ||
5271 | /* there are still entries for this ref, we can't drop it */ | |
5272 | if (ref->bytenr == bytenr) | |
5273 | goto out; | |
5274 | ||
5d4f98a2 YZ |
5275 | if (head->extent_op) { |
5276 | if (!head->must_insert_reserved) | |
5277 | goto out; | |
5278 | kfree(head->extent_op); | |
5279 | head->extent_op = NULL; | |
5280 | } | |
5281 | ||
1887be66 CM |
5282 | /* |
5283 | * waiting for the lock here would deadlock. If someone else has it | |
5284 | * locked they are already in the process of dropping it anyway | |
5285 | */ | |
5286 | if (!mutex_trylock(&head->mutex)) | |
5287 | goto out; | |
5288 | ||
5289 | /* | |
5290 | * at this point we have a head with no other entries. Go | |
5291 | * ahead and process it. | |
5292 | */ | |
5293 | head->node.in_tree = 0; | |
5294 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 5295 | |
1887be66 | 5296 | delayed_refs->num_entries--; |
cd1cfc49 | 5297 | smp_mb(); |
097b8a7c JS |
5298 | if (waitqueue_active(&root->fs_info->tree_mod_seq_wait)) |
5299 | wake_up(&root->fs_info->tree_mod_seq_wait); | |
1887be66 CM |
5300 | |
5301 | /* | |
5302 | * we don't take a ref on the node because we're removing it from the | |
5303 | * tree, so we just steal the ref the tree was holding. | |
5304 | */ | |
c3e69d58 CM |
5305 | delayed_refs->num_heads--; |
5306 | if (list_empty(&head->cluster)) | |
5307 | delayed_refs->num_heads_ready--; | |
5308 | ||
5309 | list_del_init(&head->cluster); | |
1887be66 CM |
5310 | spin_unlock(&delayed_refs->lock); |
5311 | ||
f0486c68 YZ |
5312 | BUG_ON(head->extent_op); |
5313 | if (head->must_insert_reserved) | |
5314 | ret = 1; | |
5315 | ||
5316 | mutex_unlock(&head->mutex); | |
1887be66 | 5317 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 5318 | return ret; |
1887be66 CM |
5319 | out: |
5320 | spin_unlock(&delayed_refs->lock); | |
5321 | return 0; | |
5322 | } | |
5323 | ||
f0486c68 YZ |
5324 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
5325 | struct btrfs_root *root, | |
5326 | struct extent_buffer *buf, | |
5581a51a | 5327 | u64 parent, int last_ref) |
f0486c68 | 5328 | { |
f0486c68 YZ |
5329 | struct btrfs_block_group_cache *cache = NULL; |
5330 | int ret; | |
5331 | ||
5332 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
66d7e7f0 AJ |
5333 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
5334 | buf->start, buf->len, | |
5335 | parent, root->root_key.objectid, | |
5336 | btrfs_header_level(buf), | |
5581a51a | 5337 | BTRFS_DROP_DELAYED_REF, NULL, 0); |
79787eaa | 5338 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
5339 | } |
5340 | ||
5341 | if (!last_ref) | |
5342 | return; | |
5343 | ||
f0486c68 | 5344 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); |
f0486c68 YZ |
5345 | |
5346 | if (btrfs_header_generation(buf) == trans->transid) { | |
5347 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
5348 | ret = check_ref_cleanup(trans, root, buf->start); | |
5349 | if (!ret) | |
37be25bc | 5350 | goto out; |
f0486c68 YZ |
5351 | } |
5352 | ||
5353 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
5354 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
37be25bc | 5355 | goto out; |
f0486c68 YZ |
5356 | } |
5357 | ||
5358 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
5359 | ||
5360 | btrfs_add_free_space(cache, buf->start, buf->len); | |
fb25e914 | 5361 | btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE); |
f0486c68 YZ |
5362 | } |
5363 | out: | |
a826d6dc JB |
5364 | /* |
5365 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
5366 | * anymore. | |
5367 | */ | |
5368 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
5369 | btrfs_put_block_group(cache); |
5370 | } | |
5371 | ||
79787eaa | 5372 | /* Can return -ENOMEM */ |
66d7e7f0 AJ |
5373 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
5374 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, | |
5375 | u64 owner, u64 offset, int for_cow) | |
925baedd CM |
5376 | { |
5377 | int ret; | |
66d7e7f0 | 5378 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 5379 | |
56bec294 CM |
5380 | /* |
5381 | * tree log blocks never actually go into the extent allocation | |
5382 | * tree, just update pinning info and exit early. | |
56bec294 | 5383 | */ |
5d4f98a2 YZ |
5384 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
5385 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 5386 | /* unlocks the pinned mutex */ |
11833d66 | 5387 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 5388 | ret = 0; |
5d4f98a2 | 5389 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
5390 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
5391 | num_bytes, | |
5d4f98a2 | 5392 | parent, root_objectid, (int)owner, |
66d7e7f0 | 5393 | BTRFS_DROP_DELAYED_REF, NULL, for_cow); |
5d4f98a2 | 5394 | } else { |
66d7e7f0 AJ |
5395 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
5396 | num_bytes, | |
5397 | parent, root_objectid, owner, | |
5398 | offset, BTRFS_DROP_DELAYED_REF, | |
5399 | NULL, for_cow); | |
56bec294 | 5400 | } |
925baedd CM |
5401 | return ret; |
5402 | } | |
5403 | ||
87ee04eb CM |
5404 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
5405 | { | |
5406 | u64 mask = ((u64)root->stripesize - 1); | |
5407 | u64 ret = (val + mask) & ~mask; | |
5408 | return ret; | |
5409 | } | |
5410 | ||
817d52f8 JB |
5411 | /* |
5412 | * when we wait for progress in the block group caching, its because | |
5413 | * our allocation attempt failed at least once. So, we must sleep | |
5414 | * and let some progress happen before we try again. | |
5415 | * | |
5416 | * This function will sleep at least once waiting for new free space to | |
5417 | * show up, and then it will check the block group free space numbers | |
5418 | * for our min num_bytes. Another option is to have it go ahead | |
5419 | * and look in the rbtree for a free extent of a given size, but this | |
5420 | * is a good start. | |
5421 | */ | |
5422 | static noinline int | |
5423 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
5424 | u64 num_bytes) | |
5425 | { | |
11833d66 | 5426 | struct btrfs_caching_control *caching_ctl; |
817d52f8 JB |
5427 | DEFINE_WAIT(wait); |
5428 | ||
11833d66 YZ |
5429 | caching_ctl = get_caching_control(cache); |
5430 | if (!caching_ctl) | |
817d52f8 | 5431 | return 0; |
817d52f8 | 5432 | |
11833d66 | 5433 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 5434 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
5435 | |
5436 | put_caching_control(caching_ctl); | |
5437 | return 0; | |
5438 | } | |
5439 | ||
5440 | static noinline int | |
5441 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
5442 | { | |
5443 | struct btrfs_caching_control *caching_ctl; | |
5444 | DEFINE_WAIT(wait); | |
5445 | ||
5446 | caching_ctl = get_caching_control(cache); | |
5447 | if (!caching_ctl) | |
5448 | return 0; | |
5449 | ||
5450 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
5451 | ||
5452 | put_caching_control(caching_ctl); | |
817d52f8 JB |
5453 | return 0; |
5454 | } | |
5455 | ||
7738a53a | 5456 | static int __get_block_group_index(u64 flags) |
b742bb82 YZ |
5457 | { |
5458 | int index; | |
7738a53a ID |
5459 | |
5460 | if (flags & BTRFS_BLOCK_GROUP_RAID10) | |
b742bb82 | 5461 | index = 0; |
7738a53a | 5462 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
b742bb82 | 5463 | index = 1; |
7738a53a | 5464 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
b742bb82 | 5465 | index = 2; |
7738a53a | 5466 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
b742bb82 YZ |
5467 | index = 3; |
5468 | else | |
5469 | index = 4; | |
7738a53a | 5470 | |
b742bb82 YZ |
5471 | return index; |
5472 | } | |
5473 | ||
7738a53a ID |
5474 | static int get_block_group_index(struct btrfs_block_group_cache *cache) |
5475 | { | |
5476 | return __get_block_group_index(cache->flags); | |
5477 | } | |
5478 | ||
817d52f8 | 5479 | enum btrfs_loop_type { |
285ff5af JB |
5480 | LOOP_CACHING_NOWAIT = 0, |
5481 | LOOP_CACHING_WAIT = 1, | |
5482 | LOOP_ALLOC_CHUNK = 2, | |
5483 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
5484 | }; |
5485 | ||
fec577fb CM |
5486 | /* |
5487 | * walks the btree of allocated extents and find a hole of a given size. | |
5488 | * The key ins is changed to record the hole: | |
5489 | * ins->objectid == block start | |
62e2749e | 5490 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
5491 | * ins->offset == number of blocks |
5492 | * Any available blocks before search_start are skipped. | |
5493 | */ | |
d397712b | 5494 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
5495 | struct btrfs_root *orig_root, |
5496 | u64 num_bytes, u64 empty_size, | |
98ed5174 | 5497 | u64 hint_byte, struct btrfs_key *ins, |
e0f54067 | 5498 | u64 data) |
fec577fb | 5499 | { |
80eb234a | 5500 | int ret = 0; |
d397712b | 5501 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 5502 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 5503 | struct btrfs_block_group_cache *block_group = NULL; |
274bd4fb | 5504 | struct btrfs_block_group_cache *used_block_group; |
81c9ad23 | 5505 | u64 search_start = 0; |
239b14b3 | 5506 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 5507 | int allowed_chunk_alloc = 0; |
ccf0e725 | 5508 | int done_chunk_alloc = 0; |
80eb234a | 5509 | struct btrfs_space_info *space_info; |
fa9c0d79 | 5510 | int loop = 0; |
f0486c68 | 5511 | int index = 0; |
fb25e914 JB |
5512 | int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ? |
5513 | RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC; | |
817d52f8 | 5514 | bool found_uncached_bg = false; |
0a24325e | 5515 | bool failed_cluster_refill = false; |
1cdda9b8 | 5516 | bool failed_alloc = false; |
67377734 | 5517 | bool use_cluster = true; |
60d2adbb | 5518 | bool have_caching_bg = false; |
fec577fb | 5519 | |
db94535d | 5520 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 5521 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
5522 | ins->objectid = 0; |
5523 | ins->offset = 0; | |
b1a4d965 | 5524 | |
3f7de037 JB |
5525 | trace_find_free_extent(orig_root, num_bytes, empty_size, data); |
5526 | ||
2552d17e | 5527 | space_info = __find_space_info(root->fs_info, data); |
1b1d1f66 | 5528 | if (!space_info) { |
e0f54067 | 5529 | printk(KERN_ERR "No space info for %llu\n", data); |
1b1d1f66 JB |
5530 | return -ENOSPC; |
5531 | } | |
2552d17e | 5532 | |
67377734 JB |
5533 | /* |
5534 | * If the space info is for both data and metadata it means we have a | |
5535 | * small filesystem and we can't use the clustering stuff. | |
5536 | */ | |
5537 | if (btrfs_mixed_space_info(space_info)) | |
5538 | use_cluster = false; | |
5539 | ||
0ef3e66b CM |
5540 | if (orig_root->ref_cows || empty_size) |
5541 | allowed_chunk_alloc = 1; | |
5542 | ||
67377734 | 5543 | if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) { |
fa9c0d79 | 5544 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
5545 | if (!btrfs_test_opt(root, SSD)) |
5546 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
5547 | } |
5548 | ||
67377734 JB |
5549 | if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster && |
5550 | btrfs_test_opt(root, SSD)) { | |
fa9c0d79 CM |
5551 | last_ptr = &root->fs_info->data_alloc_cluster; |
5552 | } | |
0f9dd46c | 5553 | |
239b14b3 | 5554 | if (last_ptr) { |
fa9c0d79 CM |
5555 | spin_lock(&last_ptr->lock); |
5556 | if (last_ptr->block_group) | |
5557 | hint_byte = last_ptr->window_start; | |
5558 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 5559 | } |
fa9c0d79 | 5560 | |
a061fc8d | 5561 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 5562 | search_start = max(search_start, hint_byte); |
0b86a832 | 5563 | |
817d52f8 | 5564 | if (!last_ptr) |
fa9c0d79 | 5565 | empty_cluster = 0; |
fa9c0d79 | 5566 | |
2552d17e | 5567 | if (search_start == hint_byte) { |
2552d17e JB |
5568 | block_group = btrfs_lookup_block_group(root->fs_info, |
5569 | search_start); | |
274bd4fb | 5570 | used_block_group = block_group; |
817d52f8 JB |
5571 | /* |
5572 | * we don't want to use the block group if it doesn't match our | |
5573 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
5574 | * |
5575 | * However if we are re-searching with an ideal block group | |
5576 | * picked out then we don't care that the block group is cached. | |
817d52f8 JB |
5577 | */ |
5578 | if (block_group && block_group_bits(block_group, data) && | |
285ff5af | 5579 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 5580 | down_read(&space_info->groups_sem); |
44fb5511 CM |
5581 | if (list_empty(&block_group->list) || |
5582 | block_group->ro) { | |
5583 | /* | |
5584 | * someone is removing this block group, | |
5585 | * we can't jump into the have_block_group | |
5586 | * target because our list pointers are not | |
5587 | * valid | |
5588 | */ | |
5589 | btrfs_put_block_group(block_group); | |
5590 | up_read(&space_info->groups_sem); | |
ccf0e725 | 5591 | } else { |
b742bb82 | 5592 | index = get_block_group_index(block_group); |
44fb5511 | 5593 | goto have_block_group; |
ccf0e725 | 5594 | } |
2552d17e | 5595 | } else if (block_group) { |
fa9c0d79 | 5596 | btrfs_put_block_group(block_group); |
2552d17e | 5597 | } |
42e70e7a | 5598 | } |
2552d17e | 5599 | search: |
60d2adbb | 5600 | have_caching_bg = false; |
80eb234a | 5601 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
5602 | list_for_each_entry(block_group, &space_info->block_groups[index], |
5603 | list) { | |
6226cb0a | 5604 | u64 offset; |
817d52f8 | 5605 | int cached; |
8a1413a2 | 5606 | |
274bd4fb | 5607 | used_block_group = block_group; |
11dfe35a | 5608 | btrfs_get_block_group(block_group); |
2552d17e | 5609 | search_start = block_group->key.objectid; |
42e70e7a | 5610 | |
83a50de9 CM |
5611 | /* |
5612 | * this can happen if we end up cycling through all the | |
5613 | * raid types, but we want to make sure we only allocate | |
5614 | * for the proper type. | |
5615 | */ | |
5616 | if (!block_group_bits(block_group, data)) { | |
5617 | u64 extra = BTRFS_BLOCK_GROUP_DUP | | |
5618 | BTRFS_BLOCK_GROUP_RAID1 | | |
5619 | BTRFS_BLOCK_GROUP_RAID10; | |
5620 | ||
5621 | /* | |
5622 | * if they asked for extra copies and this block group | |
5623 | * doesn't provide them, bail. This does allow us to | |
5624 | * fill raid0 from raid1. | |
5625 | */ | |
5626 | if ((data & extra) && !(block_group->flags & extra)) | |
5627 | goto loop; | |
5628 | } | |
5629 | ||
2552d17e | 5630 | have_block_group: |
291c7d2f JB |
5631 | cached = block_group_cache_done(block_group); |
5632 | if (unlikely(!cached)) { | |
291c7d2f | 5633 | found_uncached_bg = true; |
b8399dee | 5634 | ret = cache_block_group(block_group, trans, |
285ff5af | 5635 | orig_root, 0); |
1d4284bd CM |
5636 | BUG_ON(ret < 0); |
5637 | ret = 0; | |
817d52f8 JB |
5638 | } |
5639 | ||
ea6a478e | 5640 | if (unlikely(block_group->ro)) |
2552d17e | 5641 | goto loop; |
0f9dd46c | 5642 | |
0a24325e | 5643 | /* |
062c05c4 AO |
5644 | * Ok we want to try and use the cluster allocator, so |
5645 | * lets look there | |
0a24325e | 5646 | */ |
062c05c4 | 5647 | if (last_ptr) { |
fa9c0d79 CM |
5648 | /* |
5649 | * the refill lock keeps out other | |
5650 | * people trying to start a new cluster | |
5651 | */ | |
5652 | spin_lock(&last_ptr->refill_lock); | |
274bd4fb AO |
5653 | used_block_group = last_ptr->block_group; |
5654 | if (used_block_group != block_group && | |
5655 | (!used_block_group || | |
5656 | used_block_group->ro || | |
5657 | !block_group_bits(used_block_group, data))) { | |
5658 | used_block_group = block_group; | |
44fb5511 | 5659 | goto refill_cluster; |
274bd4fb AO |
5660 | } |
5661 | ||
5662 | if (used_block_group != block_group) | |
5663 | btrfs_get_block_group(used_block_group); | |
44fb5511 | 5664 | |
274bd4fb AO |
5665 | offset = btrfs_alloc_from_cluster(used_block_group, |
5666 | last_ptr, num_bytes, used_block_group->key.objectid); | |
fa9c0d79 CM |
5667 | if (offset) { |
5668 | /* we have a block, we're done */ | |
5669 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
5670 | trace_btrfs_reserve_extent_cluster(root, |
5671 | block_group, search_start, num_bytes); | |
fa9c0d79 CM |
5672 | goto checks; |
5673 | } | |
5674 | ||
274bd4fb AO |
5675 | WARN_ON(last_ptr->block_group != used_block_group); |
5676 | if (used_block_group != block_group) { | |
5677 | btrfs_put_block_group(used_block_group); | |
5678 | used_block_group = block_group; | |
fa9c0d79 | 5679 | } |
44fb5511 | 5680 | refill_cluster: |
274bd4fb | 5681 | BUG_ON(used_block_group != block_group); |
062c05c4 AO |
5682 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
5683 | * set up a new clusters, so lets just skip it | |
5684 | * and let the allocator find whatever block | |
5685 | * it can find. If we reach this point, we | |
5686 | * will have tried the cluster allocator | |
5687 | * plenty of times and not have found | |
5688 | * anything, so we are likely way too | |
5689 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
5690 | * anything. |
5691 | * | |
5692 | * However, if the cluster is taken from the | |
5693 | * current block group, release the cluster | |
5694 | * first, so that we stand a better chance of | |
5695 | * succeeding in the unclustered | |
5696 | * allocation. */ | |
5697 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
5698 | last_ptr->block_group != block_group) { | |
062c05c4 AO |
5699 | spin_unlock(&last_ptr->refill_lock); |
5700 | goto unclustered_alloc; | |
5701 | } | |
5702 | ||
fa9c0d79 CM |
5703 | /* |
5704 | * this cluster didn't work out, free it and | |
5705 | * start over | |
5706 | */ | |
5707 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
5708 | ||
a5f6f719 AO |
5709 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
5710 | spin_unlock(&last_ptr->refill_lock); | |
5711 | goto unclustered_alloc; | |
5712 | } | |
5713 | ||
fa9c0d79 | 5714 | /* allocate a cluster in this block group */ |
451d7585 | 5715 | ret = btrfs_find_space_cluster(trans, root, |
fa9c0d79 | 5716 | block_group, last_ptr, |
1b22bad7 | 5717 | search_start, num_bytes, |
fa9c0d79 CM |
5718 | empty_cluster + empty_size); |
5719 | if (ret == 0) { | |
5720 | /* | |
5721 | * now pull our allocation out of this | |
5722 | * cluster | |
5723 | */ | |
5724 | offset = btrfs_alloc_from_cluster(block_group, | |
5725 | last_ptr, num_bytes, | |
5726 | search_start); | |
5727 | if (offset) { | |
5728 | /* we found one, proceed */ | |
5729 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
5730 | trace_btrfs_reserve_extent_cluster(root, |
5731 | block_group, search_start, | |
5732 | num_bytes); | |
fa9c0d79 CM |
5733 | goto checks; |
5734 | } | |
0a24325e JB |
5735 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
5736 | && !failed_cluster_refill) { | |
817d52f8 JB |
5737 | spin_unlock(&last_ptr->refill_lock); |
5738 | ||
0a24325e | 5739 | failed_cluster_refill = true; |
817d52f8 JB |
5740 | wait_block_group_cache_progress(block_group, |
5741 | num_bytes + empty_cluster + empty_size); | |
5742 | goto have_block_group; | |
fa9c0d79 | 5743 | } |
817d52f8 | 5744 | |
fa9c0d79 CM |
5745 | /* |
5746 | * at this point we either didn't find a cluster | |
5747 | * or we weren't able to allocate a block from our | |
5748 | * cluster. Free the cluster we've been trying | |
5749 | * to use, and go to the next block group | |
5750 | */ | |
0a24325e | 5751 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 5752 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 5753 | goto loop; |
fa9c0d79 CM |
5754 | } |
5755 | ||
062c05c4 | 5756 | unclustered_alloc: |
a5f6f719 AO |
5757 | spin_lock(&block_group->free_space_ctl->tree_lock); |
5758 | if (cached && | |
5759 | block_group->free_space_ctl->free_space < | |
5760 | num_bytes + empty_cluster + empty_size) { | |
5761 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
5762 | goto loop; | |
5763 | } | |
5764 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
5765 | ||
6226cb0a JB |
5766 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
5767 | num_bytes, empty_size); | |
1cdda9b8 JB |
5768 | /* |
5769 | * If we didn't find a chunk, and we haven't failed on this | |
5770 | * block group before, and this block group is in the middle of | |
5771 | * caching and we are ok with waiting, then go ahead and wait | |
5772 | * for progress to be made, and set failed_alloc to true. | |
5773 | * | |
5774 | * If failed_alloc is true then we've already waited on this | |
5775 | * block group once and should move on to the next block group. | |
5776 | */ | |
5777 | if (!offset && !failed_alloc && !cached && | |
5778 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 5779 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
5780 | num_bytes + empty_size); |
5781 | failed_alloc = true; | |
817d52f8 | 5782 | goto have_block_group; |
1cdda9b8 | 5783 | } else if (!offset) { |
60d2adbb MX |
5784 | if (!cached) |
5785 | have_caching_bg = true; | |
1cdda9b8 | 5786 | goto loop; |
817d52f8 | 5787 | } |
fa9c0d79 | 5788 | checks: |
6226cb0a | 5789 | search_start = stripe_align(root, offset); |
25179201 | 5790 | |
2552d17e JB |
5791 | /* move on to the next group */ |
5792 | if (search_start + num_bytes > | |
274bd4fb AO |
5793 | used_block_group->key.objectid + used_block_group->key.offset) { |
5794 | btrfs_add_free_space(used_block_group, offset, num_bytes); | |
2552d17e | 5795 | goto loop; |
6226cb0a | 5796 | } |
f5a31e16 | 5797 | |
f0486c68 | 5798 | if (offset < search_start) |
274bd4fb | 5799 | btrfs_add_free_space(used_block_group, offset, |
f0486c68 YZ |
5800 | search_start - offset); |
5801 | BUG_ON(offset > search_start); | |
2552d17e | 5802 | |
274bd4fb | 5803 | ret = btrfs_update_reserved_bytes(used_block_group, num_bytes, |
fb25e914 | 5804 | alloc_type); |
f0486c68 | 5805 | if (ret == -EAGAIN) { |
274bd4fb | 5806 | btrfs_add_free_space(used_block_group, offset, num_bytes); |
2552d17e | 5807 | goto loop; |
0f9dd46c | 5808 | } |
0b86a832 | 5809 | |
f0486c68 | 5810 | /* we are all good, lets return */ |
2552d17e JB |
5811 | ins->objectid = search_start; |
5812 | ins->offset = num_bytes; | |
d2fb3437 | 5813 | |
3f7de037 JB |
5814 | trace_btrfs_reserve_extent(orig_root, block_group, |
5815 | search_start, num_bytes); | |
6226cb0a | 5816 | if (offset < search_start) |
274bd4fb | 5817 | btrfs_add_free_space(used_block_group, offset, |
6226cb0a JB |
5818 | search_start - offset); |
5819 | BUG_ON(offset > search_start); | |
274bd4fb AO |
5820 | if (used_block_group != block_group) |
5821 | btrfs_put_block_group(used_block_group); | |
d82a6f1d | 5822 | btrfs_put_block_group(block_group); |
2552d17e JB |
5823 | break; |
5824 | loop: | |
0a24325e | 5825 | failed_cluster_refill = false; |
1cdda9b8 | 5826 | failed_alloc = false; |
b742bb82 | 5827 | BUG_ON(index != get_block_group_index(block_group)); |
274bd4fb AO |
5828 | if (used_block_group != block_group) |
5829 | btrfs_put_block_group(used_block_group); | |
fa9c0d79 | 5830 | btrfs_put_block_group(block_group); |
2552d17e JB |
5831 | } |
5832 | up_read(&space_info->groups_sem); | |
5833 | ||
60d2adbb MX |
5834 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
5835 | goto search; | |
5836 | ||
b742bb82 YZ |
5837 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
5838 | goto search; | |
5839 | ||
285ff5af | 5840 | /* |
ccf0e725 JB |
5841 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
5842 | * caching kthreads as we move along | |
817d52f8 JB |
5843 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
5844 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
5845 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
5846 | * again | |
fa9c0d79 | 5847 | */ |
723bda20 | 5848 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 5849 | index = 0; |
723bda20 | 5850 | loop++; |
817d52f8 | 5851 | if (loop == LOOP_ALLOC_CHUNK) { |
723bda20 JB |
5852 | if (allowed_chunk_alloc) { |
5853 | ret = do_chunk_alloc(trans, root, num_bytes + | |
5854 | 2 * 1024 * 1024, data, | |
5855 | CHUNK_ALLOC_LIMITED); | |
06789384 LB |
5856 | /* |
5857 | * Do not bail out on ENOSPC since we | |
5858 | * can do more things. | |
5859 | */ | |
5860 | if (ret < 0 && ret != -ENOSPC) { | |
79787eaa JM |
5861 | btrfs_abort_transaction(trans, |
5862 | root, ret); | |
5863 | goto out; | |
5864 | } | |
723bda20 JB |
5865 | allowed_chunk_alloc = 0; |
5866 | if (ret == 1) | |
5867 | done_chunk_alloc = 1; | |
5868 | } else if (!done_chunk_alloc && | |
5869 | space_info->force_alloc == | |
5870 | CHUNK_ALLOC_NO_FORCE) { | |
5871 | space_info->force_alloc = CHUNK_ALLOC_LIMITED; | |
5872 | } | |
2552d17e | 5873 | |
723bda20 JB |
5874 | /* |
5875 | * We didn't allocate a chunk, go ahead and drop the | |
5876 | * empty size and loop again. | |
5877 | */ | |
5878 | if (!done_chunk_alloc) | |
5879 | loop = LOOP_NO_EMPTY_SIZE; | |
2552d17e JB |
5880 | } |
5881 | ||
723bda20 JB |
5882 | if (loop == LOOP_NO_EMPTY_SIZE) { |
5883 | empty_size = 0; | |
5884 | empty_cluster = 0; | |
fa9c0d79 | 5885 | } |
723bda20 JB |
5886 | |
5887 | goto search; | |
2552d17e JB |
5888 | } else if (!ins->objectid) { |
5889 | ret = -ENOSPC; | |
d82a6f1d | 5890 | } else if (ins->objectid) { |
80eb234a | 5891 | ret = 0; |
be744175 | 5892 | } |
79787eaa | 5893 | out: |
be744175 | 5894 | |
0f70abe2 | 5895 | return ret; |
fec577fb | 5896 | } |
ec44a35c | 5897 | |
9ed74f2d JB |
5898 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
5899 | int dump_block_groups) | |
0f9dd46c JB |
5900 | { |
5901 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 5902 | int index = 0; |
0f9dd46c | 5903 | |
9ed74f2d | 5904 | spin_lock(&info->lock); |
fb25e914 JB |
5905 | printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n", |
5906 | (unsigned long long)info->flags, | |
d397712b | 5907 | (unsigned long long)(info->total_bytes - info->bytes_used - |
9ed74f2d | 5908 | info->bytes_pinned - info->bytes_reserved - |
8929ecfa | 5909 | info->bytes_readonly), |
d397712b | 5910 | (info->full) ? "" : "not "); |
8929ecfa YZ |
5911 | printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, " |
5912 | "reserved=%llu, may_use=%llu, readonly=%llu\n", | |
21380931 | 5913 | (unsigned long long)info->total_bytes, |
8929ecfa | 5914 | (unsigned long long)info->bytes_used, |
21380931 | 5915 | (unsigned long long)info->bytes_pinned, |
8929ecfa | 5916 | (unsigned long long)info->bytes_reserved, |
21380931 | 5917 | (unsigned long long)info->bytes_may_use, |
8929ecfa | 5918 | (unsigned long long)info->bytes_readonly); |
9ed74f2d JB |
5919 | spin_unlock(&info->lock); |
5920 | ||
5921 | if (!dump_block_groups) | |
5922 | return; | |
0f9dd46c | 5923 | |
80eb234a | 5924 | down_read(&info->groups_sem); |
b742bb82 YZ |
5925 | again: |
5926 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 5927 | spin_lock(&cache->lock); |
799ffc3c | 5928 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n", |
d397712b CM |
5929 | (unsigned long long)cache->key.objectid, |
5930 | (unsigned long long)cache->key.offset, | |
5931 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
5932 | (unsigned long long)cache->pinned, | |
799ffc3c LB |
5933 | (unsigned long long)cache->reserved, |
5934 | cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
5935 | btrfs_dump_free_space(cache, bytes); |
5936 | spin_unlock(&cache->lock); | |
5937 | } | |
b742bb82 YZ |
5938 | if (++index < BTRFS_NR_RAID_TYPES) |
5939 | goto again; | |
80eb234a | 5940 | up_read(&info->groups_sem); |
0f9dd46c | 5941 | } |
e8569813 | 5942 | |
11833d66 YZ |
5943 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
5944 | struct btrfs_root *root, | |
5945 | u64 num_bytes, u64 min_alloc_size, | |
5946 | u64 empty_size, u64 hint_byte, | |
81c9ad23 | 5947 | struct btrfs_key *ins, u64 data) |
fec577fb | 5948 | { |
9e622d6b | 5949 | bool final_tried = false; |
fec577fb | 5950 | int ret; |
925baedd | 5951 | |
6a63209f | 5952 | data = btrfs_get_alloc_profile(root, data); |
98d20f67 | 5953 | again: |
0ef3e66b CM |
5954 | /* |
5955 | * the only place that sets empty_size is btrfs_realloc_node, which | |
5956 | * is not called recursively on allocations | |
5957 | */ | |
79787eaa | 5958 | if (empty_size || root->ref_cows) { |
6324fbf3 | 5959 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
5960 | num_bytes + 2 * 1024 * 1024, data, |
5961 | CHUNK_ALLOC_NO_FORCE); | |
79787eaa JM |
5962 | if (ret < 0 && ret != -ENOSPC) { |
5963 | btrfs_abort_transaction(trans, root, ret); | |
5964 | return ret; | |
5965 | } | |
5966 | } | |
0b86a832 | 5967 | |
db94535d CM |
5968 | WARN_ON(num_bytes < root->sectorsize); |
5969 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
81c9ad23 | 5970 | hint_byte, ins, data); |
3b951516 | 5971 | |
9e622d6b MX |
5972 | if (ret == -ENOSPC) { |
5973 | if (!final_tried) { | |
5974 | num_bytes = num_bytes >> 1; | |
5975 | num_bytes = num_bytes & ~(root->sectorsize - 1); | |
5976 | num_bytes = max(num_bytes, min_alloc_size); | |
79787eaa | 5977 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
9e622d6b | 5978 | num_bytes, data, CHUNK_ALLOC_FORCE); |
79787eaa JM |
5979 | if (ret < 0 && ret != -ENOSPC) { |
5980 | btrfs_abort_transaction(trans, root, ret); | |
5981 | return ret; | |
5982 | } | |
9e622d6b MX |
5983 | if (num_bytes == min_alloc_size) |
5984 | final_tried = true; | |
5985 | goto again; | |
5986 | } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
5987 | struct btrfs_space_info *sinfo; | |
5988 | ||
5989 | sinfo = __find_space_info(root->fs_info, data); | |
5990 | printk(KERN_ERR "btrfs allocation failed flags %llu, " | |
5991 | "wanted %llu\n", (unsigned long long)data, | |
5992 | (unsigned long long)num_bytes); | |
53804280 JM |
5993 | if (sinfo) |
5994 | dump_space_info(sinfo, num_bytes, 1); | |
9e622d6b | 5995 | } |
925baedd | 5996 | } |
0f9dd46c | 5997 | |
1abe9b8a | 5998 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
5999 | ||
0f9dd46c | 6000 | return ret; |
e6dcd2dc CM |
6001 | } |
6002 | ||
e688b725 CM |
6003 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, |
6004 | u64 start, u64 len, int pin) | |
65b51a00 | 6005 | { |
0f9dd46c | 6006 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 6007 | int ret = 0; |
0f9dd46c | 6008 | |
0f9dd46c JB |
6009 | cache = btrfs_lookup_block_group(root->fs_info, start); |
6010 | if (!cache) { | |
d397712b CM |
6011 | printk(KERN_ERR "Unable to find block group for %llu\n", |
6012 | (unsigned long long)start); | |
0f9dd46c JB |
6013 | return -ENOSPC; |
6014 | } | |
1f3c79a2 | 6015 | |
5378e607 LD |
6016 | if (btrfs_test_opt(root, DISCARD)) |
6017 | ret = btrfs_discard_extent(root, start, len, NULL); | |
1f3c79a2 | 6018 | |
e688b725 CM |
6019 | if (pin) |
6020 | pin_down_extent(root, cache, start, len, 1); | |
6021 | else { | |
6022 | btrfs_add_free_space(cache, start, len); | |
6023 | btrfs_update_reserved_bytes(cache, len, RESERVE_FREE); | |
6024 | } | |
fa9c0d79 | 6025 | btrfs_put_block_group(cache); |
817d52f8 | 6026 | |
1abe9b8a | 6027 | trace_btrfs_reserved_extent_free(root, start, len); |
6028 | ||
e6dcd2dc CM |
6029 | return ret; |
6030 | } | |
6031 | ||
e688b725 CM |
6032 | int btrfs_free_reserved_extent(struct btrfs_root *root, |
6033 | u64 start, u64 len) | |
6034 | { | |
6035 | return __btrfs_free_reserved_extent(root, start, len, 0); | |
6036 | } | |
6037 | ||
6038 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, | |
6039 | u64 start, u64 len) | |
6040 | { | |
6041 | return __btrfs_free_reserved_extent(root, start, len, 1); | |
6042 | } | |
6043 | ||
5d4f98a2 YZ |
6044 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
6045 | struct btrfs_root *root, | |
6046 | u64 parent, u64 root_objectid, | |
6047 | u64 flags, u64 owner, u64 offset, | |
6048 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
6049 | { |
6050 | int ret; | |
5d4f98a2 | 6051 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 6052 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 6053 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 6054 | struct btrfs_path *path; |
5d4f98a2 YZ |
6055 | struct extent_buffer *leaf; |
6056 | int type; | |
6057 | u32 size; | |
26b8003f | 6058 | |
5d4f98a2 YZ |
6059 | if (parent > 0) |
6060 | type = BTRFS_SHARED_DATA_REF_KEY; | |
6061 | else | |
6062 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 6063 | |
5d4f98a2 | 6064 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
6065 | |
6066 | path = btrfs_alloc_path(); | |
db5b493a TI |
6067 | if (!path) |
6068 | return -ENOMEM; | |
47e4bb98 | 6069 | |
b9473439 | 6070 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6071 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
6072 | ins, size); | |
79787eaa JM |
6073 | if (ret) { |
6074 | btrfs_free_path(path); | |
6075 | return ret; | |
6076 | } | |
0f9dd46c | 6077 | |
5d4f98a2 YZ |
6078 | leaf = path->nodes[0]; |
6079 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 6080 | struct btrfs_extent_item); |
5d4f98a2 YZ |
6081 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
6082 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
6083 | btrfs_set_extent_flags(leaf, extent_item, | |
6084 | flags | BTRFS_EXTENT_FLAG_DATA); | |
6085 | ||
6086 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
6087 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
6088 | if (parent > 0) { | |
6089 | struct btrfs_shared_data_ref *ref; | |
6090 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
6091 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
6092 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
6093 | } else { | |
6094 | struct btrfs_extent_data_ref *ref; | |
6095 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
6096 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
6097 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
6098 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
6099 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
6100 | } | |
47e4bb98 CM |
6101 | |
6102 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 6103 | btrfs_free_path(path); |
f510cfec | 6104 | |
f0486c68 | 6105 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
79787eaa | 6106 | if (ret) { /* -ENOENT, logic error */ |
d397712b CM |
6107 | printk(KERN_ERR "btrfs update block group failed for %llu " |
6108 | "%llu\n", (unsigned long long)ins->objectid, | |
6109 | (unsigned long long)ins->offset); | |
f5947066 CM |
6110 | BUG(); |
6111 | } | |
e6dcd2dc CM |
6112 | return ret; |
6113 | } | |
6114 | ||
5d4f98a2 YZ |
6115 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
6116 | struct btrfs_root *root, | |
6117 | u64 parent, u64 root_objectid, | |
6118 | u64 flags, struct btrfs_disk_key *key, | |
6119 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
6120 | { |
6121 | int ret; | |
5d4f98a2 YZ |
6122 | struct btrfs_fs_info *fs_info = root->fs_info; |
6123 | struct btrfs_extent_item *extent_item; | |
6124 | struct btrfs_tree_block_info *block_info; | |
6125 | struct btrfs_extent_inline_ref *iref; | |
6126 | struct btrfs_path *path; | |
6127 | struct extent_buffer *leaf; | |
6128 | u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref); | |
1c2308f8 | 6129 | |
5d4f98a2 | 6130 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6131 | if (!path) |
6132 | return -ENOMEM; | |
56bec294 | 6133 | |
5d4f98a2 YZ |
6134 | path->leave_spinning = 1; |
6135 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
6136 | ins, size); | |
79787eaa JM |
6137 | if (ret) { |
6138 | btrfs_free_path(path); | |
6139 | return ret; | |
6140 | } | |
5d4f98a2 YZ |
6141 | |
6142 | leaf = path->nodes[0]; | |
6143 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
6144 | struct btrfs_extent_item); | |
6145 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
6146 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
6147 | btrfs_set_extent_flags(leaf, extent_item, | |
6148 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
6149 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
6150 | ||
6151 | btrfs_set_tree_block_key(leaf, block_info, key); | |
6152 | btrfs_set_tree_block_level(leaf, block_info, level); | |
6153 | ||
6154 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
6155 | if (parent > 0) { | |
6156 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
6157 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6158 | BTRFS_SHARED_BLOCK_REF_KEY); | |
6159 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
6160 | } else { | |
6161 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6162 | BTRFS_TREE_BLOCK_REF_KEY); | |
6163 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
6164 | } | |
6165 | ||
6166 | btrfs_mark_buffer_dirty(leaf); | |
6167 | btrfs_free_path(path); | |
6168 | ||
f0486c68 | 6169 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
79787eaa | 6170 | if (ret) { /* -ENOENT, logic error */ |
5d4f98a2 YZ |
6171 | printk(KERN_ERR "btrfs update block group failed for %llu " |
6172 | "%llu\n", (unsigned long long)ins->objectid, | |
6173 | (unsigned long long)ins->offset); | |
6174 | BUG(); | |
6175 | } | |
6176 | return ret; | |
6177 | } | |
6178 | ||
6179 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
6180 | struct btrfs_root *root, | |
6181 | u64 root_objectid, u64 owner, | |
6182 | u64 offset, struct btrfs_key *ins) | |
6183 | { | |
6184 | int ret; | |
6185 | ||
6186 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
6187 | ||
66d7e7f0 AJ |
6188 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, |
6189 | ins->offset, 0, | |
6190 | root_objectid, owner, offset, | |
6191 | BTRFS_ADD_DELAYED_EXTENT, NULL, 0); | |
e6dcd2dc CM |
6192 | return ret; |
6193 | } | |
e02119d5 CM |
6194 | |
6195 | /* | |
6196 | * this is used by the tree logging recovery code. It records that | |
6197 | * an extent has been allocated and makes sure to clear the free | |
6198 | * space cache bits as well | |
6199 | */ | |
5d4f98a2 YZ |
6200 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
6201 | struct btrfs_root *root, | |
6202 | u64 root_objectid, u64 owner, u64 offset, | |
6203 | struct btrfs_key *ins) | |
e02119d5 CM |
6204 | { |
6205 | int ret; | |
6206 | struct btrfs_block_group_cache *block_group; | |
11833d66 YZ |
6207 | struct btrfs_caching_control *caching_ctl; |
6208 | u64 start = ins->objectid; | |
6209 | u64 num_bytes = ins->offset; | |
e02119d5 | 6210 | |
e02119d5 | 6211 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
b8399dee | 6212 | cache_block_group(block_group, trans, NULL, 0); |
11833d66 | 6213 | caching_ctl = get_caching_control(block_group); |
e02119d5 | 6214 | |
11833d66 YZ |
6215 | if (!caching_ctl) { |
6216 | BUG_ON(!block_group_cache_done(block_group)); | |
6217 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
79787eaa | 6218 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6219 | } else { |
6220 | mutex_lock(&caching_ctl->mutex); | |
6221 | ||
6222 | if (start >= caching_ctl->progress) { | |
6223 | ret = add_excluded_extent(root, start, num_bytes); | |
79787eaa | 6224 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6225 | } else if (start + num_bytes <= caching_ctl->progress) { |
6226 | ret = btrfs_remove_free_space(block_group, | |
6227 | start, num_bytes); | |
79787eaa | 6228 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6229 | } else { |
6230 | num_bytes = caching_ctl->progress - start; | |
6231 | ret = btrfs_remove_free_space(block_group, | |
6232 | start, num_bytes); | |
79787eaa | 6233 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6234 | |
6235 | start = caching_ctl->progress; | |
6236 | num_bytes = ins->objectid + ins->offset - | |
6237 | caching_ctl->progress; | |
6238 | ret = add_excluded_extent(root, start, num_bytes); | |
79787eaa | 6239 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6240 | } |
6241 | ||
6242 | mutex_unlock(&caching_ctl->mutex); | |
6243 | put_caching_control(caching_ctl); | |
6244 | } | |
6245 | ||
fb25e914 JB |
6246 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, |
6247 | RESERVE_ALLOC_NO_ACCOUNT); | |
79787eaa | 6248 | BUG_ON(ret); /* logic error */ |
fa9c0d79 | 6249 | btrfs_put_block_group(block_group); |
5d4f98a2 YZ |
6250 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
6251 | 0, owner, offset, ins, 1); | |
e02119d5 CM |
6252 | return ret; |
6253 | } | |
6254 | ||
65b51a00 CM |
6255 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, |
6256 | struct btrfs_root *root, | |
4008c04a CM |
6257 | u64 bytenr, u32 blocksize, |
6258 | int level) | |
65b51a00 CM |
6259 | { |
6260 | struct extent_buffer *buf; | |
6261 | ||
6262 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
6263 | if (!buf) | |
6264 | return ERR_PTR(-ENOMEM); | |
6265 | btrfs_set_header_generation(buf, trans->transid); | |
85d4e461 | 6266 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 CM |
6267 | btrfs_tree_lock(buf); |
6268 | clean_tree_block(trans, root, buf); | |
3083ee2e | 6269 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
6270 | |
6271 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 6272 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 6273 | |
d0c803c4 | 6274 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
8cef4e16 YZ |
6275 | /* |
6276 | * we allow two log transactions at a time, use different | |
6277 | * EXENT bit to differentiate dirty pages. | |
6278 | */ | |
6279 | if (root->log_transid % 2 == 0) | |
6280 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
6281 | buf->start + buf->len - 1, GFP_NOFS); | |
6282 | else | |
6283 | set_extent_new(&root->dirty_log_pages, buf->start, | |
6284 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 CM |
6285 | } else { |
6286 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 6287 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 6288 | } |
65b51a00 | 6289 | trans->blocks_used++; |
b4ce94de | 6290 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
6291 | return buf; |
6292 | } | |
6293 | ||
f0486c68 YZ |
6294 | static struct btrfs_block_rsv * |
6295 | use_block_rsv(struct btrfs_trans_handle *trans, | |
6296 | struct btrfs_root *root, u32 blocksize) | |
6297 | { | |
6298 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 6299 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 YZ |
6300 | int ret; |
6301 | ||
6302 | block_rsv = get_block_rsv(trans, root); | |
6303 | ||
6304 | if (block_rsv->size == 0) { | |
36ba022a | 6305 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0); |
68a82277 JB |
6306 | /* |
6307 | * If we couldn't reserve metadata bytes try and use some from | |
6308 | * the global reserve. | |
6309 | */ | |
6310 | if (ret && block_rsv != global_rsv) { | |
6311 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
6312 | if (!ret) | |
6313 | return global_rsv; | |
f0486c68 | 6314 | return ERR_PTR(ret); |
68a82277 | 6315 | } else if (ret) { |
f0486c68 | 6316 | return ERR_PTR(ret); |
68a82277 | 6317 | } |
f0486c68 YZ |
6318 | return block_rsv; |
6319 | } | |
6320 | ||
6321 | ret = block_rsv_use_bytes(block_rsv, blocksize); | |
6322 | if (!ret) | |
6323 | return block_rsv; | |
68a82277 | 6324 | if (ret) { |
dff51cd1 DS |
6325 | static DEFINE_RATELIMIT_STATE(_rs, |
6326 | DEFAULT_RATELIMIT_INTERVAL, | |
6327 | /*DEFAULT_RATELIMIT_BURST*/ 2); | |
6328 | if (__ratelimit(&_rs)) { | |
6329 | printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret); | |
6330 | WARN_ON(1); | |
6331 | } | |
36ba022a | 6332 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0); |
68a82277 | 6333 | if (!ret) { |
68a82277 JB |
6334 | return block_rsv; |
6335 | } else if (ret && block_rsv != global_rsv) { | |
6336 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
6337 | if (!ret) | |
6338 | return global_rsv; | |
6339 | } | |
6340 | } | |
f0486c68 | 6341 | |
f0486c68 YZ |
6342 | return ERR_PTR(-ENOSPC); |
6343 | } | |
6344 | ||
8c2a3ca2 JB |
6345 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
6346 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
6347 | { |
6348 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 6349 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
6350 | } |
6351 | ||
fec577fb | 6352 | /* |
f0486c68 YZ |
6353 | * finds a free extent and does all the dirty work required for allocation |
6354 | * returns the key for the extent through ins, and a tree buffer for | |
6355 | * the first block of the extent through buf. | |
6356 | * | |
fec577fb CM |
6357 | * returns the tree buffer or NULL. |
6358 | */ | |
5f39d397 | 6359 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
6360 | struct btrfs_root *root, u32 blocksize, |
6361 | u64 parent, u64 root_objectid, | |
6362 | struct btrfs_disk_key *key, int level, | |
5581a51a | 6363 | u64 hint, u64 empty_size) |
fec577fb | 6364 | { |
e2fa7227 | 6365 | struct btrfs_key ins; |
f0486c68 | 6366 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 6367 | struct extent_buffer *buf; |
f0486c68 YZ |
6368 | u64 flags = 0; |
6369 | int ret; | |
6370 | ||
fec577fb | 6371 | |
f0486c68 YZ |
6372 | block_rsv = use_block_rsv(trans, root, blocksize); |
6373 | if (IS_ERR(block_rsv)) | |
6374 | return ERR_CAST(block_rsv); | |
6375 | ||
6376 | ret = btrfs_reserve_extent(trans, root, blocksize, blocksize, | |
81c9ad23 | 6377 | empty_size, hint, &ins, 0); |
fec577fb | 6378 | if (ret) { |
8c2a3ca2 | 6379 | unuse_block_rsv(root->fs_info, block_rsv, blocksize); |
54aa1f4d | 6380 | return ERR_PTR(ret); |
fec577fb | 6381 | } |
55c69072 | 6382 | |
4008c04a CM |
6383 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
6384 | blocksize, level); | |
79787eaa | 6385 | BUG_ON(IS_ERR(buf)); /* -ENOMEM */ |
f0486c68 YZ |
6386 | |
6387 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
6388 | if (parent == 0) | |
6389 | parent = ins.objectid; | |
6390 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6391 | } else | |
6392 | BUG_ON(parent > 0); | |
6393 | ||
6394 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
6395 | struct btrfs_delayed_extent_op *extent_op; | |
6396 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
79787eaa | 6397 | BUG_ON(!extent_op); /* -ENOMEM */ |
f0486c68 YZ |
6398 | if (key) |
6399 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
6400 | else | |
6401 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
6402 | extent_op->flags_to_set = flags; | |
6403 | extent_op->update_key = 1; | |
6404 | extent_op->update_flags = 1; | |
6405 | extent_op->is_data = 0; | |
6406 | ||
66d7e7f0 AJ |
6407 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
6408 | ins.objectid, | |
f0486c68 YZ |
6409 | ins.offset, parent, root_objectid, |
6410 | level, BTRFS_ADD_DELAYED_EXTENT, | |
5581a51a | 6411 | extent_op, 0); |
79787eaa | 6412 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 | 6413 | } |
fec577fb CM |
6414 | return buf; |
6415 | } | |
a28ec197 | 6416 | |
2c47e605 YZ |
6417 | struct walk_control { |
6418 | u64 refs[BTRFS_MAX_LEVEL]; | |
6419 | u64 flags[BTRFS_MAX_LEVEL]; | |
6420 | struct btrfs_key update_progress; | |
6421 | int stage; | |
6422 | int level; | |
6423 | int shared_level; | |
6424 | int update_ref; | |
6425 | int keep_locks; | |
1c4850e2 YZ |
6426 | int reada_slot; |
6427 | int reada_count; | |
66d7e7f0 | 6428 | int for_reloc; |
2c47e605 YZ |
6429 | }; |
6430 | ||
6431 | #define DROP_REFERENCE 1 | |
6432 | #define UPDATE_BACKREF 2 | |
6433 | ||
1c4850e2 YZ |
6434 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
6435 | struct btrfs_root *root, | |
6436 | struct walk_control *wc, | |
6437 | struct btrfs_path *path) | |
6407bf6d | 6438 | { |
1c4850e2 YZ |
6439 | u64 bytenr; |
6440 | u64 generation; | |
6441 | u64 refs; | |
94fcca9f | 6442 | u64 flags; |
5d4f98a2 | 6443 | u32 nritems; |
1c4850e2 YZ |
6444 | u32 blocksize; |
6445 | struct btrfs_key key; | |
6446 | struct extent_buffer *eb; | |
6407bf6d | 6447 | int ret; |
1c4850e2 YZ |
6448 | int slot; |
6449 | int nread = 0; | |
6407bf6d | 6450 | |
1c4850e2 YZ |
6451 | if (path->slots[wc->level] < wc->reada_slot) { |
6452 | wc->reada_count = wc->reada_count * 2 / 3; | |
6453 | wc->reada_count = max(wc->reada_count, 2); | |
6454 | } else { | |
6455 | wc->reada_count = wc->reada_count * 3 / 2; | |
6456 | wc->reada_count = min_t(int, wc->reada_count, | |
6457 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
6458 | } | |
7bb86316 | 6459 | |
1c4850e2 YZ |
6460 | eb = path->nodes[wc->level]; |
6461 | nritems = btrfs_header_nritems(eb); | |
6462 | blocksize = btrfs_level_size(root, wc->level - 1); | |
bd56b302 | 6463 | |
1c4850e2 YZ |
6464 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
6465 | if (nread >= wc->reada_count) | |
6466 | break; | |
bd56b302 | 6467 | |
2dd3e67b | 6468 | cond_resched(); |
1c4850e2 YZ |
6469 | bytenr = btrfs_node_blockptr(eb, slot); |
6470 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 6471 | |
1c4850e2 YZ |
6472 | if (slot == path->slots[wc->level]) |
6473 | goto reada; | |
5d4f98a2 | 6474 | |
1c4850e2 YZ |
6475 | if (wc->stage == UPDATE_BACKREF && |
6476 | generation <= root->root_key.offset) | |
bd56b302 CM |
6477 | continue; |
6478 | ||
94fcca9f YZ |
6479 | /* We don't lock the tree block, it's OK to be racy here */ |
6480 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, | |
6481 | &refs, &flags); | |
79787eaa JM |
6482 | /* We don't care about errors in readahead. */ |
6483 | if (ret < 0) | |
6484 | continue; | |
94fcca9f YZ |
6485 | BUG_ON(refs == 0); |
6486 | ||
1c4850e2 | 6487 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
6488 | if (refs == 1) |
6489 | goto reada; | |
bd56b302 | 6490 | |
94fcca9f YZ |
6491 | if (wc->level == 1 && |
6492 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6493 | continue; | |
1c4850e2 YZ |
6494 | if (!wc->update_ref || |
6495 | generation <= root->root_key.offset) | |
6496 | continue; | |
6497 | btrfs_node_key_to_cpu(eb, &key, slot); | |
6498 | ret = btrfs_comp_cpu_keys(&key, | |
6499 | &wc->update_progress); | |
6500 | if (ret < 0) | |
6501 | continue; | |
94fcca9f YZ |
6502 | } else { |
6503 | if (wc->level == 1 && | |
6504 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6505 | continue; | |
6407bf6d | 6506 | } |
1c4850e2 YZ |
6507 | reada: |
6508 | ret = readahead_tree_block(root, bytenr, blocksize, | |
6509 | generation); | |
6510 | if (ret) | |
bd56b302 | 6511 | break; |
1c4850e2 | 6512 | nread++; |
20524f02 | 6513 | } |
1c4850e2 | 6514 | wc->reada_slot = slot; |
20524f02 | 6515 | } |
2c47e605 | 6516 | |
f82d02d9 | 6517 | /* |
2c47e605 YZ |
6518 | * hepler to process tree block while walking down the tree. |
6519 | * | |
2c47e605 YZ |
6520 | * when wc->stage == UPDATE_BACKREF, this function updates |
6521 | * back refs for pointers in the block. | |
6522 | * | |
6523 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 6524 | */ |
2c47e605 | 6525 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 6526 | struct btrfs_root *root, |
2c47e605 | 6527 | struct btrfs_path *path, |
94fcca9f | 6528 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 6529 | { |
2c47e605 YZ |
6530 | int level = wc->level; |
6531 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 6532 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
6533 | int ret; |
6534 | ||
2c47e605 YZ |
6535 | if (wc->stage == UPDATE_BACKREF && |
6536 | btrfs_header_owner(eb) != root->root_key.objectid) | |
6537 | return 1; | |
f82d02d9 | 6538 | |
2c47e605 YZ |
6539 | /* |
6540 | * when reference count of tree block is 1, it won't increase | |
6541 | * again. once full backref flag is set, we never clear it. | |
6542 | */ | |
94fcca9f YZ |
6543 | if (lookup_info && |
6544 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
6545 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
6546 | BUG_ON(!path->locks[level]); |
6547 | ret = btrfs_lookup_extent_info(trans, root, | |
6548 | eb->start, eb->len, | |
6549 | &wc->refs[level], | |
6550 | &wc->flags[level]); | |
79787eaa JM |
6551 | BUG_ON(ret == -ENOMEM); |
6552 | if (ret) | |
6553 | return ret; | |
2c47e605 YZ |
6554 | BUG_ON(wc->refs[level] == 0); |
6555 | } | |
5d4f98a2 | 6556 | |
2c47e605 YZ |
6557 | if (wc->stage == DROP_REFERENCE) { |
6558 | if (wc->refs[level] > 1) | |
6559 | return 1; | |
f82d02d9 | 6560 | |
2c47e605 | 6561 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 6562 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6563 | path->locks[level] = 0; |
6564 | } | |
6565 | return 0; | |
6566 | } | |
f82d02d9 | 6567 | |
2c47e605 YZ |
6568 | /* wc->stage == UPDATE_BACKREF */ |
6569 | if (!(wc->flags[level] & flag)) { | |
6570 | BUG_ON(!path->locks[level]); | |
66d7e7f0 | 6571 | ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc); |
79787eaa | 6572 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 6573 | ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc); |
79787eaa | 6574 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6575 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, |
6576 | eb->len, flag, 0); | |
79787eaa | 6577 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6578 | wc->flags[level] |= flag; |
6579 | } | |
6580 | ||
6581 | /* | |
6582 | * the block is shared by multiple trees, so it's not good to | |
6583 | * keep the tree lock | |
6584 | */ | |
6585 | if (path->locks[level] && level > 0) { | |
bd681513 | 6586 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6587 | path->locks[level] = 0; |
6588 | } | |
6589 | return 0; | |
6590 | } | |
6591 | ||
1c4850e2 YZ |
6592 | /* |
6593 | * hepler to process tree block pointer. | |
6594 | * | |
6595 | * when wc->stage == DROP_REFERENCE, this function checks | |
6596 | * reference count of the block pointed to. if the block | |
6597 | * is shared and we need update back refs for the subtree | |
6598 | * rooted at the block, this function changes wc->stage to | |
6599 | * UPDATE_BACKREF. if the block is shared and there is no | |
6600 | * need to update back, this function drops the reference | |
6601 | * to the block. | |
6602 | * | |
6603 | * NOTE: return value 1 means we should stop walking down. | |
6604 | */ | |
6605 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
6606 | struct btrfs_root *root, | |
6607 | struct btrfs_path *path, | |
94fcca9f | 6608 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
6609 | { |
6610 | u64 bytenr; | |
6611 | u64 generation; | |
6612 | u64 parent; | |
6613 | u32 blocksize; | |
6614 | struct btrfs_key key; | |
6615 | struct extent_buffer *next; | |
6616 | int level = wc->level; | |
6617 | int reada = 0; | |
6618 | int ret = 0; | |
6619 | ||
6620 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
6621 | path->slots[level]); | |
6622 | /* | |
6623 | * if the lower level block was created before the snapshot | |
6624 | * was created, we know there is no need to update back refs | |
6625 | * for the subtree | |
6626 | */ | |
6627 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
6628 | generation <= root->root_key.offset) { |
6629 | *lookup_info = 1; | |
1c4850e2 | 6630 | return 1; |
94fcca9f | 6631 | } |
1c4850e2 YZ |
6632 | |
6633 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
6634 | blocksize = btrfs_level_size(root, level - 1); | |
6635 | ||
6636 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
6637 | if (!next) { | |
6638 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
90d2c51d MX |
6639 | if (!next) |
6640 | return -ENOMEM; | |
1c4850e2 YZ |
6641 | reada = 1; |
6642 | } | |
6643 | btrfs_tree_lock(next); | |
6644 | btrfs_set_lock_blocking(next); | |
6645 | ||
94fcca9f YZ |
6646 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, |
6647 | &wc->refs[level - 1], | |
6648 | &wc->flags[level - 1]); | |
79787eaa JM |
6649 | if (ret < 0) { |
6650 | btrfs_tree_unlock(next); | |
6651 | return ret; | |
6652 | } | |
6653 | ||
94fcca9f YZ |
6654 | BUG_ON(wc->refs[level - 1] == 0); |
6655 | *lookup_info = 0; | |
1c4850e2 | 6656 | |
94fcca9f | 6657 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 6658 | if (wc->refs[level - 1] > 1) { |
94fcca9f YZ |
6659 | if (level == 1 && |
6660 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6661 | goto skip; | |
6662 | ||
1c4850e2 YZ |
6663 | if (!wc->update_ref || |
6664 | generation <= root->root_key.offset) | |
6665 | goto skip; | |
6666 | ||
6667 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
6668 | path->slots[level]); | |
6669 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
6670 | if (ret < 0) | |
6671 | goto skip; | |
6672 | ||
6673 | wc->stage = UPDATE_BACKREF; | |
6674 | wc->shared_level = level - 1; | |
6675 | } | |
94fcca9f YZ |
6676 | } else { |
6677 | if (level == 1 && | |
6678 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6679 | goto skip; | |
1c4850e2 YZ |
6680 | } |
6681 | ||
b9fab919 | 6682 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
6683 | btrfs_tree_unlock(next); |
6684 | free_extent_buffer(next); | |
6685 | next = NULL; | |
94fcca9f | 6686 | *lookup_info = 1; |
1c4850e2 YZ |
6687 | } |
6688 | ||
6689 | if (!next) { | |
6690 | if (reada && level == 1) | |
6691 | reada_walk_down(trans, root, wc, path); | |
6692 | next = read_tree_block(root, bytenr, blocksize, generation); | |
97d9a8a4 TI |
6693 | if (!next) |
6694 | return -EIO; | |
1c4850e2 YZ |
6695 | btrfs_tree_lock(next); |
6696 | btrfs_set_lock_blocking(next); | |
6697 | } | |
6698 | ||
6699 | level--; | |
6700 | BUG_ON(level != btrfs_header_level(next)); | |
6701 | path->nodes[level] = next; | |
6702 | path->slots[level] = 0; | |
bd681513 | 6703 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
6704 | wc->level = level; |
6705 | if (wc->level == 1) | |
6706 | wc->reada_slot = 0; | |
6707 | return 0; | |
6708 | skip: | |
6709 | wc->refs[level - 1] = 0; | |
6710 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
6711 | if (wc->stage == DROP_REFERENCE) { |
6712 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
6713 | parent = path->nodes[level]->start; | |
6714 | } else { | |
6715 | BUG_ON(root->root_key.objectid != | |
6716 | btrfs_header_owner(path->nodes[level])); | |
6717 | parent = 0; | |
6718 | } | |
1c4850e2 | 6719 | |
94fcca9f | 6720 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
66d7e7f0 | 6721 | root->root_key.objectid, level - 1, 0, 0); |
79787eaa | 6722 | BUG_ON(ret); /* -ENOMEM */ |
1c4850e2 | 6723 | } |
1c4850e2 YZ |
6724 | btrfs_tree_unlock(next); |
6725 | free_extent_buffer(next); | |
94fcca9f | 6726 | *lookup_info = 1; |
1c4850e2 YZ |
6727 | return 1; |
6728 | } | |
6729 | ||
2c47e605 YZ |
6730 | /* |
6731 | * hepler to process tree block while walking up the tree. | |
6732 | * | |
6733 | * when wc->stage == DROP_REFERENCE, this function drops | |
6734 | * reference count on the block. | |
6735 | * | |
6736 | * when wc->stage == UPDATE_BACKREF, this function changes | |
6737 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
6738 | * to UPDATE_BACKREF previously while processing the block. | |
6739 | * | |
6740 | * NOTE: return value 1 means we should stop walking up. | |
6741 | */ | |
6742 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
6743 | struct btrfs_root *root, | |
6744 | struct btrfs_path *path, | |
6745 | struct walk_control *wc) | |
6746 | { | |
f0486c68 | 6747 | int ret; |
2c47e605 YZ |
6748 | int level = wc->level; |
6749 | struct extent_buffer *eb = path->nodes[level]; | |
6750 | u64 parent = 0; | |
6751 | ||
6752 | if (wc->stage == UPDATE_BACKREF) { | |
6753 | BUG_ON(wc->shared_level < level); | |
6754 | if (level < wc->shared_level) | |
6755 | goto out; | |
6756 | ||
2c47e605 YZ |
6757 | ret = find_next_key(path, level + 1, &wc->update_progress); |
6758 | if (ret > 0) | |
6759 | wc->update_ref = 0; | |
6760 | ||
6761 | wc->stage = DROP_REFERENCE; | |
6762 | wc->shared_level = -1; | |
6763 | path->slots[level] = 0; | |
6764 | ||
6765 | /* | |
6766 | * check reference count again if the block isn't locked. | |
6767 | * we should start walking down the tree again if reference | |
6768 | * count is one. | |
6769 | */ | |
6770 | if (!path->locks[level]) { | |
6771 | BUG_ON(level == 0); | |
6772 | btrfs_tree_lock(eb); | |
6773 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6774 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6775 | |
6776 | ret = btrfs_lookup_extent_info(trans, root, | |
6777 | eb->start, eb->len, | |
6778 | &wc->refs[level], | |
6779 | &wc->flags[level]); | |
79787eaa JM |
6780 | if (ret < 0) { |
6781 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
6782 | return ret; | |
6783 | } | |
2c47e605 YZ |
6784 | BUG_ON(wc->refs[level] == 0); |
6785 | if (wc->refs[level] == 1) { | |
bd681513 | 6786 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6787 | return 1; |
6788 | } | |
f82d02d9 | 6789 | } |
2c47e605 | 6790 | } |
f82d02d9 | 6791 | |
2c47e605 YZ |
6792 | /* wc->stage == DROP_REFERENCE */ |
6793 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 6794 | |
2c47e605 YZ |
6795 | if (wc->refs[level] == 1) { |
6796 | if (level == 0) { | |
6797 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
66d7e7f0 AJ |
6798 | ret = btrfs_dec_ref(trans, root, eb, 1, |
6799 | wc->for_reloc); | |
2c47e605 | 6800 | else |
66d7e7f0 AJ |
6801 | ret = btrfs_dec_ref(trans, root, eb, 0, |
6802 | wc->for_reloc); | |
79787eaa | 6803 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6804 | } |
6805 | /* make block locked assertion in clean_tree_block happy */ | |
6806 | if (!path->locks[level] && | |
6807 | btrfs_header_generation(eb) == trans->transid) { | |
6808 | btrfs_tree_lock(eb); | |
6809 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6810 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6811 | } |
6812 | clean_tree_block(trans, root, eb); | |
6813 | } | |
6814 | ||
6815 | if (eb == root->node) { | |
6816 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6817 | parent = eb->start; | |
6818 | else | |
6819 | BUG_ON(root->root_key.objectid != | |
6820 | btrfs_header_owner(eb)); | |
6821 | } else { | |
6822 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6823 | parent = path->nodes[level + 1]->start; | |
6824 | else | |
6825 | BUG_ON(root->root_key.objectid != | |
6826 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 6827 | } |
f82d02d9 | 6828 | |
5581a51a | 6829 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
6830 | out: |
6831 | wc->refs[level] = 0; | |
6832 | wc->flags[level] = 0; | |
f0486c68 | 6833 | return 0; |
2c47e605 YZ |
6834 | } |
6835 | ||
6836 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
6837 | struct btrfs_root *root, | |
6838 | struct btrfs_path *path, | |
6839 | struct walk_control *wc) | |
6840 | { | |
2c47e605 | 6841 | int level = wc->level; |
94fcca9f | 6842 | int lookup_info = 1; |
2c47e605 YZ |
6843 | int ret; |
6844 | ||
6845 | while (level >= 0) { | |
94fcca9f | 6846 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
6847 | if (ret > 0) |
6848 | break; | |
6849 | ||
6850 | if (level == 0) | |
6851 | break; | |
6852 | ||
7a7965f8 YZ |
6853 | if (path->slots[level] >= |
6854 | btrfs_header_nritems(path->nodes[level])) | |
6855 | break; | |
6856 | ||
94fcca9f | 6857 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
6858 | if (ret > 0) { |
6859 | path->slots[level]++; | |
6860 | continue; | |
90d2c51d MX |
6861 | } else if (ret < 0) |
6862 | return ret; | |
1c4850e2 | 6863 | level = wc->level; |
f82d02d9 | 6864 | } |
f82d02d9 YZ |
6865 | return 0; |
6866 | } | |
6867 | ||
d397712b | 6868 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 6869 | struct btrfs_root *root, |
f82d02d9 | 6870 | struct btrfs_path *path, |
2c47e605 | 6871 | struct walk_control *wc, int max_level) |
20524f02 | 6872 | { |
2c47e605 | 6873 | int level = wc->level; |
20524f02 | 6874 | int ret; |
9f3a7427 | 6875 | |
2c47e605 YZ |
6876 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
6877 | while (level < max_level && path->nodes[level]) { | |
6878 | wc->level = level; | |
6879 | if (path->slots[level] + 1 < | |
6880 | btrfs_header_nritems(path->nodes[level])) { | |
6881 | path->slots[level]++; | |
20524f02 CM |
6882 | return 0; |
6883 | } else { | |
2c47e605 YZ |
6884 | ret = walk_up_proc(trans, root, path, wc); |
6885 | if (ret > 0) | |
6886 | return 0; | |
bd56b302 | 6887 | |
2c47e605 | 6888 | if (path->locks[level]) { |
bd681513 CM |
6889 | btrfs_tree_unlock_rw(path->nodes[level], |
6890 | path->locks[level]); | |
2c47e605 | 6891 | path->locks[level] = 0; |
f82d02d9 | 6892 | } |
2c47e605 YZ |
6893 | free_extent_buffer(path->nodes[level]); |
6894 | path->nodes[level] = NULL; | |
6895 | level++; | |
20524f02 CM |
6896 | } |
6897 | } | |
6898 | return 1; | |
6899 | } | |
6900 | ||
9aca1d51 | 6901 | /* |
2c47e605 YZ |
6902 | * drop a subvolume tree. |
6903 | * | |
6904 | * this function traverses the tree freeing any blocks that only | |
6905 | * referenced by the tree. | |
6906 | * | |
6907 | * when a shared tree block is found. this function decreases its | |
6908 | * reference count by one. if update_ref is true, this function | |
6909 | * also make sure backrefs for the shared block and all lower level | |
6910 | * blocks are properly updated. | |
9aca1d51 | 6911 | */ |
2c536799 | 6912 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
6913 | struct btrfs_block_rsv *block_rsv, int update_ref, |
6914 | int for_reloc) | |
20524f02 | 6915 | { |
5caf2a00 | 6916 | struct btrfs_path *path; |
2c47e605 YZ |
6917 | struct btrfs_trans_handle *trans; |
6918 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 6919 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
6920 | struct walk_control *wc; |
6921 | struct btrfs_key key; | |
6922 | int err = 0; | |
6923 | int ret; | |
6924 | int level; | |
20524f02 | 6925 | |
5caf2a00 | 6926 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
6927 | if (!path) { |
6928 | err = -ENOMEM; | |
6929 | goto out; | |
6930 | } | |
20524f02 | 6931 | |
2c47e605 | 6932 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
6933 | if (!wc) { |
6934 | btrfs_free_path(path); | |
cb1b69f4 TI |
6935 | err = -ENOMEM; |
6936 | goto out; | |
38a1a919 | 6937 | } |
2c47e605 | 6938 | |
a22285a6 | 6939 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
6940 | if (IS_ERR(trans)) { |
6941 | err = PTR_ERR(trans); | |
6942 | goto out_free; | |
6943 | } | |
98d5dc13 | 6944 | |
3fd0a558 YZ |
6945 | if (block_rsv) |
6946 | trans->block_rsv = block_rsv; | |
2c47e605 | 6947 | |
9f3a7427 | 6948 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 6949 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
6950 | path->nodes[level] = btrfs_lock_root_node(root); |
6951 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 6952 | path->slots[level] = 0; |
bd681513 | 6953 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6954 | memset(&wc->update_progress, 0, |
6955 | sizeof(wc->update_progress)); | |
9f3a7427 | 6956 | } else { |
9f3a7427 | 6957 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
6958 | memcpy(&wc->update_progress, &key, |
6959 | sizeof(wc->update_progress)); | |
6960 | ||
6702ed49 | 6961 | level = root_item->drop_level; |
2c47e605 | 6962 | BUG_ON(level == 0); |
6702ed49 | 6963 | path->lowest_level = level; |
2c47e605 YZ |
6964 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6965 | path->lowest_level = 0; | |
6966 | if (ret < 0) { | |
6967 | err = ret; | |
79787eaa | 6968 | goto out_end_trans; |
9f3a7427 | 6969 | } |
1c4850e2 | 6970 | WARN_ON(ret > 0); |
2c47e605 | 6971 | |
7d9eb12c CM |
6972 | /* |
6973 | * unlock our path, this is safe because only this | |
6974 | * function is allowed to delete this snapshot | |
6975 | */ | |
5d4f98a2 | 6976 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
6977 | |
6978 | level = btrfs_header_level(root->node); | |
6979 | while (1) { | |
6980 | btrfs_tree_lock(path->nodes[level]); | |
6981 | btrfs_set_lock_blocking(path->nodes[level]); | |
6982 | ||
6983 | ret = btrfs_lookup_extent_info(trans, root, | |
6984 | path->nodes[level]->start, | |
6985 | path->nodes[level]->len, | |
6986 | &wc->refs[level], | |
6987 | &wc->flags[level]); | |
79787eaa JM |
6988 | if (ret < 0) { |
6989 | err = ret; | |
6990 | goto out_end_trans; | |
6991 | } | |
2c47e605 YZ |
6992 | BUG_ON(wc->refs[level] == 0); |
6993 | ||
6994 | if (level == root_item->drop_level) | |
6995 | break; | |
6996 | ||
6997 | btrfs_tree_unlock(path->nodes[level]); | |
6998 | WARN_ON(wc->refs[level] != 1); | |
6999 | level--; | |
7000 | } | |
9f3a7427 | 7001 | } |
2c47e605 YZ |
7002 | |
7003 | wc->level = level; | |
7004 | wc->shared_level = -1; | |
7005 | wc->stage = DROP_REFERENCE; | |
7006 | wc->update_ref = update_ref; | |
7007 | wc->keep_locks = 0; | |
66d7e7f0 | 7008 | wc->for_reloc = for_reloc; |
1c4850e2 | 7009 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 7010 | |
d397712b | 7011 | while (1) { |
2c47e605 YZ |
7012 | ret = walk_down_tree(trans, root, path, wc); |
7013 | if (ret < 0) { | |
7014 | err = ret; | |
20524f02 | 7015 | break; |
2c47e605 | 7016 | } |
9aca1d51 | 7017 | |
2c47e605 YZ |
7018 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
7019 | if (ret < 0) { | |
7020 | err = ret; | |
20524f02 | 7021 | break; |
2c47e605 YZ |
7022 | } |
7023 | ||
7024 | if (ret > 0) { | |
7025 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
7026 | break; |
7027 | } | |
2c47e605 YZ |
7028 | |
7029 | if (wc->stage == DROP_REFERENCE) { | |
7030 | level = wc->level; | |
7031 | btrfs_node_key(path->nodes[level], | |
7032 | &root_item->drop_progress, | |
7033 | path->slots[level]); | |
7034 | root_item->drop_level = level; | |
7035 | } | |
7036 | ||
7037 | BUG_ON(wc->level == 0); | |
3fd0a558 | 7038 | if (btrfs_should_end_transaction(trans, tree_root)) { |
2c47e605 YZ |
7039 | ret = btrfs_update_root(trans, tree_root, |
7040 | &root->root_key, | |
7041 | root_item); | |
79787eaa JM |
7042 | if (ret) { |
7043 | btrfs_abort_transaction(trans, tree_root, ret); | |
7044 | err = ret; | |
7045 | goto out_end_trans; | |
7046 | } | |
2c47e605 | 7047 | |
3fd0a558 | 7048 | btrfs_end_transaction_throttle(trans, tree_root); |
a22285a6 | 7049 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
7050 | if (IS_ERR(trans)) { |
7051 | err = PTR_ERR(trans); | |
7052 | goto out_free; | |
7053 | } | |
3fd0a558 YZ |
7054 | if (block_rsv) |
7055 | trans->block_rsv = block_rsv; | |
c3e69d58 | 7056 | } |
20524f02 | 7057 | } |
b3b4aa74 | 7058 | btrfs_release_path(path); |
79787eaa JM |
7059 | if (err) |
7060 | goto out_end_trans; | |
2c47e605 YZ |
7061 | |
7062 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
79787eaa JM |
7063 | if (ret) { |
7064 | btrfs_abort_transaction(trans, tree_root, ret); | |
7065 | goto out_end_trans; | |
7066 | } | |
2c47e605 | 7067 | |
76dda93c YZ |
7068 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
7069 | ret = btrfs_find_last_root(tree_root, root->root_key.objectid, | |
7070 | NULL, NULL); | |
79787eaa JM |
7071 | if (ret < 0) { |
7072 | btrfs_abort_transaction(trans, tree_root, ret); | |
7073 | err = ret; | |
7074 | goto out_end_trans; | |
7075 | } else if (ret > 0) { | |
84cd948c JB |
7076 | /* if we fail to delete the orphan item this time |
7077 | * around, it'll get picked up the next time. | |
7078 | * | |
7079 | * The most common failure here is just -ENOENT. | |
7080 | */ | |
7081 | btrfs_del_orphan_item(trans, tree_root, | |
7082 | root->root_key.objectid); | |
76dda93c YZ |
7083 | } |
7084 | } | |
7085 | ||
7086 | if (root->in_radix) { | |
7087 | btrfs_free_fs_root(tree_root->fs_info, root); | |
7088 | } else { | |
7089 | free_extent_buffer(root->node); | |
7090 | free_extent_buffer(root->commit_root); | |
7091 | kfree(root); | |
7092 | } | |
79787eaa | 7093 | out_end_trans: |
3fd0a558 | 7094 | btrfs_end_transaction_throttle(trans, tree_root); |
79787eaa | 7095 | out_free: |
2c47e605 | 7096 | kfree(wc); |
5caf2a00 | 7097 | btrfs_free_path(path); |
cb1b69f4 TI |
7098 | out: |
7099 | if (err) | |
7100 | btrfs_std_error(root->fs_info, err); | |
2c536799 | 7101 | return err; |
20524f02 | 7102 | } |
9078a3e1 | 7103 | |
2c47e605 YZ |
7104 | /* |
7105 | * drop subtree rooted at tree block 'node'. | |
7106 | * | |
7107 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 7108 | * only used by relocation code |
2c47e605 | 7109 | */ |
f82d02d9 YZ |
7110 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
7111 | struct btrfs_root *root, | |
7112 | struct extent_buffer *node, | |
7113 | struct extent_buffer *parent) | |
7114 | { | |
7115 | struct btrfs_path *path; | |
2c47e605 | 7116 | struct walk_control *wc; |
f82d02d9 YZ |
7117 | int level; |
7118 | int parent_level; | |
7119 | int ret = 0; | |
7120 | int wret; | |
7121 | ||
2c47e605 YZ |
7122 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
7123 | ||
f82d02d9 | 7124 | path = btrfs_alloc_path(); |
db5b493a TI |
7125 | if (!path) |
7126 | return -ENOMEM; | |
f82d02d9 | 7127 | |
2c47e605 | 7128 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
7129 | if (!wc) { |
7130 | btrfs_free_path(path); | |
7131 | return -ENOMEM; | |
7132 | } | |
2c47e605 | 7133 | |
b9447ef8 | 7134 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
7135 | parent_level = btrfs_header_level(parent); |
7136 | extent_buffer_get(parent); | |
7137 | path->nodes[parent_level] = parent; | |
7138 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
7139 | ||
b9447ef8 | 7140 | btrfs_assert_tree_locked(node); |
f82d02d9 | 7141 | level = btrfs_header_level(node); |
f82d02d9 YZ |
7142 | path->nodes[level] = node; |
7143 | path->slots[level] = 0; | |
bd681513 | 7144 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7145 | |
7146 | wc->refs[parent_level] = 1; | |
7147 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
7148 | wc->level = level; | |
7149 | wc->shared_level = -1; | |
7150 | wc->stage = DROP_REFERENCE; | |
7151 | wc->update_ref = 0; | |
7152 | wc->keep_locks = 1; | |
66d7e7f0 | 7153 | wc->for_reloc = 1; |
1c4850e2 | 7154 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
7155 | |
7156 | while (1) { | |
2c47e605 YZ |
7157 | wret = walk_down_tree(trans, root, path, wc); |
7158 | if (wret < 0) { | |
f82d02d9 | 7159 | ret = wret; |
f82d02d9 | 7160 | break; |
2c47e605 | 7161 | } |
f82d02d9 | 7162 | |
2c47e605 | 7163 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
7164 | if (wret < 0) |
7165 | ret = wret; | |
7166 | if (wret != 0) | |
7167 | break; | |
7168 | } | |
7169 | ||
2c47e605 | 7170 | kfree(wc); |
f82d02d9 YZ |
7171 | btrfs_free_path(path); |
7172 | return ret; | |
7173 | } | |
7174 | ||
ec44a35c CM |
7175 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
7176 | { | |
7177 | u64 num_devices; | |
fc67c450 | 7178 | u64 stripped; |
e4d8ec0f | 7179 | |
fc67c450 ID |
7180 | /* |
7181 | * if restripe for this chunk_type is on pick target profile and | |
7182 | * return, otherwise do the usual balance | |
7183 | */ | |
7184 | stripped = get_restripe_target(root->fs_info, flags); | |
7185 | if (stripped) | |
7186 | return extended_to_chunk(stripped); | |
e4d8ec0f | 7187 | |
cd02dca5 CM |
7188 | /* |
7189 | * we add in the count of missing devices because we want | |
7190 | * to make sure that any RAID levels on a degraded FS | |
7191 | * continue to be honored. | |
7192 | */ | |
7193 | num_devices = root->fs_info->fs_devices->rw_devices + | |
7194 | root->fs_info->fs_devices->missing_devices; | |
7195 | ||
fc67c450 ID |
7196 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
7197 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
7198 | ||
ec44a35c CM |
7199 | if (num_devices == 1) { |
7200 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7201 | stripped = flags & ~stripped; | |
7202 | ||
7203 | /* turn raid0 into single device chunks */ | |
7204 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
7205 | return stripped; | |
7206 | ||
7207 | /* turn mirroring into duplication */ | |
7208 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7209 | BTRFS_BLOCK_GROUP_RAID10)) | |
7210 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
7211 | } else { |
7212 | /* they already had raid on here, just return */ | |
ec44a35c CM |
7213 | if (flags & stripped) |
7214 | return flags; | |
7215 | ||
7216 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7217 | stripped = flags & ~stripped; | |
7218 | ||
7219 | /* switch duplicated blocks with raid1 */ | |
7220 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
7221 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
7222 | ||
e3176ca2 | 7223 | /* this is drive concat, leave it alone */ |
ec44a35c | 7224 | } |
e3176ca2 | 7225 | |
ec44a35c CM |
7226 | return flags; |
7227 | } | |
7228 | ||
199c36ea | 7229 | static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 7230 | { |
f0486c68 YZ |
7231 | struct btrfs_space_info *sinfo = cache->space_info; |
7232 | u64 num_bytes; | |
199c36ea | 7233 | u64 min_allocable_bytes; |
f0486c68 | 7234 | int ret = -ENOSPC; |
0ef3e66b | 7235 | |
c286ac48 | 7236 | |
199c36ea MX |
7237 | /* |
7238 | * We need some metadata space and system metadata space for | |
7239 | * allocating chunks in some corner cases until we force to set | |
7240 | * it to be readonly. | |
7241 | */ | |
7242 | if ((sinfo->flags & | |
7243 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
7244 | !force) | |
7245 | min_allocable_bytes = 1 * 1024 * 1024; | |
7246 | else | |
7247 | min_allocable_bytes = 0; | |
7248 | ||
f0486c68 YZ |
7249 | spin_lock(&sinfo->lock); |
7250 | spin_lock(&cache->lock); | |
61cfea9b W |
7251 | |
7252 | if (cache->ro) { | |
7253 | ret = 0; | |
7254 | goto out; | |
7255 | } | |
7256 | ||
f0486c68 YZ |
7257 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
7258 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7259 | ||
7260 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
37be25bc JB |
7261 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + |
7262 | min_allocable_bytes <= sinfo->total_bytes) { | |
f0486c68 | 7263 | sinfo->bytes_readonly += num_bytes; |
f0486c68 YZ |
7264 | cache->ro = 1; |
7265 | ret = 0; | |
7266 | } | |
61cfea9b | 7267 | out: |
f0486c68 YZ |
7268 | spin_unlock(&cache->lock); |
7269 | spin_unlock(&sinfo->lock); | |
7270 | return ret; | |
7271 | } | |
7d9eb12c | 7272 | |
f0486c68 YZ |
7273 | int btrfs_set_block_group_ro(struct btrfs_root *root, |
7274 | struct btrfs_block_group_cache *cache) | |
c286ac48 | 7275 | |
f0486c68 YZ |
7276 | { |
7277 | struct btrfs_trans_handle *trans; | |
7278 | u64 alloc_flags; | |
7279 | int ret; | |
7d9eb12c | 7280 | |
f0486c68 | 7281 | BUG_ON(cache->ro); |
0ef3e66b | 7282 | |
ff5714cc | 7283 | trans = btrfs_join_transaction(root); |
79787eaa JM |
7284 | if (IS_ERR(trans)) |
7285 | return PTR_ERR(trans); | |
5d4f98a2 | 7286 | |
f0486c68 | 7287 | alloc_flags = update_block_group_flags(root, cache->flags); |
79787eaa JM |
7288 | if (alloc_flags != cache->flags) { |
7289 | ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, | |
7290 | CHUNK_ALLOC_FORCE); | |
7291 | if (ret < 0) | |
7292 | goto out; | |
7293 | } | |
5d4f98a2 | 7294 | |
199c36ea | 7295 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7296 | if (!ret) |
7297 | goto out; | |
7298 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
0e4f8f88 CM |
7299 | ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
7300 | CHUNK_ALLOC_FORCE); | |
f0486c68 YZ |
7301 | if (ret < 0) |
7302 | goto out; | |
199c36ea | 7303 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7304 | out: |
7305 | btrfs_end_transaction(trans, root); | |
7306 | return ret; | |
7307 | } | |
5d4f98a2 | 7308 | |
c87f08ca CM |
7309 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
7310 | struct btrfs_root *root, u64 type) | |
7311 | { | |
7312 | u64 alloc_flags = get_alloc_profile(root, type); | |
0e4f8f88 CM |
7313 | return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
7314 | CHUNK_ALLOC_FORCE); | |
c87f08ca CM |
7315 | } |
7316 | ||
6d07bcec MX |
7317 | /* |
7318 | * helper to account the unused space of all the readonly block group in the | |
7319 | * list. takes mirrors into account. | |
7320 | */ | |
7321 | static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list) | |
7322 | { | |
7323 | struct btrfs_block_group_cache *block_group; | |
7324 | u64 free_bytes = 0; | |
7325 | int factor; | |
7326 | ||
7327 | list_for_each_entry(block_group, groups_list, list) { | |
7328 | spin_lock(&block_group->lock); | |
7329 | ||
7330 | if (!block_group->ro) { | |
7331 | spin_unlock(&block_group->lock); | |
7332 | continue; | |
7333 | } | |
7334 | ||
7335 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7336 | BTRFS_BLOCK_GROUP_RAID10 | | |
7337 | BTRFS_BLOCK_GROUP_DUP)) | |
7338 | factor = 2; | |
7339 | else | |
7340 | factor = 1; | |
7341 | ||
7342 | free_bytes += (block_group->key.offset - | |
7343 | btrfs_block_group_used(&block_group->item)) * | |
7344 | factor; | |
7345 | ||
7346 | spin_unlock(&block_group->lock); | |
7347 | } | |
7348 | ||
7349 | return free_bytes; | |
7350 | } | |
7351 | ||
7352 | /* | |
7353 | * helper to account the unused space of all the readonly block group in the | |
7354 | * space_info. takes mirrors into account. | |
7355 | */ | |
7356 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) | |
7357 | { | |
7358 | int i; | |
7359 | u64 free_bytes = 0; | |
7360 | ||
7361 | spin_lock(&sinfo->lock); | |
7362 | ||
7363 | for(i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
7364 | if (!list_empty(&sinfo->block_groups[i])) | |
7365 | free_bytes += __btrfs_get_ro_block_group_free_space( | |
7366 | &sinfo->block_groups[i]); | |
7367 | ||
7368 | spin_unlock(&sinfo->lock); | |
7369 | ||
7370 | return free_bytes; | |
7371 | } | |
7372 | ||
143bede5 | 7373 | void btrfs_set_block_group_rw(struct btrfs_root *root, |
f0486c68 | 7374 | struct btrfs_block_group_cache *cache) |
5d4f98a2 | 7375 | { |
f0486c68 YZ |
7376 | struct btrfs_space_info *sinfo = cache->space_info; |
7377 | u64 num_bytes; | |
7378 | ||
7379 | BUG_ON(!cache->ro); | |
7380 | ||
7381 | spin_lock(&sinfo->lock); | |
7382 | spin_lock(&cache->lock); | |
7383 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
7384 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7385 | sinfo->bytes_readonly -= num_bytes; | |
7386 | cache->ro = 0; | |
7387 | spin_unlock(&cache->lock); | |
7388 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
7389 | } |
7390 | ||
ba1bf481 JB |
7391 | /* |
7392 | * checks to see if its even possible to relocate this block group. | |
7393 | * | |
7394 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
7395 | * ok to go ahead and try. | |
7396 | */ | |
7397 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 7398 | { |
ba1bf481 JB |
7399 | struct btrfs_block_group_cache *block_group; |
7400 | struct btrfs_space_info *space_info; | |
7401 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
7402 | struct btrfs_device *device; | |
cdcb725c | 7403 | u64 min_free; |
6719db6a JB |
7404 | u64 dev_min = 1; |
7405 | u64 dev_nr = 0; | |
4a5e98f5 | 7406 | u64 target; |
cdcb725c | 7407 | int index; |
ba1bf481 JB |
7408 | int full = 0; |
7409 | int ret = 0; | |
1a40e23b | 7410 | |
ba1bf481 | 7411 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 7412 | |
ba1bf481 JB |
7413 | /* odd, couldn't find the block group, leave it alone */ |
7414 | if (!block_group) | |
7415 | return -1; | |
1a40e23b | 7416 | |
cdcb725c | 7417 | min_free = btrfs_block_group_used(&block_group->item); |
7418 | ||
ba1bf481 | 7419 | /* no bytes used, we're good */ |
cdcb725c | 7420 | if (!min_free) |
1a40e23b ZY |
7421 | goto out; |
7422 | ||
ba1bf481 JB |
7423 | space_info = block_group->space_info; |
7424 | spin_lock(&space_info->lock); | |
17d217fe | 7425 | |
ba1bf481 | 7426 | full = space_info->full; |
17d217fe | 7427 | |
ba1bf481 JB |
7428 | /* |
7429 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
7430 | * relocate it unless we're able to allocate a new chunk below. |
7431 | * | |
7432 | * Otherwise, we need to make sure we have room in the space to handle | |
7433 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 7434 | */ |
7ce618db | 7435 | if ((space_info->total_bytes != block_group->key.offset) && |
cdcb725c | 7436 | (space_info->bytes_used + space_info->bytes_reserved + |
7437 | space_info->bytes_pinned + space_info->bytes_readonly + | |
7438 | min_free < space_info->total_bytes)) { | |
ba1bf481 JB |
7439 | spin_unlock(&space_info->lock); |
7440 | goto out; | |
17d217fe | 7441 | } |
ba1bf481 | 7442 | spin_unlock(&space_info->lock); |
ea8c2819 | 7443 | |
ba1bf481 JB |
7444 | /* |
7445 | * ok we don't have enough space, but maybe we have free space on our | |
7446 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
7447 | * alloc devices and guess if we have enough space. if this block |
7448 | * group is going to be restriped, run checks against the target | |
7449 | * profile instead of the current one. | |
ba1bf481 JB |
7450 | */ |
7451 | ret = -1; | |
ea8c2819 | 7452 | |
cdcb725c | 7453 | /* |
7454 | * index: | |
7455 | * 0: raid10 | |
7456 | * 1: raid1 | |
7457 | * 2: dup | |
7458 | * 3: raid0 | |
7459 | * 4: single | |
7460 | */ | |
4a5e98f5 ID |
7461 | target = get_restripe_target(root->fs_info, block_group->flags); |
7462 | if (target) { | |
7463 | index = __get_block_group_index(extended_to_chunk(target)); | |
7464 | } else { | |
7465 | /* | |
7466 | * this is just a balance, so if we were marked as full | |
7467 | * we know there is no space for a new chunk | |
7468 | */ | |
7469 | if (full) | |
7470 | goto out; | |
7471 | ||
7472 | index = get_block_group_index(block_group); | |
7473 | } | |
7474 | ||
cdcb725c | 7475 | if (index == 0) { |
7476 | dev_min = 4; | |
6719db6a JB |
7477 | /* Divide by 2 */ |
7478 | min_free >>= 1; | |
cdcb725c | 7479 | } else if (index == 1) { |
7480 | dev_min = 2; | |
7481 | } else if (index == 2) { | |
6719db6a JB |
7482 | /* Multiply by 2 */ |
7483 | min_free <<= 1; | |
cdcb725c | 7484 | } else if (index == 3) { |
7485 | dev_min = fs_devices->rw_devices; | |
6719db6a | 7486 | do_div(min_free, dev_min); |
cdcb725c | 7487 | } |
7488 | ||
ba1bf481 JB |
7489 | mutex_lock(&root->fs_info->chunk_mutex); |
7490 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7bfc837d | 7491 | u64 dev_offset; |
56bec294 | 7492 | |
ba1bf481 JB |
7493 | /* |
7494 | * check to make sure we can actually find a chunk with enough | |
7495 | * space to fit our block group in. | |
7496 | */ | |
7497 | if (device->total_bytes > device->bytes_used + min_free) { | |
125ccb0a | 7498 | ret = find_free_dev_extent(device, min_free, |
7bfc837d | 7499 | &dev_offset, NULL); |
ba1bf481 | 7500 | if (!ret) |
cdcb725c | 7501 | dev_nr++; |
7502 | ||
7503 | if (dev_nr >= dev_min) | |
73e48b27 | 7504 | break; |
cdcb725c | 7505 | |
ba1bf481 | 7506 | ret = -1; |
725c8463 | 7507 | } |
edbd8d4e | 7508 | } |
ba1bf481 | 7509 | mutex_unlock(&root->fs_info->chunk_mutex); |
edbd8d4e | 7510 | out: |
ba1bf481 | 7511 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
7512 | return ret; |
7513 | } | |
7514 | ||
b2950863 CH |
7515 | static int find_first_block_group(struct btrfs_root *root, |
7516 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 7517 | { |
925baedd | 7518 | int ret = 0; |
0b86a832 CM |
7519 | struct btrfs_key found_key; |
7520 | struct extent_buffer *leaf; | |
7521 | int slot; | |
edbd8d4e | 7522 | |
0b86a832 CM |
7523 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
7524 | if (ret < 0) | |
925baedd CM |
7525 | goto out; |
7526 | ||
d397712b | 7527 | while (1) { |
0b86a832 | 7528 | slot = path->slots[0]; |
edbd8d4e | 7529 | leaf = path->nodes[0]; |
0b86a832 CM |
7530 | if (slot >= btrfs_header_nritems(leaf)) { |
7531 | ret = btrfs_next_leaf(root, path); | |
7532 | if (ret == 0) | |
7533 | continue; | |
7534 | if (ret < 0) | |
925baedd | 7535 | goto out; |
0b86a832 | 7536 | break; |
edbd8d4e | 7537 | } |
0b86a832 | 7538 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 7539 | |
0b86a832 | 7540 | if (found_key.objectid >= key->objectid && |
925baedd CM |
7541 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
7542 | ret = 0; | |
7543 | goto out; | |
7544 | } | |
0b86a832 | 7545 | path->slots[0]++; |
edbd8d4e | 7546 | } |
925baedd | 7547 | out: |
0b86a832 | 7548 | return ret; |
edbd8d4e CM |
7549 | } |
7550 | ||
0af3d00b JB |
7551 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
7552 | { | |
7553 | struct btrfs_block_group_cache *block_group; | |
7554 | u64 last = 0; | |
7555 | ||
7556 | while (1) { | |
7557 | struct inode *inode; | |
7558 | ||
7559 | block_group = btrfs_lookup_first_block_group(info, last); | |
7560 | while (block_group) { | |
7561 | spin_lock(&block_group->lock); | |
7562 | if (block_group->iref) | |
7563 | break; | |
7564 | spin_unlock(&block_group->lock); | |
7565 | block_group = next_block_group(info->tree_root, | |
7566 | block_group); | |
7567 | } | |
7568 | if (!block_group) { | |
7569 | if (last == 0) | |
7570 | break; | |
7571 | last = 0; | |
7572 | continue; | |
7573 | } | |
7574 | ||
7575 | inode = block_group->inode; | |
7576 | block_group->iref = 0; | |
7577 | block_group->inode = NULL; | |
7578 | spin_unlock(&block_group->lock); | |
7579 | iput(inode); | |
7580 | last = block_group->key.objectid + block_group->key.offset; | |
7581 | btrfs_put_block_group(block_group); | |
7582 | } | |
7583 | } | |
7584 | ||
1a40e23b ZY |
7585 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
7586 | { | |
7587 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 7588 | struct btrfs_space_info *space_info; |
11833d66 | 7589 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
7590 | struct rb_node *n; |
7591 | ||
11833d66 YZ |
7592 | down_write(&info->extent_commit_sem); |
7593 | while (!list_empty(&info->caching_block_groups)) { | |
7594 | caching_ctl = list_entry(info->caching_block_groups.next, | |
7595 | struct btrfs_caching_control, list); | |
7596 | list_del(&caching_ctl->list); | |
7597 | put_caching_control(caching_ctl); | |
7598 | } | |
7599 | up_write(&info->extent_commit_sem); | |
7600 | ||
1a40e23b ZY |
7601 | spin_lock(&info->block_group_cache_lock); |
7602 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
7603 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
7604 | cache_node); | |
1a40e23b ZY |
7605 | rb_erase(&block_group->cache_node, |
7606 | &info->block_group_cache_tree); | |
d899e052 YZ |
7607 | spin_unlock(&info->block_group_cache_lock); |
7608 | ||
80eb234a | 7609 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 7610 | list_del(&block_group->list); |
80eb234a | 7611 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 7612 | |
817d52f8 | 7613 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 7614 | wait_block_group_cache_done(block_group); |
817d52f8 | 7615 | |
3c14874a JB |
7616 | /* |
7617 | * We haven't cached this block group, which means we could | |
7618 | * possibly have excluded extents on this block group. | |
7619 | */ | |
7620 | if (block_group->cached == BTRFS_CACHE_NO) | |
7621 | free_excluded_extents(info->extent_root, block_group); | |
7622 | ||
817d52f8 | 7623 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 7624 | btrfs_put_block_group(block_group); |
d899e052 YZ |
7625 | |
7626 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
7627 | } |
7628 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
7629 | |
7630 | /* now that all the block groups are freed, go through and | |
7631 | * free all the space_info structs. This is only called during | |
7632 | * the final stages of unmount, and so we know nobody is | |
7633 | * using them. We call synchronize_rcu() once before we start, | |
7634 | * just to be on the safe side. | |
7635 | */ | |
7636 | synchronize_rcu(); | |
7637 | ||
8929ecfa YZ |
7638 | release_global_block_rsv(info); |
7639 | ||
4184ea7f CM |
7640 | while(!list_empty(&info->space_info)) { |
7641 | space_info = list_entry(info->space_info.next, | |
7642 | struct btrfs_space_info, | |
7643 | list); | |
f0486c68 | 7644 | if (space_info->bytes_pinned > 0 || |
fb25e914 JB |
7645 | space_info->bytes_reserved > 0 || |
7646 | space_info->bytes_may_use > 0) { | |
f0486c68 YZ |
7647 | WARN_ON(1); |
7648 | dump_space_info(space_info, 0, 0); | |
7649 | } | |
4184ea7f CM |
7650 | list_del(&space_info->list); |
7651 | kfree(space_info); | |
7652 | } | |
1a40e23b ZY |
7653 | return 0; |
7654 | } | |
7655 | ||
b742bb82 YZ |
7656 | static void __link_block_group(struct btrfs_space_info *space_info, |
7657 | struct btrfs_block_group_cache *cache) | |
7658 | { | |
7659 | int index = get_block_group_index(cache); | |
7660 | ||
7661 | down_write(&space_info->groups_sem); | |
7662 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
7663 | up_write(&space_info->groups_sem); | |
7664 | } | |
7665 | ||
9078a3e1 CM |
7666 | int btrfs_read_block_groups(struct btrfs_root *root) |
7667 | { | |
7668 | struct btrfs_path *path; | |
7669 | int ret; | |
9078a3e1 | 7670 | struct btrfs_block_group_cache *cache; |
be744175 | 7671 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 7672 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
7673 | struct btrfs_key key; |
7674 | struct btrfs_key found_key; | |
5f39d397 | 7675 | struct extent_buffer *leaf; |
0af3d00b JB |
7676 | int need_clear = 0; |
7677 | u64 cache_gen; | |
96b5179d | 7678 | |
be744175 | 7679 | root = info->extent_root; |
9078a3e1 | 7680 | key.objectid = 0; |
0b86a832 | 7681 | key.offset = 0; |
9078a3e1 | 7682 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
7683 | path = btrfs_alloc_path(); |
7684 | if (!path) | |
7685 | return -ENOMEM; | |
026fd317 | 7686 | path->reada = 1; |
9078a3e1 | 7687 | |
6c41761f | 7688 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); |
73bc1876 | 7689 | if (btrfs_test_opt(root, SPACE_CACHE) && |
6c41761f | 7690 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) |
0af3d00b | 7691 | need_clear = 1; |
88c2ba3b JB |
7692 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
7693 | need_clear = 1; | |
0af3d00b | 7694 | |
d397712b | 7695 | while (1) { |
0b86a832 | 7696 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
7697 | if (ret > 0) |
7698 | break; | |
0b86a832 CM |
7699 | if (ret != 0) |
7700 | goto error; | |
5f39d397 CM |
7701 | leaf = path->nodes[0]; |
7702 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 7703 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 7704 | if (!cache) { |
0b86a832 | 7705 | ret = -ENOMEM; |
f0486c68 | 7706 | goto error; |
9078a3e1 | 7707 | } |
34d52cb6 LZ |
7708 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
7709 | GFP_NOFS); | |
7710 | if (!cache->free_space_ctl) { | |
7711 | kfree(cache); | |
7712 | ret = -ENOMEM; | |
7713 | goto error; | |
7714 | } | |
3e1ad54f | 7715 | |
d2fb3437 | 7716 | atomic_set(&cache->count, 1); |
c286ac48 | 7717 | spin_lock_init(&cache->lock); |
817d52f8 | 7718 | cache->fs_info = info; |
0f9dd46c | 7719 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7720 | INIT_LIST_HEAD(&cache->cluster_list); |
96303081 | 7721 | |
cf7c1ef6 LB |
7722 | if (need_clear) { |
7723 | /* | |
7724 | * When we mount with old space cache, we need to | |
7725 | * set BTRFS_DC_CLEAR and set dirty flag. | |
7726 | * | |
7727 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
7728 | * truncate the old free space cache inode and | |
7729 | * setup a new one. | |
7730 | * b) Setting 'dirty flag' makes sure that we flush | |
7731 | * the new space cache info onto disk. | |
7732 | */ | |
0af3d00b | 7733 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 LB |
7734 | if (btrfs_test_opt(root, SPACE_CACHE)) |
7735 | cache->dirty = 1; | |
7736 | } | |
0af3d00b | 7737 | |
5f39d397 CM |
7738 | read_extent_buffer(leaf, &cache->item, |
7739 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
7740 | sizeof(cache->item)); | |
9078a3e1 | 7741 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 7742 | |
9078a3e1 | 7743 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 7744 | btrfs_release_path(path); |
0b86a832 | 7745 | cache->flags = btrfs_block_group_flags(&cache->item); |
817d52f8 JB |
7746 | cache->sectorsize = root->sectorsize; |
7747 | ||
34d52cb6 LZ |
7748 | btrfs_init_free_space_ctl(cache); |
7749 | ||
3c14874a JB |
7750 | /* |
7751 | * We need to exclude the super stripes now so that the space | |
7752 | * info has super bytes accounted for, otherwise we'll think | |
7753 | * we have more space than we actually do. | |
7754 | */ | |
7755 | exclude_super_stripes(root, cache); | |
7756 | ||
817d52f8 JB |
7757 | /* |
7758 | * check for two cases, either we are full, and therefore | |
7759 | * don't need to bother with the caching work since we won't | |
7760 | * find any space, or we are empty, and we can just add all | |
7761 | * the space in and be done with it. This saves us _alot_ of | |
7762 | * time, particularly in the full case. | |
7763 | */ | |
7764 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 7765 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7766 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 7767 | free_excluded_extents(root, cache); |
817d52f8 | 7768 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 7769 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
7770 | cache->cached = BTRFS_CACHE_FINISHED; |
7771 | add_new_free_space(cache, root->fs_info, | |
7772 | found_key.objectid, | |
7773 | found_key.objectid + | |
7774 | found_key.offset); | |
11833d66 | 7775 | free_excluded_extents(root, cache); |
817d52f8 | 7776 | } |
96b5179d | 7777 | |
6324fbf3 CM |
7778 | ret = update_space_info(info, cache->flags, found_key.offset, |
7779 | btrfs_block_group_used(&cache->item), | |
7780 | &space_info); | |
79787eaa | 7781 | BUG_ON(ret); /* -ENOMEM */ |
6324fbf3 | 7782 | cache->space_info = space_info; |
1b2da372 | 7783 | spin_lock(&cache->space_info->lock); |
f0486c68 | 7784 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7785 | spin_unlock(&cache->space_info->lock); |
7786 | ||
b742bb82 | 7787 | __link_block_group(space_info, cache); |
0f9dd46c JB |
7788 | |
7789 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
79787eaa | 7790 | BUG_ON(ret); /* Logic error */ |
75ccf47d CM |
7791 | |
7792 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
2b82032c | 7793 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
199c36ea | 7794 | set_block_group_ro(cache, 1); |
9078a3e1 | 7795 | } |
b742bb82 YZ |
7796 | |
7797 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
7798 | if (!(get_alloc_profile(root, space_info->flags) & | |
7799 | (BTRFS_BLOCK_GROUP_RAID10 | | |
7800 | BTRFS_BLOCK_GROUP_RAID1 | | |
7801 | BTRFS_BLOCK_GROUP_DUP))) | |
7802 | continue; | |
7803 | /* | |
7804 | * avoid allocating from un-mirrored block group if there are | |
7805 | * mirrored block groups. | |
7806 | */ | |
7807 | list_for_each_entry(cache, &space_info->block_groups[3], list) | |
199c36ea | 7808 | set_block_group_ro(cache, 1); |
b742bb82 | 7809 | list_for_each_entry(cache, &space_info->block_groups[4], list) |
199c36ea | 7810 | set_block_group_ro(cache, 1); |
9078a3e1 | 7811 | } |
f0486c68 YZ |
7812 | |
7813 | init_global_block_rsv(info); | |
0b86a832 CM |
7814 | ret = 0; |
7815 | error: | |
9078a3e1 | 7816 | btrfs_free_path(path); |
0b86a832 | 7817 | return ret; |
9078a3e1 | 7818 | } |
6324fbf3 CM |
7819 | |
7820 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
7821 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 7822 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
7823 | u64 size) |
7824 | { | |
7825 | int ret; | |
6324fbf3 CM |
7826 | struct btrfs_root *extent_root; |
7827 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
7828 | |
7829 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 7830 | |
12fcfd22 | 7831 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 7832 | |
8f18cf13 | 7833 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
7834 | if (!cache) |
7835 | return -ENOMEM; | |
34d52cb6 LZ |
7836 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
7837 | GFP_NOFS); | |
7838 | if (!cache->free_space_ctl) { | |
7839 | kfree(cache); | |
7840 | return -ENOMEM; | |
7841 | } | |
0f9dd46c | 7842 | |
e17cade2 | 7843 | cache->key.objectid = chunk_offset; |
6324fbf3 | 7844 | cache->key.offset = size; |
d2fb3437 | 7845 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
96303081 | 7846 | cache->sectorsize = root->sectorsize; |
0af3d00b | 7847 | cache->fs_info = root->fs_info; |
96303081 | 7848 | |
d2fb3437 | 7849 | atomic_set(&cache->count, 1); |
c286ac48 | 7850 | spin_lock_init(&cache->lock); |
0f9dd46c | 7851 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7852 | INIT_LIST_HEAD(&cache->cluster_list); |
0ef3e66b | 7853 | |
34d52cb6 LZ |
7854 | btrfs_init_free_space_ctl(cache); |
7855 | ||
6324fbf3 | 7856 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
7857 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
7858 | cache->flags = type; | |
7859 | btrfs_set_block_group_flags(&cache->item, type); | |
7860 | ||
11833d66 | 7861 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7862 | cache->cached = BTRFS_CACHE_FINISHED; |
11833d66 | 7863 | exclude_super_stripes(root, cache); |
96303081 | 7864 | |
817d52f8 JB |
7865 | add_new_free_space(cache, root->fs_info, chunk_offset, |
7866 | chunk_offset + size); | |
7867 | ||
11833d66 YZ |
7868 | free_excluded_extents(root, cache); |
7869 | ||
6324fbf3 CM |
7870 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
7871 | &cache->space_info); | |
79787eaa | 7872 | BUG_ON(ret); /* -ENOMEM */ |
c7c144db | 7873 | update_global_block_rsv(root->fs_info); |
1b2da372 JB |
7874 | |
7875 | spin_lock(&cache->space_info->lock); | |
f0486c68 | 7876 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7877 | spin_unlock(&cache->space_info->lock); |
7878 | ||
b742bb82 | 7879 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 7880 | |
0f9dd46c | 7881 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
79787eaa | 7882 | BUG_ON(ret); /* Logic error */ |
c286ac48 | 7883 | |
6324fbf3 CM |
7884 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
7885 | sizeof(cache->item)); | |
79787eaa JM |
7886 | if (ret) { |
7887 | btrfs_abort_transaction(trans, extent_root, ret); | |
7888 | return ret; | |
7889 | } | |
6324fbf3 | 7890 | |
d18a2c44 | 7891 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 7892 | |
6324fbf3 CM |
7893 | return 0; |
7894 | } | |
1a40e23b | 7895 | |
10ea00f5 ID |
7896 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
7897 | { | |
899c81ea ID |
7898 | u64 extra_flags = chunk_to_extended(flags) & |
7899 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 ID |
7900 | |
7901 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
7902 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
7903 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
7904 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
7905 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
7906 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
7907 | } | |
7908 | ||
1a40e23b ZY |
7909 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
7910 | struct btrfs_root *root, u64 group_start) | |
7911 | { | |
7912 | struct btrfs_path *path; | |
7913 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 7914 | struct btrfs_free_cluster *cluster; |
0af3d00b | 7915 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 7916 | struct btrfs_key key; |
0af3d00b | 7917 | struct inode *inode; |
1a40e23b | 7918 | int ret; |
10ea00f5 | 7919 | int index; |
89a55897 | 7920 | int factor; |
1a40e23b | 7921 | |
1a40e23b ZY |
7922 | root = root->fs_info->extent_root; |
7923 | ||
7924 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
7925 | BUG_ON(!block_group); | |
c146afad | 7926 | BUG_ON(!block_group->ro); |
1a40e23b | 7927 | |
9f7c43c9 | 7928 | /* |
7929 | * Free the reserved super bytes from this block group before | |
7930 | * remove it. | |
7931 | */ | |
7932 | free_excluded_extents(root, block_group); | |
7933 | ||
1a40e23b | 7934 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 7935 | index = get_block_group_index(block_group); |
89a55897 JB |
7936 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
7937 | BTRFS_BLOCK_GROUP_RAID1 | | |
7938 | BTRFS_BLOCK_GROUP_RAID10)) | |
7939 | factor = 2; | |
7940 | else | |
7941 | factor = 1; | |
1a40e23b | 7942 | |
44fb5511 CM |
7943 | /* make sure this block group isn't part of an allocation cluster */ |
7944 | cluster = &root->fs_info->data_alloc_cluster; | |
7945 | spin_lock(&cluster->refill_lock); | |
7946 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7947 | spin_unlock(&cluster->refill_lock); | |
7948 | ||
7949 | /* | |
7950 | * make sure this block group isn't part of a metadata | |
7951 | * allocation cluster | |
7952 | */ | |
7953 | cluster = &root->fs_info->meta_alloc_cluster; | |
7954 | spin_lock(&cluster->refill_lock); | |
7955 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7956 | spin_unlock(&cluster->refill_lock); | |
7957 | ||
1a40e23b | 7958 | path = btrfs_alloc_path(); |
d8926bb3 MF |
7959 | if (!path) { |
7960 | ret = -ENOMEM; | |
7961 | goto out; | |
7962 | } | |
1a40e23b | 7963 | |
10b2f34d | 7964 | inode = lookup_free_space_inode(tree_root, block_group, path); |
0af3d00b | 7965 | if (!IS_ERR(inode)) { |
b532402e | 7966 | ret = btrfs_orphan_add(trans, inode); |
79787eaa JM |
7967 | if (ret) { |
7968 | btrfs_add_delayed_iput(inode); | |
7969 | goto out; | |
7970 | } | |
0af3d00b JB |
7971 | clear_nlink(inode); |
7972 | /* One for the block groups ref */ | |
7973 | spin_lock(&block_group->lock); | |
7974 | if (block_group->iref) { | |
7975 | block_group->iref = 0; | |
7976 | block_group->inode = NULL; | |
7977 | spin_unlock(&block_group->lock); | |
7978 | iput(inode); | |
7979 | } else { | |
7980 | spin_unlock(&block_group->lock); | |
7981 | } | |
7982 | /* One for our lookup ref */ | |
455757c3 | 7983 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
7984 | } |
7985 | ||
7986 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
7987 | key.offset = block_group->key.objectid; | |
7988 | key.type = 0; | |
7989 | ||
7990 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
7991 | if (ret < 0) | |
7992 | goto out; | |
7993 | if (ret > 0) | |
b3b4aa74 | 7994 | btrfs_release_path(path); |
0af3d00b JB |
7995 | if (ret == 0) { |
7996 | ret = btrfs_del_item(trans, tree_root, path); | |
7997 | if (ret) | |
7998 | goto out; | |
b3b4aa74 | 7999 | btrfs_release_path(path); |
0af3d00b JB |
8000 | } |
8001 | ||
3dfdb934 | 8002 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
8003 | rb_erase(&block_group->cache_node, |
8004 | &root->fs_info->block_group_cache_tree); | |
3dfdb934 | 8005 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 8006 | |
80eb234a | 8007 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
8008 | /* |
8009 | * we must use list_del_init so people can check to see if they | |
8010 | * are still on the list after taking the semaphore | |
8011 | */ | |
8012 | list_del_init(&block_group->list); | |
10ea00f5 ID |
8013 | if (list_empty(&block_group->space_info->block_groups[index])) |
8014 | clear_avail_alloc_bits(root->fs_info, block_group->flags); | |
80eb234a | 8015 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 8016 | |
817d52f8 | 8017 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 8018 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
8019 | |
8020 | btrfs_remove_free_space_cache(block_group); | |
8021 | ||
c146afad YZ |
8022 | spin_lock(&block_group->space_info->lock); |
8023 | block_group->space_info->total_bytes -= block_group->key.offset; | |
8024 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 8025 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
c146afad | 8026 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 8027 | |
0af3d00b JB |
8028 | memcpy(&key, &block_group->key, sizeof(key)); |
8029 | ||
283bb197 | 8030 | btrfs_clear_space_info_full(root->fs_info); |
c146afad | 8031 | |
fa9c0d79 CM |
8032 | btrfs_put_block_group(block_group); |
8033 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
8034 | |
8035 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
8036 | if (ret > 0) | |
8037 | ret = -EIO; | |
8038 | if (ret < 0) | |
8039 | goto out; | |
8040 | ||
8041 | ret = btrfs_del_item(trans, root, path); | |
8042 | out: | |
8043 | btrfs_free_path(path); | |
8044 | return ret; | |
8045 | } | |
acce952b | 8046 | |
c59021f8 | 8047 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
8048 | { | |
8049 | struct btrfs_space_info *space_info; | |
1aba86d6 | 8050 | struct btrfs_super_block *disk_super; |
8051 | u64 features; | |
8052 | u64 flags; | |
8053 | int mixed = 0; | |
c59021f8 | 8054 | int ret; |
8055 | ||
6c41761f | 8056 | disk_super = fs_info->super_copy; |
1aba86d6 | 8057 | if (!btrfs_super_root(disk_super)) |
8058 | return 1; | |
c59021f8 | 8059 | |
1aba86d6 | 8060 | features = btrfs_super_incompat_flags(disk_super); |
8061 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
8062 | mixed = 1; | |
c59021f8 | 8063 | |
1aba86d6 | 8064 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
8065 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
c59021f8 | 8066 | if (ret) |
1aba86d6 | 8067 | goto out; |
c59021f8 | 8068 | |
1aba86d6 | 8069 | if (mixed) { |
8070 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
8071 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8072 | } else { | |
8073 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
8074 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8075 | if (ret) | |
8076 | goto out; | |
8077 | ||
8078 | flags = BTRFS_BLOCK_GROUP_DATA; | |
8079 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8080 | } | |
8081 | out: | |
c59021f8 | 8082 | return ret; |
8083 | } | |
8084 | ||
acce952b | 8085 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
8086 | { | |
8087 | return unpin_extent_range(root, start, end); | |
8088 | } | |
8089 | ||
8090 | int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 8091 | u64 num_bytes, u64 *actual_bytes) |
acce952b | 8092 | { |
5378e607 | 8093 | return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes); |
acce952b | 8094 | } |
f7039b1d LD |
8095 | |
8096 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) | |
8097 | { | |
8098 | struct btrfs_fs_info *fs_info = root->fs_info; | |
8099 | struct btrfs_block_group_cache *cache = NULL; | |
8100 | u64 group_trimmed; | |
8101 | u64 start; | |
8102 | u64 end; | |
8103 | u64 trimmed = 0; | |
2cac13e4 | 8104 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
8105 | int ret = 0; |
8106 | ||
2cac13e4 LB |
8107 | /* |
8108 | * try to trim all FS space, our block group may start from non-zero. | |
8109 | */ | |
8110 | if (range->len == total_bytes) | |
8111 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
8112 | else | |
8113 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
8114 | |
8115 | while (cache) { | |
8116 | if (cache->key.objectid >= (range->start + range->len)) { | |
8117 | btrfs_put_block_group(cache); | |
8118 | break; | |
8119 | } | |
8120 | ||
8121 | start = max(range->start, cache->key.objectid); | |
8122 | end = min(range->start + range->len, | |
8123 | cache->key.objectid + cache->key.offset); | |
8124 | ||
8125 | if (end - start >= range->minlen) { | |
8126 | if (!block_group_cache_done(cache)) { | |
8127 | ret = cache_block_group(cache, NULL, root, 0); | |
8128 | if (!ret) | |
8129 | wait_block_group_cache_done(cache); | |
8130 | } | |
8131 | ret = btrfs_trim_block_group(cache, | |
8132 | &group_trimmed, | |
8133 | start, | |
8134 | end, | |
8135 | range->minlen); | |
8136 | ||
8137 | trimmed += group_trimmed; | |
8138 | if (ret) { | |
8139 | btrfs_put_block_group(cache); | |
8140 | break; | |
8141 | } | |
8142 | } | |
8143 | ||
8144 | cache = next_block_group(fs_info->tree_root, cache); | |
8145 | } | |
8146 | ||
8147 | range->len = trimmed; | |
8148 | return ret; | |
8149 | } |