]>
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 | |
ae1e206b JB |
2254 | /* |
2255 | * We need to try and merge add/drops of the same ref since we | |
2256 | * can run into issues with relocate dropping the implicit ref | |
2257 | * and then it being added back again before the drop can | |
2258 | * finish. If we merged anything we need to re-loop so we can | |
2259 | * get a good ref. | |
2260 | */ | |
2261 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, | |
2262 | locked_ref); | |
2263 | ||
d1270cd9 AJ |
2264 | /* |
2265 | * locked_ref is the head node, so we have to go one | |
2266 | * node back for any delayed ref updates | |
2267 | */ | |
2268 | ref = select_delayed_ref(locked_ref); | |
2269 | ||
2270 | if (ref && ref->seq && | |
097b8a7c | 2271 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d1270cd9 AJ |
2272 | /* |
2273 | * there are still refs with lower seq numbers in the | |
2274 | * process of being added. Don't run this ref yet. | |
2275 | */ | |
2276 | list_del_init(&locked_ref->cluster); | |
2277 | mutex_unlock(&locked_ref->mutex); | |
2278 | locked_ref = NULL; | |
2279 | delayed_refs->num_heads_ready++; | |
2280 | spin_unlock(&delayed_refs->lock); | |
2281 | cond_resched(); | |
2282 | spin_lock(&delayed_refs->lock); | |
2283 | continue; | |
2284 | } | |
2285 | ||
56bec294 CM |
2286 | /* |
2287 | * record the must insert reserved flag before we | |
2288 | * drop the spin lock. | |
2289 | */ | |
2290 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2291 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2292 | |
5d4f98a2 YZ |
2293 | extent_op = locked_ref->extent_op; |
2294 | locked_ref->extent_op = NULL; | |
2295 | ||
56bec294 CM |
2296 | if (!ref) { |
2297 | /* All delayed refs have been processed, Go ahead | |
2298 | * and send the head node to run_one_delayed_ref, | |
2299 | * so that any accounting fixes can happen | |
2300 | */ | |
2301 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2302 | |
2303 | if (extent_op && must_insert_reserved) { | |
2304 | kfree(extent_op); | |
2305 | extent_op = NULL; | |
2306 | } | |
2307 | ||
2308 | if (extent_op) { | |
2309 | spin_unlock(&delayed_refs->lock); | |
2310 | ||
2311 | ret = run_delayed_extent_op(trans, root, | |
2312 | ref, extent_op); | |
5d4f98a2 YZ |
2313 | kfree(extent_op); |
2314 | ||
79787eaa JM |
2315 | if (ret) { |
2316 | printk(KERN_DEBUG "btrfs: run_delayed_extent_op returned %d\n", ret); | |
253beebd | 2317 | spin_lock(&delayed_refs->lock); |
79787eaa JM |
2318 | return ret; |
2319 | } | |
2320 | ||
203bf287 | 2321 | goto next; |
5d4f98a2 YZ |
2322 | } |
2323 | ||
c3e69d58 | 2324 | list_del_init(&locked_ref->cluster); |
56bec294 CM |
2325 | locked_ref = NULL; |
2326 | } | |
02217ed2 | 2327 | |
56bec294 CM |
2328 | ref->in_tree = 0; |
2329 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
2330 | delayed_refs->num_entries--; | |
22cd2e7d AJ |
2331 | if (locked_ref) { |
2332 | /* | |
2333 | * when we play the delayed ref, also correct the | |
2334 | * ref_mod on head | |
2335 | */ | |
2336 | switch (ref->action) { | |
2337 | case BTRFS_ADD_DELAYED_REF: | |
2338 | case BTRFS_ADD_DELAYED_EXTENT: | |
2339 | locked_ref->node.ref_mod -= ref->ref_mod; | |
2340 | break; | |
2341 | case BTRFS_DROP_DELAYED_REF: | |
2342 | locked_ref->node.ref_mod += ref->ref_mod; | |
2343 | break; | |
2344 | default: | |
2345 | WARN_ON(1); | |
2346 | } | |
2347 | } | |
56bec294 | 2348 | spin_unlock(&delayed_refs->lock); |
925baedd | 2349 | |
5d4f98a2 | 2350 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 | 2351 | must_insert_reserved); |
eb099670 | 2352 | |
5d4f98a2 YZ |
2353 | btrfs_put_delayed_ref(ref); |
2354 | kfree(extent_op); | |
c3e69d58 | 2355 | count++; |
79787eaa JM |
2356 | |
2357 | if (ret) { | |
2358 | printk(KERN_DEBUG "btrfs: run_one_delayed_ref returned %d\n", ret); | |
253beebd | 2359 | spin_lock(&delayed_refs->lock); |
79787eaa JM |
2360 | return ret; |
2361 | } | |
2362 | ||
203bf287 | 2363 | next: |
097b8a7c | 2364 | do_chunk_alloc(trans, fs_info->extent_root, |
203bf287 CM |
2365 | 2 * 1024 * 1024, |
2366 | btrfs_get_alloc_profile(root, 0), | |
2367 | CHUNK_ALLOC_NO_FORCE); | |
c3e69d58 CM |
2368 | cond_resched(); |
2369 | spin_lock(&delayed_refs->lock); | |
2370 | } | |
2371 | return count; | |
2372 | } | |
2373 | ||
709c0486 AJ |
2374 | #ifdef SCRAMBLE_DELAYED_REFS |
2375 | /* | |
2376 | * Normally delayed refs get processed in ascending bytenr order. This | |
2377 | * correlates in most cases to the order added. To expose dependencies on this | |
2378 | * order, we start to process the tree in the middle instead of the beginning | |
2379 | */ | |
2380 | static u64 find_middle(struct rb_root *root) | |
2381 | { | |
2382 | struct rb_node *n = root->rb_node; | |
2383 | struct btrfs_delayed_ref_node *entry; | |
2384 | int alt = 1; | |
2385 | u64 middle; | |
2386 | u64 first = 0, last = 0; | |
2387 | ||
2388 | n = rb_first(root); | |
2389 | if (n) { | |
2390 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2391 | first = entry->bytenr; | |
2392 | } | |
2393 | n = rb_last(root); | |
2394 | if (n) { | |
2395 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2396 | last = entry->bytenr; | |
2397 | } | |
2398 | n = root->rb_node; | |
2399 | ||
2400 | while (n) { | |
2401 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2402 | WARN_ON(!entry->in_tree); | |
2403 | ||
2404 | middle = entry->bytenr; | |
2405 | ||
2406 | if (alt) | |
2407 | n = n->rb_left; | |
2408 | else | |
2409 | n = n->rb_right; | |
2410 | ||
2411 | alt = 1 - alt; | |
2412 | } | |
2413 | return middle; | |
2414 | } | |
2415 | #endif | |
2416 | ||
bed92eae AJ |
2417 | int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans, |
2418 | struct btrfs_fs_info *fs_info) | |
2419 | { | |
2420 | struct qgroup_update *qgroup_update; | |
2421 | int ret = 0; | |
2422 | ||
2423 | if (list_empty(&trans->qgroup_ref_list) != | |
2424 | !trans->delayed_ref_elem.seq) { | |
2425 | /* list without seq or seq without list */ | |
2426 | printk(KERN_ERR "btrfs: qgroup accounting update error, list is%s empty, seq is %llu\n", | |
2427 | list_empty(&trans->qgroup_ref_list) ? "" : " not", | |
2428 | trans->delayed_ref_elem.seq); | |
2429 | BUG(); | |
2430 | } | |
2431 | ||
2432 | if (!trans->delayed_ref_elem.seq) | |
2433 | return 0; | |
2434 | ||
2435 | while (!list_empty(&trans->qgroup_ref_list)) { | |
2436 | qgroup_update = list_first_entry(&trans->qgroup_ref_list, | |
2437 | struct qgroup_update, list); | |
2438 | list_del(&qgroup_update->list); | |
2439 | if (!ret) | |
2440 | ret = btrfs_qgroup_account_ref( | |
2441 | trans, fs_info, qgroup_update->node, | |
2442 | qgroup_update->extent_op); | |
2443 | kfree(qgroup_update); | |
2444 | } | |
2445 | ||
2446 | btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem); | |
2447 | ||
2448 | return ret; | |
2449 | } | |
2450 | ||
c3e69d58 CM |
2451 | /* |
2452 | * this starts processing the delayed reference count updates and | |
2453 | * extent insertions we have queued up so far. count can be | |
2454 | * 0, which means to process everything in the tree at the start | |
2455 | * of the run (but not newly added entries), or it can be some target | |
2456 | * number you'd like to process. | |
79787eaa JM |
2457 | * |
2458 | * Returns 0 on success or if called with an aborted transaction | |
2459 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2460 | */ |
2461 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2462 | struct btrfs_root *root, unsigned long count) | |
2463 | { | |
2464 | struct rb_node *node; | |
2465 | struct btrfs_delayed_ref_root *delayed_refs; | |
2466 | struct btrfs_delayed_ref_node *ref; | |
2467 | struct list_head cluster; | |
2468 | int ret; | |
a168650c | 2469 | u64 delayed_start; |
c3e69d58 CM |
2470 | int run_all = count == (unsigned long)-1; |
2471 | int run_most = 0; | |
1fa11e26 | 2472 | int loops; |
c3e69d58 | 2473 | |
79787eaa JM |
2474 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2475 | if (trans->aborted) | |
2476 | return 0; | |
2477 | ||
c3e69d58 CM |
2478 | if (root == root->fs_info->extent_root) |
2479 | root = root->fs_info->tree_root; | |
2480 | ||
203bf287 CM |
2481 | do_chunk_alloc(trans, root->fs_info->extent_root, |
2482 | 2 * 1024 * 1024, btrfs_get_alloc_profile(root, 0), | |
2483 | CHUNK_ALLOC_NO_FORCE); | |
2484 | ||
edf39272 JS |
2485 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); |
2486 | ||
c3e69d58 CM |
2487 | delayed_refs = &trans->transaction->delayed_refs; |
2488 | INIT_LIST_HEAD(&cluster); | |
2489 | again: | |
1fa11e26 | 2490 | loops = 0; |
c3e69d58 | 2491 | spin_lock(&delayed_refs->lock); |
097b8a7c | 2492 | |
709c0486 AJ |
2493 | #ifdef SCRAMBLE_DELAYED_REFS |
2494 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2495 | #endif | |
2496 | ||
c3e69d58 CM |
2497 | if (count == 0) { |
2498 | count = delayed_refs->num_entries * 2; | |
2499 | run_most = 1; | |
2500 | } | |
2501 | while (1) { | |
2502 | if (!(run_all || run_most) && | |
2503 | delayed_refs->num_heads_ready < 64) | |
2504 | break; | |
eb099670 | 2505 | |
56bec294 | 2506 | /* |
c3e69d58 CM |
2507 | * go find something we can process in the rbtree. We start at |
2508 | * the beginning of the tree, and then build a cluster | |
2509 | * of refs to process starting at the first one we are able to | |
2510 | * lock | |
56bec294 | 2511 | */ |
a168650c | 2512 | delayed_start = delayed_refs->run_delayed_start; |
c3e69d58 CM |
2513 | ret = btrfs_find_ref_cluster(trans, &cluster, |
2514 | delayed_refs->run_delayed_start); | |
2515 | if (ret) | |
56bec294 CM |
2516 | break; |
2517 | ||
c3e69d58 | 2518 | ret = run_clustered_refs(trans, root, &cluster); |
79787eaa JM |
2519 | if (ret < 0) { |
2520 | spin_unlock(&delayed_refs->lock); | |
2521 | btrfs_abort_transaction(trans, root, ret); | |
2522 | return ret; | |
2523 | } | |
c3e69d58 CM |
2524 | |
2525 | count -= min_t(unsigned long, ret, count); | |
2526 | ||
2527 | if (count == 0) | |
2528 | break; | |
a168650c | 2529 | |
1fa11e26 AJ |
2530 | if (delayed_start >= delayed_refs->run_delayed_start) { |
2531 | if (loops == 0) { | |
2532 | /* | |
2533 | * btrfs_find_ref_cluster looped. let's do one | |
2534 | * more cycle. if we don't run any delayed ref | |
2535 | * during that cycle (because we can't because | |
2536 | * all of them are blocked), bail out. | |
2537 | */ | |
2538 | loops = 1; | |
2539 | } else { | |
2540 | /* | |
2541 | * no runnable refs left, stop trying | |
2542 | */ | |
2543 | BUG_ON(run_all); | |
2544 | break; | |
2545 | } | |
2546 | } | |
2547 | if (ret) { | |
a168650c | 2548 | /* refs were run, let's reset staleness detection */ |
1fa11e26 | 2549 | loops = 0; |
a168650c | 2550 | } |
eb099670 | 2551 | } |
c3e69d58 | 2552 | |
56bec294 | 2553 | if (run_all) { |
56bec294 | 2554 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 2555 | if (!node) |
56bec294 | 2556 | goto out; |
c3e69d58 | 2557 | count = (unsigned long)-1; |
e9d0b13b | 2558 | |
56bec294 CM |
2559 | while (node) { |
2560 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2561 | rb_node); | |
2562 | if (btrfs_delayed_ref_is_head(ref)) { | |
2563 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 2564 | |
56bec294 CM |
2565 | head = btrfs_delayed_node_to_head(ref); |
2566 | atomic_inc(&ref->refs); | |
2567 | ||
2568 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2569 | /* |
2570 | * Mutex was contended, block until it's | |
2571 | * released and try again | |
2572 | */ | |
56bec294 CM |
2573 | mutex_lock(&head->mutex); |
2574 | mutex_unlock(&head->mutex); | |
2575 | ||
2576 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2577 | cond_resched(); |
56bec294 CM |
2578 | goto again; |
2579 | } | |
2580 | node = rb_next(node); | |
2581 | } | |
2582 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
2583 | schedule_timeout(1); |
2584 | goto again; | |
5f39d397 | 2585 | } |
54aa1f4d | 2586 | out: |
c3e69d58 | 2587 | spin_unlock(&delayed_refs->lock); |
edf39272 | 2588 | assert_qgroups_uptodate(trans); |
a28ec197 CM |
2589 | return 0; |
2590 | } | |
2591 | ||
5d4f98a2 YZ |
2592 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2593 | struct btrfs_root *root, | |
2594 | u64 bytenr, u64 num_bytes, u64 flags, | |
2595 | int is_data) | |
2596 | { | |
2597 | struct btrfs_delayed_extent_op *extent_op; | |
2598 | int ret; | |
2599 | ||
2600 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
2601 | if (!extent_op) | |
2602 | return -ENOMEM; | |
2603 | ||
2604 | extent_op->flags_to_set = flags; | |
2605 | extent_op->update_flags = 1; | |
2606 | extent_op->update_key = 0; | |
2607 | extent_op->is_data = is_data ? 1 : 0; | |
2608 | ||
66d7e7f0 AJ |
2609 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, |
2610 | num_bytes, extent_op); | |
5d4f98a2 YZ |
2611 | if (ret) |
2612 | kfree(extent_op); | |
2613 | return ret; | |
2614 | } | |
2615 | ||
2616 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2617 | struct btrfs_root *root, | |
2618 | struct btrfs_path *path, | |
2619 | u64 objectid, u64 offset, u64 bytenr) | |
2620 | { | |
2621 | struct btrfs_delayed_ref_head *head; | |
2622 | struct btrfs_delayed_ref_node *ref; | |
2623 | struct btrfs_delayed_data_ref *data_ref; | |
2624 | struct btrfs_delayed_ref_root *delayed_refs; | |
2625 | struct rb_node *node; | |
2626 | int ret = 0; | |
2627 | ||
2628 | ret = -ENOENT; | |
2629 | delayed_refs = &trans->transaction->delayed_refs; | |
2630 | spin_lock(&delayed_refs->lock); | |
2631 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2632 | if (!head) | |
2633 | goto out; | |
2634 | ||
2635 | if (!mutex_trylock(&head->mutex)) { | |
2636 | atomic_inc(&head->node.refs); | |
2637 | spin_unlock(&delayed_refs->lock); | |
2638 | ||
b3b4aa74 | 2639 | btrfs_release_path(path); |
5d4f98a2 | 2640 | |
8cc33e5c DS |
2641 | /* |
2642 | * Mutex was contended, block until it's released and let | |
2643 | * caller try again | |
2644 | */ | |
5d4f98a2 YZ |
2645 | mutex_lock(&head->mutex); |
2646 | mutex_unlock(&head->mutex); | |
2647 | btrfs_put_delayed_ref(&head->node); | |
2648 | return -EAGAIN; | |
2649 | } | |
2650 | ||
2651 | node = rb_prev(&head->node.rb_node); | |
2652 | if (!node) | |
2653 | goto out_unlock; | |
2654 | ||
2655 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2656 | ||
2657 | if (ref->bytenr != bytenr) | |
2658 | goto out_unlock; | |
2659 | ||
2660 | ret = 1; | |
2661 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2662 | goto out_unlock; | |
2663 | ||
2664 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2665 | ||
2666 | node = rb_prev(node); | |
2667 | if (node) { | |
df57dbe6 LB |
2668 | int seq = ref->seq; |
2669 | ||
5d4f98a2 | 2670 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); |
df57dbe6 | 2671 | if (ref->bytenr == bytenr && ref->seq == seq) |
5d4f98a2 YZ |
2672 | goto out_unlock; |
2673 | } | |
2674 | ||
2675 | if (data_ref->root != root->root_key.objectid || | |
2676 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2677 | goto out_unlock; | |
2678 | ||
2679 | ret = 0; | |
2680 | out_unlock: | |
2681 | mutex_unlock(&head->mutex); | |
2682 | out: | |
2683 | spin_unlock(&delayed_refs->lock); | |
2684 | return ret; | |
2685 | } | |
2686 | ||
2687 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2688 | struct btrfs_root *root, | |
2689 | struct btrfs_path *path, | |
2690 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2691 | { |
2692 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2693 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2694 | struct btrfs_extent_data_ref *ref; |
2695 | struct btrfs_extent_inline_ref *iref; | |
2696 | struct btrfs_extent_item *ei; | |
f321e491 | 2697 | struct btrfs_key key; |
5d4f98a2 | 2698 | u32 item_size; |
be20aa9d | 2699 | int ret; |
925baedd | 2700 | |
be20aa9d | 2701 | key.objectid = bytenr; |
31840ae1 | 2702 | key.offset = (u64)-1; |
f321e491 | 2703 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2704 | |
be20aa9d CM |
2705 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2706 | if (ret < 0) | |
2707 | goto out; | |
79787eaa | 2708 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2709 | |
2710 | ret = -ENOENT; | |
2711 | if (path->slots[0] == 0) | |
31840ae1 | 2712 | goto out; |
be20aa9d | 2713 | |
31840ae1 | 2714 | path->slots[0]--; |
f321e491 | 2715 | leaf = path->nodes[0]; |
5d4f98a2 | 2716 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2717 | |
5d4f98a2 | 2718 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2719 | goto out; |
f321e491 | 2720 | |
5d4f98a2 YZ |
2721 | ret = 1; |
2722 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2723 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2724 | if (item_size < sizeof(*ei)) { | |
2725 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2726 | goto out; | |
2727 | } | |
2728 | #endif | |
2729 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2730 | |
5d4f98a2 YZ |
2731 | if (item_size != sizeof(*ei) + |
2732 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2733 | goto out; | |
be20aa9d | 2734 | |
5d4f98a2 YZ |
2735 | if (btrfs_extent_generation(leaf, ei) <= |
2736 | btrfs_root_last_snapshot(&root->root_item)) | |
2737 | goto out; | |
2738 | ||
2739 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2740 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2741 | BTRFS_EXTENT_DATA_REF_KEY) | |
2742 | goto out; | |
2743 | ||
2744 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2745 | if (btrfs_extent_refs(leaf, ei) != | |
2746 | btrfs_extent_data_ref_count(leaf, ref) || | |
2747 | btrfs_extent_data_ref_root(leaf, ref) != | |
2748 | root->root_key.objectid || | |
2749 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2750 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2751 | goto out; | |
2752 | ||
2753 | ret = 0; | |
2754 | out: | |
2755 | return ret; | |
2756 | } | |
2757 | ||
2758 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2759 | struct btrfs_root *root, | |
2760 | u64 objectid, u64 offset, u64 bytenr) | |
2761 | { | |
2762 | struct btrfs_path *path; | |
2763 | int ret; | |
2764 | int ret2; | |
2765 | ||
2766 | path = btrfs_alloc_path(); | |
2767 | if (!path) | |
2768 | return -ENOENT; | |
2769 | ||
2770 | do { | |
2771 | ret = check_committed_ref(trans, root, path, objectid, | |
2772 | offset, bytenr); | |
2773 | if (ret && ret != -ENOENT) | |
f321e491 | 2774 | goto out; |
80ff3856 | 2775 | |
5d4f98a2 YZ |
2776 | ret2 = check_delayed_ref(trans, root, path, objectid, |
2777 | offset, bytenr); | |
2778 | } while (ret2 == -EAGAIN); | |
2779 | ||
2780 | if (ret2 && ret2 != -ENOENT) { | |
2781 | ret = ret2; | |
2782 | goto out; | |
f321e491 | 2783 | } |
5d4f98a2 YZ |
2784 | |
2785 | if (ret != -ENOENT || ret2 != -ENOENT) | |
2786 | ret = 0; | |
be20aa9d | 2787 | out: |
80ff3856 | 2788 | btrfs_free_path(path); |
f0486c68 YZ |
2789 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2790 | WARN_ON(ret > 0); | |
f321e491 | 2791 | return ret; |
be20aa9d | 2792 | } |
c5739bba | 2793 | |
5d4f98a2 | 2794 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2795 | struct btrfs_root *root, |
5d4f98a2 | 2796 | struct extent_buffer *buf, |
66d7e7f0 | 2797 | int full_backref, int inc, int for_cow) |
31840ae1 ZY |
2798 | { |
2799 | u64 bytenr; | |
5d4f98a2 YZ |
2800 | u64 num_bytes; |
2801 | u64 parent; | |
31840ae1 | 2802 | u64 ref_root; |
31840ae1 | 2803 | u32 nritems; |
31840ae1 ZY |
2804 | struct btrfs_key key; |
2805 | struct btrfs_file_extent_item *fi; | |
2806 | int i; | |
2807 | int level; | |
2808 | int ret = 0; | |
31840ae1 | 2809 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
66d7e7f0 | 2810 | u64, u64, u64, u64, u64, u64, int); |
31840ae1 ZY |
2811 | |
2812 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
2813 | nritems = btrfs_header_nritems(buf); |
2814 | level = btrfs_header_level(buf); | |
2815 | ||
5d4f98a2 YZ |
2816 | if (!root->ref_cows && level == 0) |
2817 | return 0; | |
31840ae1 | 2818 | |
5d4f98a2 YZ |
2819 | if (inc) |
2820 | process_func = btrfs_inc_extent_ref; | |
2821 | else | |
2822 | process_func = btrfs_free_extent; | |
31840ae1 | 2823 | |
5d4f98a2 YZ |
2824 | if (full_backref) |
2825 | parent = buf->start; | |
2826 | else | |
2827 | parent = 0; | |
2828 | ||
2829 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2830 | if (level == 0) { |
5d4f98a2 | 2831 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
2832 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
2833 | continue; | |
5d4f98a2 | 2834 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2835 | struct btrfs_file_extent_item); |
2836 | if (btrfs_file_extent_type(buf, fi) == | |
2837 | BTRFS_FILE_EXTENT_INLINE) | |
2838 | continue; | |
2839 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2840 | if (bytenr == 0) | |
2841 | continue; | |
5d4f98a2 YZ |
2842 | |
2843 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2844 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2845 | ret = process_func(trans, root, bytenr, num_bytes, | |
2846 | parent, ref_root, key.objectid, | |
66d7e7f0 | 2847 | key.offset, for_cow); |
31840ae1 ZY |
2848 | if (ret) |
2849 | goto fail; | |
2850 | } else { | |
5d4f98a2 YZ |
2851 | bytenr = btrfs_node_blockptr(buf, i); |
2852 | num_bytes = btrfs_level_size(root, level - 1); | |
2853 | ret = process_func(trans, root, bytenr, num_bytes, | |
66d7e7f0 AJ |
2854 | parent, ref_root, level - 1, 0, |
2855 | for_cow); | |
31840ae1 ZY |
2856 | if (ret) |
2857 | goto fail; | |
2858 | } | |
2859 | } | |
2860 | return 0; | |
2861 | fail: | |
5d4f98a2 YZ |
2862 | return ret; |
2863 | } | |
2864 | ||
2865 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 2866 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 2867 | { |
66d7e7f0 | 2868 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow); |
5d4f98a2 YZ |
2869 | } |
2870 | ||
2871 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 2872 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 2873 | { |
66d7e7f0 | 2874 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow); |
31840ae1 ZY |
2875 | } |
2876 | ||
9078a3e1 CM |
2877 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2878 | struct btrfs_root *root, | |
2879 | struct btrfs_path *path, | |
2880 | struct btrfs_block_group_cache *cache) | |
2881 | { | |
2882 | int ret; | |
9078a3e1 | 2883 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
2884 | unsigned long bi; |
2885 | struct extent_buffer *leaf; | |
9078a3e1 | 2886 | |
9078a3e1 | 2887 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
2888 | if (ret < 0) |
2889 | goto fail; | |
79787eaa | 2890 | BUG_ON(ret); /* Corruption */ |
5f39d397 CM |
2891 | |
2892 | leaf = path->nodes[0]; | |
2893 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2894 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
2895 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 2896 | btrfs_release_path(path); |
54aa1f4d | 2897 | fail: |
79787eaa JM |
2898 | if (ret) { |
2899 | btrfs_abort_transaction(trans, root, ret); | |
9078a3e1 | 2900 | return ret; |
79787eaa | 2901 | } |
9078a3e1 CM |
2902 | return 0; |
2903 | ||
2904 | } | |
2905 | ||
4a8c9a62 YZ |
2906 | static struct btrfs_block_group_cache * |
2907 | next_block_group(struct btrfs_root *root, | |
2908 | struct btrfs_block_group_cache *cache) | |
2909 | { | |
2910 | struct rb_node *node; | |
2911 | spin_lock(&root->fs_info->block_group_cache_lock); | |
2912 | node = rb_next(&cache->cache_node); | |
2913 | btrfs_put_block_group(cache); | |
2914 | if (node) { | |
2915 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
2916 | cache_node); | |
11dfe35a | 2917 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
2918 | } else |
2919 | cache = NULL; | |
2920 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
2921 | return cache; | |
2922 | } | |
2923 | ||
0af3d00b JB |
2924 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
2925 | struct btrfs_trans_handle *trans, | |
2926 | struct btrfs_path *path) | |
2927 | { | |
2928 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
2929 | struct inode *inode = NULL; | |
2930 | u64 alloc_hint = 0; | |
2b20982e | 2931 | int dcs = BTRFS_DC_ERROR; |
0af3d00b JB |
2932 | int num_pages = 0; |
2933 | int retries = 0; | |
2934 | int ret = 0; | |
2935 | ||
2936 | /* | |
2937 | * If this block group is smaller than 100 megs don't bother caching the | |
2938 | * block group. | |
2939 | */ | |
2940 | if (block_group->key.offset < (100 * 1024 * 1024)) { | |
2941 | spin_lock(&block_group->lock); | |
2942 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
2943 | spin_unlock(&block_group->lock); | |
2944 | return 0; | |
2945 | } | |
2946 | ||
2947 | again: | |
2948 | inode = lookup_free_space_inode(root, block_group, path); | |
2949 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
2950 | ret = PTR_ERR(inode); | |
b3b4aa74 | 2951 | btrfs_release_path(path); |
0af3d00b JB |
2952 | goto out; |
2953 | } | |
2954 | ||
2955 | if (IS_ERR(inode)) { | |
2956 | BUG_ON(retries); | |
2957 | retries++; | |
2958 | ||
2959 | if (block_group->ro) | |
2960 | goto out_free; | |
2961 | ||
2962 | ret = create_free_space_inode(root, trans, block_group, path); | |
2963 | if (ret) | |
2964 | goto out_free; | |
2965 | goto again; | |
2966 | } | |
2967 | ||
5b0e95bf JB |
2968 | /* We've already setup this transaction, go ahead and exit */ |
2969 | if (block_group->cache_generation == trans->transid && | |
2970 | i_size_read(inode)) { | |
2971 | dcs = BTRFS_DC_SETUP; | |
2972 | goto out_put; | |
2973 | } | |
2974 | ||
0af3d00b JB |
2975 | /* |
2976 | * We want to set the generation to 0, that way if anything goes wrong | |
2977 | * from here on out we know not to trust this cache when we load up next | |
2978 | * time. | |
2979 | */ | |
2980 | BTRFS_I(inode)->generation = 0; | |
2981 | ret = btrfs_update_inode(trans, root, inode); | |
2982 | WARN_ON(ret); | |
2983 | ||
2984 | if (i_size_read(inode) > 0) { | |
2985 | ret = btrfs_truncate_free_space_cache(root, trans, path, | |
2986 | inode); | |
2987 | if (ret) | |
2988 | goto out_put; | |
2989 | } | |
2990 | ||
2991 | spin_lock(&block_group->lock); | |
cf7c1ef6 LB |
2992 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
2993 | !btrfs_test_opt(root, SPACE_CACHE)) { | |
2994 | /* | |
2995 | * don't bother trying to write stuff out _if_ | |
2996 | * a) we're not cached, | |
2997 | * b) we're with nospace_cache mount option. | |
2998 | */ | |
2b20982e | 2999 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3000 | spin_unlock(&block_group->lock); |
3001 | goto out_put; | |
3002 | } | |
3003 | spin_unlock(&block_group->lock); | |
3004 | ||
6fc823b1 JB |
3005 | /* |
3006 | * Try to preallocate enough space based on how big the block group is. | |
3007 | * Keep in mind this has to include any pinned space which could end up | |
3008 | * taking up quite a bit since it's not folded into the other space | |
3009 | * cache. | |
3010 | */ | |
3011 | num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024); | |
0af3d00b JB |
3012 | if (!num_pages) |
3013 | num_pages = 1; | |
3014 | ||
0af3d00b JB |
3015 | num_pages *= 16; |
3016 | num_pages *= PAGE_CACHE_SIZE; | |
3017 | ||
3018 | ret = btrfs_check_data_free_space(inode, num_pages); | |
3019 | if (ret) | |
3020 | goto out_put; | |
3021 | ||
3022 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3023 | num_pages, num_pages, | |
3024 | &alloc_hint); | |
2b20982e JB |
3025 | if (!ret) |
3026 | dcs = BTRFS_DC_SETUP; | |
0af3d00b | 3027 | btrfs_free_reserved_data_space(inode, num_pages); |
c09544e0 | 3028 | |
0af3d00b JB |
3029 | out_put: |
3030 | iput(inode); | |
3031 | out_free: | |
b3b4aa74 | 3032 | btrfs_release_path(path); |
0af3d00b JB |
3033 | out: |
3034 | spin_lock(&block_group->lock); | |
e65cbb94 | 3035 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3036 | block_group->cache_generation = trans->transid; |
2b20982e | 3037 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3038 | spin_unlock(&block_group->lock); |
3039 | ||
3040 | return ret; | |
3041 | } | |
3042 | ||
96b5179d CM |
3043 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
3044 | struct btrfs_root *root) | |
9078a3e1 | 3045 | { |
4a8c9a62 | 3046 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 3047 | int err = 0; |
9078a3e1 | 3048 | struct btrfs_path *path; |
96b5179d | 3049 | u64 last = 0; |
9078a3e1 CM |
3050 | |
3051 | path = btrfs_alloc_path(); | |
3052 | if (!path) | |
3053 | return -ENOMEM; | |
3054 | ||
0af3d00b JB |
3055 | again: |
3056 | while (1) { | |
3057 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
3058 | while (cache) { | |
3059 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3060 | break; | |
3061 | cache = next_block_group(root, cache); | |
3062 | } | |
3063 | if (!cache) { | |
3064 | if (last == 0) | |
3065 | break; | |
3066 | last = 0; | |
3067 | continue; | |
3068 | } | |
3069 | err = cache_save_setup(cache, trans, path); | |
3070 | last = cache->key.objectid + cache->key.offset; | |
3071 | btrfs_put_block_group(cache); | |
3072 | } | |
3073 | ||
d397712b | 3074 | while (1) { |
4a8c9a62 YZ |
3075 | if (last == 0) { |
3076 | err = btrfs_run_delayed_refs(trans, root, | |
3077 | (unsigned long)-1); | |
79787eaa JM |
3078 | if (err) /* File system offline */ |
3079 | goto out; | |
0f9dd46c | 3080 | } |
54aa1f4d | 3081 | |
4a8c9a62 YZ |
3082 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
3083 | while (cache) { | |
0af3d00b JB |
3084 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) { |
3085 | btrfs_put_block_group(cache); | |
3086 | goto again; | |
3087 | } | |
3088 | ||
4a8c9a62 YZ |
3089 | if (cache->dirty) |
3090 | break; | |
3091 | cache = next_block_group(root, cache); | |
3092 | } | |
3093 | if (!cache) { | |
3094 | if (last == 0) | |
3095 | break; | |
3096 | last = 0; | |
3097 | continue; | |
3098 | } | |
0f9dd46c | 3099 | |
0cb59c99 JB |
3100 | if (cache->disk_cache_state == BTRFS_DC_SETUP) |
3101 | cache->disk_cache_state = BTRFS_DC_NEED_WRITE; | |
e8569813 | 3102 | cache->dirty = 0; |
4a8c9a62 | 3103 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 3104 | |
4a8c9a62 | 3105 | err = write_one_cache_group(trans, root, path, cache); |
79787eaa JM |
3106 | if (err) /* File system offline */ |
3107 | goto out; | |
3108 | ||
4a8c9a62 | 3109 | btrfs_put_block_group(cache); |
9078a3e1 | 3110 | } |
4a8c9a62 | 3111 | |
0cb59c99 JB |
3112 | while (1) { |
3113 | /* | |
3114 | * I don't think this is needed since we're just marking our | |
3115 | * preallocated extent as written, but just in case it can't | |
3116 | * hurt. | |
3117 | */ | |
3118 | if (last == 0) { | |
3119 | err = btrfs_run_delayed_refs(trans, root, | |
3120 | (unsigned long)-1); | |
79787eaa JM |
3121 | if (err) /* File system offline */ |
3122 | goto out; | |
0cb59c99 JB |
3123 | } |
3124 | ||
3125 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
3126 | while (cache) { | |
3127 | /* | |
3128 | * Really this shouldn't happen, but it could if we | |
3129 | * couldn't write the entire preallocated extent and | |
3130 | * splitting the extent resulted in a new block. | |
3131 | */ | |
3132 | if (cache->dirty) { | |
3133 | btrfs_put_block_group(cache); | |
3134 | goto again; | |
3135 | } | |
3136 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
3137 | break; | |
3138 | cache = next_block_group(root, cache); | |
3139 | } | |
3140 | if (!cache) { | |
3141 | if (last == 0) | |
3142 | break; | |
3143 | last = 0; | |
3144 | continue; | |
3145 | } | |
3146 | ||
79787eaa | 3147 | err = btrfs_write_out_cache(root, trans, cache, path); |
0cb59c99 JB |
3148 | |
3149 | /* | |
3150 | * If we didn't have an error then the cache state is still | |
3151 | * NEED_WRITE, so we can set it to WRITTEN. | |
3152 | */ | |
79787eaa | 3153 | if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE) |
0cb59c99 JB |
3154 | cache->disk_cache_state = BTRFS_DC_WRITTEN; |
3155 | last = cache->key.objectid + cache->key.offset; | |
3156 | btrfs_put_block_group(cache); | |
3157 | } | |
79787eaa | 3158 | out: |
0cb59c99 | 3159 | |
9078a3e1 | 3160 | btrfs_free_path(path); |
79787eaa | 3161 | return err; |
9078a3e1 CM |
3162 | } |
3163 | ||
d2fb3437 YZ |
3164 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
3165 | { | |
3166 | struct btrfs_block_group_cache *block_group; | |
3167 | int readonly = 0; | |
3168 | ||
3169 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
3170 | if (!block_group || block_group->ro) | |
3171 | readonly = 1; | |
3172 | if (block_group) | |
fa9c0d79 | 3173 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3174 | return readonly; |
3175 | } | |
3176 | ||
593060d7 CM |
3177 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3178 | u64 total_bytes, u64 bytes_used, | |
3179 | struct btrfs_space_info **space_info) | |
3180 | { | |
3181 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3182 | int i; |
3183 | int factor; | |
3184 | ||
3185 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3186 | BTRFS_BLOCK_GROUP_RAID10)) | |
3187 | factor = 2; | |
3188 | else | |
3189 | factor = 1; | |
593060d7 CM |
3190 | |
3191 | found = __find_space_info(info, flags); | |
3192 | if (found) { | |
25179201 | 3193 | spin_lock(&found->lock); |
593060d7 | 3194 | found->total_bytes += total_bytes; |
89a55897 | 3195 | found->disk_total += total_bytes * factor; |
593060d7 | 3196 | found->bytes_used += bytes_used; |
b742bb82 | 3197 | found->disk_used += bytes_used * factor; |
8f18cf13 | 3198 | found->full = 0; |
25179201 | 3199 | spin_unlock(&found->lock); |
593060d7 CM |
3200 | *space_info = found; |
3201 | return 0; | |
3202 | } | |
c146afad | 3203 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3204 | if (!found) |
3205 | return -ENOMEM; | |
3206 | ||
b742bb82 YZ |
3207 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
3208 | INIT_LIST_HEAD(&found->block_groups[i]); | |
80eb234a | 3209 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3210 | spin_lock_init(&found->lock); |
52ba6929 | 3211 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
593060d7 | 3212 | found->total_bytes = total_bytes; |
89a55897 | 3213 | found->disk_total = total_bytes * factor; |
593060d7 | 3214 | found->bytes_used = bytes_used; |
b742bb82 | 3215 | found->disk_used = bytes_used * factor; |
593060d7 | 3216 | found->bytes_pinned = 0; |
e8569813 | 3217 | found->bytes_reserved = 0; |
c146afad | 3218 | found->bytes_readonly = 0; |
f0486c68 | 3219 | found->bytes_may_use = 0; |
593060d7 | 3220 | found->full = 0; |
0e4f8f88 | 3221 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3222 | found->chunk_alloc = 0; |
fdb5effd JB |
3223 | found->flush = 0; |
3224 | init_waitqueue_head(&found->wait); | |
593060d7 | 3225 | *space_info = found; |
4184ea7f | 3226 | list_add_rcu(&found->list, &info->space_info); |
b4d7c3c9 LZ |
3227 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3228 | info->data_sinfo = found; | |
593060d7 CM |
3229 | return 0; |
3230 | } | |
3231 | ||
8790d502 CM |
3232 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3233 | { | |
899c81ea ID |
3234 | u64 extra_flags = chunk_to_extended(flags) & |
3235 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 ID |
3236 | |
3237 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3238 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3239 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3240 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3241 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3242 | fs_info->avail_system_alloc_bits |= extra_flags; | |
8790d502 | 3243 | } |
593060d7 | 3244 | |
fc67c450 ID |
3245 | /* |
3246 | * returns target flags in extended format or 0 if restripe for this | |
3247 | * chunk_type is not in progress | |
c6664b42 ID |
3248 | * |
3249 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
3250 | */ |
3251 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
3252 | { | |
3253 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3254 | u64 target = 0; | |
3255 | ||
fc67c450 ID |
3256 | if (!bctl) |
3257 | return 0; | |
3258 | ||
3259 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
3260 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3261 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3262 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
3263 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3264 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3265 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
3266 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3267 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3268 | } | |
3269 | ||
3270 | return target; | |
3271 | } | |
3272 | ||
a46d11a8 ID |
3273 | /* |
3274 | * @flags: available profiles in extended format (see ctree.h) | |
3275 | * | |
e4d8ec0f ID |
3276 | * Returns reduced profile in chunk format. If profile changing is in |
3277 | * progress (either running or paused) picks the target profile (if it's | |
3278 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 3279 | */ |
2b82032c | 3280 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 3281 | { |
cd02dca5 CM |
3282 | /* |
3283 | * we add in the count of missing devices because we want | |
3284 | * to make sure that any RAID levels on a degraded FS | |
3285 | * continue to be honored. | |
3286 | */ | |
3287 | u64 num_devices = root->fs_info->fs_devices->rw_devices + | |
3288 | root->fs_info->fs_devices->missing_devices; | |
fc67c450 | 3289 | u64 target; |
a061fc8d | 3290 | |
fc67c450 ID |
3291 | /* |
3292 | * see if restripe for this chunk_type is in progress, if so | |
3293 | * try to reduce to the target profile | |
3294 | */ | |
e4d8ec0f | 3295 | spin_lock(&root->fs_info->balance_lock); |
fc67c450 ID |
3296 | target = get_restripe_target(root->fs_info, flags); |
3297 | if (target) { | |
3298 | /* pick target profile only if it's already available */ | |
3299 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
e4d8ec0f | 3300 | spin_unlock(&root->fs_info->balance_lock); |
fc67c450 | 3301 | return extended_to_chunk(target); |
e4d8ec0f ID |
3302 | } |
3303 | } | |
3304 | spin_unlock(&root->fs_info->balance_lock); | |
3305 | ||
a061fc8d CM |
3306 | if (num_devices == 1) |
3307 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
3308 | if (num_devices < 4) | |
3309 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
3310 | ||
ec44a35c CM |
3311 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
3312 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 3313 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 3314 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 3315 | } |
ec44a35c CM |
3316 | |
3317 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 3318 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 3319 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 3320 | } |
ec44a35c CM |
3321 | |
3322 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
3323 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
3324 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
a46d11a8 | 3325 | (flags & BTRFS_BLOCK_GROUP_DUP))) { |
ec44a35c | 3326 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; |
a46d11a8 ID |
3327 | } |
3328 | ||
899c81ea | 3329 | return extended_to_chunk(flags); |
ec44a35c CM |
3330 | } |
3331 | ||
b742bb82 | 3332 | static u64 get_alloc_profile(struct btrfs_root *root, u64 flags) |
6a63209f | 3333 | { |
b742bb82 | 3334 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
6fef8df1 | 3335 | flags |= root->fs_info->avail_data_alloc_bits; |
b742bb82 | 3336 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
6fef8df1 | 3337 | flags |= root->fs_info->avail_system_alloc_bits; |
b742bb82 | 3338 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
6fef8df1 ID |
3339 | flags |= root->fs_info->avail_metadata_alloc_bits; |
3340 | ||
b742bb82 | 3341 | return btrfs_reduce_alloc_profile(root, flags); |
6a63209f JB |
3342 | } |
3343 | ||
6d07bcec | 3344 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 3345 | { |
b742bb82 | 3346 | u64 flags; |
9ed74f2d | 3347 | |
b742bb82 YZ |
3348 | if (data) |
3349 | flags = BTRFS_BLOCK_GROUP_DATA; | |
3350 | else if (root == root->fs_info->chunk_root) | |
3351 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 3352 | else |
b742bb82 | 3353 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3354 | |
b742bb82 | 3355 | return get_alloc_profile(root, flags); |
6a63209f | 3356 | } |
9ed74f2d | 3357 | |
6a63209f | 3358 | /* |
6a63209f JB |
3359 | * This will check the space that the inode allocates from to make sure we have |
3360 | * enough space for bytes. | |
6a63209f | 3361 | */ |
0ca1f7ce | 3362 | int btrfs_check_data_free_space(struct inode *inode, u64 bytes) |
6a63209f | 3363 | { |
6a63209f | 3364 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 3365 | struct btrfs_root *root = BTRFS_I(inode)->root; |
b4d7c3c9 | 3366 | struct btrfs_fs_info *fs_info = root->fs_info; |
ab6e2410 | 3367 | u64 used; |
0af3d00b | 3368 | int ret = 0, committed = 0, alloc_chunk = 1; |
6a63209f | 3369 | |
6a63209f JB |
3370 | /* make sure bytes are sectorsize aligned */ |
3371 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
6a63209f | 3372 | |
82d5902d LZ |
3373 | if (root == root->fs_info->tree_root || |
3374 | BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) { | |
0af3d00b JB |
3375 | alloc_chunk = 0; |
3376 | committed = 1; | |
3377 | } | |
3378 | ||
b4d7c3c9 | 3379 | data_sinfo = fs_info->data_sinfo; |
33b4d47f CM |
3380 | if (!data_sinfo) |
3381 | goto alloc; | |
9ed74f2d | 3382 | |
6a63209f JB |
3383 | again: |
3384 | /* make sure we have enough space to handle the data first */ | |
3385 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
3386 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
3387 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
3388 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
3389 | |
3390 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 3391 | struct btrfs_trans_handle *trans; |
9ed74f2d | 3392 | |
6a63209f JB |
3393 | /* |
3394 | * if we don't have enough free bytes in this space then we need | |
3395 | * to alloc a new chunk. | |
3396 | */ | |
0af3d00b | 3397 | if (!data_sinfo->full && alloc_chunk) { |
6a63209f | 3398 | u64 alloc_target; |
9ed74f2d | 3399 | |
0e4f8f88 | 3400 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 3401 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 3402 | alloc: |
6a63209f | 3403 | alloc_target = btrfs_get_alloc_profile(root, 1); |
7a7eaa40 | 3404 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3405 | if (IS_ERR(trans)) |
3406 | return PTR_ERR(trans); | |
9ed74f2d | 3407 | |
6a63209f JB |
3408 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
3409 | bytes + 2 * 1024 * 1024, | |
0e4f8f88 CM |
3410 | alloc_target, |
3411 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 3412 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
3413 | if (ret < 0) { |
3414 | if (ret != -ENOSPC) | |
3415 | return ret; | |
3416 | else | |
3417 | goto commit_trans; | |
3418 | } | |
9ed74f2d | 3419 | |
b4d7c3c9 LZ |
3420 | if (!data_sinfo) |
3421 | data_sinfo = fs_info->data_sinfo; | |
3422 | ||
6a63209f JB |
3423 | goto again; |
3424 | } | |
f2bb8f5c JB |
3425 | |
3426 | /* | |
3427 | * If we have less pinned bytes than we want to allocate then | |
3428 | * don't bother committing the transaction, it won't help us. | |
3429 | */ | |
3430 | if (data_sinfo->bytes_pinned < bytes) | |
3431 | committed = 1; | |
6a63209f | 3432 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3433 | |
4e06bdd6 | 3434 | /* commit the current transaction and try again */ |
d52a5b5f | 3435 | commit_trans: |
a4abeea4 JB |
3436 | if (!committed && |
3437 | !atomic_read(&root->fs_info->open_ioctl_trans)) { | |
4e06bdd6 | 3438 | committed = 1; |
7a7eaa40 | 3439 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3440 | if (IS_ERR(trans)) |
3441 | return PTR_ERR(trans); | |
4e06bdd6 JB |
3442 | ret = btrfs_commit_transaction(trans, root); |
3443 | if (ret) | |
3444 | return ret; | |
3445 | goto again; | |
3446 | } | |
9ed74f2d | 3447 | |
6a63209f JB |
3448 | return -ENOSPC; |
3449 | } | |
3450 | data_sinfo->bytes_may_use += bytes; | |
8c2a3ca2 | 3451 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3452 | data_sinfo->flags, bytes, 1); |
6a63209f | 3453 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3454 | |
9ed74f2d | 3455 | return 0; |
9ed74f2d | 3456 | } |
6a63209f | 3457 | |
6a63209f | 3458 | /* |
fb25e914 | 3459 | * Called if we need to clear a data reservation for this inode. |
6a63209f | 3460 | */ |
0ca1f7ce | 3461 | void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes) |
e3ccfa98 | 3462 | { |
0ca1f7ce | 3463 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6a63209f | 3464 | struct btrfs_space_info *data_sinfo; |
e3ccfa98 | 3465 | |
6a63209f JB |
3466 | /* make sure bytes are sectorsize aligned */ |
3467 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
e3ccfa98 | 3468 | |
b4d7c3c9 | 3469 | data_sinfo = root->fs_info->data_sinfo; |
6a63209f JB |
3470 | spin_lock(&data_sinfo->lock); |
3471 | data_sinfo->bytes_may_use -= bytes; | |
8c2a3ca2 | 3472 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3473 | data_sinfo->flags, bytes, 0); |
6a63209f | 3474 | spin_unlock(&data_sinfo->lock); |
e3ccfa98 JB |
3475 | } |
3476 | ||
97e728d4 | 3477 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 3478 | { |
97e728d4 JB |
3479 | struct list_head *head = &info->space_info; |
3480 | struct btrfs_space_info *found; | |
e3ccfa98 | 3481 | |
97e728d4 JB |
3482 | rcu_read_lock(); |
3483 | list_for_each_entry_rcu(found, head, list) { | |
3484 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 3485 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 3486 | } |
97e728d4 | 3487 | rcu_read_unlock(); |
e3ccfa98 JB |
3488 | } |
3489 | ||
e5bc2458 | 3490 | static int should_alloc_chunk(struct btrfs_root *root, |
0e4f8f88 CM |
3491 | struct btrfs_space_info *sinfo, u64 alloc_bytes, |
3492 | int force) | |
32c00aff | 3493 | { |
fb25e914 | 3494 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
424499db | 3495 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 3496 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 3497 | u64 thresh; |
e3ccfa98 | 3498 | |
0e4f8f88 CM |
3499 | if (force == CHUNK_ALLOC_FORCE) |
3500 | return 1; | |
3501 | ||
fb25e914 JB |
3502 | /* |
3503 | * We need to take into account the global rsv because for all intents | |
3504 | * and purposes it's used space. Don't worry about locking the | |
3505 | * global_rsv, it doesn't change except when the transaction commits. | |
3506 | */ | |
54338b5c JB |
3507 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
3508 | num_allocated += global_rsv->size; | |
fb25e914 | 3509 | |
0e4f8f88 CM |
3510 | /* |
3511 | * in limited mode, we want to have some free space up to | |
3512 | * about 1% of the FS size. | |
3513 | */ | |
3514 | if (force == CHUNK_ALLOC_LIMITED) { | |
6c41761f | 3515 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 CM |
3516 | thresh = max_t(u64, 64 * 1024 * 1024, |
3517 | div_factor_fine(thresh, 1)); | |
3518 | ||
3519 | if (num_bytes - num_allocated < thresh) | |
3520 | return 1; | |
3521 | } | |
0e4f8f88 | 3522 | |
54338b5c | 3523 | if (num_allocated + alloc_bytes < div_factor(num_bytes, 8)) |
14ed0ca6 | 3524 | return 0; |
424499db | 3525 | return 1; |
32c00aff JB |
3526 | } |
3527 | ||
15d1ff81 LB |
3528 | static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type) |
3529 | { | |
3530 | u64 num_dev; | |
3531 | ||
3532 | if (type & BTRFS_BLOCK_GROUP_RAID10 || | |
3533 | type & BTRFS_BLOCK_GROUP_RAID0) | |
3534 | num_dev = root->fs_info->fs_devices->rw_devices; | |
3535 | else if (type & BTRFS_BLOCK_GROUP_RAID1) | |
3536 | num_dev = 2; | |
3537 | else | |
3538 | num_dev = 1; /* DUP or single */ | |
3539 | ||
3540 | /* metadata for updaing devices and chunk tree */ | |
3541 | return btrfs_calc_trans_metadata_size(root, num_dev + 1); | |
3542 | } | |
3543 | ||
3544 | static void check_system_chunk(struct btrfs_trans_handle *trans, | |
3545 | struct btrfs_root *root, u64 type) | |
3546 | { | |
3547 | struct btrfs_space_info *info; | |
3548 | u64 left; | |
3549 | u64 thresh; | |
3550 | ||
3551 | info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
3552 | spin_lock(&info->lock); | |
3553 | left = info->total_bytes - info->bytes_used - info->bytes_pinned - | |
3554 | info->bytes_reserved - info->bytes_readonly; | |
3555 | spin_unlock(&info->lock); | |
3556 | ||
3557 | thresh = get_system_chunk_thresh(root, type); | |
3558 | if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
3559 | printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n", | |
3560 | left, thresh, type); | |
3561 | dump_space_info(info, 0, 0); | |
3562 | } | |
3563 | ||
3564 | if (left < thresh) { | |
3565 | u64 flags; | |
3566 | ||
3567 | flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0); | |
3568 | btrfs_alloc_chunk(trans, root, flags); | |
3569 | } | |
3570 | } | |
3571 | ||
6324fbf3 CM |
3572 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
3573 | struct btrfs_root *extent_root, u64 alloc_bytes, | |
0ef3e66b | 3574 | u64 flags, int force) |
9ed74f2d | 3575 | { |
6324fbf3 | 3576 | struct btrfs_space_info *space_info; |
97e728d4 | 3577 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 3578 | int wait_for_alloc = 0; |
9ed74f2d | 3579 | int ret = 0; |
9ed74f2d | 3580 | |
6324fbf3 | 3581 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
3582 | if (!space_info) { |
3583 | ret = update_space_info(extent_root->fs_info, flags, | |
3584 | 0, 0, &space_info); | |
79787eaa | 3585 | BUG_ON(ret); /* -ENOMEM */ |
9ed74f2d | 3586 | } |
79787eaa | 3587 | BUG_ON(!space_info); /* Logic error */ |
9ed74f2d | 3588 | |
6d74119f | 3589 | again: |
25179201 | 3590 | spin_lock(&space_info->lock); |
9e622d6b | 3591 | if (force < space_info->force_alloc) |
0e4f8f88 | 3592 | force = space_info->force_alloc; |
25179201 JB |
3593 | if (space_info->full) { |
3594 | spin_unlock(&space_info->lock); | |
6d74119f | 3595 | return 0; |
9ed74f2d JB |
3596 | } |
3597 | ||
0e4f8f88 | 3598 | if (!should_alloc_chunk(extent_root, space_info, alloc_bytes, force)) { |
25179201 | 3599 | spin_unlock(&space_info->lock); |
6d74119f JB |
3600 | return 0; |
3601 | } else if (space_info->chunk_alloc) { | |
3602 | wait_for_alloc = 1; | |
3603 | } else { | |
3604 | space_info->chunk_alloc = 1; | |
9ed74f2d | 3605 | } |
0e4f8f88 | 3606 | |
25179201 | 3607 | spin_unlock(&space_info->lock); |
9ed74f2d | 3608 | |
6d74119f JB |
3609 | mutex_lock(&fs_info->chunk_mutex); |
3610 | ||
3611 | /* | |
3612 | * The chunk_mutex is held throughout the entirety of a chunk | |
3613 | * allocation, so once we've acquired the chunk_mutex we know that the | |
3614 | * other guy is done and we need to recheck and see if we should | |
3615 | * allocate. | |
3616 | */ | |
3617 | if (wait_for_alloc) { | |
3618 | mutex_unlock(&fs_info->chunk_mutex); | |
3619 | wait_for_alloc = 0; | |
3620 | goto again; | |
3621 | } | |
3622 | ||
67377734 JB |
3623 | /* |
3624 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3625 | * allocating mixed chunks instead of individual chunks. | |
3626 | */ | |
3627 | if (btrfs_mixed_space_info(space_info)) | |
3628 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3629 | ||
97e728d4 JB |
3630 | /* |
3631 | * if we're doing a data chunk, go ahead and make sure that | |
3632 | * we keep a reasonable number of metadata chunks allocated in the | |
3633 | * FS as well. | |
3634 | */ | |
9ed74f2d | 3635 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
3636 | fs_info->data_chunk_allocations++; |
3637 | if (!(fs_info->data_chunk_allocations % | |
3638 | fs_info->metadata_ratio)) | |
3639 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
3640 | } |
3641 | ||
15d1ff81 LB |
3642 | /* |
3643 | * Check if we have enough space in SYSTEM chunk because we may need | |
3644 | * to update devices. | |
3645 | */ | |
3646 | check_system_chunk(trans, extent_root, flags); | |
3647 | ||
2b82032c | 3648 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
92b8e897 MF |
3649 | if (ret < 0 && ret != -ENOSPC) |
3650 | goto out; | |
3651 | ||
9ed74f2d | 3652 | spin_lock(&space_info->lock); |
9ed74f2d | 3653 | if (ret) |
6324fbf3 | 3654 | space_info->full = 1; |
424499db YZ |
3655 | else |
3656 | ret = 1; | |
6d74119f | 3657 | |
0e4f8f88 | 3658 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3659 | space_info->chunk_alloc = 0; |
9ed74f2d | 3660 | spin_unlock(&space_info->lock); |
92b8e897 | 3661 | out: |
a25c75d5 | 3662 | mutex_unlock(&fs_info->chunk_mutex); |
0f9dd46c | 3663 | return ret; |
6324fbf3 | 3664 | } |
9ed74f2d | 3665 | |
9ed74f2d | 3666 | /* |
5da9d01b | 3667 | * shrink metadata reservation for delalloc |
9ed74f2d | 3668 | */ |
f4c738c2 JB |
3669 | static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig, |
3670 | bool wait_ordered) | |
5da9d01b | 3671 | { |
0ca1f7ce | 3672 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 3673 | struct btrfs_space_info *space_info; |
663350ac | 3674 | struct btrfs_trans_handle *trans; |
f4c738c2 | 3675 | u64 delalloc_bytes; |
5da9d01b | 3676 | u64 max_reclaim; |
b1953bce | 3677 | long time_left; |
877da174 | 3678 | unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT; |
b1953bce | 3679 | int loops = 0; |
5da9d01b | 3680 | |
663350ac | 3681 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0ca1f7ce | 3682 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 3683 | space_info = block_rsv->space_info; |
bf9022e0 CM |
3684 | |
3685 | smp_mb(); | |
f4c738c2 JB |
3686 | delalloc_bytes = root->fs_info->delalloc_bytes; |
3687 | if (delalloc_bytes == 0) { | |
fdb5effd | 3688 | if (trans) |
f4c738c2 | 3689 | return; |
fdb5effd | 3690 | btrfs_wait_ordered_extents(root, 0, 0); |
f4c738c2 | 3691 | return; |
fdb5effd JB |
3692 | } |
3693 | ||
f4c738c2 JB |
3694 | while (delalloc_bytes && loops < 3) { |
3695 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
3696 | nr_pages = max_reclaim >> PAGE_CACHE_SHIFT; | |
0e175a18 | 3697 | writeback_inodes_sb_nr_if_idle(root->fs_info->sb, nr_pages, |
f4c738c2 | 3698 | WB_REASON_FS_FREE_SPACE); |
5da9d01b | 3699 | |
0019f10d | 3700 | spin_lock(&space_info->lock); |
f4c738c2 JB |
3701 | if (space_info->bytes_used + space_info->bytes_reserved + |
3702 | space_info->bytes_pinned + space_info->bytes_readonly + | |
3703 | space_info->bytes_may_use + orig <= | |
3704 | space_info->total_bytes) { | |
3705 | spin_unlock(&space_info->lock); | |
3706 | break; | |
3707 | } | |
0019f10d | 3708 | spin_unlock(&space_info->lock); |
5da9d01b | 3709 | |
36e39c40 | 3710 | loops++; |
f104d044 JB |
3711 | if (wait_ordered && !trans) { |
3712 | btrfs_wait_ordered_extents(root, 0, 0); | |
3713 | } else { | |
f4c738c2 | 3714 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
3715 | if (time_left) |
3716 | break; | |
3717 | } | |
f4c738c2 JB |
3718 | smp_mb(); |
3719 | delalloc_bytes = root->fs_info->delalloc_bytes; | |
5da9d01b | 3720 | } |
5da9d01b YZ |
3721 | } |
3722 | ||
663350ac JB |
3723 | /** |
3724 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
3725 | * @root - the root we're allocating for | |
3726 | * @bytes - the number of bytes we want to reserve | |
3727 | * @force - force the commit | |
8bb8ab2e | 3728 | * |
663350ac JB |
3729 | * This will check to make sure that committing the transaction will actually |
3730 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
3731 | * will return -ENOSPC. | |
8bb8ab2e | 3732 | */ |
663350ac JB |
3733 | static int may_commit_transaction(struct btrfs_root *root, |
3734 | struct btrfs_space_info *space_info, | |
3735 | u64 bytes, int force) | |
3736 | { | |
3737 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; | |
3738 | struct btrfs_trans_handle *trans; | |
3739 | ||
3740 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
3741 | if (trans) | |
3742 | return -EAGAIN; | |
3743 | ||
3744 | if (force) | |
3745 | goto commit; | |
3746 | ||
3747 | /* See if there is enough pinned space to make this reservation */ | |
3748 | spin_lock(&space_info->lock); | |
3749 | if (space_info->bytes_pinned >= bytes) { | |
3750 | spin_unlock(&space_info->lock); | |
3751 | goto commit; | |
3752 | } | |
3753 | spin_unlock(&space_info->lock); | |
3754 | ||
3755 | /* | |
3756 | * See if there is some space in the delayed insertion reservation for | |
3757 | * this reservation. | |
3758 | */ | |
3759 | if (space_info != delayed_rsv->space_info) | |
3760 | return -ENOSPC; | |
3761 | ||
d9b0218f | 3762 | spin_lock(&space_info->lock); |
663350ac | 3763 | spin_lock(&delayed_rsv->lock); |
d9b0218f | 3764 | if (space_info->bytes_pinned + delayed_rsv->size < bytes) { |
663350ac | 3765 | spin_unlock(&delayed_rsv->lock); |
d9b0218f | 3766 | spin_unlock(&space_info->lock); |
663350ac JB |
3767 | return -ENOSPC; |
3768 | } | |
3769 | spin_unlock(&delayed_rsv->lock); | |
d9b0218f | 3770 | spin_unlock(&space_info->lock); |
663350ac JB |
3771 | |
3772 | commit: | |
3773 | trans = btrfs_join_transaction(root); | |
3774 | if (IS_ERR(trans)) | |
3775 | return -ENOSPC; | |
3776 | ||
3777 | return btrfs_commit_transaction(trans, root); | |
3778 | } | |
3779 | ||
96c3f433 JB |
3780 | enum flush_state { |
3781 | FLUSH_DELALLOC = 1, | |
3782 | FLUSH_DELALLOC_WAIT = 2, | |
3783 | FLUSH_DELAYED_ITEMS_NR = 3, | |
3784 | FLUSH_DELAYED_ITEMS = 4, | |
3785 | COMMIT_TRANS = 5, | |
3786 | }; | |
3787 | ||
3788 | static int flush_space(struct btrfs_root *root, | |
3789 | struct btrfs_space_info *space_info, u64 num_bytes, | |
3790 | u64 orig_bytes, int state) | |
3791 | { | |
3792 | struct btrfs_trans_handle *trans; | |
3793 | int nr; | |
f4c738c2 | 3794 | int ret = 0; |
96c3f433 JB |
3795 | |
3796 | switch (state) { | |
3797 | case FLUSH_DELALLOC: | |
3798 | case FLUSH_DELALLOC_WAIT: | |
f4c738c2 JB |
3799 | shrink_delalloc(root, num_bytes, orig_bytes, |
3800 | state == FLUSH_DELALLOC_WAIT); | |
96c3f433 JB |
3801 | break; |
3802 | case FLUSH_DELAYED_ITEMS_NR: | |
3803 | case FLUSH_DELAYED_ITEMS: | |
3804 | if (state == FLUSH_DELAYED_ITEMS_NR) { | |
3805 | u64 bytes = btrfs_calc_trans_metadata_size(root, 1); | |
3806 | ||
3807 | nr = (int)div64_u64(num_bytes, bytes); | |
3808 | if (!nr) | |
3809 | nr = 1; | |
3810 | nr *= 2; | |
3811 | } else { | |
3812 | nr = -1; | |
3813 | } | |
3814 | trans = btrfs_join_transaction(root); | |
3815 | if (IS_ERR(trans)) { | |
3816 | ret = PTR_ERR(trans); | |
3817 | break; | |
3818 | } | |
3819 | ret = btrfs_run_delayed_items_nr(trans, root, nr); | |
3820 | btrfs_end_transaction(trans, root); | |
3821 | break; | |
3822 | case COMMIT_TRANS: | |
3823 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); | |
3824 | break; | |
3825 | default: | |
3826 | ret = -ENOSPC; | |
3827 | break; | |
3828 | } | |
3829 | ||
3830 | return ret; | |
3831 | } | |
4a92b1b8 JB |
3832 | /** |
3833 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
3834 | * @root - the root we're allocating for | |
3835 | * @block_rsv - the block_rsv we're allocating for | |
3836 | * @orig_bytes - the number of bytes we want | |
3837 | * @flush - wether or not we can flush to make our reservation | |
8bb8ab2e | 3838 | * |
4a92b1b8 JB |
3839 | * This will reserve orgi_bytes number of bytes from the space info associated |
3840 | * with the block_rsv. If there is not enough space it will make an attempt to | |
3841 | * flush out space to make room. It will do this by flushing delalloc if | |
3842 | * possible or committing the transaction. If flush is 0 then no attempts to | |
3843 | * regain reservations will be made and this will fail if there is not enough | |
3844 | * space already. | |
8bb8ab2e | 3845 | */ |
4a92b1b8 | 3846 | static int reserve_metadata_bytes(struct btrfs_root *root, |
8bb8ab2e JB |
3847 | struct btrfs_block_rsv *block_rsv, |
3848 | u64 orig_bytes, int flush) | |
9ed74f2d | 3849 | { |
f0486c68 | 3850 | struct btrfs_space_info *space_info = block_rsv->space_info; |
2bf64758 | 3851 | u64 used; |
8bb8ab2e | 3852 | u64 num_bytes = orig_bytes; |
96c3f433 | 3853 | int flush_state = FLUSH_DELALLOC; |
8bb8ab2e | 3854 | int ret = 0; |
fdb5effd | 3855 | bool flushing = false; |
96c3f433 | 3856 | bool committed = false; |
9ed74f2d | 3857 | |
8bb8ab2e | 3858 | again: |
fdb5effd | 3859 | ret = 0; |
8bb8ab2e | 3860 | spin_lock(&space_info->lock); |
fdb5effd JB |
3861 | /* |
3862 | * We only want to wait if somebody other than us is flushing and we are | |
3863 | * actually alloed to flush. | |
3864 | */ | |
3865 | while (flush && !flushing && space_info->flush) { | |
3866 | spin_unlock(&space_info->lock); | |
3867 | /* | |
3868 | * If we have a trans handle we can't wait because the flusher | |
3869 | * may have to commit the transaction, which would mean we would | |
3870 | * deadlock since we are waiting for the flusher to finish, but | |
3871 | * hold the current transaction open. | |
3872 | */ | |
663350ac | 3873 | if (current->journal_info) |
fdb5effd | 3874 | return -EAGAIN; |
b9688bb8 AJ |
3875 | ret = wait_event_killable(space_info->wait, !space_info->flush); |
3876 | /* Must have been killed, return */ | |
3877 | if (ret) | |
fdb5effd JB |
3878 | return -EINTR; |
3879 | ||
3880 | spin_lock(&space_info->lock); | |
3881 | } | |
3882 | ||
3883 | ret = -ENOSPC; | |
2bf64758 JB |
3884 | used = space_info->bytes_used + space_info->bytes_reserved + |
3885 | space_info->bytes_pinned + space_info->bytes_readonly + | |
3886 | space_info->bytes_may_use; | |
9ed74f2d | 3887 | |
8bb8ab2e JB |
3888 | /* |
3889 | * The idea here is that we've not already over-reserved the block group | |
3890 | * then we can go ahead and save our reservation first and then start | |
3891 | * flushing if we need to. Otherwise if we've already overcommitted | |
3892 | * lets start flushing stuff first and then come back and try to make | |
3893 | * our reservation. | |
3894 | */ | |
2bf64758 JB |
3895 | if (used <= space_info->total_bytes) { |
3896 | if (used + orig_bytes <= space_info->total_bytes) { | |
fb25e914 | 3897 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 3898 | trace_btrfs_space_reservation(root->fs_info, |
2bcc0328 | 3899 | "space_info", space_info->flags, orig_bytes, 1); |
8bb8ab2e JB |
3900 | ret = 0; |
3901 | } else { | |
3902 | /* | |
3903 | * Ok set num_bytes to orig_bytes since we aren't | |
3904 | * overocmmitted, this way we only try and reclaim what | |
3905 | * we need. | |
3906 | */ | |
3907 | num_bytes = orig_bytes; | |
3908 | } | |
3909 | } else { | |
3910 | /* | |
3911 | * Ok we're over committed, set num_bytes to the overcommitted | |
3912 | * amount plus the amount of bytes that we need for this | |
3913 | * reservation. | |
3914 | */ | |
2bf64758 | 3915 | num_bytes = used - space_info->total_bytes + |
96c3f433 | 3916 | (orig_bytes * 2); |
8bb8ab2e | 3917 | } |
9ed74f2d | 3918 | |
36ba022a | 3919 | if (ret) { |
2bf64758 JB |
3920 | u64 profile = btrfs_get_alloc_profile(root, 0); |
3921 | u64 avail; | |
3922 | ||
7e355b83 JB |
3923 | /* |
3924 | * If we have a lot of space that's pinned, don't bother doing | |
3925 | * the overcommit dance yet and just commit the transaction. | |
3926 | */ | |
3927 | avail = (space_info->total_bytes - space_info->bytes_used) * 8; | |
3928 | do_div(avail, 10); | |
663350ac | 3929 | if (space_info->bytes_pinned >= avail && flush && !committed) { |
7e355b83 JB |
3930 | space_info->flush = 1; |
3931 | flushing = true; | |
3932 | spin_unlock(&space_info->lock); | |
663350ac JB |
3933 | ret = may_commit_transaction(root, space_info, |
3934 | orig_bytes, 1); | |
3935 | if (ret) | |
3936 | goto out; | |
3937 | committed = true; | |
3938 | goto again; | |
7e355b83 JB |
3939 | } |
3940 | ||
2bf64758 JB |
3941 | spin_lock(&root->fs_info->free_chunk_lock); |
3942 | avail = root->fs_info->free_chunk_space; | |
3943 | ||
3944 | /* | |
3945 | * If we have dup, raid1 or raid10 then only half of the free | |
3946 | * space is actually useable. | |
3947 | */ | |
3948 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
3949 | BTRFS_BLOCK_GROUP_RAID1 | | |
3950 | BTRFS_BLOCK_GROUP_RAID10)) | |
3951 | avail >>= 1; | |
3952 | ||
3953 | /* | |
3954 | * If we aren't flushing don't let us overcommit too much, say | |
3955 | * 1/8th of the space. If we can flush, let it overcommit up to | |
3956 | * 1/2 of the space. | |
3957 | */ | |
3958 | if (flush) | |
3959 | avail >>= 3; | |
3960 | else | |
3961 | avail >>= 1; | |
3962 | spin_unlock(&root->fs_info->free_chunk_lock); | |
3963 | ||
9a82ca65 | 3964 | if (used + num_bytes < space_info->total_bytes + avail) { |
2bf64758 | 3965 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 3966 | trace_btrfs_space_reservation(root->fs_info, |
2bcc0328 | 3967 | "space_info", space_info->flags, orig_bytes, 1); |
2bf64758 JB |
3968 | ret = 0; |
3969 | } | |
3970 | } | |
3971 | ||
8bb8ab2e JB |
3972 | /* |
3973 | * Couldn't make our reservation, save our place so while we're trying | |
3974 | * to reclaim space we can actually use it instead of somebody else | |
3975 | * stealing it from us. | |
3976 | */ | |
fdb5effd JB |
3977 | if (ret && flush) { |
3978 | flushing = true; | |
3979 | space_info->flush = 1; | |
8bb8ab2e | 3980 | } |
9ed74f2d | 3981 | |
f0486c68 | 3982 | spin_unlock(&space_info->lock); |
9ed74f2d | 3983 | |
fdb5effd | 3984 | if (!ret || !flush) |
8bb8ab2e | 3985 | goto out; |
f0486c68 | 3986 | |
96c3f433 JB |
3987 | ret = flush_space(root, space_info, num_bytes, orig_bytes, |
3988 | flush_state); | |
3989 | flush_state++; | |
3990 | if (!ret) | |
8bb8ab2e | 3991 | goto again; |
96c3f433 | 3992 | else if (flush_state <= COMMIT_TRANS) |
8bb8ab2e JB |
3993 | goto again; |
3994 | ||
3995 | out: | |
fdb5effd | 3996 | if (flushing) { |
8bb8ab2e | 3997 | spin_lock(&space_info->lock); |
fdb5effd JB |
3998 | space_info->flush = 0; |
3999 | wake_up_all(&space_info->wait); | |
8bb8ab2e | 4000 | spin_unlock(&space_info->lock); |
f0486c68 | 4001 | } |
f0486c68 YZ |
4002 | return ret; |
4003 | } | |
4004 | ||
79787eaa JM |
4005 | static struct btrfs_block_rsv *get_block_rsv( |
4006 | const struct btrfs_trans_handle *trans, | |
4007 | const struct btrfs_root *root) | |
f0486c68 | 4008 | { |
4c13d758 JB |
4009 | struct btrfs_block_rsv *block_rsv = NULL; |
4010 | ||
0e721106 JB |
4011 | if (root->ref_cows) |
4012 | block_rsv = trans->block_rsv; | |
4013 | ||
4014 | if (root == root->fs_info->csum_root && trans->adding_csums) | |
f0486c68 | 4015 | block_rsv = trans->block_rsv; |
4c13d758 JB |
4016 | |
4017 | if (!block_rsv) | |
f0486c68 YZ |
4018 | block_rsv = root->block_rsv; |
4019 | ||
4020 | if (!block_rsv) | |
4021 | block_rsv = &root->fs_info->empty_block_rsv; | |
4022 | ||
4023 | return block_rsv; | |
4024 | } | |
4025 | ||
4026 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
4027 | u64 num_bytes) | |
4028 | { | |
4029 | int ret = -ENOSPC; | |
4030 | spin_lock(&block_rsv->lock); | |
4031 | if (block_rsv->reserved >= num_bytes) { | |
4032 | block_rsv->reserved -= num_bytes; | |
4033 | if (block_rsv->reserved < block_rsv->size) | |
4034 | block_rsv->full = 0; | |
4035 | ret = 0; | |
4036 | } | |
4037 | spin_unlock(&block_rsv->lock); | |
4038 | return ret; | |
4039 | } | |
4040 | ||
4041 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
4042 | u64 num_bytes, int update_size) | |
4043 | { | |
4044 | spin_lock(&block_rsv->lock); | |
4045 | block_rsv->reserved += num_bytes; | |
4046 | if (update_size) | |
4047 | block_rsv->size += num_bytes; | |
4048 | else if (block_rsv->reserved >= block_rsv->size) | |
4049 | block_rsv->full = 1; | |
4050 | spin_unlock(&block_rsv->lock); | |
4051 | } | |
4052 | ||
8c2a3ca2 JB |
4053 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
4054 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 4055 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
4056 | { |
4057 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
4058 | ||
4059 | spin_lock(&block_rsv->lock); | |
4060 | if (num_bytes == (u64)-1) | |
4061 | num_bytes = block_rsv->size; | |
4062 | block_rsv->size -= num_bytes; | |
4063 | if (block_rsv->reserved >= block_rsv->size) { | |
4064 | num_bytes = block_rsv->reserved - block_rsv->size; | |
4065 | block_rsv->reserved = block_rsv->size; | |
4066 | block_rsv->full = 1; | |
4067 | } else { | |
4068 | num_bytes = 0; | |
4069 | } | |
4070 | spin_unlock(&block_rsv->lock); | |
4071 | ||
4072 | if (num_bytes > 0) { | |
4073 | if (dest) { | |
e9e22899 JB |
4074 | spin_lock(&dest->lock); |
4075 | if (!dest->full) { | |
4076 | u64 bytes_to_add; | |
4077 | ||
4078 | bytes_to_add = dest->size - dest->reserved; | |
4079 | bytes_to_add = min(num_bytes, bytes_to_add); | |
4080 | dest->reserved += bytes_to_add; | |
4081 | if (dest->reserved >= dest->size) | |
4082 | dest->full = 1; | |
4083 | num_bytes -= bytes_to_add; | |
4084 | } | |
4085 | spin_unlock(&dest->lock); | |
4086 | } | |
4087 | if (num_bytes) { | |
f0486c68 | 4088 | spin_lock(&space_info->lock); |
fb25e914 | 4089 | space_info->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4090 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4091 | space_info->flags, num_bytes, 0); |
36e39c40 | 4092 | space_info->reservation_progress++; |
f0486c68 | 4093 | spin_unlock(&space_info->lock); |
4e06bdd6 | 4094 | } |
9ed74f2d | 4095 | } |
f0486c68 | 4096 | } |
4e06bdd6 | 4097 | |
f0486c68 YZ |
4098 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
4099 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
4100 | { | |
4101 | int ret; | |
9ed74f2d | 4102 | |
f0486c68 YZ |
4103 | ret = block_rsv_use_bytes(src, num_bytes); |
4104 | if (ret) | |
4105 | return ret; | |
9ed74f2d | 4106 | |
f0486c68 | 4107 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
4108 | return 0; |
4109 | } | |
4110 | ||
f0486c68 | 4111 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv) |
9ed74f2d | 4112 | { |
f0486c68 YZ |
4113 | memset(rsv, 0, sizeof(*rsv)); |
4114 | spin_lock_init(&rsv->lock); | |
f0486c68 YZ |
4115 | } |
4116 | ||
4117 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root) | |
4118 | { | |
4119 | struct btrfs_block_rsv *block_rsv; | |
4120 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 4121 | |
f0486c68 YZ |
4122 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
4123 | if (!block_rsv) | |
4124 | return NULL; | |
9ed74f2d | 4125 | |
f0486c68 | 4126 | btrfs_init_block_rsv(block_rsv); |
f0486c68 YZ |
4127 | block_rsv->space_info = __find_space_info(fs_info, |
4128 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
4129 | return block_rsv; |
4130 | } | |
9ed74f2d | 4131 | |
f0486c68 YZ |
4132 | void btrfs_free_block_rsv(struct btrfs_root *root, |
4133 | struct btrfs_block_rsv *rsv) | |
4134 | { | |
2aaa6655 JB |
4135 | if (!rsv) |
4136 | return; | |
dabdb640 JB |
4137 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
4138 | kfree(rsv); | |
9ed74f2d JB |
4139 | } |
4140 | ||
61b520a9 MX |
4141 | static inline int __block_rsv_add(struct btrfs_root *root, |
4142 | struct btrfs_block_rsv *block_rsv, | |
4143 | u64 num_bytes, int flush) | |
9ed74f2d | 4144 | { |
f0486c68 | 4145 | int ret; |
9ed74f2d | 4146 | |
f0486c68 YZ |
4147 | if (num_bytes == 0) |
4148 | return 0; | |
8bb8ab2e | 4149 | |
61b520a9 | 4150 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
4151 | if (!ret) { |
4152 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
4153 | return 0; | |
4154 | } | |
9ed74f2d | 4155 | |
f0486c68 | 4156 | return ret; |
f0486c68 | 4157 | } |
9ed74f2d | 4158 | |
61b520a9 MX |
4159 | int btrfs_block_rsv_add(struct btrfs_root *root, |
4160 | struct btrfs_block_rsv *block_rsv, | |
4161 | u64 num_bytes) | |
4162 | { | |
4163 | return __block_rsv_add(root, block_rsv, num_bytes, 1); | |
4164 | } | |
4165 | ||
c06a0e12 JB |
4166 | int btrfs_block_rsv_add_noflush(struct btrfs_root *root, |
4167 | struct btrfs_block_rsv *block_rsv, | |
4168 | u64 num_bytes) | |
f0486c68 | 4169 | { |
61b520a9 | 4170 | return __block_rsv_add(root, block_rsv, num_bytes, 0); |
f0486c68 | 4171 | } |
9ed74f2d | 4172 | |
4a92b1b8 | 4173 | int btrfs_block_rsv_check(struct btrfs_root *root, |
36ba022a | 4174 | struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
4175 | { |
4176 | u64 num_bytes = 0; | |
f0486c68 | 4177 | int ret = -ENOSPC; |
9ed74f2d | 4178 | |
f0486c68 YZ |
4179 | if (!block_rsv) |
4180 | return 0; | |
9ed74f2d | 4181 | |
f0486c68 | 4182 | spin_lock(&block_rsv->lock); |
36ba022a JB |
4183 | num_bytes = div_factor(block_rsv->size, min_factor); |
4184 | if (block_rsv->reserved >= num_bytes) | |
4185 | ret = 0; | |
4186 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 4187 | |
36ba022a JB |
4188 | return ret; |
4189 | } | |
4190 | ||
aa38a711 MX |
4191 | static inline int __btrfs_block_rsv_refill(struct btrfs_root *root, |
4192 | struct btrfs_block_rsv *block_rsv, | |
4193 | u64 min_reserved, int flush) | |
36ba022a JB |
4194 | { |
4195 | u64 num_bytes = 0; | |
4196 | int ret = -ENOSPC; | |
4197 | ||
4198 | if (!block_rsv) | |
4199 | return 0; | |
4200 | ||
4201 | spin_lock(&block_rsv->lock); | |
4202 | num_bytes = min_reserved; | |
13553e52 | 4203 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 4204 | ret = 0; |
13553e52 | 4205 | else |
f0486c68 | 4206 | num_bytes -= block_rsv->reserved; |
f0486c68 | 4207 | spin_unlock(&block_rsv->lock); |
13553e52 | 4208 | |
f0486c68 YZ |
4209 | if (!ret) |
4210 | return 0; | |
4211 | ||
aa38a711 | 4212 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
4213 | if (!ret) { |
4214 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 4215 | return 0; |
6a63209f | 4216 | } |
9ed74f2d | 4217 | |
13553e52 | 4218 | return ret; |
f0486c68 YZ |
4219 | } |
4220 | ||
aa38a711 MX |
4221 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
4222 | struct btrfs_block_rsv *block_rsv, | |
4223 | u64 min_reserved) | |
4224 | { | |
4225 | return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 1); | |
4226 | } | |
4227 | ||
4228 | int btrfs_block_rsv_refill_noflush(struct btrfs_root *root, | |
4229 | struct btrfs_block_rsv *block_rsv, | |
4230 | u64 min_reserved) | |
4231 | { | |
4232 | return __btrfs_block_rsv_refill(root, block_rsv, min_reserved, 0); | |
4233 | } | |
4234 | ||
f0486c68 YZ |
4235 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, |
4236 | struct btrfs_block_rsv *dst_rsv, | |
4237 | u64 num_bytes) | |
4238 | { | |
4239 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
4240 | } | |
4241 | ||
4242 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
4243 | struct btrfs_block_rsv *block_rsv, | |
4244 | u64 num_bytes) | |
4245 | { | |
4246 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
4247 | if (global_rsv->full || global_rsv == block_rsv || | |
4248 | block_rsv->space_info != global_rsv->space_info) | |
4249 | global_rsv = NULL; | |
8c2a3ca2 JB |
4250 | block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv, |
4251 | num_bytes); | |
6a63209f JB |
4252 | } |
4253 | ||
4254 | /* | |
8929ecfa YZ |
4255 | * helper to calculate size of global block reservation. |
4256 | * the desired value is sum of space used by extent tree, | |
4257 | * checksum tree and root tree | |
6a63209f | 4258 | */ |
8929ecfa | 4259 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 4260 | { |
8929ecfa YZ |
4261 | struct btrfs_space_info *sinfo; |
4262 | u64 num_bytes; | |
4263 | u64 meta_used; | |
4264 | u64 data_used; | |
6c41761f | 4265 | int csum_size = btrfs_super_csum_size(fs_info->super_copy); |
6a63209f | 4266 | |
8929ecfa YZ |
4267 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
4268 | spin_lock(&sinfo->lock); | |
4269 | data_used = sinfo->bytes_used; | |
4270 | spin_unlock(&sinfo->lock); | |
33b4d47f | 4271 | |
8929ecfa YZ |
4272 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
4273 | spin_lock(&sinfo->lock); | |
6d48755d JB |
4274 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
4275 | data_used = 0; | |
8929ecfa YZ |
4276 | meta_used = sinfo->bytes_used; |
4277 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 4278 | |
8929ecfa YZ |
4279 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
4280 | csum_size * 2; | |
4281 | num_bytes += div64_u64(data_used + meta_used, 50); | |
4e06bdd6 | 4282 | |
8929ecfa | 4283 | if (num_bytes * 3 > meta_used) |
8e62c2de | 4284 | num_bytes = div64_u64(meta_used, 3); |
ab6e2410 | 4285 | |
8929ecfa YZ |
4286 | return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10); |
4287 | } | |
6a63209f | 4288 | |
8929ecfa YZ |
4289 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
4290 | { | |
4291 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
4292 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
4293 | u64 num_bytes; | |
6a63209f | 4294 | |
8929ecfa | 4295 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 4296 | |
8929ecfa | 4297 | spin_lock(&sinfo->lock); |
1f699d38 | 4298 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 4299 | |
8929ecfa | 4300 | block_rsv->size = num_bytes; |
4e06bdd6 | 4301 | |
8929ecfa | 4302 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
6d48755d JB |
4303 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
4304 | sinfo->bytes_may_use; | |
8929ecfa YZ |
4305 | |
4306 | if (sinfo->total_bytes > num_bytes) { | |
4307 | num_bytes = sinfo->total_bytes - num_bytes; | |
4308 | block_rsv->reserved += num_bytes; | |
fb25e914 | 4309 | sinfo->bytes_may_use += num_bytes; |
8c2a3ca2 | 4310 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4311 | sinfo->flags, num_bytes, 1); |
6a63209f | 4312 | } |
6a63209f | 4313 | |
8929ecfa YZ |
4314 | if (block_rsv->reserved >= block_rsv->size) { |
4315 | num_bytes = block_rsv->reserved - block_rsv->size; | |
fb25e914 | 4316 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4317 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4318 | sinfo->flags, num_bytes, 0); |
36e39c40 | 4319 | sinfo->reservation_progress++; |
8929ecfa YZ |
4320 | block_rsv->reserved = block_rsv->size; |
4321 | block_rsv->full = 1; | |
4322 | } | |
182608c8 | 4323 | |
8929ecfa | 4324 | spin_unlock(&block_rsv->lock); |
1f699d38 | 4325 | spin_unlock(&sinfo->lock); |
6a63209f JB |
4326 | } |
4327 | ||
f0486c68 | 4328 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4329 | { |
f0486c68 | 4330 | struct btrfs_space_info *space_info; |
6a63209f | 4331 | |
f0486c68 YZ |
4332 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
4333 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 4334 | |
f0486c68 | 4335 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 4336 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 4337 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
4338 | fs_info->trans_block_rsv.space_info = space_info; |
4339 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 4340 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 4341 | |
8929ecfa YZ |
4342 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
4343 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
4344 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
4345 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 4346 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 4347 | |
8929ecfa | 4348 | update_global_block_rsv(fs_info); |
6a63209f JB |
4349 | } |
4350 | ||
8929ecfa | 4351 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4352 | { |
8c2a3ca2 JB |
4353 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
4354 | (u64)-1); | |
8929ecfa YZ |
4355 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
4356 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
4357 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
4358 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
4359 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
4360 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
4361 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
4362 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
4363 | } |
4364 | ||
a22285a6 YZ |
4365 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
4366 | struct btrfs_root *root) | |
6a63209f | 4367 | { |
0e721106 JB |
4368 | if (!trans->block_rsv) |
4369 | return; | |
4370 | ||
a22285a6 YZ |
4371 | if (!trans->bytes_reserved) |
4372 | return; | |
6a63209f | 4373 | |
e77266e4 | 4374 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 4375 | trans->transid, trans->bytes_reserved, 0); |
b24e03db | 4376 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); |
a22285a6 YZ |
4377 | trans->bytes_reserved = 0; |
4378 | } | |
6a63209f | 4379 | |
79787eaa | 4380 | /* Can only return 0 or -ENOSPC */ |
d68fc57b YZ |
4381 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
4382 | struct inode *inode) | |
4383 | { | |
4384 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4385 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4386 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
4387 | ||
4388 | /* | |
fcb80c2a JB |
4389 | * We need to hold space in order to delete our orphan item once we've |
4390 | * added it, so this takes the reservation so we can release it later | |
4391 | * when we are truly done with the orphan item. | |
d68fc57b | 4392 | */ |
ff5714cc | 4393 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4394 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4395 | btrfs_ino(inode), num_bytes, 1); | |
d68fc57b | 4396 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
6a63209f JB |
4397 | } |
4398 | ||
d68fc57b | 4399 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 4400 | { |
d68fc57b | 4401 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 4402 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4403 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4404 | btrfs_ino(inode), num_bytes, 0); | |
d68fc57b YZ |
4405 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
4406 | } | |
97e728d4 | 4407 | |
a22285a6 YZ |
4408 | int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans, |
4409 | struct btrfs_pending_snapshot *pending) | |
4410 | { | |
4411 | struct btrfs_root *root = pending->root; | |
4412 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4413 | struct btrfs_block_rsv *dst_rsv = &pending->block_rsv; | |
4414 | /* | |
4415 | * two for root back/forward refs, two for directory entries | |
4416 | * and one for root of the snapshot. | |
4417 | */ | |
16cdcec7 | 4418 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 5); |
a22285a6 YZ |
4419 | dst_rsv->space_info = src_rsv->space_info; |
4420 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
97e728d4 JB |
4421 | } |
4422 | ||
7709cde3 JB |
4423 | /** |
4424 | * drop_outstanding_extent - drop an outstanding extent | |
4425 | * @inode: the inode we're dropping the extent for | |
4426 | * | |
4427 | * This is called when we are freeing up an outstanding extent, either called | |
4428 | * after an error or after an extent is written. This will return the number of | |
4429 | * reserved extents that need to be freed. This must be called with | |
4430 | * BTRFS_I(inode)->lock held. | |
4431 | */ | |
9e0baf60 JB |
4432 | static unsigned drop_outstanding_extent(struct inode *inode) |
4433 | { | |
7fd2ae21 | 4434 | unsigned drop_inode_space = 0; |
9e0baf60 JB |
4435 | unsigned dropped_extents = 0; |
4436 | ||
9e0baf60 JB |
4437 | BUG_ON(!BTRFS_I(inode)->outstanding_extents); |
4438 | BTRFS_I(inode)->outstanding_extents--; | |
4439 | ||
7fd2ae21 | 4440 | if (BTRFS_I(inode)->outstanding_extents == 0 && |
72ac3c0d JB |
4441 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4442 | &BTRFS_I(inode)->runtime_flags)) | |
7fd2ae21 | 4443 | drop_inode_space = 1; |
7fd2ae21 | 4444 | |
9e0baf60 JB |
4445 | /* |
4446 | * If we have more or the same amount of outsanding extents than we have | |
4447 | * reserved then we need to leave the reserved extents count alone. | |
4448 | */ | |
4449 | if (BTRFS_I(inode)->outstanding_extents >= | |
4450 | BTRFS_I(inode)->reserved_extents) | |
7fd2ae21 | 4451 | return drop_inode_space; |
9e0baf60 JB |
4452 | |
4453 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
4454 | BTRFS_I(inode)->outstanding_extents; | |
4455 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
7fd2ae21 | 4456 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
4457 | } |
4458 | ||
7709cde3 JB |
4459 | /** |
4460 | * calc_csum_metadata_size - return the amount of metada space that must be | |
4461 | * reserved/free'd for the given bytes. | |
4462 | * @inode: the inode we're manipulating | |
4463 | * @num_bytes: the number of bytes in question | |
4464 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
4465 | * | |
4466 | * This adjusts the number of csum_bytes in the inode and then returns the | |
4467 | * correct amount of metadata that must either be reserved or freed. We | |
4468 | * calculate how many checksums we can fit into one leaf and then divide the | |
4469 | * number of bytes that will need to be checksumed by this value to figure out | |
4470 | * how many checksums will be required. If we are adding bytes then the number | |
4471 | * may go up and we will return the number of additional bytes that must be | |
4472 | * reserved. If it is going down we will return the number of bytes that must | |
4473 | * be freed. | |
4474 | * | |
4475 | * This must be called with BTRFS_I(inode)->lock held. | |
4476 | */ | |
4477 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, | |
4478 | int reserve) | |
6324fbf3 | 4479 | { |
7709cde3 JB |
4480 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4481 | u64 csum_size; | |
4482 | int num_csums_per_leaf; | |
4483 | int num_csums; | |
4484 | int old_csums; | |
4485 | ||
4486 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && | |
4487 | BTRFS_I(inode)->csum_bytes == 0) | |
4488 | return 0; | |
4489 | ||
4490 | old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4491 | if (reserve) | |
4492 | BTRFS_I(inode)->csum_bytes += num_bytes; | |
4493 | else | |
4494 | BTRFS_I(inode)->csum_bytes -= num_bytes; | |
4495 | csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item); | |
4496 | num_csums_per_leaf = (int)div64_u64(csum_size, | |
4497 | sizeof(struct btrfs_csum_item) + | |
4498 | sizeof(struct btrfs_disk_key)); | |
4499 | num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4500 | num_csums = num_csums + num_csums_per_leaf - 1; | |
4501 | num_csums = num_csums / num_csums_per_leaf; | |
4502 | ||
4503 | old_csums = old_csums + num_csums_per_leaf - 1; | |
4504 | old_csums = old_csums / num_csums_per_leaf; | |
4505 | ||
4506 | /* No change, no need to reserve more */ | |
4507 | if (old_csums == num_csums) | |
4508 | return 0; | |
4509 | ||
4510 | if (reserve) | |
4511 | return btrfs_calc_trans_metadata_size(root, | |
4512 | num_csums - old_csums); | |
4513 | ||
4514 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); | |
0ca1f7ce | 4515 | } |
c146afad | 4516 | |
0ca1f7ce YZ |
4517 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
4518 | { | |
4519 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4520 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 | 4521 | u64 to_reserve = 0; |
660d3f6c | 4522 | u64 csum_bytes; |
9e0baf60 | 4523 | unsigned nr_extents = 0; |
660d3f6c | 4524 | int extra_reserve = 0; |
c09544e0 | 4525 | int flush = 1; |
0ca1f7ce | 4526 | int ret; |
6324fbf3 | 4527 | |
660d3f6c | 4528 | /* Need to be holding the i_mutex here if we aren't free space cache */ |
83eea1f1 | 4529 | if (btrfs_is_free_space_inode(inode)) |
c09544e0 JB |
4530 | flush = 0; |
4531 | ||
4532 | if (flush && btrfs_transaction_in_commit(root->fs_info)) | |
0ca1f7ce | 4533 | schedule_timeout(1); |
ec44a35c | 4534 | |
f248679e | 4535 | mutex_lock(&BTRFS_I(inode)->delalloc_mutex); |
0ca1f7ce | 4536 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 4537 | |
9e0baf60 JB |
4538 | spin_lock(&BTRFS_I(inode)->lock); |
4539 | BTRFS_I(inode)->outstanding_extents++; | |
4540 | ||
4541 | if (BTRFS_I(inode)->outstanding_extents > | |
660d3f6c | 4542 | BTRFS_I(inode)->reserved_extents) |
9e0baf60 JB |
4543 | nr_extents = BTRFS_I(inode)->outstanding_extents - |
4544 | BTRFS_I(inode)->reserved_extents; | |
57a45ced | 4545 | |
7fd2ae21 JB |
4546 | /* |
4547 | * Add an item to reserve for updating the inode when we complete the | |
4548 | * delalloc io. | |
4549 | */ | |
72ac3c0d JB |
4550 | if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4551 | &BTRFS_I(inode)->runtime_flags)) { | |
7fd2ae21 | 4552 | nr_extents++; |
660d3f6c | 4553 | extra_reserve = 1; |
593060d7 | 4554 | } |
7fd2ae21 JB |
4555 | |
4556 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); | |
7709cde3 | 4557 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
660d3f6c | 4558 | csum_bytes = BTRFS_I(inode)->csum_bytes; |
9e0baf60 | 4559 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 4560 | |
c5567237 AJ |
4561 | if (root->fs_info->quota_enabled) { |
4562 | ret = btrfs_qgroup_reserve(root, num_bytes + | |
4563 | nr_extents * root->leafsize); | |
55e591ff DC |
4564 | if (ret) { |
4565 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
c5567237 | 4566 | return ret; |
55e591ff | 4567 | } |
c5567237 AJ |
4568 | } |
4569 | ||
36ba022a | 4570 | ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush); |
9e0baf60 | 4571 | if (ret) { |
7ed49f18 | 4572 | u64 to_free = 0; |
9e0baf60 | 4573 | unsigned dropped; |
7ed49f18 | 4574 | |
7709cde3 | 4575 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 4576 | dropped = drop_outstanding_extent(inode); |
9e0baf60 | 4577 | /* |
660d3f6c JB |
4578 | * If the inodes csum_bytes is the same as the original |
4579 | * csum_bytes then we know we haven't raced with any free()ers | |
4580 | * so we can just reduce our inodes csum bytes and carry on. | |
4581 | * Otherwise we have to do the normal free thing to account for | |
4582 | * the case that the free side didn't free up its reserve | |
4583 | * because of this outstanding reservation. | |
9e0baf60 | 4584 | */ |
660d3f6c JB |
4585 | if (BTRFS_I(inode)->csum_bytes == csum_bytes) |
4586 | calc_csum_metadata_size(inode, num_bytes, 0); | |
4587 | else | |
4588 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
4589 | spin_unlock(&BTRFS_I(inode)->lock); | |
4590 | if (dropped) | |
4591 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
4592 | ||
8c2a3ca2 | 4593 | if (to_free) { |
7ed49f18 | 4594 | btrfs_block_rsv_release(root, block_rsv, to_free); |
8c2a3ca2 JB |
4595 | trace_btrfs_space_reservation(root->fs_info, |
4596 | "delalloc", | |
4597 | btrfs_ino(inode), | |
4598 | to_free, 0); | |
4599 | } | |
f248679e | 4600 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); |
0ca1f7ce | 4601 | return ret; |
9e0baf60 | 4602 | } |
25179201 | 4603 | |
660d3f6c JB |
4604 | spin_lock(&BTRFS_I(inode)->lock); |
4605 | if (extra_reserve) { | |
72ac3c0d JB |
4606 | set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4607 | &BTRFS_I(inode)->runtime_flags); | |
660d3f6c JB |
4608 | nr_extents--; |
4609 | } | |
4610 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
4611 | spin_unlock(&BTRFS_I(inode)->lock); | |
f248679e | 4612 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); |
660d3f6c | 4613 | |
8c2a3ca2 JB |
4614 | if (to_reserve) |
4615 | trace_btrfs_space_reservation(root->fs_info,"delalloc", | |
4616 | btrfs_ino(inode), to_reserve, 1); | |
0ca1f7ce YZ |
4617 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
4618 | ||
0ca1f7ce YZ |
4619 | return 0; |
4620 | } | |
4621 | ||
7709cde3 JB |
4622 | /** |
4623 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
4624 | * @inode: the inode to release the reservation for | |
4625 | * @num_bytes: the number of bytes we're releasing | |
4626 | * | |
4627 | * This will release the metadata reservation for an inode. This can be called | |
4628 | * once we complete IO for a given set of bytes to release their metadata | |
4629 | * reservations. | |
4630 | */ | |
0ca1f7ce YZ |
4631 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) |
4632 | { | |
4633 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
4634 | u64 to_free = 0; |
4635 | unsigned dropped; | |
0ca1f7ce YZ |
4636 | |
4637 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
7709cde3 | 4638 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 4639 | dropped = drop_outstanding_extent(inode); |
97e728d4 | 4640 | |
7709cde3 JB |
4641 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); |
4642 | spin_unlock(&BTRFS_I(inode)->lock); | |
9e0baf60 JB |
4643 | if (dropped > 0) |
4644 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce | 4645 | |
8c2a3ca2 JB |
4646 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
4647 | btrfs_ino(inode), to_free, 0); | |
c5567237 AJ |
4648 | if (root->fs_info->quota_enabled) { |
4649 | btrfs_qgroup_free(root, num_bytes + | |
4650 | dropped * root->leafsize); | |
4651 | } | |
4652 | ||
0ca1f7ce YZ |
4653 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, |
4654 | to_free); | |
4655 | } | |
4656 | ||
7709cde3 JB |
4657 | /** |
4658 | * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc | |
4659 | * @inode: inode we're writing to | |
4660 | * @num_bytes: the number of bytes we want to allocate | |
4661 | * | |
4662 | * This will do the following things | |
4663 | * | |
4664 | * o reserve space in the data space info for num_bytes | |
4665 | * o reserve space in the metadata space info based on number of outstanding | |
4666 | * extents and how much csums will be needed | |
4667 | * o add to the inodes ->delalloc_bytes | |
4668 | * o add it to the fs_info's delalloc inodes list. | |
4669 | * | |
4670 | * This will return 0 for success and -ENOSPC if there is no space left. | |
4671 | */ | |
0ca1f7ce YZ |
4672 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes) |
4673 | { | |
4674 | int ret; | |
4675 | ||
4676 | ret = btrfs_check_data_free_space(inode, num_bytes); | |
d397712b | 4677 | if (ret) |
0ca1f7ce YZ |
4678 | return ret; |
4679 | ||
4680 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes); | |
4681 | if (ret) { | |
4682 | btrfs_free_reserved_data_space(inode, num_bytes); | |
4683 | return ret; | |
4684 | } | |
4685 | ||
4686 | return 0; | |
4687 | } | |
4688 | ||
7709cde3 JB |
4689 | /** |
4690 | * btrfs_delalloc_release_space - release data and metadata space for delalloc | |
4691 | * @inode: inode we're releasing space for | |
4692 | * @num_bytes: the number of bytes we want to free up | |
4693 | * | |
4694 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
4695 | * called in the case that we don't need the metadata AND data reservations | |
4696 | * anymore. So if there is an error or we insert an inline extent. | |
4697 | * | |
4698 | * This function will release the metadata space that was not used and will | |
4699 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
4700 | * list if there are no delalloc bytes left. | |
4701 | */ | |
0ca1f7ce YZ |
4702 | void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes) |
4703 | { | |
4704 | btrfs_delalloc_release_metadata(inode, num_bytes); | |
4705 | btrfs_free_reserved_data_space(inode, num_bytes); | |
6324fbf3 CM |
4706 | } |
4707 | ||
9078a3e1 CM |
4708 | static int update_block_group(struct btrfs_trans_handle *trans, |
4709 | struct btrfs_root *root, | |
f0486c68 | 4710 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 4711 | { |
0af3d00b | 4712 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 4713 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 4714 | u64 total = num_bytes; |
9078a3e1 | 4715 | u64 old_val; |
db94535d | 4716 | u64 byte_in_group; |
0af3d00b | 4717 | int factor; |
3e1ad54f | 4718 | |
5d4f98a2 YZ |
4719 | /* block accounting for super block */ |
4720 | spin_lock(&info->delalloc_lock); | |
6c41761f | 4721 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
4722 | if (alloc) |
4723 | old_val += num_bytes; | |
4724 | else | |
4725 | old_val -= num_bytes; | |
6c41761f | 4726 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
5d4f98a2 YZ |
4727 | spin_unlock(&info->delalloc_lock); |
4728 | ||
d397712b | 4729 | while (total) { |
db94535d | 4730 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 4731 | if (!cache) |
79787eaa | 4732 | return -ENOENT; |
b742bb82 YZ |
4733 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
4734 | BTRFS_BLOCK_GROUP_RAID1 | | |
4735 | BTRFS_BLOCK_GROUP_RAID10)) | |
4736 | factor = 2; | |
4737 | else | |
4738 | factor = 1; | |
9d66e233 JB |
4739 | /* |
4740 | * If this block group has free space cache written out, we | |
4741 | * need to make sure to load it if we are removing space. This | |
4742 | * is because we need the unpinning stage to actually add the | |
4743 | * space back to the block group, otherwise we will leak space. | |
4744 | */ | |
4745 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
b8399dee | 4746 | cache_block_group(cache, trans, NULL, 1); |
0af3d00b | 4747 | |
db94535d CM |
4748 | byte_in_group = bytenr - cache->key.objectid; |
4749 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 4750 | |
25179201 | 4751 | spin_lock(&cache->space_info->lock); |
c286ac48 | 4752 | spin_lock(&cache->lock); |
0af3d00b | 4753 | |
73bc1876 | 4754 | if (btrfs_test_opt(root, SPACE_CACHE) && |
0af3d00b JB |
4755 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
4756 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
4757 | ||
0f9dd46c | 4758 | cache->dirty = 1; |
9078a3e1 | 4759 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 4760 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 4761 | if (alloc) { |
db94535d | 4762 | old_val += num_bytes; |
11833d66 YZ |
4763 | btrfs_set_block_group_used(&cache->item, old_val); |
4764 | cache->reserved -= num_bytes; | |
11833d66 | 4765 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
4766 | cache->space_info->bytes_used += num_bytes; |
4767 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 4768 | spin_unlock(&cache->lock); |
25179201 | 4769 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 4770 | } else { |
db94535d | 4771 | old_val -= num_bytes; |
c286ac48 | 4772 | btrfs_set_block_group_used(&cache->item, old_val); |
f0486c68 YZ |
4773 | cache->pinned += num_bytes; |
4774 | cache->space_info->bytes_pinned += num_bytes; | |
6324fbf3 | 4775 | cache->space_info->bytes_used -= num_bytes; |
b742bb82 | 4776 | cache->space_info->disk_used -= num_bytes * factor; |
c286ac48 | 4777 | spin_unlock(&cache->lock); |
25179201 | 4778 | spin_unlock(&cache->space_info->lock); |
1f3c79a2 | 4779 | |
f0486c68 YZ |
4780 | set_extent_dirty(info->pinned_extents, |
4781 | bytenr, bytenr + num_bytes - 1, | |
4782 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 4783 | } |
fa9c0d79 | 4784 | btrfs_put_block_group(cache); |
db94535d CM |
4785 | total -= num_bytes; |
4786 | bytenr += num_bytes; | |
9078a3e1 CM |
4787 | } |
4788 | return 0; | |
4789 | } | |
6324fbf3 | 4790 | |
a061fc8d CM |
4791 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
4792 | { | |
0f9dd46c | 4793 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 4794 | u64 bytenr; |
0f9dd46c JB |
4795 | |
4796 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); | |
4797 | if (!cache) | |
a061fc8d | 4798 | return 0; |
0f9dd46c | 4799 | |
d2fb3437 | 4800 | bytenr = cache->key.objectid; |
fa9c0d79 | 4801 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
4802 | |
4803 | return bytenr; | |
a061fc8d CM |
4804 | } |
4805 | ||
f0486c68 YZ |
4806 | static int pin_down_extent(struct btrfs_root *root, |
4807 | struct btrfs_block_group_cache *cache, | |
4808 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 4809 | { |
11833d66 YZ |
4810 | spin_lock(&cache->space_info->lock); |
4811 | spin_lock(&cache->lock); | |
4812 | cache->pinned += num_bytes; | |
4813 | cache->space_info->bytes_pinned += num_bytes; | |
4814 | if (reserved) { | |
4815 | cache->reserved -= num_bytes; | |
4816 | cache->space_info->bytes_reserved -= num_bytes; | |
4817 | } | |
4818 | spin_unlock(&cache->lock); | |
4819 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 4820 | |
f0486c68 YZ |
4821 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
4822 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
4823 | return 0; | |
4824 | } | |
68b38550 | 4825 | |
f0486c68 YZ |
4826 | /* |
4827 | * this function must be called within transaction | |
4828 | */ | |
4829 | int btrfs_pin_extent(struct btrfs_root *root, | |
4830 | u64 bytenr, u64 num_bytes, int reserved) | |
4831 | { | |
4832 | struct btrfs_block_group_cache *cache; | |
68b38550 | 4833 | |
f0486c68 | 4834 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
79787eaa | 4835 | BUG_ON(!cache); /* Logic error */ |
f0486c68 YZ |
4836 | |
4837 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
4838 | ||
4839 | btrfs_put_block_group(cache); | |
11833d66 YZ |
4840 | return 0; |
4841 | } | |
4842 | ||
f0486c68 | 4843 | /* |
e688b725 CM |
4844 | * this function must be called within transaction |
4845 | */ | |
4846 | int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans, | |
4847 | struct btrfs_root *root, | |
4848 | u64 bytenr, u64 num_bytes) | |
4849 | { | |
4850 | struct btrfs_block_group_cache *cache; | |
4851 | ||
4852 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
79787eaa | 4853 | BUG_ON(!cache); /* Logic error */ |
e688b725 CM |
4854 | |
4855 | /* | |
4856 | * pull in the free space cache (if any) so that our pin | |
4857 | * removes the free space from the cache. We have load_only set | |
4858 | * to one because the slow code to read in the free extents does check | |
4859 | * the pinned extents. | |
4860 | */ | |
4861 | cache_block_group(cache, trans, root, 1); | |
4862 | ||
4863 | pin_down_extent(root, cache, bytenr, num_bytes, 0); | |
4864 | ||
4865 | /* remove us from the free space cache (if we're there at all) */ | |
4866 | btrfs_remove_free_space(cache, bytenr, num_bytes); | |
4867 | btrfs_put_block_group(cache); | |
4868 | return 0; | |
4869 | } | |
4870 | ||
fb25e914 JB |
4871 | /** |
4872 | * btrfs_update_reserved_bytes - update the block_group and space info counters | |
4873 | * @cache: The cache we are manipulating | |
4874 | * @num_bytes: The number of bytes in question | |
4875 | * @reserve: One of the reservation enums | |
4876 | * | |
4877 | * This is called by the allocator when it reserves space, or by somebody who is | |
4878 | * freeing space that was never actually used on disk. For example if you | |
4879 | * reserve some space for a new leaf in transaction A and before transaction A | |
4880 | * commits you free that leaf, you call this with reserve set to 0 in order to | |
4881 | * clear the reservation. | |
4882 | * | |
4883 | * Metadata reservations should be called with RESERVE_ALLOC so we do the proper | |
4884 | * ENOSPC accounting. For data we handle the reservation through clearing the | |
4885 | * delalloc bits in the io_tree. We have to do this since we could end up | |
4886 | * allocating less disk space for the amount of data we have reserved in the | |
4887 | * case of compression. | |
4888 | * | |
4889 | * If this is a reservation and the block group has become read only we cannot | |
4890 | * make the reservation and return -EAGAIN, otherwise this function always | |
4891 | * succeeds. | |
f0486c68 | 4892 | */ |
fb25e914 JB |
4893 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
4894 | u64 num_bytes, int reserve) | |
11833d66 | 4895 | { |
fb25e914 | 4896 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 4897 | int ret = 0; |
79787eaa | 4898 | |
fb25e914 JB |
4899 | spin_lock(&space_info->lock); |
4900 | spin_lock(&cache->lock); | |
4901 | if (reserve != RESERVE_FREE) { | |
f0486c68 YZ |
4902 | if (cache->ro) { |
4903 | ret = -EAGAIN; | |
4904 | } else { | |
fb25e914 JB |
4905 | cache->reserved += num_bytes; |
4906 | space_info->bytes_reserved += num_bytes; | |
4907 | if (reserve == RESERVE_ALLOC) { | |
8c2a3ca2 | 4908 | trace_btrfs_space_reservation(cache->fs_info, |
2bcc0328 LB |
4909 | "space_info", space_info->flags, |
4910 | num_bytes, 0); | |
fb25e914 JB |
4911 | space_info->bytes_may_use -= num_bytes; |
4912 | } | |
f0486c68 | 4913 | } |
fb25e914 JB |
4914 | } else { |
4915 | if (cache->ro) | |
4916 | space_info->bytes_readonly += num_bytes; | |
4917 | cache->reserved -= num_bytes; | |
4918 | space_info->bytes_reserved -= num_bytes; | |
4919 | space_info->reservation_progress++; | |
324ae4df | 4920 | } |
fb25e914 JB |
4921 | spin_unlock(&cache->lock); |
4922 | spin_unlock(&space_info->lock); | |
f0486c68 | 4923 | return ret; |
324ae4df | 4924 | } |
9078a3e1 | 4925 | |
143bede5 | 4926 | void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
11833d66 | 4927 | struct btrfs_root *root) |
e8569813 | 4928 | { |
e8569813 | 4929 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
4930 | struct btrfs_caching_control *next; |
4931 | struct btrfs_caching_control *caching_ctl; | |
4932 | struct btrfs_block_group_cache *cache; | |
e8569813 | 4933 | |
11833d66 | 4934 | down_write(&fs_info->extent_commit_sem); |
25179201 | 4935 | |
11833d66 YZ |
4936 | list_for_each_entry_safe(caching_ctl, next, |
4937 | &fs_info->caching_block_groups, list) { | |
4938 | cache = caching_ctl->block_group; | |
4939 | if (block_group_cache_done(cache)) { | |
4940 | cache->last_byte_to_unpin = (u64)-1; | |
4941 | list_del_init(&caching_ctl->list); | |
4942 | put_caching_control(caching_ctl); | |
e8569813 | 4943 | } else { |
11833d66 | 4944 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 4945 | } |
e8569813 | 4946 | } |
11833d66 YZ |
4947 | |
4948 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
4949 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
4950 | else | |
4951 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
4952 | ||
4953 | up_write(&fs_info->extent_commit_sem); | |
8929ecfa YZ |
4954 | |
4955 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
4956 | } |
4957 | ||
11833d66 | 4958 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
ccd467d6 | 4959 | { |
11833d66 YZ |
4960 | struct btrfs_fs_info *fs_info = root->fs_info; |
4961 | struct btrfs_block_group_cache *cache = NULL; | |
4962 | u64 len; | |
ccd467d6 | 4963 | |
11833d66 YZ |
4964 | while (start <= end) { |
4965 | if (!cache || | |
4966 | start >= cache->key.objectid + cache->key.offset) { | |
4967 | if (cache) | |
4968 | btrfs_put_block_group(cache); | |
4969 | cache = btrfs_lookup_block_group(fs_info, start); | |
79787eaa | 4970 | BUG_ON(!cache); /* Logic error */ |
11833d66 YZ |
4971 | } |
4972 | ||
4973 | len = cache->key.objectid + cache->key.offset - start; | |
4974 | len = min(len, end + 1 - start); | |
4975 | ||
4976 | if (start < cache->last_byte_to_unpin) { | |
4977 | len = min(len, cache->last_byte_to_unpin - start); | |
4978 | btrfs_add_free_space(cache, start, len); | |
4979 | } | |
4980 | ||
f0486c68 YZ |
4981 | start += len; |
4982 | ||
11833d66 YZ |
4983 | spin_lock(&cache->space_info->lock); |
4984 | spin_lock(&cache->lock); | |
4985 | cache->pinned -= len; | |
4986 | cache->space_info->bytes_pinned -= len; | |
37be25bc | 4987 | if (cache->ro) |
f0486c68 | 4988 | cache->space_info->bytes_readonly += len; |
11833d66 YZ |
4989 | spin_unlock(&cache->lock); |
4990 | spin_unlock(&cache->space_info->lock); | |
ccd467d6 | 4991 | } |
11833d66 YZ |
4992 | |
4993 | if (cache) | |
4994 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
4995 | return 0; |
4996 | } | |
4997 | ||
4998 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 4999 | struct btrfs_root *root) |
a28ec197 | 5000 | { |
11833d66 YZ |
5001 | struct btrfs_fs_info *fs_info = root->fs_info; |
5002 | struct extent_io_tree *unpin; | |
1a5bc167 CM |
5003 | u64 start; |
5004 | u64 end; | |
a28ec197 | 5005 | int ret; |
a28ec197 | 5006 | |
79787eaa JM |
5007 | if (trans->aborted) |
5008 | return 0; | |
5009 | ||
11833d66 YZ |
5010 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
5011 | unpin = &fs_info->freed_extents[1]; | |
5012 | else | |
5013 | unpin = &fs_info->freed_extents[0]; | |
5014 | ||
d397712b | 5015 | while (1) { |
1a5bc167 CM |
5016 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
5017 | EXTENT_DIRTY); | |
5018 | if (ret) | |
a28ec197 | 5019 | break; |
1f3c79a2 | 5020 | |
5378e607 LD |
5021 | if (btrfs_test_opt(root, DISCARD)) |
5022 | ret = btrfs_discard_extent(root, start, | |
5023 | end + 1 - start, NULL); | |
1f3c79a2 | 5024 | |
1a5bc167 | 5025 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
11833d66 | 5026 | unpin_extent_range(root, start, end); |
b9473439 | 5027 | cond_resched(); |
a28ec197 | 5028 | } |
817d52f8 | 5029 | |
e20d96d6 CM |
5030 | return 0; |
5031 | } | |
5032 | ||
5d4f98a2 YZ |
5033 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
5034 | struct btrfs_root *root, | |
5035 | u64 bytenr, u64 num_bytes, u64 parent, | |
5036 | u64 root_objectid, u64 owner_objectid, | |
5037 | u64 owner_offset, int refs_to_drop, | |
5038 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 5039 | { |
e2fa7227 | 5040 | struct btrfs_key key; |
5d4f98a2 | 5041 | struct btrfs_path *path; |
1261ec42 CM |
5042 | struct btrfs_fs_info *info = root->fs_info; |
5043 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 5044 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
5045 | struct btrfs_extent_item *ei; |
5046 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 5047 | int ret; |
5d4f98a2 | 5048 | int is_data; |
952fccac CM |
5049 | int extent_slot = 0; |
5050 | int found_extent = 0; | |
5051 | int num_to_del = 1; | |
5d4f98a2 YZ |
5052 | u32 item_size; |
5053 | u64 refs; | |
037e6390 | 5054 | |
5caf2a00 | 5055 | path = btrfs_alloc_path(); |
54aa1f4d CM |
5056 | if (!path) |
5057 | return -ENOMEM; | |
5f26f772 | 5058 | |
3c12ac72 | 5059 | path->reada = 1; |
b9473439 | 5060 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5061 | |
5062 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
5063 | BUG_ON(!is_data && refs_to_drop != 1); | |
5064 | ||
5065 | ret = lookup_extent_backref(trans, extent_root, path, &iref, | |
5066 | bytenr, num_bytes, parent, | |
5067 | root_objectid, owner_objectid, | |
5068 | owner_offset); | |
7bb86316 | 5069 | if (ret == 0) { |
952fccac | 5070 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
5071 | while (extent_slot >= 0) { |
5072 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 5073 | extent_slot); |
5d4f98a2 | 5074 | if (key.objectid != bytenr) |
952fccac | 5075 | break; |
5d4f98a2 YZ |
5076 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
5077 | key.offset == num_bytes) { | |
952fccac CM |
5078 | found_extent = 1; |
5079 | break; | |
5080 | } | |
5081 | if (path->slots[0] - extent_slot > 5) | |
5082 | break; | |
5d4f98a2 | 5083 | extent_slot--; |
952fccac | 5084 | } |
5d4f98a2 YZ |
5085 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
5086 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
5087 | if (found_extent && item_size < sizeof(*ei)) | |
5088 | found_extent = 0; | |
5089 | #endif | |
31840ae1 | 5090 | if (!found_extent) { |
5d4f98a2 | 5091 | BUG_ON(iref); |
56bec294 | 5092 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
5093 | NULL, refs_to_drop, |
5094 | is_data); | |
79787eaa JM |
5095 | if (ret) |
5096 | goto abort; | |
b3b4aa74 | 5097 | btrfs_release_path(path); |
b9473439 | 5098 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5099 | |
5100 | key.objectid = bytenr; | |
5101 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5102 | key.offset = num_bytes; | |
5103 | ||
31840ae1 ZY |
5104 | ret = btrfs_search_slot(trans, extent_root, |
5105 | &key, path, -1, 1); | |
f3465ca4 JB |
5106 | if (ret) { |
5107 | printk(KERN_ERR "umm, got %d back from search" | |
d397712b CM |
5108 | ", was looking for %llu\n", ret, |
5109 | (unsigned long long)bytenr); | |
b783e62d JB |
5110 | if (ret > 0) |
5111 | btrfs_print_leaf(extent_root, | |
5112 | path->nodes[0]); | |
f3465ca4 | 5113 | } |
79787eaa JM |
5114 | if (ret < 0) |
5115 | goto abort; | |
31840ae1 ZY |
5116 | extent_slot = path->slots[0]; |
5117 | } | |
79787eaa | 5118 | } else if (ret == -ENOENT) { |
7bb86316 CM |
5119 | btrfs_print_leaf(extent_root, path->nodes[0]); |
5120 | WARN_ON(1); | |
d397712b | 5121 | printk(KERN_ERR "btrfs unable to find ref byte nr %llu " |
5d4f98a2 | 5122 | "parent %llu root %llu owner %llu offset %llu\n", |
d397712b | 5123 | (unsigned long long)bytenr, |
56bec294 | 5124 | (unsigned long long)parent, |
d397712b | 5125 | (unsigned long long)root_objectid, |
5d4f98a2 YZ |
5126 | (unsigned long long)owner_objectid, |
5127 | (unsigned long long)owner_offset); | |
79787eaa JM |
5128 | } else { |
5129 | goto abort; | |
7bb86316 | 5130 | } |
5f39d397 CM |
5131 | |
5132 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
5133 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
5134 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
5135 | if (item_size < sizeof(*ei)) { | |
5136 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
5137 | ret = convert_extent_item_v0(trans, extent_root, path, | |
5138 | owner_objectid, 0); | |
79787eaa JM |
5139 | if (ret < 0) |
5140 | goto abort; | |
5d4f98a2 | 5141 | |
b3b4aa74 | 5142 | btrfs_release_path(path); |
5d4f98a2 YZ |
5143 | path->leave_spinning = 1; |
5144 | ||
5145 | key.objectid = bytenr; | |
5146 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5147 | key.offset = num_bytes; | |
5148 | ||
5149 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
5150 | -1, 1); | |
5151 | if (ret) { | |
5152 | printk(KERN_ERR "umm, got %d back from search" | |
5153 | ", was looking for %llu\n", ret, | |
5154 | (unsigned long long)bytenr); | |
5155 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
5156 | } | |
79787eaa JM |
5157 | if (ret < 0) |
5158 | goto abort; | |
5d4f98a2 YZ |
5159 | extent_slot = path->slots[0]; |
5160 | leaf = path->nodes[0]; | |
5161 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
5162 | } | |
5163 | #endif | |
5164 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 5165 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 5166 | struct btrfs_extent_item); |
5d4f98a2 YZ |
5167 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { |
5168 | struct btrfs_tree_block_info *bi; | |
5169 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
5170 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
5171 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
5172 | } | |
56bec294 | 5173 | |
5d4f98a2 | 5174 | refs = btrfs_extent_refs(leaf, ei); |
56bec294 CM |
5175 | BUG_ON(refs < refs_to_drop); |
5176 | refs -= refs_to_drop; | |
5f39d397 | 5177 | |
5d4f98a2 YZ |
5178 | if (refs > 0) { |
5179 | if (extent_op) | |
5180 | __run_delayed_extent_op(extent_op, leaf, ei); | |
5181 | /* | |
5182 | * In the case of inline back ref, reference count will | |
5183 | * be updated by remove_extent_backref | |
952fccac | 5184 | */ |
5d4f98a2 YZ |
5185 | if (iref) { |
5186 | BUG_ON(!found_extent); | |
5187 | } else { | |
5188 | btrfs_set_extent_refs(leaf, ei, refs); | |
5189 | btrfs_mark_buffer_dirty(leaf); | |
5190 | } | |
5191 | if (found_extent) { | |
5192 | ret = remove_extent_backref(trans, extent_root, path, | |
5193 | iref, refs_to_drop, | |
5194 | is_data); | |
79787eaa JM |
5195 | if (ret) |
5196 | goto abort; | |
952fccac | 5197 | } |
5d4f98a2 | 5198 | } else { |
5d4f98a2 YZ |
5199 | if (found_extent) { |
5200 | BUG_ON(is_data && refs_to_drop != | |
5201 | extent_data_ref_count(root, path, iref)); | |
5202 | if (iref) { | |
5203 | BUG_ON(path->slots[0] != extent_slot); | |
5204 | } else { | |
5205 | BUG_ON(path->slots[0] != extent_slot + 1); | |
5206 | path->slots[0] = extent_slot; | |
5207 | num_to_del = 2; | |
5208 | } | |
78fae27e | 5209 | } |
b9473439 | 5210 | |
952fccac CM |
5211 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
5212 | num_to_del); | |
79787eaa JM |
5213 | if (ret) |
5214 | goto abort; | |
b3b4aa74 | 5215 | btrfs_release_path(path); |
21af804c | 5216 | |
5d4f98a2 | 5217 | if (is_data) { |
459931ec | 5218 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
79787eaa JM |
5219 | if (ret) |
5220 | goto abort; | |
459931ec CM |
5221 | } |
5222 | ||
f0486c68 | 5223 | ret = update_block_group(trans, root, bytenr, num_bytes, 0); |
79787eaa JM |
5224 | if (ret) |
5225 | goto abort; | |
a28ec197 | 5226 | } |
79787eaa | 5227 | out: |
5caf2a00 | 5228 | btrfs_free_path(path); |
a28ec197 | 5229 | return ret; |
79787eaa JM |
5230 | |
5231 | abort: | |
5232 | btrfs_abort_transaction(trans, extent_root, ret); | |
5233 | goto out; | |
a28ec197 CM |
5234 | } |
5235 | ||
1887be66 | 5236 | /* |
f0486c68 | 5237 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
5238 | * delayed ref for that extent as well. This searches the delayed ref tree for |
5239 | * a given extent, and if there are no other delayed refs to be processed, it | |
5240 | * removes it from the tree. | |
5241 | */ | |
5242 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
5243 | struct btrfs_root *root, u64 bytenr) | |
5244 | { | |
5245 | struct btrfs_delayed_ref_head *head; | |
5246 | struct btrfs_delayed_ref_root *delayed_refs; | |
5247 | struct btrfs_delayed_ref_node *ref; | |
5248 | struct rb_node *node; | |
f0486c68 | 5249 | int ret = 0; |
1887be66 CM |
5250 | |
5251 | delayed_refs = &trans->transaction->delayed_refs; | |
5252 | spin_lock(&delayed_refs->lock); | |
5253 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
5254 | if (!head) | |
5255 | goto out; | |
5256 | ||
5257 | node = rb_prev(&head->node.rb_node); | |
5258 | if (!node) | |
5259 | goto out; | |
5260 | ||
5261 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
5262 | ||
5263 | /* there are still entries for this ref, we can't drop it */ | |
5264 | if (ref->bytenr == bytenr) | |
5265 | goto out; | |
5266 | ||
5d4f98a2 YZ |
5267 | if (head->extent_op) { |
5268 | if (!head->must_insert_reserved) | |
5269 | goto out; | |
5270 | kfree(head->extent_op); | |
5271 | head->extent_op = NULL; | |
5272 | } | |
5273 | ||
1887be66 CM |
5274 | /* |
5275 | * waiting for the lock here would deadlock. If someone else has it | |
5276 | * locked they are already in the process of dropping it anyway | |
5277 | */ | |
5278 | if (!mutex_trylock(&head->mutex)) | |
5279 | goto out; | |
5280 | ||
5281 | /* | |
5282 | * at this point we have a head with no other entries. Go | |
5283 | * ahead and process it. | |
5284 | */ | |
5285 | head->node.in_tree = 0; | |
5286 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 5287 | |
1887be66 CM |
5288 | delayed_refs->num_entries--; |
5289 | ||
5290 | /* | |
5291 | * we don't take a ref on the node because we're removing it from the | |
5292 | * tree, so we just steal the ref the tree was holding. | |
5293 | */ | |
c3e69d58 CM |
5294 | delayed_refs->num_heads--; |
5295 | if (list_empty(&head->cluster)) | |
5296 | delayed_refs->num_heads_ready--; | |
5297 | ||
5298 | list_del_init(&head->cluster); | |
1887be66 CM |
5299 | spin_unlock(&delayed_refs->lock); |
5300 | ||
f0486c68 YZ |
5301 | BUG_ON(head->extent_op); |
5302 | if (head->must_insert_reserved) | |
5303 | ret = 1; | |
5304 | ||
5305 | mutex_unlock(&head->mutex); | |
1887be66 | 5306 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 5307 | return ret; |
1887be66 CM |
5308 | out: |
5309 | spin_unlock(&delayed_refs->lock); | |
5310 | return 0; | |
5311 | } | |
5312 | ||
f0486c68 YZ |
5313 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
5314 | struct btrfs_root *root, | |
5315 | struct extent_buffer *buf, | |
5581a51a | 5316 | u64 parent, int last_ref) |
f0486c68 | 5317 | { |
f0486c68 YZ |
5318 | struct btrfs_block_group_cache *cache = NULL; |
5319 | int ret; | |
5320 | ||
5321 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
66d7e7f0 AJ |
5322 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
5323 | buf->start, buf->len, | |
5324 | parent, root->root_key.objectid, | |
5325 | btrfs_header_level(buf), | |
5581a51a | 5326 | BTRFS_DROP_DELAYED_REF, NULL, 0); |
79787eaa | 5327 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
5328 | } |
5329 | ||
5330 | if (!last_ref) | |
5331 | return; | |
5332 | ||
f0486c68 | 5333 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); |
f0486c68 YZ |
5334 | |
5335 | if (btrfs_header_generation(buf) == trans->transid) { | |
5336 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
5337 | ret = check_ref_cleanup(trans, root, buf->start); | |
5338 | if (!ret) | |
37be25bc | 5339 | goto out; |
f0486c68 YZ |
5340 | } |
5341 | ||
5342 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
5343 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
37be25bc | 5344 | goto out; |
f0486c68 YZ |
5345 | } |
5346 | ||
5347 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
5348 | ||
5349 | btrfs_add_free_space(cache, buf->start, buf->len); | |
fb25e914 | 5350 | btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE); |
f0486c68 YZ |
5351 | } |
5352 | out: | |
a826d6dc JB |
5353 | /* |
5354 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
5355 | * anymore. | |
5356 | */ | |
5357 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
5358 | btrfs_put_block_group(cache); |
5359 | } | |
5360 | ||
79787eaa | 5361 | /* Can return -ENOMEM */ |
66d7e7f0 AJ |
5362 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
5363 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, | |
5364 | u64 owner, u64 offset, int for_cow) | |
925baedd CM |
5365 | { |
5366 | int ret; | |
66d7e7f0 | 5367 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 5368 | |
56bec294 CM |
5369 | /* |
5370 | * tree log blocks never actually go into the extent allocation | |
5371 | * tree, just update pinning info and exit early. | |
56bec294 | 5372 | */ |
5d4f98a2 YZ |
5373 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
5374 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 5375 | /* unlocks the pinned mutex */ |
11833d66 | 5376 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 5377 | ret = 0; |
5d4f98a2 | 5378 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
5379 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
5380 | num_bytes, | |
5d4f98a2 | 5381 | parent, root_objectid, (int)owner, |
66d7e7f0 | 5382 | BTRFS_DROP_DELAYED_REF, NULL, for_cow); |
5d4f98a2 | 5383 | } else { |
66d7e7f0 AJ |
5384 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
5385 | num_bytes, | |
5386 | parent, root_objectid, owner, | |
5387 | offset, BTRFS_DROP_DELAYED_REF, | |
5388 | NULL, for_cow); | |
56bec294 | 5389 | } |
925baedd CM |
5390 | return ret; |
5391 | } | |
5392 | ||
87ee04eb CM |
5393 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
5394 | { | |
5395 | u64 mask = ((u64)root->stripesize - 1); | |
5396 | u64 ret = (val + mask) & ~mask; | |
5397 | return ret; | |
5398 | } | |
5399 | ||
817d52f8 JB |
5400 | /* |
5401 | * when we wait for progress in the block group caching, its because | |
5402 | * our allocation attempt failed at least once. So, we must sleep | |
5403 | * and let some progress happen before we try again. | |
5404 | * | |
5405 | * This function will sleep at least once waiting for new free space to | |
5406 | * show up, and then it will check the block group free space numbers | |
5407 | * for our min num_bytes. Another option is to have it go ahead | |
5408 | * and look in the rbtree for a free extent of a given size, but this | |
5409 | * is a good start. | |
5410 | */ | |
5411 | static noinline int | |
5412 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
5413 | u64 num_bytes) | |
5414 | { | |
11833d66 | 5415 | struct btrfs_caching_control *caching_ctl; |
817d52f8 JB |
5416 | DEFINE_WAIT(wait); |
5417 | ||
11833d66 YZ |
5418 | caching_ctl = get_caching_control(cache); |
5419 | if (!caching_ctl) | |
817d52f8 | 5420 | return 0; |
817d52f8 | 5421 | |
11833d66 | 5422 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 5423 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
5424 | |
5425 | put_caching_control(caching_ctl); | |
5426 | return 0; | |
5427 | } | |
5428 | ||
5429 | static noinline int | |
5430 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
5431 | { | |
5432 | struct btrfs_caching_control *caching_ctl; | |
5433 | DEFINE_WAIT(wait); | |
5434 | ||
5435 | caching_ctl = get_caching_control(cache); | |
5436 | if (!caching_ctl) | |
5437 | return 0; | |
5438 | ||
5439 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
5440 | ||
5441 | put_caching_control(caching_ctl); | |
817d52f8 JB |
5442 | return 0; |
5443 | } | |
5444 | ||
7738a53a | 5445 | static int __get_block_group_index(u64 flags) |
b742bb82 YZ |
5446 | { |
5447 | int index; | |
7738a53a ID |
5448 | |
5449 | if (flags & BTRFS_BLOCK_GROUP_RAID10) | |
b742bb82 | 5450 | index = 0; |
7738a53a | 5451 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
b742bb82 | 5452 | index = 1; |
7738a53a | 5453 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
b742bb82 | 5454 | index = 2; |
7738a53a | 5455 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
b742bb82 YZ |
5456 | index = 3; |
5457 | else | |
5458 | index = 4; | |
7738a53a | 5459 | |
b742bb82 YZ |
5460 | return index; |
5461 | } | |
5462 | ||
7738a53a ID |
5463 | static int get_block_group_index(struct btrfs_block_group_cache *cache) |
5464 | { | |
5465 | return __get_block_group_index(cache->flags); | |
5466 | } | |
5467 | ||
817d52f8 | 5468 | enum btrfs_loop_type { |
285ff5af JB |
5469 | LOOP_CACHING_NOWAIT = 0, |
5470 | LOOP_CACHING_WAIT = 1, | |
5471 | LOOP_ALLOC_CHUNK = 2, | |
5472 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
5473 | }; |
5474 | ||
fec577fb CM |
5475 | /* |
5476 | * walks the btree of allocated extents and find a hole of a given size. | |
5477 | * The key ins is changed to record the hole: | |
5478 | * ins->objectid == block start | |
62e2749e | 5479 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
5480 | * ins->offset == number of blocks |
5481 | * Any available blocks before search_start are skipped. | |
5482 | */ | |
d397712b | 5483 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
5484 | struct btrfs_root *orig_root, |
5485 | u64 num_bytes, u64 empty_size, | |
98ed5174 | 5486 | u64 hint_byte, struct btrfs_key *ins, |
e0f54067 | 5487 | u64 data) |
fec577fb | 5488 | { |
80eb234a | 5489 | int ret = 0; |
d397712b | 5490 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 5491 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 5492 | struct btrfs_block_group_cache *block_group = NULL; |
274bd4fb | 5493 | struct btrfs_block_group_cache *used_block_group; |
81c9ad23 | 5494 | u64 search_start = 0; |
239b14b3 | 5495 | int empty_cluster = 2 * 1024 * 1024; |
0ef3e66b | 5496 | int allowed_chunk_alloc = 0; |
ccf0e725 | 5497 | int done_chunk_alloc = 0; |
80eb234a | 5498 | struct btrfs_space_info *space_info; |
fa9c0d79 | 5499 | int loop = 0; |
f0486c68 | 5500 | int index = 0; |
fb25e914 JB |
5501 | int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ? |
5502 | RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC; | |
817d52f8 | 5503 | bool found_uncached_bg = false; |
0a24325e | 5504 | bool failed_cluster_refill = false; |
1cdda9b8 | 5505 | bool failed_alloc = false; |
67377734 | 5506 | bool use_cluster = true; |
60d2adbb | 5507 | bool have_caching_bg = false; |
fec577fb | 5508 | |
db94535d | 5509 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 5510 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
5511 | ins->objectid = 0; |
5512 | ins->offset = 0; | |
b1a4d965 | 5513 | |
3f7de037 JB |
5514 | trace_find_free_extent(orig_root, num_bytes, empty_size, data); |
5515 | ||
2552d17e | 5516 | space_info = __find_space_info(root->fs_info, data); |
1b1d1f66 | 5517 | if (!space_info) { |
e0f54067 | 5518 | printk(KERN_ERR "No space info for %llu\n", data); |
1b1d1f66 JB |
5519 | return -ENOSPC; |
5520 | } | |
2552d17e | 5521 | |
67377734 JB |
5522 | /* |
5523 | * If the space info is for both data and metadata it means we have a | |
5524 | * small filesystem and we can't use the clustering stuff. | |
5525 | */ | |
5526 | if (btrfs_mixed_space_info(space_info)) | |
5527 | use_cluster = false; | |
5528 | ||
0ef3e66b CM |
5529 | if (orig_root->ref_cows || empty_size) |
5530 | allowed_chunk_alloc = 1; | |
5531 | ||
67377734 | 5532 | if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) { |
fa9c0d79 | 5533 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
5534 | if (!btrfs_test_opt(root, SSD)) |
5535 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
5536 | } |
5537 | ||
67377734 JB |
5538 | if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster && |
5539 | btrfs_test_opt(root, SSD)) { | |
fa9c0d79 CM |
5540 | last_ptr = &root->fs_info->data_alloc_cluster; |
5541 | } | |
0f9dd46c | 5542 | |
239b14b3 | 5543 | if (last_ptr) { |
fa9c0d79 CM |
5544 | spin_lock(&last_ptr->lock); |
5545 | if (last_ptr->block_group) | |
5546 | hint_byte = last_ptr->window_start; | |
5547 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 5548 | } |
fa9c0d79 | 5549 | |
a061fc8d | 5550 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 5551 | search_start = max(search_start, hint_byte); |
0b86a832 | 5552 | |
817d52f8 | 5553 | if (!last_ptr) |
fa9c0d79 | 5554 | empty_cluster = 0; |
fa9c0d79 | 5555 | |
2552d17e | 5556 | if (search_start == hint_byte) { |
2552d17e JB |
5557 | block_group = btrfs_lookup_block_group(root->fs_info, |
5558 | search_start); | |
274bd4fb | 5559 | used_block_group = block_group; |
817d52f8 JB |
5560 | /* |
5561 | * we don't want to use the block group if it doesn't match our | |
5562 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
5563 | * |
5564 | * However if we are re-searching with an ideal block group | |
5565 | * picked out then we don't care that the block group is cached. | |
817d52f8 JB |
5566 | */ |
5567 | if (block_group && block_group_bits(block_group, data) && | |
285ff5af | 5568 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 5569 | down_read(&space_info->groups_sem); |
44fb5511 CM |
5570 | if (list_empty(&block_group->list) || |
5571 | block_group->ro) { | |
5572 | /* | |
5573 | * someone is removing this block group, | |
5574 | * we can't jump into the have_block_group | |
5575 | * target because our list pointers are not | |
5576 | * valid | |
5577 | */ | |
5578 | btrfs_put_block_group(block_group); | |
5579 | up_read(&space_info->groups_sem); | |
ccf0e725 | 5580 | } else { |
b742bb82 | 5581 | index = get_block_group_index(block_group); |
44fb5511 | 5582 | goto have_block_group; |
ccf0e725 | 5583 | } |
2552d17e | 5584 | } else if (block_group) { |
fa9c0d79 | 5585 | btrfs_put_block_group(block_group); |
2552d17e | 5586 | } |
42e70e7a | 5587 | } |
2552d17e | 5588 | search: |
60d2adbb | 5589 | have_caching_bg = false; |
80eb234a | 5590 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
5591 | list_for_each_entry(block_group, &space_info->block_groups[index], |
5592 | list) { | |
6226cb0a | 5593 | u64 offset; |
817d52f8 | 5594 | int cached; |
8a1413a2 | 5595 | |
274bd4fb | 5596 | used_block_group = block_group; |
11dfe35a | 5597 | btrfs_get_block_group(block_group); |
2552d17e | 5598 | search_start = block_group->key.objectid; |
42e70e7a | 5599 | |
83a50de9 CM |
5600 | /* |
5601 | * this can happen if we end up cycling through all the | |
5602 | * raid types, but we want to make sure we only allocate | |
5603 | * for the proper type. | |
5604 | */ | |
5605 | if (!block_group_bits(block_group, data)) { | |
5606 | u64 extra = BTRFS_BLOCK_GROUP_DUP | | |
5607 | BTRFS_BLOCK_GROUP_RAID1 | | |
5608 | BTRFS_BLOCK_GROUP_RAID10; | |
5609 | ||
5610 | /* | |
5611 | * if they asked for extra copies and this block group | |
5612 | * doesn't provide them, bail. This does allow us to | |
5613 | * fill raid0 from raid1. | |
5614 | */ | |
5615 | if ((data & extra) && !(block_group->flags & extra)) | |
5616 | goto loop; | |
5617 | } | |
5618 | ||
2552d17e | 5619 | have_block_group: |
291c7d2f JB |
5620 | cached = block_group_cache_done(block_group); |
5621 | if (unlikely(!cached)) { | |
291c7d2f | 5622 | found_uncached_bg = true; |
b8399dee | 5623 | ret = cache_block_group(block_group, trans, |
285ff5af | 5624 | orig_root, 0); |
1d4284bd CM |
5625 | BUG_ON(ret < 0); |
5626 | ret = 0; | |
817d52f8 JB |
5627 | } |
5628 | ||
ea6a478e | 5629 | if (unlikely(block_group->ro)) |
2552d17e | 5630 | goto loop; |
0f9dd46c | 5631 | |
0a24325e | 5632 | /* |
062c05c4 AO |
5633 | * Ok we want to try and use the cluster allocator, so |
5634 | * lets look there | |
0a24325e | 5635 | */ |
062c05c4 | 5636 | if (last_ptr) { |
fa9c0d79 CM |
5637 | /* |
5638 | * the refill lock keeps out other | |
5639 | * people trying to start a new cluster | |
5640 | */ | |
5641 | spin_lock(&last_ptr->refill_lock); | |
274bd4fb AO |
5642 | used_block_group = last_ptr->block_group; |
5643 | if (used_block_group != block_group && | |
5644 | (!used_block_group || | |
5645 | used_block_group->ro || | |
5646 | !block_group_bits(used_block_group, data))) { | |
5647 | used_block_group = block_group; | |
44fb5511 | 5648 | goto refill_cluster; |
274bd4fb AO |
5649 | } |
5650 | ||
5651 | if (used_block_group != block_group) | |
5652 | btrfs_get_block_group(used_block_group); | |
44fb5511 | 5653 | |
274bd4fb AO |
5654 | offset = btrfs_alloc_from_cluster(used_block_group, |
5655 | last_ptr, num_bytes, used_block_group->key.objectid); | |
fa9c0d79 CM |
5656 | if (offset) { |
5657 | /* we have a block, we're done */ | |
5658 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
5659 | trace_btrfs_reserve_extent_cluster(root, |
5660 | block_group, search_start, num_bytes); | |
fa9c0d79 CM |
5661 | goto checks; |
5662 | } | |
5663 | ||
274bd4fb AO |
5664 | WARN_ON(last_ptr->block_group != used_block_group); |
5665 | if (used_block_group != block_group) { | |
5666 | btrfs_put_block_group(used_block_group); | |
5667 | used_block_group = block_group; | |
fa9c0d79 | 5668 | } |
44fb5511 | 5669 | refill_cluster: |
274bd4fb | 5670 | BUG_ON(used_block_group != block_group); |
062c05c4 AO |
5671 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
5672 | * set up a new clusters, so lets just skip it | |
5673 | * and let the allocator find whatever block | |
5674 | * it can find. If we reach this point, we | |
5675 | * will have tried the cluster allocator | |
5676 | * plenty of times and not have found | |
5677 | * anything, so we are likely way too | |
5678 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
5679 | * anything. |
5680 | * | |
5681 | * However, if the cluster is taken from the | |
5682 | * current block group, release the cluster | |
5683 | * first, so that we stand a better chance of | |
5684 | * succeeding in the unclustered | |
5685 | * allocation. */ | |
5686 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
5687 | last_ptr->block_group != block_group) { | |
062c05c4 AO |
5688 | spin_unlock(&last_ptr->refill_lock); |
5689 | goto unclustered_alloc; | |
5690 | } | |
5691 | ||
fa9c0d79 CM |
5692 | /* |
5693 | * this cluster didn't work out, free it and | |
5694 | * start over | |
5695 | */ | |
5696 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
5697 | ||
a5f6f719 AO |
5698 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
5699 | spin_unlock(&last_ptr->refill_lock); | |
5700 | goto unclustered_alloc; | |
5701 | } | |
5702 | ||
fa9c0d79 | 5703 | /* allocate a cluster in this block group */ |
451d7585 | 5704 | ret = btrfs_find_space_cluster(trans, root, |
fa9c0d79 | 5705 | block_group, last_ptr, |
1b22bad7 | 5706 | search_start, num_bytes, |
fa9c0d79 CM |
5707 | empty_cluster + empty_size); |
5708 | if (ret == 0) { | |
5709 | /* | |
5710 | * now pull our allocation out of this | |
5711 | * cluster | |
5712 | */ | |
5713 | offset = btrfs_alloc_from_cluster(block_group, | |
5714 | last_ptr, num_bytes, | |
5715 | search_start); | |
5716 | if (offset) { | |
5717 | /* we found one, proceed */ | |
5718 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
5719 | trace_btrfs_reserve_extent_cluster(root, |
5720 | block_group, search_start, | |
5721 | num_bytes); | |
fa9c0d79 CM |
5722 | goto checks; |
5723 | } | |
0a24325e JB |
5724 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
5725 | && !failed_cluster_refill) { | |
817d52f8 JB |
5726 | spin_unlock(&last_ptr->refill_lock); |
5727 | ||
0a24325e | 5728 | failed_cluster_refill = true; |
817d52f8 JB |
5729 | wait_block_group_cache_progress(block_group, |
5730 | num_bytes + empty_cluster + empty_size); | |
5731 | goto have_block_group; | |
fa9c0d79 | 5732 | } |
817d52f8 | 5733 | |
fa9c0d79 CM |
5734 | /* |
5735 | * at this point we either didn't find a cluster | |
5736 | * or we weren't able to allocate a block from our | |
5737 | * cluster. Free the cluster we've been trying | |
5738 | * to use, and go to the next block group | |
5739 | */ | |
0a24325e | 5740 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 5741 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 5742 | goto loop; |
fa9c0d79 CM |
5743 | } |
5744 | ||
062c05c4 | 5745 | unclustered_alloc: |
a5f6f719 AO |
5746 | spin_lock(&block_group->free_space_ctl->tree_lock); |
5747 | if (cached && | |
5748 | block_group->free_space_ctl->free_space < | |
5749 | num_bytes + empty_cluster + empty_size) { | |
5750 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
5751 | goto loop; | |
5752 | } | |
5753 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
5754 | ||
6226cb0a JB |
5755 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
5756 | num_bytes, empty_size); | |
1cdda9b8 JB |
5757 | /* |
5758 | * If we didn't find a chunk, and we haven't failed on this | |
5759 | * block group before, and this block group is in the middle of | |
5760 | * caching and we are ok with waiting, then go ahead and wait | |
5761 | * for progress to be made, and set failed_alloc to true. | |
5762 | * | |
5763 | * If failed_alloc is true then we've already waited on this | |
5764 | * block group once and should move on to the next block group. | |
5765 | */ | |
5766 | if (!offset && !failed_alloc && !cached && | |
5767 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 5768 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
5769 | num_bytes + empty_size); |
5770 | failed_alloc = true; | |
817d52f8 | 5771 | goto have_block_group; |
1cdda9b8 | 5772 | } else if (!offset) { |
60d2adbb MX |
5773 | if (!cached) |
5774 | have_caching_bg = true; | |
1cdda9b8 | 5775 | goto loop; |
817d52f8 | 5776 | } |
fa9c0d79 | 5777 | checks: |
6226cb0a | 5778 | search_start = stripe_align(root, offset); |
25179201 | 5779 | |
2552d17e JB |
5780 | /* move on to the next group */ |
5781 | if (search_start + num_bytes > | |
274bd4fb AO |
5782 | used_block_group->key.objectid + used_block_group->key.offset) { |
5783 | btrfs_add_free_space(used_block_group, offset, num_bytes); | |
2552d17e | 5784 | goto loop; |
6226cb0a | 5785 | } |
f5a31e16 | 5786 | |
f0486c68 | 5787 | if (offset < search_start) |
274bd4fb | 5788 | btrfs_add_free_space(used_block_group, offset, |
f0486c68 YZ |
5789 | search_start - offset); |
5790 | BUG_ON(offset > search_start); | |
2552d17e | 5791 | |
274bd4fb | 5792 | ret = btrfs_update_reserved_bytes(used_block_group, num_bytes, |
fb25e914 | 5793 | alloc_type); |
f0486c68 | 5794 | if (ret == -EAGAIN) { |
274bd4fb | 5795 | btrfs_add_free_space(used_block_group, offset, num_bytes); |
2552d17e | 5796 | goto loop; |
0f9dd46c | 5797 | } |
0b86a832 | 5798 | |
f0486c68 | 5799 | /* we are all good, lets return */ |
2552d17e JB |
5800 | ins->objectid = search_start; |
5801 | ins->offset = num_bytes; | |
d2fb3437 | 5802 | |
3f7de037 JB |
5803 | trace_btrfs_reserve_extent(orig_root, block_group, |
5804 | search_start, num_bytes); | |
6226cb0a | 5805 | if (offset < search_start) |
274bd4fb | 5806 | btrfs_add_free_space(used_block_group, offset, |
6226cb0a JB |
5807 | search_start - offset); |
5808 | BUG_ON(offset > search_start); | |
274bd4fb AO |
5809 | if (used_block_group != block_group) |
5810 | btrfs_put_block_group(used_block_group); | |
d82a6f1d | 5811 | btrfs_put_block_group(block_group); |
2552d17e JB |
5812 | break; |
5813 | loop: | |
0a24325e | 5814 | failed_cluster_refill = false; |
1cdda9b8 | 5815 | failed_alloc = false; |
b742bb82 | 5816 | BUG_ON(index != get_block_group_index(block_group)); |
274bd4fb AO |
5817 | if (used_block_group != block_group) |
5818 | btrfs_put_block_group(used_block_group); | |
fa9c0d79 | 5819 | btrfs_put_block_group(block_group); |
2552d17e JB |
5820 | } |
5821 | up_read(&space_info->groups_sem); | |
5822 | ||
60d2adbb MX |
5823 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
5824 | goto search; | |
5825 | ||
b742bb82 YZ |
5826 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
5827 | goto search; | |
5828 | ||
285ff5af | 5829 | /* |
ccf0e725 JB |
5830 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
5831 | * caching kthreads as we move along | |
817d52f8 JB |
5832 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
5833 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
5834 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
5835 | * again | |
fa9c0d79 | 5836 | */ |
723bda20 | 5837 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 5838 | index = 0; |
723bda20 | 5839 | loop++; |
817d52f8 | 5840 | if (loop == LOOP_ALLOC_CHUNK) { |
723bda20 JB |
5841 | if (allowed_chunk_alloc) { |
5842 | ret = do_chunk_alloc(trans, root, num_bytes + | |
5843 | 2 * 1024 * 1024, data, | |
5844 | CHUNK_ALLOC_LIMITED); | |
06789384 LB |
5845 | /* |
5846 | * Do not bail out on ENOSPC since we | |
5847 | * can do more things. | |
5848 | */ | |
5849 | if (ret < 0 && ret != -ENOSPC) { | |
79787eaa JM |
5850 | btrfs_abort_transaction(trans, |
5851 | root, ret); | |
5852 | goto out; | |
5853 | } | |
723bda20 JB |
5854 | allowed_chunk_alloc = 0; |
5855 | if (ret == 1) | |
5856 | done_chunk_alloc = 1; | |
5857 | } else if (!done_chunk_alloc && | |
5858 | space_info->force_alloc == | |
5859 | CHUNK_ALLOC_NO_FORCE) { | |
5860 | space_info->force_alloc = CHUNK_ALLOC_LIMITED; | |
5861 | } | |
2552d17e | 5862 | |
723bda20 JB |
5863 | /* |
5864 | * We didn't allocate a chunk, go ahead and drop the | |
5865 | * empty size and loop again. | |
5866 | */ | |
5867 | if (!done_chunk_alloc) | |
5868 | loop = LOOP_NO_EMPTY_SIZE; | |
2552d17e JB |
5869 | } |
5870 | ||
723bda20 JB |
5871 | if (loop == LOOP_NO_EMPTY_SIZE) { |
5872 | empty_size = 0; | |
5873 | empty_cluster = 0; | |
fa9c0d79 | 5874 | } |
723bda20 JB |
5875 | |
5876 | goto search; | |
2552d17e JB |
5877 | } else if (!ins->objectid) { |
5878 | ret = -ENOSPC; | |
d82a6f1d | 5879 | } else if (ins->objectid) { |
80eb234a | 5880 | ret = 0; |
be744175 | 5881 | } |
79787eaa | 5882 | out: |
be744175 | 5883 | |
0f70abe2 | 5884 | return ret; |
fec577fb | 5885 | } |
ec44a35c | 5886 | |
9ed74f2d JB |
5887 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
5888 | int dump_block_groups) | |
0f9dd46c JB |
5889 | { |
5890 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 5891 | int index = 0; |
0f9dd46c | 5892 | |
9ed74f2d | 5893 | spin_lock(&info->lock); |
fb25e914 JB |
5894 | printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n", |
5895 | (unsigned long long)info->flags, | |
d397712b | 5896 | (unsigned long long)(info->total_bytes - info->bytes_used - |
9ed74f2d | 5897 | info->bytes_pinned - info->bytes_reserved - |
8929ecfa | 5898 | info->bytes_readonly), |
d397712b | 5899 | (info->full) ? "" : "not "); |
8929ecfa YZ |
5900 | printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, " |
5901 | "reserved=%llu, may_use=%llu, readonly=%llu\n", | |
21380931 | 5902 | (unsigned long long)info->total_bytes, |
8929ecfa | 5903 | (unsigned long long)info->bytes_used, |
21380931 | 5904 | (unsigned long long)info->bytes_pinned, |
8929ecfa | 5905 | (unsigned long long)info->bytes_reserved, |
21380931 | 5906 | (unsigned long long)info->bytes_may_use, |
8929ecfa | 5907 | (unsigned long long)info->bytes_readonly); |
9ed74f2d JB |
5908 | spin_unlock(&info->lock); |
5909 | ||
5910 | if (!dump_block_groups) | |
5911 | return; | |
0f9dd46c | 5912 | |
80eb234a | 5913 | down_read(&info->groups_sem); |
b742bb82 YZ |
5914 | again: |
5915 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 5916 | spin_lock(&cache->lock); |
799ffc3c | 5917 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n", |
d397712b CM |
5918 | (unsigned long long)cache->key.objectid, |
5919 | (unsigned long long)cache->key.offset, | |
5920 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
5921 | (unsigned long long)cache->pinned, | |
799ffc3c LB |
5922 | (unsigned long long)cache->reserved, |
5923 | cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
5924 | btrfs_dump_free_space(cache, bytes); |
5925 | spin_unlock(&cache->lock); | |
5926 | } | |
b742bb82 YZ |
5927 | if (++index < BTRFS_NR_RAID_TYPES) |
5928 | goto again; | |
80eb234a | 5929 | up_read(&info->groups_sem); |
0f9dd46c | 5930 | } |
e8569813 | 5931 | |
11833d66 YZ |
5932 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
5933 | struct btrfs_root *root, | |
5934 | u64 num_bytes, u64 min_alloc_size, | |
5935 | u64 empty_size, u64 hint_byte, | |
81c9ad23 | 5936 | struct btrfs_key *ins, u64 data) |
fec577fb | 5937 | { |
9e622d6b | 5938 | bool final_tried = false; |
fec577fb | 5939 | int ret; |
925baedd | 5940 | |
6a63209f | 5941 | data = btrfs_get_alloc_profile(root, data); |
98d20f67 | 5942 | again: |
0ef3e66b CM |
5943 | /* |
5944 | * the only place that sets empty_size is btrfs_realloc_node, which | |
5945 | * is not called recursively on allocations | |
5946 | */ | |
79787eaa | 5947 | if (empty_size || root->ref_cows) { |
6324fbf3 | 5948 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
5949 | num_bytes + 2 * 1024 * 1024, data, |
5950 | CHUNK_ALLOC_NO_FORCE); | |
79787eaa JM |
5951 | if (ret < 0 && ret != -ENOSPC) { |
5952 | btrfs_abort_transaction(trans, root, ret); | |
5953 | return ret; | |
5954 | } | |
5955 | } | |
0b86a832 | 5956 | |
db94535d CM |
5957 | WARN_ON(num_bytes < root->sectorsize); |
5958 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
81c9ad23 | 5959 | hint_byte, ins, data); |
3b951516 | 5960 | |
9e622d6b MX |
5961 | if (ret == -ENOSPC) { |
5962 | if (!final_tried) { | |
5963 | num_bytes = num_bytes >> 1; | |
5964 | num_bytes = num_bytes & ~(root->sectorsize - 1); | |
5965 | num_bytes = max(num_bytes, min_alloc_size); | |
79787eaa | 5966 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
9e622d6b | 5967 | num_bytes, data, CHUNK_ALLOC_FORCE); |
79787eaa JM |
5968 | if (ret < 0 && ret != -ENOSPC) { |
5969 | btrfs_abort_transaction(trans, root, ret); | |
5970 | return ret; | |
5971 | } | |
9e622d6b MX |
5972 | if (num_bytes == min_alloc_size) |
5973 | final_tried = true; | |
5974 | goto again; | |
5975 | } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
5976 | struct btrfs_space_info *sinfo; | |
5977 | ||
5978 | sinfo = __find_space_info(root->fs_info, data); | |
5979 | printk(KERN_ERR "btrfs allocation failed flags %llu, " | |
5980 | "wanted %llu\n", (unsigned long long)data, | |
5981 | (unsigned long long)num_bytes); | |
53804280 JM |
5982 | if (sinfo) |
5983 | dump_space_info(sinfo, num_bytes, 1); | |
9e622d6b | 5984 | } |
925baedd | 5985 | } |
0f9dd46c | 5986 | |
1abe9b8a | 5987 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
5988 | ||
0f9dd46c | 5989 | return ret; |
e6dcd2dc CM |
5990 | } |
5991 | ||
e688b725 CM |
5992 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, |
5993 | u64 start, u64 len, int pin) | |
65b51a00 | 5994 | { |
0f9dd46c | 5995 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 5996 | int ret = 0; |
0f9dd46c | 5997 | |
0f9dd46c JB |
5998 | cache = btrfs_lookup_block_group(root->fs_info, start); |
5999 | if (!cache) { | |
d397712b CM |
6000 | printk(KERN_ERR "Unable to find block group for %llu\n", |
6001 | (unsigned long long)start); | |
0f9dd46c JB |
6002 | return -ENOSPC; |
6003 | } | |
1f3c79a2 | 6004 | |
5378e607 LD |
6005 | if (btrfs_test_opt(root, DISCARD)) |
6006 | ret = btrfs_discard_extent(root, start, len, NULL); | |
1f3c79a2 | 6007 | |
e688b725 CM |
6008 | if (pin) |
6009 | pin_down_extent(root, cache, start, len, 1); | |
6010 | else { | |
6011 | btrfs_add_free_space(cache, start, len); | |
6012 | btrfs_update_reserved_bytes(cache, len, RESERVE_FREE); | |
6013 | } | |
fa9c0d79 | 6014 | btrfs_put_block_group(cache); |
817d52f8 | 6015 | |
1abe9b8a | 6016 | trace_btrfs_reserved_extent_free(root, start, len); |
6017 | ||
e6dcd2dc CM |
6018 | return ret; |
6019 | } | |
6020 | ||
e688b725 CM |
6021 | int btrfs_free_reserved_extent(struct btrfs_root *root, |
6022 | u64 start, u64 len) | |
6023 | { | |
6024 | return __btrfs_free_reserved_extent(root, start, len, 0); | |
6025 | } | |
6026 | ||
6027 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, | |
6028 | u64 start, u64 len) | |
6029 | { | |
6030 | return __btrfs_free_reserved_extent(root, start, len, 1); | |
6031 | } | |
6032 | ||
5d4f98a2 YZ |
6033 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
6034 | struct btrfs_root *root, | |
6035 | u64 parent, u64 root_objectid, | |
6036 | u64 flags, u64 owner, u64 offset, | |
6037 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
6038 | { |
6039 | int ret; | |
5d4f98a2 | 6040 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 6041 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 6042 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 6043 | struct btrfs_path *path; |
5d4f98a2 YZ |
6044 | struct extent_buffer *leaf; |
6045 | int type; | |
6046 | u32 size; | |
26b8003f | 6047 | |
5d4f98a2 YZ |
6048 | if (parent > 0) |
6049 | type = BTRFS_SHARED_DATA_REF_KEY; | |
6050 | else | |
6051 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 6052 | |
5d4f98a2 | 6053 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
6054 | |
6055 | path = btrfs_alloc_path(); | |
db5b493a TI |
6056 | if (!path) |
6057 | return -ENOMEM; | |
47e4bb98 | 6058 | |
b9473439 | 6059 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6060 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
6061 | ins, size); | |
79787eaa JM |
6062 | if (ret) { |
6063 | btrfs_free_path(path); | |
6064 | return ret; | |
6065 | } | |
0f9dd46c | 6066 | |
5d4f98a2 YZ |
6067 | leaf = path->nodes[0]; |
6068 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 6069 | struct btrfs_extent_item); |
5d4f98a2 YZ |
6070 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
6071 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
6072 | btrfs_set_extent_flags(leaf, extent_item, | |
6073 | flags | BTRFS_EXTENT_FLAG_DATA); | |
6074 | ||
6075 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
6076 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
6077 | if (parent > 0) { | |
6078 | struct btrfs_shared_data_ref *ref; | |
6079 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
6080 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
6081 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
6082 | } else { | |
6083 | struct btrfs_extent_data_ref *ref; | |
6084 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
6085 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
6086 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
6087 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
6088 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
6089 | } | |
47e4bb98 CM |
6090 | |
6091 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 6092 | btrfs_free_path(path); |
f510cfec | 6093 | |
f0486c68 | 6094 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
79787eaa | 6095 | if (ret) { /* -ENOENT, logic error */ |
d397712b CM |
6096 | printk(KERN_ERR "btrfs update block group failed for %llu " |
6097 | "%llu\n", (unsigned long long)ins->objectid, | |
6098 | (unsigned long long)ins->offset); | |
f5947066 CM |
6099 | BUG(); |
6100 | } | |
e6dcd2dc CM |
6101 | return ret; |
6102 | } | |
6103 | ||
5d4f98a2 YZ |
6104 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
6105 | struct btrfs_root *root, | |
6106 | u64 parent, u64 root_objectid, | |
6107 | u64 flags, struct btrfs_disk_key *key, | |
6108 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
6109 | { |
6110 | int ret; | |
5d4f98a2 YZ |
6111 | struct btrfs_fs_info *fs_info = root->fs_info; |
6112 | struct btrfs_extent_item *extent_item; | |
6113 | struct btrfs_tree_block_info *block_info; | |
6114 | struct btrfs_extent_inline_ref *iref; | |
6115 | struct btrfs_path *path; | |
6116 | struct extent_buffer *leaf; | |
6117 | u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref); | |
1c2308f8 | 6118 | |
5d4f98a2 | 6119 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6120 | if (!path) |
6121 | return -ENOMEM; | |
56bec294 | 6122 | |
5d4f98a2 YZ |
6123 | path->leave_spinning = 1; |
6124 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
6125 | ins, size); | |
79787eaa JM |
6126 | if (ret) { |
6127 | btrfs_free_path(path); | |
6128 | return ret; | |
6129 | } | |
5d4f98a2 YZ |
6130 | |
6131 | leaf = path->nodes[0]; | |
6132 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
6133 | struct btrfs_extent_item); | |
6134 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
6135 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
6136 | btrfs_set_extent_flags(leaf, extent_item, | |
6137 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
6138 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
6139 | ||
6140 | btrfs_set_tree_block_key(leaf, block_info, key); | |
6141 | btrfs_set_tree_block_level(leaf, block_info, level); | |
6142 | ||
6143 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
6144 | if (parent > 0) { | |
6145 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
6146 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6147 | BTRFS_SHARED_BLOCK_REF_KEY); | |
6148 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
6149 | } else { | |
6150 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6151 | BTRFS_TREE_BLOCK_REF_KEY); | |
6152 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
6153 | } | |
6154 | ||
6155 | btrfs_mark_buffer_dirty(leaf); | |
6156 | btrfs_free_path(path); | |
6157 | ||
f0486c68 | 6158 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
79787eaa | 6159 | if (ret) { /* -ENOENT, logic error */ |
5d4f98a2 YZ |
6160 | printk(KERN_ERR "btrfs update block group failed for %llu " |
6161 | "%llu\n", (unsigned long long)ins->objectid, | |
6162 | (unsigned long long)ins->offset); | |
6163 | BUG(); | |
6164 | } | |
6165 | return ret; | |
6166 | } | |
6167 | ||
6168 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
6169 | struct btrfs_root *root, | |
6170 | u64 root_objectid, u64 owner, | |
6171 | u64 offset, struct btrfs_key *ins) | |
6172 | { | |
6173 | int ret; | |
6174 | ||
6175 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
6176 | ||
66d7e7f0 AJ |
6177 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, |
6178 | ins->offset, 0, | |
6179 | root_objectid, owner, offset, | |
6180 | BTRFS_ADD_DELAYED_EXTENT, NULL, 0); | |
e6dcd2dc CM |
6181 | return ret; |
6182 | } | |
e02119d5 CM |
6183 | |
6184 | /* | |
6185 | * this is used by the tree logging recovery code. It records that | |
6186 | * an extent has been allocated and makes sure to clear the free | |
6187 | * space cache bits as well | |
6188 | */ | |
5d4f98a2 YZ |
6189 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
6190 | struct btrfs_root *root, | |
6191 | u64 root_objectid, u64 owner, u64 offset, | |
6192 | struct btrfs_key *ins) | |
e02119d5 CM |
6193 | { |
6194 | int ret; | |
6195 | struct btrfs_block_group_cache *block_group; | |
11833d66 YZ |
6196 | struct btrfs_caching_control *caching_ctl; |
6197 | u64 start = ins->objectid; | |
6198 | u64 num_bytes = ins->offset; | |
e02119d5 | 6199 | |
e02119d5 | 6200 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
b8399dee | 6201 | cache_block_group(block_group, trans, NULL, 0); |
11833d66 | 6202 | caching_ctl = get_caching_control(block_group); |
e02119d5 | 6203 | |
11833d66 YZ |
6204 | if (!caching_ctl) { |
6205 | BUG_ON(!block_group_cache_done(block_group)); | |
6206 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
79787eaa | 6207 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6208 | } else { |
6209 | mutex_lock(&caching_ctl->mutex); | |
6210 | ||
6211 | if (start >= caching_ctl->progress) { | |
6212 | ret = add_excluded_extent(root, start, num_bytes); | |
79787eaa | 6213 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6214 | } else if (start + num_bytes <= caching_ctl->progress) { |
6215 | ret = btrfs_remove_free_space(block_group, | |
6216 | start, num_bytes); | |
79787eaa | 6217 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6218 | } else { |
6219 | num_bytes = caching_ctl->progress - start; | |
6220 | ret = btrfs_remove_free_space(block_group, | |
6221 | start, num_bytes); | |
79787eaa | 6222 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6223 | |
6224 | start = caching_ctl->progress; | |
6225 | num_bytes = ins->objectid + ins->offset - | |
6226 | caching_ctl->progress; | |
6227 | ret = add_excluded_extent(root, start, num_bytes); | |
79787eaa | 6228 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6229 | } |
6230 | ||
6231 | mutex_unlock(&caching_ctl->mutex); | |
6232 | put_caching_control(caching_ctl); | |
6233 | } | |
6234 | ||
fb25e914 JB |
6235 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, |
6236 | RESERVE_ALLOC_NO_ACCOUNT); | |
79787eaa | 6237 | BUG_ON(ret); /* logic error */ |
fa9c0d79 | 6238 | btrfs_put_block_group(block_group); |
5d4f98a2 YZ |
6239 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
6240 | 0, owner, offset, ins, 1); | |
e02119d5 CM |
6241 | return ret; |
6242 | } | |
6243 | ||
65b51a00 CM |
6244 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, |
6245 | struct btrfs_root *root, | |
4008c04a CM |
6246 | u64 bytenr, u32 blocksize, |
6247 | int level) | |
65b51a00 CM |
6248 | { |
6249 | struct extent_buffer *buf; | |
6250 | ||
6251 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
6252 | if (!buf) | |
6253 | return ERR_PTR(-ENOMEM); | |
6254 | btrfs_set_header_generation(buf, trans->transid); | |
85d4e461 | 6255 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 CM |
6256 | btrfs_tree_lock(buf); |
6257 | clean_tree_block(trans, root, buf); | |
3083ee2e | 6258 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
6259 | |
6260 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 6261 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 6262 | |
d0c803c4 | 6263 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
8cef4e16 YZ |
6264 | /* |
6265 | * we allow two log transactions at a time, use different | |
6266 | * EXENT bit to differentiate dirty pages. | |
6267 | */ | |
6268 | if (root->log_transid % 2 == 0) | |
6269 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
6270 | buf->start + buf->len - 1, GFP_NOFS); | |
6271 | else | |
6272 | set_extent_new(&root->dirty_log_pages, buf->start, | |
6273 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 CM |
6274 | } else { |
6275 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 6276 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 6277 | } |
65b51a00 | 6278 | trans->blocks_used++; |
b4ce94de | 6279 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
6280 | return buf; |
6281 | } | |
6282 | ||
f0486c68 YZ |
6283 | static struct btrfs_block_rsv * |
6284 | use_block_rsv(struct btrfs_trans_handle *trans, | |
6285 | struct btrfs_root *root, u32 blocksize) | |
6286 | { | |
6287 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 6288 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 YZ |
6289 | int ret; |
6290 | ||
6291 | block_rsv = get_block_rsv(trans, root); | |
6292 | ||
6293 | if (block_rsv->size == 0) { | |
36ba022a | 6294 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0); |
68a82277 JB |
6295 | /* |
6296 | * If we couldn't reserve metadata bytes try and use some from | |
6297 | * the global reserve. | |
6298 | */ | |
6299 | if (ret && block_rsv != global_rsv) { | |
6300 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
6301 | if (!ret) | |
6302 | return global_rsv; | |
f0486c68 | 6303 | return ERR_PTR(ret); |
68a82277 | 6304 | } else if (ret) { |
f0486c68 | 6305 | return ERR_PTR(ret); |
68a82277 | 6306 | } |
f0486c68 YZ |
6307 | return block_rsv; |
6308 | } | |
6309 | ||
6310 | ret = block_rsv_use_bytes(block_rsv, blocksize); | |
6311 | if (!ret) | |
6312 | return block_rsv; | |
ca7e70f5 | 6313 | if (ret && !block_rsv->failfast) { |
dff51cd1 DS |
6314 | static DEFINE_RATELIMIT_STATE(_rs, |
6315 | DEFAULT_RATELIMIT_INTERVAL, | |
6316 | /*DEFAULT_RATELIMIT_BURST*/ 2); | |
6317 | if (__ratelimit(&_rs)) { | |
6318 | printk(KERN_DEBUG "btrfs: block rsv returned %d\n", ret); | |
6319 | WARN_ON(1); | |
6320 | } | |
36ba022a | 6321 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, 0); |
68a82277 | 6322 | if (!ret) { |
68a82277 JB |
6323 | return block_rsv; |
6324 | } else if (ret && block_rsv != global_rsv) { | |
6325 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
6326 | if (!ret) | |
6327 | return global_rsv; | |
6328 | } | |
6329 | } | |
f0486c68 | 6330 | |
f0486c68 YZ |
6331 | return ERR_PTR(-ENOSPC); |
6332 | } | |
6333 | ||
8c2a3ca2 JB |
6334 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
6335 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
6336 | { |
6337 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 6338 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
6339 | } |
6340 | ||
fec577fb | 6341 | /* |
f0486c68 YZ |
6342 | * finds a free extent and does all the dirty work required for allocation |
6343 | * returns the key for the extent through ins, and a tree buffer for | |
6344 | * the first block of the extent through buf. | |
6345 | * | |
fec577fb CM |
6346 | * returns the tree buffer or NULL. |
6347 | */ | |
5f39d397 | 6348 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
6349 | struct btrfs_root *root, u32 blocksize, |
6350 | u64 parent, u64 root_objectid, | |
6351 | struct btrfs_disk_key *key, int level, | |
5581a51a | 6352 | u64 hint, u64 empty_size) |
fec577fb | 6353 | { |
e2fa7227 | 6354 | struct btrfs_key ins; |
f0486c68 | 6355 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 6356 | struct extent_buffer *buf; |
f0486c68 YZ |
6357 | u64 flags = 0; |
6358 | int ret; | |
6359 | ||
fec577fb | 6360 | |
f0486c68 YZ |
6361 | block_rsv = use_block_rsv(trans, root, blocksize); |
6362 | if (IS_ERR(block_rsv)) | |
6363 | return ERR_CAST(block_rsv); | |
6364 | ||
6365 | ret = btrfs_reserve_extent(trans, root, blocksize, blocksize, | |
81c9ad23 | 6366 | empty_size, hint, &ins, 0); |
fec577fb | 6367 | if (ret) { |
8c2a3ca2 | 6368 | unuse_block_rsv(root->fs_info, block_rsv, blocksize); |
54aa1f4d | 6369 | return ERR_PTR(ret); |
fec577fb | 6370 | } |
55c69072 | 6371 | |
4008c04a CM |
6372 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
6373 | blocksize, level); | |
79787eaa | 6374 | BUG_ON(IS_ERR(buf)); /* -ENOMEM */ |
f0486c68 YZ |
6375 | |
6376 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
6377 | if (parent == 0) | |
6378 | parent = ins.objectid; | |
6379 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6380 | } else | |
6381 | BUG_ON(parent > 0); | |
6382 | ||
6383 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
6384 | struct btrfs_delayed_extent_op *extent_op; | |
6385 | extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); | |
79787eaa | 6386 | BUG_ON(!extent_op); /* -ENOMEM */ |
f0486c68 YZ |
6387 | if (key) |
6388 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
6389 | else | |
6390 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
6391 | extent_op->flags_to_set = flags; | |
6392 | extent_op->update_key = 1; | |
6393 | extent_op->update_flags = 1; | |
6394 | extent_op->is_data = 0; | |
6395 | ||
66d7e7f0 AJ |
6396 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
6397 | ins.objectid, | |
f0486c68 YZ |
6398 | ins.offset, parent, root_objectid, |
6399 | level, BTRFS_ADD_DELAYED_EXTENT, | |
5581a51a | 6400 | extent_op, 0); |
79787eaa | 6401 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 | 6402 | } |
fec577fb CM |
6403 | return buf; |
6404 | } | |
a28ec197 | 6405 | |
2c47e605 YZ |
6406 | struct walk_control { |
6407 | u64 refs[BTRFS_MAX_LEVEL]; | |
6408 | u64 flags[BTRFS_MAX_LEVEL]; | |
6409 | struct btrfs_key update_progress; | |
6410 | int stage; | |
6411 | int level; | |
6412 | int shared_level; | |
6413 | int update_ref; | |
6414 | int keep_locks; | |
1c4850e2 YZ |
6415 | int reada_slot; |
6416 | int reada_count; | |
66d7e7f0 | 6417 | int for_reloc; |
2c47e605 YZ |
6418 | }; |
6419 | ||
6420 | #define DROP_REFERENCE 1 | |
6421 | #define UPDATE_BACKREF 2 | |
6422 | ||
1c4850e2 YZ |
6423 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
6424 | struct btrfs_root *root, | |
6425 | struct walk_control *wc, | |
6426 | struct btrfs_path *path) | |
6407bf6d | 6427 | { |
1c4850e2 YZ |
6428 | u64 bytenr; |
6429 | u64 generation; | |
6430 | u64 refs; | |
94fcca9f | 6431 | u64 flags; |
5d4f98a2 | 6432 | u32 nritems; |
1c4850e2 YZ |
6433 | u32 blocksize; |
6434 | struct btrfs_key key; | |
6435 | struct extent_buffer *eb; | |
6407bf6d | 6436 | int ret; |
1c4850e2 YZ |
6437 | int slot; |
6438 | int nread = 0; | |
6407bf6d | 6439 | |
1c4850e2 YZ |
6440 | if (path->slots[wc->level] < wc->reada_slot) { |
6441 | wc->reada_count = wc->reada_count * 2 / 3; | |
6442 | wc->reada_count = max(wc->reada_count, 2); | |
6443 | } else { | |
6444 | wc->reada_count = wc->reada_count * 3 / 2; | |
6445 | wc->reada_count = min_t(int, wc->reada_count, | |
6446 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
6447 | } | |
7bb86316 | 6448 | |
1c4850e2 YZ |
6449 | eb = path->nodes[wc->level]; |
6450 | nritems = btrfs_header_nritems(eb); | |
6451 | blocksize = btrfs_level_size(root, wc->level - 1); | |
bd56b302 | 6452 | |
1c4850e2 YZ |
6453 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
6454 | if (nread >= wc->reada_count) | |
6455 | break; | |
bd56b302 | 6456 | |
2dd3e67b | 6457 | cond_resched(); |
1c4850e2 YZ |
6458 | bytenr = btrfs_node_blockptr(eb, slot); |
6459 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 6460 | |
1c4850e2 YZ |
6461 | if (slot == path->slots[wc->level]) |
6462 | goto reada; | |
5d4f98a2 | 6463 | |
1c4850e2 YZ |
6464 | if (wc->stage == UPDATE_BACKREF && |
6465 | generation <= root->root_key.offset) | |
bd56b302 CM |
6466 | continue; |
6467 | ||
94fcca9f YZ |
6468 | /* We don't lock the tree block, it's OK to be racy here */ |
6469 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, | |
6470 | &refs, &flags); | |
79787eaa JM |
6471 | /* We don't care about errors in readahead. */ |
6472 | if (ret < 0) | |
6473 | continue; | |
94fcca9f YZ |
6474 | BUG_ON(refs == 0); |
6475 | ||
1c4850e2 | 6476 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
6477 | if (refs == 1) |
6478 | goto reada; | |
bd56b302 | 6479 | |
94fcca9f YZ |
6480 | if (wc->level == 1 && |
6481 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6482 | continue; | |
1c4850e2 YZ |
6483 | if (!wc->update_ref || |
6484 | generation <= root->root_key.offset) | |
6485 | continue; | |
6486 | btrfs_node_key_to_cpu(eb, &key, slot); | |
6487 | ret = btrfs_comp_cpu_keys(&key, | |
6488 | &wc->update_progress); | |
6489 | if (ret < 0) | |
6490 | continue; | |
94fcca9f YZ |
6491 | } else { |
6492 | if (wc->level == 1 && | |
6493 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6494 | continue; | |
6407bf6d | 6495 | } |
1c4850e2 YZ |
6496 | reada: |
6497 | ret = readahead_tree_block(root, bytenr, blocksize, | |
6498 | generation); | |
6499 | if (ret) | |
bd56b302 | 6500 | break; |
1c4850e2 | 6501 | nread++; |
20524f02 | 6502 | } |
1c4850e2 | 6503 | wc->reada_slot = slot; |
20524f02 | 6504 | } |
2c47e605 | 6505 | |
f82d02d9 | 6506 | /* |
2c47e605 YZ |
6507 | * hepler to process tree block while walking down the tree. |
6508 | * | |
2c47e605 YZ |
6509 | * when wc->stage == UPDATE_BACKREF, this function updates |
6510 | * back refs for pointers in the block. | |
6511 | * | |
6512 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 6513 | */ |
2c47e605 | 6514 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 6515 | struct btrfs_root *root, |
2c47e605 | 6516 | struct btrfs_path *path, |
94fcca9f | 6517 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 6518 | { |
2c47e605 YZ |
6519 | int level = wc->level; |
6520 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 6521 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
6522 | int ret; |
6523 | ||
2c47e605 YZ |
6524 | if (wc->stage == UPDATE_BACKREF && |
6525 | btrfs_header_owner(eb) != root->root_key.objectid) | |
6526 | return 1; | |
f82d02d9 | 6527 | |
2c47e605 YZ |
6528 | /* |
6529 | * when reference count of tree block is 1, it won't increase | |
6530 | * again. once full backref flag is set, we never clear it. | |
6531 | */ | |
94fcca9f YZ |
6532 | if (lookup_info && |
6533 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
6534 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
6535 | BUG_ON(!path->locks[level]); |
6536 | ret = btrfs_lookup_extent_info(trans, root, | |
6537 | eb->start, eb->len, | |
6538 | &wc->refs[level], | |
6539 | &wc->flags[level]); | |
79787eaa JM |
6540 | BUG_ON(ret == -ENOMEM); |
6541 | if (ret) | |
6542 | return ret; | |
2c47e605 YZ |
6543 | BUG_ON(wc->refs[level] == 0); |
6544 | } | |
5d4f98a2 | 6545 | |
2c47e605 YZ |
6546 | if (wc->stage == DROP_REFERENCE) { |
6547 | if (wc->refs[level] > 1) | |
6548 | return 1; | |
f82d02d9 | 6549 | |
2c47e605 | 6550 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 6551 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6552 | path->locks[level] = 0; |
6553 | } | |
6554 | return 0; | |
6555 | } | |
f82d02d9 | 6556 | |
2c47e605 YZ |
6557 | /* wc->stage == UPDATE_BACKREF */ |
6558 | if (!(wc->flags[level] & flag)) { | |
6559 | BUG_ON(!path->locks[level]); | |
66d7e7f0 | 6560 | ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc); |
79787eaa | 6561 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 6562 | ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc); |
79787eaa | 6563 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6564 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, |
6565 | eb->len, flag, 0); | |
79787eaa | 6566 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6567 | wc->flags[level] |= flag; |
6568 | } | |
6569 | ||
6570 | /* | |
6571 | * the block is shared by multiple trees, so it's not good to | |
6572 | * keep the tree lock | |
6573 | */ | |
6574 | if (path->locks[level] && level > 0) { | |
bd681513 | 6575 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6576 | path->locks[level] = 0; |
6577 | } | |
6578 | return 0; | |
6579 | } | |
6580 | ||
1c4850e2 YZ |
6581 | /* |
6582 | * hepler to process tree block pointer. | |
6583 | * | |
6584 | * when wc->stage == DROP_REFERENCE, this function checks | |
6585 | * reference count of the block pointed to. if the block | |
6586 | * is shared and we need update back refs for the subtree | |
6587 | * rooted at the block, this function changes wc->stage to | |
6588 | * UPDATE_BACKREF. if the block is shared and there is no | |
6589 | * need to update back, this function drops the reference | |
6590 | * to the block. | |
6591 | * | |
6592 | * NOTE: return value 1 means we should stop walking down. | |
6593 | */ | |
6594 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
6595 | struct btrfs_root *root, | |
6596 | struct btrfs_path *path, | |
94fcca9f | 6597 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
6598 | { |
6599 | u64 bytenr; | |
6600 | u64 generation; | |
6601 | u64 parent; | |
6602 | u32 blocksize; | |
6603 | struct btrfs_key key; | |
6604 | struct extent_buffer *next; | |
6605 | int level = wc->level; | |
6606 | int reada = 0; | |
6607 | int ret = 0; | |
6608 | ||
6609 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
6610 | path->slots[level]); | |
6611 | /* | |
6612 | * if the lower level block was created before the snapshot | |
6613 | * was created, we know there is no need to update back refs | |
6614 | * for the subtree | |
6615 | */ | |
6616 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
6617 | generation <= root->root_key.offset) { |
6618 | *lookup_info = 1; | |
1c4850e2 | 6619 | return 1; |
94fcca9f | 6620 | } |
1c4850e2 YZ |
6621 | |
6622 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
6623 | blocksize = btrfs_level_size(root, level - 1); | |
6624 | ||
6625 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
6626 | if (!next) { | |
6627 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
90d2c51d MX |
6628 | if (!next) |
6629 | return -ENOMEM; | |
1c4850e2 YZ |
6630 | reada = 1; |
6631 | } | |
6632 | btrfs_tree_lock(next); | |
6633 | btrfs_set_lock_blocking(next); | |
6634 | ||
94fcca9f YZ |
6635 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, |
6636 | &wc->refs[level - 1], | |
6637 | &wc->flags[level - 1]); | |
79787eaa JM |
6638 | if (ret < 0) { |
6639 | btrfs_tree_unlock(next); | |
6640 | return ret; | |
6641 | } | |
6642 | ||
94fcca9f YZ |
6643 | BUG_ON(wc->refs[level - 1] == 0); |
6644 | *lookup_info = 0; | |
1c4850e2 | 6645 | |
94fcca9f | 6646 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 6647 | if (wc->refs[level - 1] > 1) { |
94fcca9f YZ |
6648 | if (level == 1 && |
6649 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6650 | goto skip; | |
6651 | ||
1c4850e2 YZ |
6652 | if (!wc->update_ref || |
6653 | generation <= root->root_key.offset) | |
6654 | goto skip; | |
6655 | ||
6656 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
6657 | path->slots[level]); | |
6658 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
6659 | if (ret < 0) | |
6660 | goto skip; | |
6661 | ||
6662 | wc->stage = UPDATE_BACKREF; | |
6663 | wc->shared_level = level - 1; | |
6664 | } | |
94fcca9f YZ |
6665 | } else { |
6666 | if (level == 1 && | |
6667 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6668 | goto skip; | |
1c4850e2 YZ |
6669 | } |
6670 | ||
b9fab919 | 6671 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
6672 | btrfs_tree_unlock(next); |
6673 | free_extent_buffer(next); | |
6674 | next = NULL; | |
94fcca9f | 6675 | *lookup_info = 1; |
1c4850e2 YZ |
6676 | } |
6677 | ||
6678 | if (!next) { | |
6679 | if (reada && level == 1) | |
6680 | reada_walk_down(trans, root, wc, path); | |
6681 | next = read_tree_block(root, bytenr, blocksize, generation); | |
97d9a8a4 TI |
6682 | if (!next) |
6683 | return -EIO; | |
1c4850e2 YZ |
6684 | btrfs_tree_lock(next); |
6685 | btrfs_set_lock_blocking(next); | |
6686 | } | |
6687 | ||
6688 | level--; | |
6689 | BUG_ON(level != btrfs_header_level(next)); | |
6690 | path->nodes[level] = next; | |
6691 | path->slots[level] = 0; | |
bd681513 | 6692 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
6693 | wc->level = level; |
6694 | if (wc->level == 1) | |
6695 | wc->reada_slot = 0; | |
6696 | return 0; | |
6697 | skip: | |
6698 | wc->refs[level - 1] = 0; | |
6699 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
6700 | if (wc->stage == DROP_REFERENCE) { |
6701 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
6702 | parent = path->nodes[level]->start; | |
6703 | } else { | |
6704 | BUG_ON(root->root_key.objectid != | |
6705 | btrfs_header_owner(path->nodes[level])); | |
6706 | parent = 0; | |
6707 | } | |
1c4850e2 | 6708 | |
94fcca9f | 6709 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
66d7e7f0 | 6710 | root->root_key.objectid, level - 1, 0, 0); |
79787eaa | 6711 | BUG_ON(ret); /* -ENOMEM */ |
1c4850e2 | 6712 | } |
1c4850e2 YZ |
6713 | btrfs_tree_unlock(next); |
6714 | free_extent_buffer(next); | |
94fcca9f | 6715 | *lookup_info = 1; |
1c4850e2 YZ |
6716 | return 1; |
6717 | } | |
6718 | ||
2c47e605 YZ |
6719 | /* |
6720 | * hepler to process tree block while walking up the tree. | |
6721 | * | |
6722 | * when wc->stage == DROP_REFERENCE, this function drops | |
6723 | * reference count on the block. | |
6724 | * | |
6725 | * when wc->stage == UPDATE_BACKREF, this function changes | |
6726 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
6727 | * to UPDATE_BACKREF previously while processing the block. | |
6728 | * | |
6729 | * NOTE: return value 1 means we should stop walking up. | |
6730 | */ | |
6731 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
6732 | struct btrfs_root *root, | |
6733 | struct btrfs_path *path, | |
6734 | struct walk_control *wc) | |
6735 | { | |
f0486c68 | 6736 | int ret; |
2c47e605 YZ |
6737 | int level = wc->level; |
6738 | struct extent_buffer *eb = path->nodes[level]; | |
6739 | u64 parent = 0; | |
6740 | ||
6741 | if (wc->stage == UPDATE_BACKREF) { | |
6742 | BUG_ON(wc->shared_level < level); | |
6743 | if (level < wc->shared_level) | |
6744 | goto out; | |
6745 | ||
2c47e605 YZ |
6746 | ret = find_next_key(path, level + 1, &wc->update_progress); |
6747 | if (ret > 0) | |
6748 | wc->update_ref = 0; | |
6749 | ||
6750 | wc->stage = DROP_REFERENCE; | |
6751 | wc->shared_level = -1; | |
6752 | path->slots[level] = 0; | |
6753 | ||
6754 | /* | |
6755 | * check reference count again if the block isn't locked. | |
6756 | * we should start walking down the tree again if reference | |
6757 | * count is one. | |
6758 | */ | |
6759 | if (!path->locks[level]) { | |
6760 | BUG_ON(level == 0); | |
6761 | btrfs_tree_lock(eb); | |
6762 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6763 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6764 | |
6765 | ret = btrfs_lookup_extent_info(trans, root, | |
6766 | eb->start, eb->len, | |
6767 | &wc->refs[level], | |
6768 | &wc->flags[level]); | |
79787eaa JM |
6769 | if (ret < 0) { |
6770 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
6771 | return ret; | |
6772 | } | |
2c47e605 YZ |
6773 | BUG_ON(wc->refs[level] == 0); |
6774 | if (wc->refs[level] == 1) { | |
bd681513 | 6775 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6776 | return 1; |
6777 | } | |
f82d02d9 | 6778 | } |
2c47e605 | 6779 | } |
f82d02d9 | 6780 | |
2c47e605 YZ |
6781 | /* wc->stage == DROP_REFERENCE */ |
6782 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 6783 | |
2c47e605 YZ |
6784 | if (wc->refs[level] == 1) { |
6785 | if (level == 0) { | |
6786 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
66d7e7f0 AJ |
6787 | ret = btrfs_dec_ref(trans, root, eb, 1, |
6788 | wc->for_reloc); | |
2c47e605 | 6789 | else |
66d7e7f0 AJ |
6790 | ret = btrfs_dec_ref(trans, root, eb, 0, |
6791 | wc->for_reloc); | |
79787eaa | 6792 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6793 | } |
6794 | /* make block locked assertion in clean_tree_block happy */ | |
6795 | if (!path->locks[level] && | |
6796 | btrfs_header_generation(eb) == trans->transid) { | |
6797 | btrfs_tree_lock(eb); | |
6798 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6799 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6800 | } |
6801 | clean_tree_block(trans, root, eb); | |
6802 | } | |
6803 | ||
6804 | if (eb == root->node) { | |
6805 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6806 | parent = eb->start; | |
6807 | else | |
6808 | BUG_ON(root->root_key.objectid != | |
6809 | btrfs_header_owner(eb)); | |
6810 | } else { | |
6811 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6812 | parent = path->nodes[level + 1]->start; | |
6813 | else | |
6814 | BUG_ON(root->root_key.objectid != | |
6815 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 6816 | } |
f82d02d9 | 6817 | |
5581a51a | 6818 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
6819 | out: |
6820 | wc->refs[level] = 0; | |
6821 | wc->flags[level] = 0; | |
f0486c68 | 6822 | return 0; |
2c47e605 YZ |
6823 | } |
6824 | ||
6825 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
6826 | struct btrfs_root *root, | |
6827 | struct btrfs_path *path, | |
6828 | struct walk_control *wc) | |
6829 | { | |
2c47e605 | 6830 | int level = wc->level; |
94fcca9f | 6831 | int lookup_info = 1; |
2c47e605 YZ |
6832 | int ret; |
6833 | ||
6834 | while (level >= 0) { | |
94fcca9f | 6835 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
6836 | if (ret > 0) |
6837 | break; | |
6838 | ||
6839 | if (level == 0) | |
6840 | break; | |
6841 | ||
7a7965f8 YZ |
6842 | if (path->slots[level] >= |
6843 | btrfs_header_nritems(path->nodes[level])) | |
6844 | break; | |
6845 | ||
94fcca9f | 6846 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
6847 | if (ret > 0) { |
6848 | path->slots[level]++; | |
6849 | continue; | |
90d2c51d MX |
6850 | } else if (ret < 0) |
6851 | return ret; | |
1c4850e2 | 6852 | level = wc->level; |
f82d02d9 | 6853 | } |
f82d02d9 YZ |
6854 | return 0; |
6855 | } | |
6856 | ||
d397712b | 6857 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 6858 | struct btrfs_root *root, |
f82d02d9 | 6859 | struct btrfs_path *path, |
2c47e605 | 6860 | struct walk_control *wc, int max_level) |
20524f02 | 6861 | { |
2c47e605 | 6862 | int level = wc->level; |
20524f02 | 6863 | int ret; |
9f3a7427 | 6864 | |
2c47e605 YZ |
6865 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
6866 | while (level < max_level && path->nodes[level]) { | |
6867 | wc->level = level; | |
6868 | if (path->slots[level] + 1 < | |
6869 | btrfs_header_nritems(path->nodes[level])) { | |
6870 | path->slots[level]++; | |
20524f02 CM |
6871 | return 0; |
6872 | } else { | |
2c47e605 YZ |
6873 | ret = walk_up_proc(trans, root, path, wc); |
6874 | if (ret > 0) | |
6875 | return 0; | |
bd56b302 | 6876 | |
2c47e605 | 6877 | if (path->locks[level]) { |
bd681513 CM |
6878 | btrfs_tree_unlock_rw(path->nodes[level], |
6879 | path->locks[level]); | |
2c47e605 | 6880 | path->locks[level] = 0; |
f82d02d9 | 6881 | } |
2c47e605 YZ |
6882 | free_extent_buffer(path->nodes[level]); |
6883 | path->nodes[level] = NULL; | |
6884 | level++; | |
20524f02 CM |
6885 | } |
6886 | } | |
6887 | return 1; | |
6888 | } | |
6889 | ||
9aca1d51 | 6890 | /* |
2c47e605 YZ |
6891 | * drop a subvolume tree. |
6892 | * | |
6893 | * this function traverses the tree freeing any blocks that only | |
6894 | * referenced by the tree. | |
6895 | * | |
6896 | * when a shared tree block is found. this function decreases its | |
6897 | * reference count by one. if update_ref is true, this function | |
6898 | * also make sure backrefs for the shared block and all lower level | |
6899 | * blocks are properly updated. | |
9aca1d51 | 6900 | */ |
2c536799 | 6901 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
6902 | struct btrfs_block_rsv *block_rsv, int update_ref, |
6903 | int for_reloc) | |
20524f02 | 6904 | { |
5caf2a00 | 6905 | struct btrfs_path *path; |
2c47e605 YZ |
6906 | struct btrfs_trans_handle *trans; |
6907 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 6908 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
6909 | struct walk_control *wc; |
6910 | struct btrfs_key key; | |
6911 | int err = 0; | |
6912 | int ret; | |
6913 | int level; | |
20524f02 | 6914 | |
5caf2a00 | 6915 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
6916 | if (!path) { |
6917 | err = -ENOMEM; | |
6918 | goto out; | |
6919 | } | |
20524f02 | 6920 | |
2c47e605 | 6921 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
6922 | if (!wc) { |
6923 | btrfs_free_path(path); | |
cb1b69f4 TI |
6924 | err = -ENOMEM; |
6925 | goto out; | |
38a1a919 | 6926 | } |
2c47e605 | 6927 | |
a22285a6 | 6928 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
6929 | if (IS_ERR(trans)) { |
6930 | err = PTR_ERR(trans); | |
6931 | goto out_free; | |
6932 | } | |
98d5dc13 | 6933 | |
3fd0a558 YZ |
6934 | if (block_rsv) |
6935 | trans->block_rsv = block_rsv; | |
2c47e605 | 6936 | |
9f3a7427 | 6937 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 6938 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
6939 | path->nodes[level] = btrfs_lock_root_node(root); |
6940 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 6941 | path->slots[level] = 0; |
bd681513 | 6942 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6943 | memset(&wc->update_progress, 0, |
6944 | sizeof(wc->update_progress)); | |
9f3a7427 | 6945 | } else { |
9f3a7427 | 6946 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
6947 | memcpy(&wc->update_progress, &key, |
6948 | sizeof(wc->update_progress)); | |
6949 | ||
6702ed49 | 6950 | level = root_item->drop_level; |
2c47e605 | 6951 | BUG_ON(level == 0); |
6702ed49 | 6952 | path->lowest_level = level; |
2c47e605 YZ |
6953 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
6954 | path->lowest_level = 0; | |
6955 | if (ret < 0) { | |
6956 | err = ret; | |
79787eaa | 6957 | goto out_end_trans; |
9f3a7427 | 6958 | } |
1c4850e2 | 6959 | WARN_ON(ret > 0); |
2c47e605 | 6960 | |
7d9eb12c CM |
6961 | /* |
6962 | * unlock our path, this is safe because only this | |
6963 | * function is allowed to delete this snapshot | |
6964 | */ | |
5d4f98a2 | 6965 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
6966 | |
6967 | level = btrfs_header_level(root->node); | |
6968 | while (1) { | |
6969 | btrfs_tree_lock(path->nodes[level]); | |
6970 | btrfs_set_lock_blocking(path->nodes[level]); | |
6971 | ||
6972 | ret = btrfs_lookup_extent_info(trans, root, | |
6973 | path->nodes[level]->start, | |
6974 | path->nodes[level]->len, | |
6975 | &wc->refs[level], | |
6976 | &wc->flags[level]); | |
79787eaa JM |
6977 | if (ret < 0) { |
6978 | err = ret; | |
6979 | goto out_end_trans; | |
6980 | } | |
2c47e605 YZ |
6981 | BUG_ON(wc->refs[level] == 0); |
6982 | ||
6983 | if (level == root_item->drop_level) | |
6984 | break; | |
6985 | ||
6986 | btrfs_tree_unlock(path->nodes[level]); | |
6987 | WARN_ON(wc->refs[level] != 1); | |
6988 | level--; | |
6989 | } | |
9f3a7427 | 6990 | } |
2c47e605 YZ |
6991 | |
6992 | wc->level = level; | |
6993 | wc->shared_level = -1; | |
6994 | wc->stage = DROP_REFERENCE; | |
6995 | wc->update_ref = update_ref; | |
6996 | wc->keep_locks = 0; | |
66d7e7f0 | 6997 | wc->for_reloc = for_reloc; |
1c4850e2 | 6998 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 6999 | |
d397712b | 7000 | while (1) { |
2c47e605 YZ |
7001 | ret = walk_down_tree(trans, root, path, wc); |
7002 | if (ret < 0) { | |
7003 | err = ret; | |
20524f02 | 7004 | break; |
2c47e605 | 7005 | } |
9aca1d51 | 7006 | |
2c47e605 YZ |
7007 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
7008 | if (ret < 0) { | |
7009 | err = ret; | |
20524f02 | 7010 | break; |
2c47e605 YZ |
7011 | } |
7012 | ||
7013 | if (ret > 0) { | |
7014 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
7015 | break; |
7016 | } | |
2c47e605 YZ |
7017 | |
7018 | if (wc->stage == DROP_REFERENCE) { | |
7019 | level = wc->level; | |
7020 | btrfs_node_key(path->nodes[level], | |
7021 | &root_item->drop_progress, | |
7022 | path->slots[level]); | |
7023 | root_item->drop_level = level; | |
7024 | } | |
7025 | ||
7026 | BUG_ON(wc->level == 0); | |
3fd0a558 | 7027 | if (btrfs_should_end_transaction(trans, tree_root)) { |
2c47e605 YZ |
7028 | ret = btrfs_update_root(trans, tree_root, |
7029 | &root->root_key, | |
7030 | root_item); | |
79787eaa JM |
7031 | if (ret) { |
7032 | btrfs_abort_transaction(trans, tree_root, ret); | |
7033 | err = ret; | |
7034 | goto out_end_trans; | |
7035 | } | |
2c47e605 | 7036 | |
3fd0a558 | 7037 | btrfs_end_transaction_throttle(trans, tree_root); |
a22285a6 | 7038 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
7039 | if (IS_ERR(trans)) { |
7040 | err = PTR_ERR(trans); | |
7041 | goto out_free; | |
7042 | } | |
3fd0a558 YZ |
7043 | if (block_rsv) |
7044 | trans->block_rsv = block_rsv; | |
c3e69d58 | 7045 | } |
20524f02 | 7046 | } |
b3b4aa74 | 7047 | btrfs_release_path(path); |
79787eaa JM |
7048 | if (err) |
7049 | goto out_end_trans; | |
2c47e605 YZ |
7050 | |
7051 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
79787eaa JM |
7052 | if (ret) { |
7053 | btrfs_abort_transaction(trans, tree_root, ret); | |
7054 | goto out_end_trans; | |
7055 | } | |
2c47e605 | 7056 | |
76dda93c YZ |
7057 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
7058 | ret = btrfs_find_last_root(tree_root, root->root_key.objectid, | |
7059 | NULL, NULL); | |
79787eaa JM |
7060 | if (ret < 0) { |
7061 | btrfs_abort_transaction(trans, tree_root, ret); | |
7062 | err = ret; | |
7063 | goto out_end_trans; | |
7064 | } else if (ret > 0) { | |
84cd948c JB |
7065 | /* if we fail to delete the orphan item this time |
7066 | * around, it'll get picked up the next time. | |
7067 | * | |
7068 | * The most common failure here is just -ENOENT. | |
7069 | */ | |
7070 | btrfs_del_orphan_item(trans, tree_root, | |
7071 | root->root_key.objectid); | |
76dda93c YZ |
7072 | } |
7073 | } | |
7074 | ||
7075 | if (root->in_radix) { | |
7076 | btrfs_free_fs_root(tree_root->fs_info, root); | |
7077 | } else { | |
7078 | free_extent_buffer(root->node); | |
7079 | free_extent_buffer(root->commit_root); | |
7080 | kfree(root); | |
7081 | } | |
79787eaa | 7082 | out_end_trans: |
3fd0a558 | 7083 | btrfs_end_transaction_throttle(trans, tree_root); |
79787eaa | 7084 | out_free: |
2c47e605 | 7085 | kfree(wc); |
5caf2a00 | 7086 | btrfs_free_path(path); |
cb1b69f4 TI |
7087 | out: |
7088 | if (err) | |
7089 | btrfs_std_error(root->fs_info, err); | |
2c536799 | 7090 | return err; |
20524f02 | 7091 | } |
9078a3e1 | 7092 | |
2c47e605 YZ |
7093 | /* |
7094 | * drop subtree rooted at tree block 'node'. | |
7095 | * | |
7096 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 7097 | * only used by relocation code |
2c47e605 | 7098 | */ |
f82d02d9 YZ |
7099 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
7100 | struct btrfs_root *root, | |
7101 | struct extent_buffer *node, | |
7102 | struct extent_buffer *parent) | |
7103 | { | |
7104 | struct btrfs_path *path; | |
2c47e605 | 7105 | struct walk_control *wc; |
f82d02d9 YZ |
7106 | int level; |
7107 | int parent_level; | |
7108 | int ret = 0; | |
7109 | int wret; | |
7110 | ||
2c47e605 YZ |
7111 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
7112 | ||
f82d02d9 | 7113 | path = btrfs_alloc_path(); |
db5b493a TI |
7114 | if (!path) |
7115 | return -ENOMEM; | |
f82d02d9 | 7116 | |
2c47e605 | 7117 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
7118 | if (!wc) { |
7119 | btrfs_free_path(path); | |
7120 | return -ENOMEM; | |
7121 | } | |
2c47e605 | 7122 | |
b9447ef8 | 7123 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
7124 | parent_level = btrfs_header_level(parent); |
7125 | extent_buffer_get(parent); | |
7126 | path->nodes[parent_level] = parent; | |
7127 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
7128 | ||
b9447ef8 | 7129 | btrfs_assert_tree_locked(node); |
f82d02d9 | 7130 | level = btrfs_header_level(node); |
f82d02d9 YZ |
7131 | path->nodes[level] = node; |
7132 | path->slots[level] = 0; | |
bd681513 | 7133 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7134 | |
7135 | wc->refs[parent_level] = 1; | |
7136 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
7137 | wc->level = level; | |
7138 | wc->shared_level = -1; | |
7139 | wc->stage = DROP_REFERENCE; | |
7140 | wc->update_ref = 0; | |
7141 | wc->keep_locks = 1; | |
66d7e7f0 | 7142 | wc->for_reloc = 1; |
1c4850e2 | 7143 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
7144 | |
7145 | while (1) { | |
2c47e605 YZ |
7146 | wret = walk_down_tree(trans, root, path, wc); |
7147 | if (wret < 0) { | |
f82d02d9 | 7148 | ret = wret; |
f82d02d9 | 7149 | break; |
2c47e605 | 7150 | } |
f82d02d9 | 7151 | |
2c47e605 | 7152 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
7153 | if (wret < 0) |
7154 | ret = wret; | |
7155 | if (wret != 0) | |
7156 | break; | |
7157 | } | |
7158 | ||
2c47e605 | 7159 | kfree(wc); |
f82d02d9 YZ |
7160 | btrfs_free_path(path); |
7161 | return ret; | |
7162 | } | |
7163 | ||
ec44a35c CM |
7164 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
7165 | { | |
7166 | u64 num_devices; | |
fc67c450 | 7167 | u64 stripped; |
e4d8ec0f | 7168 | |
fc67c450 ID |
7169 | /* |
7170 | * if restripe for this chunk_type is on pick target profile and | |
7171 | * return, otherwise do the usual balance | |
7172 | */ | |
7173 | stripped = get_restripe_target(root->fs_info, flags); | |
7174 | if (stripped) | |
7175 | return extended_to_chunk(stripped); | |
e4d8ec0f | 7176 | |
cd02dca5 CM |
7177 | /* |
7178 | * we add in the count of missing devices because we want | |
7179 | * to make sure that any RAID levels on a degraded FS | |
7180 | * continue to be honored. | |
7181 | */ | |
7182 | num_devices = root->fs_info->fs_devices->rw_devices + | |
7183 | root->fs_info->fs_devices->missing_devices; | |
7184 | ||
fc67c450 ID |
7185 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
7186 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
7187 | ||
ec44a35c CM |
7188 | if (num_devices == 1) { |
7189 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7190 | stripped = flags & ~stripped; | |
7191 | ||
7192 | /* turn raid0 into single device chunks */ | |
7193 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
7194 | return stripped; | |
7195 | ||
7196 | /* turn mirroring into duplication */ | |
7197 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7198 | BTRFS_BLOCK_GROUP_RAID10)) | |
7199 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
7200 | } else { |
7201 | /* they already had raid on here, just return */ | |
ec44a35c CM |
7202 | if (flags & stripped) |
7203 | return flags; | |
7204 | ||
7205 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7206 | stripped = flags & ~stripped; | |
7207 | ||
7208 | /* switch duplicated blocks with raid1 */ | |
7209 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
7210 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
7211 | ||
e3176ca2 | 7212 | /* this is drive concat, leave it alone */ |
ec44a35c | 7213 | } |
e3176ca2 | 7214 | |
ec44a35c CM |
7215 | return flags; |
7216 | } | |
7217 | ||
199c36ea | 7218 | static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 7219 | { |
f0486c68 YZ |
7220 | struct btrfs_space_info *sinfo = cache->space_info; |
7221 | u64 num_bytes; | |
199c36ea | 7222 | u64 min_allocable_bytes; |
f0486c68 | 7223 | int ret = -ENOSPC; |
0ef3e66b | 7224 | |
c286ac48 | 7225 | |
199c36ea MX |
7226 | /* |
7227 | * We need some metadata space and system metadata space for | |
7228 | * allocating chunks in some corner cases until we force to set | |
7229 | * it to be readonly. | |
7230 | */ | |
7231 | if ((sinfo->flags & | |
7232 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
7233 | !force) | |
7234 | min_allocable_bytes = 1 * 1024 * 1024; | |
7235 | else | |
7236 | min_allocable_bytes = 0; | |
7237 | ||
f0486c68 YZ |
7238 | spin_lock(&sinfo->lock); |
7239 | spin_lock(&cache->lock); | |
61cfea9b W |
7240 | |
7241 | if (cache->ro) { | |
7242 | ret = 0; | |
7243 | goto out; | |
7244 | } | |
7245 | ||
f0486c68 YZ |
7246 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
7247 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7248 | ||
7249 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
37be25bc JB |
7250 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + |
7251 | min_allocable_bytes <= sinfo->total_bytes) { | |
f0486c68 | 7252 | sinfo->bytes_readonly += num_bytes; |
f0486c68 YZ |
7253 | cache->ro = 1; |
7254 | ret = 0; | |
7255 | } | |
61cfea9b | 7256 | out: |
f0486c68 YZ |
7257 | spin_unlock(&cache->lock); |
7258 | spin_unlock(&sinfo->lock); | |
7259 | return ret; | |
7260 | } | |
7d9eb12c | 7261 | |
f0486c68 YZ |
7262 | int btrfs_set_block_group_ro(struct btrfs_root *root, |
7263 | struct btrfs_block_group_cache *cache) | |
c286ac48 | 7264 | |
f0486c68 YZ |
7265 | { |
7266 | struct btrfs_trans_handle *trans; | |
7267 | u64 alloc_flags; | |
7268 | int ret; | |
7d9eb12c | 7269 | |
f0486c68 | 7270 | BUG_ON(cache->ro); |
0ef3e66b | 7271 | |
ff5714cc | 7272 | trans = btrfs_join_transaction(root); |
79787eaa JM |
7273 | if (IS_ERR(trans)) |
7274 | return PTR_ERR(trans); | |
5d4f98a2 | 7275 | |
f0486c68 | 7276 | alloc_flags = update_block_group_flags(root, cache->flags); |
79787eaa JM |
7277 | if (alloc_flags != cache->flags) { |
7278 | ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, | |
7279 | CHUNK_ALLOC_FORCE); | |
7280 | if (ret < 0) | |
7281 | goto out; | |
7282 | } | |
5d4f98a2 | 7283 | |
199c36ea | 7284 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7285 | if (!ret) |
7286 | goto out; | |
7287 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
0e4f8f88 CM |
7288 | ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
7289 | CHUNK_ALLOC_FORCE); | |
f0486c68 YZ |
7290 | if (ret < 0) |
7291 | goto out; | |
199c36ea | 7292 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7293 | out: |
7294 | btrfs_end_transaction(trans, root); | |
7295 | return ret; | |
7296 | } | |
5d4f98a2 | 7297 | |
c87f08ca CM |
7298 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
7299 | struct btrfs_root *root, u64 type) | |
7300 | { | |
7301 | u64 alloc_flags = get_alloc_profile(root, type); | |
0e4f8f88 CM |
7302 | return do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags, |
7303 | CHUNK_ALLOC_FORCE); | |
c87f08ca CM |
7304 | } |
7305 | ||
6d07bcec MX |
7306 | /* |
7307 | * helper to account the unused space of all the readonly block group in the | |
7308 | * list. takes mirrors into account. | |
7309 | */ | |
7310 | static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list) | |
7311 | { | |
7312 | struct btrfs_block_group_cache *block_group; | |
7313 | u64 free_bytes = 0; | |
7314 | int factor; | |
7315 | ||
7316 | list_for_each_entry(block_group, groups_list, list) { | |
7317 | spin_lock(&block_group->lock); | |
7318 | ||
7319 | if (!block_group->ro) { | |
7320 | spin_unlock(&block_group->lock); | |
7321 | continue; | |
7322 | } | |
7323 | ||
7324 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7325 | BTRFS_BLOCK_GROUP_RAID10 | | |
7326 | BTRFS_BLOCK_GROUP_DUP)) | |
7327 | factor = 2; | |
7328 | else | |
7329 | factor = 1; | |
7330 | ||
7331 | free_bytes += (block_group->key.offset - | |
7332 | btrfs_block_group_used(&block_group->item)) * | |
7333 | factor; | |
7334 | ||
7335 | spin_unlock(&block_group->lock); | |
7336 | } | |
7337 | ||
7338 | return free_bytes; | |
7339 | } | |
7340 | ||
7341 | /* | |
7342 | * helper to account the unused space of all the readonly block group in the | |
7343 | * space_info. takes mirrors into account. | |
7344 | */ | |
7345 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) | |
7346 | { | |
7347 | int i; | |
7348 | u64 free_bytes = 0; | |
7349 | ||
7350 | spin_lock(&sinfo->lock); | |
7351 | ||
7352 | for(i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
7353 | if (!list_empty(&sinfo->block_groups[i])) | |
7354 | free_bytes += __btrfs_get_ro_block_group_free_space( | |
7355 | &sinfo->block_groups[i]); | |
7356 | ||
7357 | spin_unlock(&sinfo->lock); | |
7358 | ||
7359 | return free_bytes; | |
7360 | } | |
7361 | ||
143bede5 | 7362 | void btrfs_set_block_group_rw(struct btrfs_root *root, |
f0486c68 | 7363 | struct btrfs_block_group_cache *cache) |
5d4f98a2 | 7364 | { |
f0486c68 YZ |
7365 | struct btrfs_space_info *sinfo = cache->space_info; |
7366 | u64 num_bytes; | |
7367 | ||
7368 | BUG_ON(!cache->ro); | |
7369 | ||
7370 | spin_lock(&sinfo->lock); | |
7371 | spin_lock(&cache->lock); | |
7372 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
7373 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7374 | sinfo->bytes_readonly -= num_bytes; | |
7375 | cache->ro = 0; | |
7376 | spin_unlock(&cache->lock); | |
7377 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
7378 | } |
7379 | ||
ba1bf481 JB |
7380 | /* |
7381 | * checks to see if its even possible to relocate this block group. | |
7382 | * | |
7383 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
7384 | * ok to go ahead and try. | |
7385 | */ | |
7386 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 7387 | { |
ba1bf481 JB |
7388 | struct btrfs_block_group_cache *block_group; |
7389 | struct btrfs_space_info *space_info; | |
7390 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
7391 | struct btrfs_device *device; | |
cdcb725c | 7392 | u64 min_free; |
6719db6a JB |
7393 | u64 dev_min = 1; |
7394 | u64 dev_nr = 0; | |
4a5e98f5 | 7395 | u64 target; |
cdcb725c | 7396 | int index; |
ba1bf481 JB |
7397 | int full = 0; |
7398 | int ret = 0; | |
1a40e23b | 7399 | |
ba1bf481 | 7400 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 7401 | |
ba1bf481 JB |
7402 | /* odd, couldn't find the block group, leave it alone */ |
7403 | if (!block_group) | |
7404 | return -1; | |
1a40e23b | 7405 | |
cdcb725c | 7406 | min_free = btrfs_block_group_used(&block_group->item); |
7407 | ||
ba1bf481 | 7408 | /* no bytes used, we're good */ |
cdcb725c | 7409 | if (!min_free) |
1a40e23b ZY |
7410 | goto out; |
7411 | ||
ba1bf481 JB |
7412 | space_info = block_group->space_info; |
7413 | spin_lock(&space_info->lock); | |
17d217fe | 7414 | |
ba1bf481 | 7415 | full = space_info->full; |
17d217fe | 7416 | |
ba1bf481 JB |
7417 | /* |
7418 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
7419 | * relocate it unless we're able to allocate a new chunk below. |
7420 | * | |
7421 | * Otherwise, we need to make sure we have room in the space to handle | |
7422 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 7423 | */ |
7ce618db | 7424 | if ((space_info->total_bytes != block_group->key.offset) && |
cdcb725c | 7425 | (space_info->bytes_used + space_info->bytes_reserved + |
7426 | space_info->bytes_pinned + space_info->bytes_readonly + | |
7427 | min_free < space_info->total_bytes)) { | |
ba1bf481 JB |
7428 | spin_unlock(&space_info->lock); |
7429 | goto out; | |
17d217fe | 7430 | } |
ba1bf481 | 7431 | spin_unlock(&space_info->lock); |
ea8c2819 | 7432 | |
ba1bf481 JB |
7433 | /* |
7434 | * ok we don't have enough space, but maybe we have free space on our | |
7435 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
7436 | * alloc devices and guess if we have enough space. if this block |
7437 | * group is going to be restriped, run checks against the target | |
7438 | * profile instead of the current one. | |
ba1bf481 JB |
7439 | */ |
7440 | ret = -1; | |
ea8c2819 | 7441 | |
cdcb725c | 7442 | /* |
7443 | * index: | |
7444 | * 0: raid10 | |
7445 | * 1: raid1 | |
7446 | * 2: dup | |
7447 | * 3: raid0 | |
7448 | * 4: single | |
7449 | */ | |
4a5e98f5 ID |
7450 | target = get_restripe_target(root->fs_info, block_group->flags); |
7451 | if (target) { | |
7452 | index = __get_block_group_index(extended_to_chunk(target)); | |
7453 | } else { | |
7454 | /* | |
7455 | * this is just a balance, so if we were marked as full | |
7456 | * we know there is no space for a new chunk | |
7457 | */ | |
7458 | if (full) | |
7459 | goto out; | |
7460 | ||
7461 | index = get_block_group_index(block_group); | |
7462 | } | |
7463 | ||
cdcb725c | 7464 | if (index == 0) { |
7465 | dev_min = 4; | |
6719db6a JB |
7466 | /* Divide by 2 */ |
7467 | min_free >>= 1; | |
cdcb725c | 7468 | } else if (index == 1) { |
7469 | dev_min = 2; | |
7470 | } else if (index == 2) { | |
6719db6a JB |
7471 | /* Multiply by 2 */ |
7472 | min_free <<= 1; | |
cdcb725c | 7473 | } else if (index == 3) { |
7474 | dev_min = fs_devices->rw_devices; | |
6719db6a | 7475 | do_div(min_free, dev_min); |
cdcb725c | 7476 | } |
7477 | ||
ba1bf481 JB |
7478 | mutex_lock(&root->fs_info->chunk_mutex); |
7479 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7bfc837d | 7480 | u64 dev_offset; |
56bec294 | 7481 | |
ba1bf481 JB |
7482 | /* |
7483 | * check to make sure we can actually find a chunk with enough | |
7484 | * space to fit our block group in. | |
7485 | */ | |
7486 | if (device->total_bytes > device->bytes_used + min_free) { | |
125ccb0a | 7487 | ret = find_free_dev_extent(device, min_free, |
7bfc837d | 7488 | &dev_offset, NULL); |
ba1bf481 | 7489 | if (!ret) |
cdcb725c | 7490 | dev_nr++; |
7491 | ||
7492 | if (dev_nr >= dev_min) | |
73e48b27 | 7493 | break; |
cdcb725c | 7494 | |
ba1bf481 | 7495 | ret = -1; |
725c8463 | 7496 | } |
edbd8d4e | 7497 | } |
ba1bf481 | 7498 | mutex_unlock(&root->fs_info->chunk_mutex); |
edbd8d4e | 7499 | out: |
ba1bf481 | 7500 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
7501 | return ret; |
7502 | } | |
7503 | ||
b2950863 CH |
7504 | static int find_first_block_group(struct btrfs_root *root, |
7505 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 7506 | { |
925baedd | 7507 | int ret = 0; |
0b86a832 CM |
7508 | struct btrfs_key found_key; |
7509 | struct extent_buffer *leaf; | |
7510 | int slot; | |
edbd8d4e | 7511 | |
0b86a832 CM |
7512 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
7513 | if (ret < 0) | |
925baedd CM |
7514 | goto out; |
7515 | ||
d397712b | 7516 | while (1) { |
0b86a832 | 7517 | slot = path->slots[0]; |
edbd8d4e | 7518 | leaf = path->nodes[0]; |
0b86a832 CM |
7519 | if (slot >= btrfs_header_nritems(leaf)) { |
7520 | ret = btrfs_next_leaf(root, path); | |
7521 | if (ret == 0) | |
7522 | continue; | |
7523 | if (ret < 0) | |
925baedd | 7524 | goto out; |
0b86a832 | 7525 | break; |
edbd8d4e | 7526 | } |
0b86a832 | 7527 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 7528 | |
0b86a832 | 7529 | if (found_key.objectid >= key->objectid && |
925baedd CM |
7530 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
7531 | ret = 0; | |
7532 | goto out; | |
7533 | } | |
0b86a832 | 7534 | path->slots[0]++; |
edbd8d4e | 7535 | } |
925baedd | 7536 | out: |
0b86a832 | 7537 | return ret; |
edbd8d4e CM |
7538 | } |
7539 | ||
0af3d00b JB |
7540 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
7541 | { | |
7542 | struct btrfs_block_group_cache *block_group; | |
7543 | u64 last = 0; | |
7544 | ||
7545 | while (1) { | |
7546 | struct inode *inode; | |
7547 | ||
7548 | block_group = btrfs_lookup_first_block_group(info, last); | |
7549 | while (block_group) { | |
7550 | spin_lock(&block_group->lock); | |
7551 | if (block_group->iref) | |
7552 | break; | |
7553 | spin_unlock(&block_group->lock); | |
7554 | block_group = next_block_group(info->tree_root, | |
7555 | block_group); | |
7556 | } | |
7557 | if (!block_group) { | |
7558 | if (last == 0) | |
7559 | break; | |
7560 | last = 0; | |
7561 | continue; | |
7562 | } | |
7563 | ||
7564 | inode = block_group->inode; | |
7565 | block_group->iref = 0; | |
7566 | block_group->inode = NULL; | |
7567 | spin_unlock(&block_group->lock); | |
7568 | iput(inode); | |
7569 | last = block_group->key.objectid + block_group->key.offset; | |
7570 | btrfs_put_block_group(block_group); | |
7571 | } | |
7572 | } | |
7573 | ||
1a40e23b ZY |
7574 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
7575 | { | |
7576 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 7577 | struct btrfs_space_info *space_info; |
11833d66 | 7578 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
7579 | struct rb_node *n; |
7580 | ||
11833d66 YZ |
7581 | down_write(&info->extent_commit_sem); |
7582 | while (!list_empty(&info->caching_block_groups)) { | |
7583 | caching_ctl = list_entry(info->caching_block_groups.next, | |
7584 | struct btrfs_caching_control, list); | |
7585 | list_del(&caching_ctl->list); | |
7586 | put_caching_control(caching_ctl); | |
7587 | } | |
7588 | up_write(&info->extent_commit_sem); | |
7589 | ||
1a40e23b ZY |
7590 | spin_lock(&info->block_group_cache_lock); |
7591 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
7592 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
7593 | cache_node); | |
1a40e23b ZY |
7594 | rb_erase(&block_group->cache_node, |
7595 | &info->block_group_cache_tree); | |
d899e052 YZ |
7596 | spin_unlock(&info->block_group_cache_lock); |
7597 | ||
80eb234a | 7598 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 7599 | list_del(&block_group->list); |
80eb234a | 7600 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 7601 | |
817d52f8 | 7602 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 7603 | wait_block_group_cache_done(block_group); |
817d52f8 | 7604 | |
3c14874a JB |
7605 | /* |
7606 | * We haven't cached this block group, which means we could | |
7607 | * possibly have excluded extents on this block group. | |
7608 | */ | |
7609 | if (block_group->cached == BTRFS_CACHE_NO) | |
7610 | free_excluded_extents(info->extent_root, block_group); | |
7611 | ||
817d52f8 | 7612 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 7613 | btrfs_put_block_group(block_group); |
d899e052 YZ |
7614 | |
7615 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
7616 | } |
7617 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
7618 | |
7619 | /* now that all the block groups are freed, go through and | |
7620 | * free all the space_info structs. This is only called during | |
7621 | * the final stages of unmount, and so we know nobody is | |
7622 | * using them. We call synchronize_rcu() once before we start, | |
7623 | * just to be on the safe side. | |
7624 | */ | |
7625 | synchronize_rcu(); | |
7626 | ||
8929ecfa YZ |
7627 | release_global_block_rsv(info); |
7628 | ||
4184ea7f CM |
7629 | while(!list_empty(&info->space_info)) { |
7630 | space_info = list_entry(info->space_info.next, | |
7631 | struct btrfs_space_info, | |
7632 | list); | |
f0486c68 | 7633 | if (space_info->bytes_pinned > 0 || |
fb25e914 JB |
7634 | space_info->bytes_reserved > 0 || |
7635 | space_info->bytes_may_use > 0) { | |
f0486c68 YZ |
7636 | WARN_ON(1); |
7637 | dump_space_info(space_info, 0, 0); | |
7638 | } | |
4184ea7f CM |
7639 | list_del(&space_info->list); |
7640 | kfree(space_info); | |
7641 | } | |
1a40e23b ZY |
7642 | return 0; |
7643 | } | |
7644 | ||
b742bb82 YZ |
7645 | static void __link_block_group(struct btrfs_space_info *space_info, |
7646 | struct btrfs_block_group_cache *cache) | |
7647 | { | |
7648 | int index = get_block_group_index(cache); | |
7649 | ||
7650 | down_write(&space_info->groups_sem); | |
7651 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
7652 | up_write(&space_info->groups_sem); | |
7653 | } | |
7654 | ||
9078a3e1 CM |
7655 | int btrfs_read_block_groups(struct btrfs_root *root) |
7656 | { | |
7657 | struct btrfs_path *path; | |
7658 | int ret; | |
9078a3e1 | 7659 | struct btrfs_block_group_cache *cache; |
be744175 | 7660 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 7661 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
7662 | struct btrfs_key key; |
7663 | struct btrfs_key found_key; | |
5f39d397 | 7664 | struct extent_buffer *leaf; |
0af3d00b JB |
7665 | int need_clear = 0; |
7666 | u64 cache_gen; | |
96b5179d | 7667 | |
be744175 | 7668 | root = info->extent_root; |
9078a3e1 | 7669 | key.objectid = 0; |
0b86a832 | 7670 | key.offset = 0; |
9078a3e1 | 7671 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
7672 | path = btrfs_alloc_path(); |
7673 | if (!path) | |
7674 | return -ENOMEM; | |
026fd317 | 7675 | path->reada = 1; |
9078a3e1 | 7676 | |
6c41761f | 7677 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); |
73bc1876 | 7678 | if (btrfs_test_opt(root, SPACE_CACHE) && |
6c41761f | 7679 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) |
0af3d00b | 7680 | need_clear = 1; |
88c2ba3b JB |
7681 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
7682 | need_clear = 1; | |
0af3d00b | 7683 | |
d397712b | 7684 | while (1) { |
0b86a832 | 7685 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
7686 | if (ret > 0) |
7687 | break; | |
0b86a832 CM |
7688 | if (ret != 0) |
7689 | goto error; | |
5f39d397 CM |
7690 | leaf = path->nodes[0]; |
7691 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 7692 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 7693 | if (!cache) { |
0b86a832 | 7694 | ret = -ENOMEM; |
f0486c68 | 7695 | goto error; |
9078a3e1 | 7696 | } |
34d52cb6 LZ |
7697 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
7698 | GFP_NOFS); | |
7699 | if (!cache->free_space_ctl) { | |
7700 | kfree(cache); | |
7701 | ret = -ENOMEM; | |
7702 | goto error; | |
7703 | } | |
3e1ad54f | 7704 | |
d2fb3437 | 7705 | atomic_set(&cache->count, 1); |
c286ac48 | 7706 | spin_lock_init(&cache->lock); |
817d52f8 | 7707 | cache->fs_info = info; |
0f9dd46c | 7708 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7709 | INIT_LIST_HEAD(&cache->cluster_list); |
96303081 | 7710 | |
cf7c1ef6 LB |
7711 | if (need_clear) { |
7712 | /* | |
7713 | * When we mount with old space cache, we need to | |
7714 | * set BTRFS_DC_CLEAR and set dirty flag. | |
7715 | * | |
7716 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
7717 | * truncate the old free space cache inode and | |
7718 | * setup a new one. | |
7719 | * b) Setting 'dirty flag' makes sure that we flush | |
7720 | * the new space cache info onto disk. | |
7721 | */ | |
0af3d00b | 7722 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 LB |
7723 | if (btrfs_test_opt(root, SPACE_CACHE)) |
7724 | cache->dirty = 1; | |
7725 | } | |
0af3d00b | 7726 | |
5f39d397 CM |
7727 | read_extent_buffer(leaf, &cache->item, |
7728 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
7729 | sizeof(cache->item)); | |
9078a3e1 | 7730 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 7731 | |
9078a3e1 | 7732 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 7733 | btrfs_release_path(path); |
0b86a832 | 7734 | cache->flags = btrfs_block_group_flags(&cache->item); |
817d52f8 JB |
7735 | cache->sectorsize = root->sectorsize; |
7736 | ||
34d52cb6 LZ |
7737 | btrfs_init_free_space_ctl(cache); |
7738 | ||
3c14874a JB |
7739 | /* |
7740 | * We need to exclude the super stripes now so that the space | |
7741 | * info has super bytes accounted for, otherwise we'll think | |
7742 | * we have more space than we actually do. | |
7743 | */ | |
7744 | exclude_super_stripes(root, cache); | |
7745 | ||
817d52f8 JB |
7746 | /* |
7747 | * check for two cases, either we are full, and therefore | |
7748 | * don't need to bother with the caching work since we won't | |
7749 | * find any space, or we are empty, and we can just add all | |
7750 | * the space in and be done with it. This saves us _alot_ of | |
7751 | * time, particularly in the full case. | |
7752 | */ | |
7753 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 7754 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7755 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 7756 | free_excluded_extents(root, cache); |
817d52f8 | 7757 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 7758 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
7759 | cache->cached = BTRFS_CACHE_FINISHED; |
7760 | add_new_free_space(cache, root->fs_info, | |
7761 | found_key.objectid, | |
7762 | found_key.objectid + | |
7763 | found_key.offset); | |
11833d66 | 7764 | free_excluded_extents(root, cache); |
817d52f8 | 7765 | } |
96b5179d | 7766 | |
6324fbf3 CM |
7767 | ret = update_space_info(info, cache->flags, found_key.offset, |
7768 | btrfs_block_group_used(&cache->item), | |
7769 | &space_info); | |
79787eaa | 7770 | BUG_ON(ret); /* -ENOMEM */ |
6324fbf3 | 7771 | cache->space_info = space_info; |
1b2da372 | 7772 | spin_lock(&cache->space_info->lock); |
f0486c68 | 7773 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7774 | spin_unlock(&cache->space_info->lock); |
7775 | ||
b742bb82 | 7776 | __link_block_group(space_info, cache); |
0f9dd46c JB |
7777 | |
7778 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
79787eaa | 7779 | BUG_ON(ret); /* Logic error */ |
75ccf47d CM |
7780 | |
7781 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
2b82032c | 7782 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
199c36ea | 7783 | set_block_group_ro(cache, 1); |
9078a3e1 | 7784 | } |
b742bb82 YZ |
7785 | |
7786 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
7787 | if (!(get_alloc_profile(root, space_info->flags) & | |
7788 | (BTRFS_BLOCK_GROUP_RAID10 | | |
7789 | BTRFS_BLOCK_GROUP_RAID1 | | |
7790 | BTRFS_BLOCK_GROUP_DUP))) | |
7791 | continue; | |
7792 | /* | |
7793 | * avoid allocating from un-mirrored block group if there are | |
7794 | * mirrored block groups. | |
7795 | */ | |
7796 | list_for_each_entry(cache, &space_info->block_groups[3], list) | |
199c36ea | 7797 | set_block_group_ro(cache, 1); |
b742bb82 | 7798 | list_for_each_entry(cache, &space_info->block_groups[4], list) |
199c36ea | 7799 | set_block_group_ro(cache, 1); |
9078a3e1 | 7800 | } |
f0486c68 YZ |
7801 | |
7802 | init_global_block_rsv(info); | |
0b86a832 CM |
7803 | ret = 0; |
7804 | error: | |
9078a3e1 | 7805 | btrfs_free_path(path); |
0b86a832 | 7806 | return ret; |
9078a3e1 | 7807 | } |
6324fbf3 CM |
7808 | |
7809 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, | |
7810 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 7811 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
7812 | u64 size) |
7813 | { | |
7814 | int ret; | |
6324fbf3 CM |
7815 | struct btrfs_root *extent_root; |
7816 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
7817 | |
7818 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 7819 | |
12fcfd22 | 7820 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 7821 | |
8f18cf13 | 7822 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
7823 | if (!cache) |
7824 | return -ENOMEM; | |
34d52cb6 LZ |
7825 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
7826 | GFP_NOFS); | |
7827 | if (!cache->free_space_ctl) { | |
7828 | kfree(cache); | |
7829 | return -ENOMEM; | |
7830 | } | |
0f9dd46c | 7831 | |
e17cade2 | 7832 | cache->key.objectid = chunk_offset; |
6324fbf3 | 7833 | cache->key.offset = size; |
d2fb3437 | 7834 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
96303081 | 7835 | cache->sectorsize = root->sectorsize; |
0af3d00b | 7836 | cache->fs_info = root->fs_info; |
96303081 | 7837 | |
d2fb3437 | 7838 | atomic_set(&cache->count, 1); |
c286ac48 | 7839 | spin_lock_init(&cache->lock); |
0f9dd46c | 7840 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7841 | INIT_LIST_HEAD(&cache->cluster_list); |
0ef3e66b | 7842 | |
34d52cb6 LZ |
7843 | btrfs_init_free_space_ctl(cache); |
7844 | ||
6324fbf3 | 7845 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
7846 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
7847 | cache->flags = type; | |
7848 | btrfs_set_block_group_flags(&cache->item, type); | |
7849 | ||
11833d66 | 7850 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7851 | cache->cached = BTRFS_CACHE_FINISHED; |
11833d66 | 7852 | exclude_super_stripes(root, cache); |
96303081 | 7853 | |
817d52f8 JB |
7854 | add_new_free_space(cache, root->fs_info, chunk_offset, |
7855 | chunk_offset + size); | |
7856 | ||
11833d66 YZ |
7857 | free_excluded_extents(root, cache); |
7858 | ||
6324fbf3 CM |
7859 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
7860 | &cache->space_info); | |
79787eaa | 7861 | BUG_ON(ret); /* -ENOMEM */ |
c7c144db | 7862 | update_global_block_rsv(root->fs_info); |
1b2da372 JB |
7863 | |
7864 | spin_lock(&cache->space_info->lock); | |
f0486c68 | 7865 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7866 | spin_unlock(&cache->space_info->lock); |
7867 | ||
b742bb82 | 7868 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 7869 | |
0f9dd46c | 7870 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
79787eaa | 7871 | BUG_ON(ret); /* Logic error */ |
c286ac48 | 7872 | |
6324fbf3 CM |
7873 | ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item, |
7874 | sizeof(cache->item)); | |
79787eaa JM |
7875 | if (ret) { |
7876 | btrfs_abort_transaction(trans, extent_root, ret); | |
7877 | return ret; | |
7878 | } | |
6324fbf3 | 7879 | |
d18a2c44 | 7880 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 7881 | |
6324fbf3 CM |
7882 | return 0; |
7883 | } | |
1a40e23b | 7884 | |
10ea00f5 ID |
7885 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
7886 | { | |
899c81ea ID |
7887 | u64 extra_flags = chunk_to_extended(flags) & |
7888 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 ID |
7889 | |
7890 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
7891 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
7892 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
7893 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
7894 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
7895 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
7896 | } | |
7897 | ||
1a40e23b ZY |
7898 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
7899 | struct btrfs_root *root, u64 group_start) | |
7900 | { | |
7901 | struct btrfs_path *path; | |
7902 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 7903 | struct btrfs_free_cluster *cluster; |
0af3d00b | 7904 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 7905 | struct btrfs_key key; |
0af3d00b | 7906 | struct inode *inode; |
1a40e23b | 7907 | int ret; |
10ea00f5 | 7908 | int index; |
89a55897 | 7909 | int factor; |
1a40e23b | 7910 | |
1a40e23b ZY |
7911 | root = root->fs_info->extent_root; |
7912 | ||
7913 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
7914 | BUG_ON(!block_group); | |
c146afad | 7915 | BUG_ON(!block_group->ro); |
1a40e23b | 7916 | |
9f7c43c9 | 7917 | /* |
7918 | * Free the reserved super bytes from this block group before | |
7919 | * remove it. | |
7920 | */ | |
7921 | free_excluded_extents(root, block_group); | |
7922 | ||
1a40e23b | 7923 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 7924 | index = get_block_group_index(block_group); |
89a55897 JB |
7925 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
7926 | BTRFS_BLOCK_GROUP_RAID1 | | |
7927 | BTRFS_BLOCK_GROUP_RAID10)) | |
7928 | factor = 2; | |
7929 | else | |
7930 | factor = 1; | |
1a40e23b | 7931 | |
44fb5511 CM |
7932 | /* make sure this block group isn't part of an allocation cluster */ |
7933 | cluster = &root->fs_info->data_alloc_cluster; | |
7934 | spin_lock(&cluster->refill_lock); | |
7935 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7936 | spin_unlock(&cluster->refill_lock); | |
7937 | ||
7938 | /* | |
7939 | * make sure this block group isn't part of a metadata | |
7940 | * allocation cluster | |
7941 | */ | |
7942 | cluster = &root->fs_info->meta_alloc_cluster; | |
7943 | spin_lock(&cluster->refill_lock); | |
7944 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
7945 | spin_unlock(&cluster->refill_lock); | |
7946 | ||
1a40e23b | 7947 | path = btrfs_alloc_path(); |
d8926bb3 MF |
7948 | if (!path) { |
7949 | ret = -ENOMEM; | |
7950 | goto out; | |
7951 | } | |
1a40e23b | 7952 | |
10b2f34d | 7953 | inode = lookup_free_space_inode(tree_root, block_group, path); |
0af3d00b | 7954 | if (!IS_ERR(inode)) { |
b532402e | 7955 | ret = btrfs_orphan_add(trans, inode); |
79787eaa JM |
7956 | if (ret) { |
7957 | btrfs_add_delayed_iput(inode); | |
7958 | goto out; | |
7959 | } | |
0af3d00b JB |
7960 | clear_nlink(inode); |
7961 | /* One for the block groups ref */ | |
7962 | spin_lock(&block_group->lock); | |
7963 | if (block_group->iref) { | |
7964 | block_group->iref = 0; | |
7965 | block_group->inode = NULL; | |
7966 | spin_unlock(&block_group->lock); | |
7967 | iput(inode); | |
7968 | } else { | |
7969 | spin_unlock(&block_group->lock); | |
7970 | } | |
7971 | /* One for our lookup ref */ | |
455757c3 | 7972 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
7973 | } |
7974 | ||
7975 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
7976 | key.offset = block_group->key.objectid; | |
7977 | key.type = 0; | |
7978 | ||
7979 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
7980 | if (ret < 0) | |
7981 | goto out; | |
7982 | if (ret > 0) | |
b3b4aa74 | 7983 | btrfs_release_path(path); |
0af3d00b JB |
7984 | if (ret == 0) { |
7985 | ret = btrfs_del_item(trans, tree_root, path); | |
7986 | if (ret) | |
7987 | goto out; | |
b3b4aa74 | 7988 | btrfs_release_path(path); |
0af3d00b JB |
7989 | } |
7990 | ||
3dfdb934 | 7991 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
7992 | rb_erase(&block_group->cache_node, |
7993 | &root->fs_info->block_group_cache_tree); | |
3dfdb934 | 7994 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 7995 | |
80eb234a | 7996 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
7997 | /* |
7998 | * we must use list_del_init so people can check to see if they | |
7999 | * are still on the list after taking the semaphore | |
8000 | */ | |
8001 | list_del_init(&block_group->list); | |
10ea00f5 ID |
8002 | if (list_empty(&block_group->space_info->block_groups[index])) |
8003 | clear_avail_alloc_bits(root->fs_info, block_group->flags); | |
80eb234a | 8004 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 8005 | |
817d52f8 | 8006 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 8007 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
8008 | |
8009 | btrfs_remove_free_space_cache(block_group); | |
8010 | ||
c146afad YZ |
8011 | spin_lock(&block_group->space_info->lock); |
8012 | block_group->space_info->total_bytes -= block_group->key.offset; | |
8013 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 8014 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
c146afad | 8015 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 8016 | |
0af3d00b JB |
8017 | memcpy(&key, &block_group->key, sizeof(key)); |
8018 | ||
283bb197 | 8019 | btrfs_clear_space_info_full(root->fs_info); |
c146afad | 8020 | |
fa9c0d79 CM |
8021 | btrfs_put_block_group(block_group); |
8022 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
8023 | |
8024 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
8025 | if (ret > 0) | |
8026 | ret = -EIO; | |
8027 | if (ret < 0) | |
8028 | goto out; | |
8029 | ||
8030 | ret = btrfs_del_item(trans, root, path); | |
8031 | out: | |
8032 | btrfs_free_path(path); | |
8033 | return ret; | |
8034 | } | |
acce952b | 8035 | |
c59021f8 | 8036 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
8037 | { | |
8038 | struct btrfs_space_info *space_info; | |
1aba86d6 | 8039 | struct btrfs_super_block *disk_super; |
8040 | u64 features; | |
8041 | u64 flags; | |
8042 | int mixed = 0; | |
c59021f8 | 8043 | int ret; |
8044 | ||
6c41761f | 8045 | disk_super = fs_info->super_copy; |
1aba86d6 | 8046 | if (!btrfs_super_root(disk_super)) |
8047 | return 1; | |
c59021f8 | 8048 | |
1aba86d6 | 8049 | features = btrfs_super_incompat_flags(disk_super); |
8050 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
8051 | mixed = 1; | |
c59021f8 | 8052 | |
1aba86d6 | 8053 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
8054 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
c59021f8 | 8055 | if (ret) |
1aba86d6 | 8056 | goto out; |
c59021f8 | 8057 | |
1aba86d6 | 8058 | if (mixed) { |
8059 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
8060 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8061 | } else { | |
8062 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
8063 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8064 | if (ret) | |
8065 | goto out; | |
8066 | ||
8067 | flags = BTRFS_BLOCK_GROUP_DATA; | |
8068 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8069 | } | |
8070 | out: | |
c59021f8 | 8071 | return ret; |
8072 | } | |
8073 | ||
acce952b | 8074 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
8075 | { | |
8076 | return unpin_extent_range(root, start, end); | |
8077 | } | |
8078 | ||
8079 | int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 8080 | u64 num_bytes, u64 *actual_bytes) |
acce952b | 8081 | { |
5378e607 | 8082 | return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes); |
acce952b | 8083 | } |
f7039b1d LD |
8084 | |
8085 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) | |
8086 | { | |
8087 | struct btrfs_fs_info *fs_info = root->fs_info; | |
8088 | struct btrfs_block_group_cache *cache = NULL; | |
8089 | u64 group_trimmed; | |
8090 | u64 start; | |
8091 | u64 end; | |
8092 | u64 trimmed = 0; | |
2cac13e4 | 8093 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
8094 | int ret = 0; |
8095 | ||
2cac13e4 LB |
8096 | /* |
8097 | * try to trim all FS space, our block group may start from non-zero. | |
8098 | */ | |
8099 | if (range->len == total_bytes) | |
8100 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
8101 | else | |
8102 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
8103 | |
8104 | while (cache) { | |
8105 | if (cache->key.objectid >= (range->start + range->len)) { | |
8106 | btrfs_put_block_group(cache); | |
8107 | break; | |
8108 | } | |
8109 | ||
8110 | start = max(range->start, cache->key.objectid); | |
8111 | end = min(range->start + range->len, | |
8112 | cache->key.objectid + cache->key.offset); | |
8113 | ||
8114 | if (end - start >= range->minlen) { | |
8115 | if (!block_group_cache_done(cache)) { | |
8116 | ret = cache_block_group(cache, NULL, root, 0); | |
8117 | if (!ret) | |
8118 | wait_block_group_cache_done(cache); | |
8119 | } | |
8120 | ret = btrfs_trim_block_group(cache, | |
8121 | &group_trimmed, | |
8122 | start, | |
8123 | end, | |
8124 | range->minlen); | |
8125 | ||
8126 | trimmed += group_trimmed; | |
8127 | if (ret) { | |
8128 | btrfs_put_block_group(cache); | |
8129 | break; | |
8130 | } | |
8131 | } | |
8132 | ||
8133 | cache = next_block_group(fs_info->tree_root, cache); | |
8134 | } | |
8135 | ||
8136 | range->len = trimmed; | |
8137 | return ret; | |
8138 | } |