]>
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" |
3fed40cc | 36 | #include "math.h" |
fec577fb | 37 | |
709c0486 AJ |
38 | #undef SCRAMBLE_DELAYED_REFS |
39 | ||
9e622d6b MX |
40 | /* |
41 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
42 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
43 | * if we really need one. | |
44 | * | |
0e4f8f88 CM |
45 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
46 | * if we have very few chunks already allocated. This is | |
47 | * used as part of the clustering code to help make sure | |
48 | * we have a good pool of storage to cluster in, without | |
49 | * filling the FS with empty chunks | |
50 | * | |
9e622d6b MX |
51 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
52 | * | |
0e4f8f88 CM |
53 | */ |
54 | enum { | |
55 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
56 | CHUNK_ALLOC_LIMITED = 1, |
57 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
58 | }; |
59 | ||
fb25e914 JB |
60 | /* |
61 | * Control how reservations are dealt with. | |
62 | * | |
63 | * RESERVE_FREE - freeing a reservation. | |
64 | * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for | |
65 | * ENOSPC accounting | |
66 | * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update | |
67 | * bytes_may_use as the ENOSPC accounting is done elsewhere | |
68 | */ | |
69 | enum { | |
70 | RESERVE_FREE = 0, | |
71 | RESERVE_ALLOC = 1, | |
72 | RESERVE_ALLOC_NO_ACCOUNT = 2, | |
73 | }; | |
74 | ||
f3465ca4 JB |
75 | static int update_block_group(struct btrfs_trans_handle *trans, |
76 | struct btrfs_root *root, | |
f0486c68 | 77 | u64 bytenr, u64 num_bytes, int alloc); |
5d4f98a2 YZ |
78 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
79 | struct btrfs_root *root, | |
80 | u64 bytenr, u64 num_bytes, u64 parent, | |
81 | u64 root_objectid, u64 owner_objectid, | |
82 | u64 owner_offset, int refs_to_drop, | |
83 | struct btrfs_delayed_extent_op *extra_op); | |
84 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
85 | struct extent_buffer *leaf, | |
86 | struct btrfs_extent_item *ei); | |
87 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
88 | struct btrfs_root *root, | |
89 | u64 parent, u64 root_objectid, | |
90 | u64 flags, u64 owner, u64 offset, | |
91 | struct btrfs_key *ins, int ref_mod); | |
92 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
93 | struct btrfs_root *root, | |
94 | u64 parent, u64 root_objectid, | |
95 | u64 flags, struct btrfs_disk_key *key, | |
96 | int level, struct btrfs_key *ins); | |
6a63209f | 97 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 JB |
98 | struct btrfs_root *extent_root, u64 flags, |
99 | int force); | |
11833d66 YZ |
100 | static int find_next_key(struct btrfs_path *path, int level, |
101 | struct btrfs_key *key); | |
9ed74f2d JB |
102 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
103 | int dump_block_groups); | |
fb25e914 JB |
104 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
105 | u64 num_bytes, int reserve); | |
6a63209f | 106 | |
817d52f8 JB |
107 | static noinline int |
108 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
109 | { | |
110 | smp_mb(); | |
111 | return cache->cached == BTRFS_CACHE_FINISHED; | |
112 | } | |
113 | ||
0f9dd46c JB |
114 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
115 | { | |
116 | return (cache->flags & bits) == bits; | |
117 | } | |
118 | ||
62a45b60 | 119 | static void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
120 | { |
121 | atomic_inc(&cache->count); | |
122 | } | |
123 | ||
124 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
125 | { | |
f0486c68 YZ |
126 | if (atomic_dec_and_test(&cache->count)) { |
127 | WARN_ON(cache->pinned > 0); | |
128 | WARN_ON(cache->reserved > 0); | |
34d52cb6 | 129 | kfree(cache->free_space_ctl); |
11dfe35a | 130 | kfree(cache); |
f0486c68 | 131 | } |
11dfe35a JB |
132 | } |
133 | ||
0f9dd46c JB |
134 | /* |
135 | * this adds the block group to the fs_info rb tree for the block group | |
136 | * cache | |
137 | */ | |
b2950863 | 138 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
139 | struct btrfs_block_group_cache *block_group) |
140 | { | |
141 | struct rb_node **p; | |
142 | struct rb_node *parent = NULL; | |
143 | struct btrfs_block_group_cache *cache; | |
144 | ||
145 | spin_lock(&info->block_group_cache_lock); | |
146 | p = &info->block_group_cache_tree.rb_node; | |
147 | ||
148 | while (*p) { | |
149 | parent = *p; | |
150 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
151 | cache_node); | |
152 | if (block_group->key.objectid < cache->key.objectid) { | |
153 | p = &(*p)->rb_left; | |
154 | } else if (block_group->key.objectid > cache->key.objectid) { | |
155 | p = &(*p)->rb_right; | |
156 | } else { | |
157 | spin_unlock(&info->block_group_cache_lock); | |
158 | return -EEXIST; | |
159 | } | |
160 | } | |
161 | ||
162 | rb_link_node(&block_group->cache_node, parent, p); | |
163 | rb_insert_color(&block_group->cache_node, | |
164 | &info->block_group_cache_tree); | |
165 | spin_unlock(&info->block_group_cache_lock); | |
166 | ||
167 | return 0; | |
168 | } | |
169 | ||
170 | /* | |
171 | * This will return the block group at or after bytenr if contains is 0, else | |
172 | * it will return the block group that contains the bytenr | |
173 | */ | |
174 | static struct btrfs_block_group_cache * | |
175 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
176 | int contains) | |
177 | { | |
178 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
179 | struct rb_node *n; | |
180 | u64 end, start; | |
181 | ||
182 | spin_lock(&info->block_group_cache_lock); | |
183 | n = info->block_group_cache_tree.rb_node; | |
184 | ||
185 | while (n) { | |
186 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
187 | cache_node); | |
188 | end = cache->key.objectid + cache->key.offset - 1; | |
189 | start = cache->key.objectid; | |
190 | ||
191 | if (bytenr < start) { | |
192 | if (!contains && (!ret || start < ret->key.objectid)) | |
193 | ret = cache; | |
194 | n = n->rb_left; | |
195 | } else if (bytenr > start) { | |
196 | if (contains && bytenr <= end) { | |
197 | ret = cache; | |
198 | break; | |
199 | } | |
200 | n = n->rb_right; | |
201 | } else { | |
202 | ret = cache; | |
203 | break; | |
204 | } | |
205 | } | |
d2fb3437 | 206 | if (ret) |
11dfe35a | 207 | btrfs_get_block_group(ret); |
0f9dd46c JB |
208 | spin_unlock(&info->block_group_cache_lock); |
209 | ||
210 | return ret; | |
211 | } | |
212 | ||
11833d66 YZ |
213 | static int add_excluded_extent(struct btrfs_root *root, |
214 | u64 start, u64 num_bytes) | |
817d52f8 | 215 | { |
11833d66 YZ |
216 | u64 end = start + num_bytes - 1; |
217 | set_extent_bits(&root->fs_info->freed_extents[0], | |
218 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
219 | set_extent_bits(&root->fs_info->freed_extents[1], | |
220 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
221 | return 0; | |
222 | } | |
817d52f8 | 223 | |
11833d66 YZ |
224 | static void free_excluded_extents(struct btrfs_root *root, |
225 | struct btrfs_block_group_cache *cache) | |
226 | { | |
227 | u64 start, end; | |
817d52f8 | 228 | |
11833d66 YZ |
229 | start = cache->key.objectid; |
230 | end = start + cache->key.offset - 1; | |
231 | ||
232 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
233 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
234 | clear_extent_bits(&root->fs_info->freed_extents[1], | |
235 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
817d52f8 JB |
236 | } |
237 | ||
11833d66 YZ |
238 | static int exclude_super_stripes(struct btrfs_root *root, |
239 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 240 | { |
817d52f8 JB |
241 | u64 bytenr; |
242 | u64 *logical; | |
243 | int stripe_len; | |
244 | int i, nr, ret; | |
245 | ||
06b2331f YZ |
246 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
247 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
248 | cache->bytes_super += stripe_len; | |
249 | ret = add_excluded_extent(root, cache->key.objectid, | |
250 | stripe_len); | |
79787eaa | 251 | BUG_ON(ret); /* -ENOMEM */ |
06b2331f YZ |
252 | } |
253 | ||
817d52f8 JB |
254 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
255 | bytenr = btrfs_sb_offset(i); | |
256 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
257 | cache->key.objectid, bytenr, | |
258 | 0, &logical, &nr, &stripe_len); | |
79787eaa | 259 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 | 260 | |
817d52f8 | 261 | while (nr--) { |
1b2da372 | 262 | cache->bytes_super += stripe_len; |
11833d66 YZ |
263 | ret = add_excluded_extent(root, logical[nr], |
264 | stripe_len); | |
79787eaa | 265 | BUG_ON(ret); /* -ENOMEM */ |
817d52f8 | 266 | } |
11833d66 | 267 | |
817d52f8 JB |
268 | kfree(logical); |
269 | } | |
817d52f8 JB |
270 | return 0; |
271 | } | |
272 | ||
11833d66 YZ |
273 | static struct btrfs_caching_control * |
274 | get_caching_control(struct btrfs_block_group_cache *cache) | |
275 | { | |
276 | struct btrfs_caching_control *ctl; | |
277 | ||
278 | spin_lock(&cache->lock); | |
279 | if (cache->cached != BTRFS_CACHE_STARTED) { | |
280 | spin_unlock(&cache->lock); | |
281 | return NULL; | |
282 | } | |
283 | ||
dde5abee JB |
284 | /* We're loading it the fast way, so we don't have a caching_ctl. */ |
285 | if (!cache->caching_ctl) { | |
286 | spin_unlock(&cache->lock); | |
11833d66 YZ |
287 | return NULL; |
288 | } | |
289 | ||
290 | ctl = cache->caching_ctl; | |
291 | atomic_inc(&ctl->count); | |
292 | spin_unlock(&cache->lock); | |
293 | return ctl; | |
294 | } | |
295 | ||
296 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
297 | { | |
298 | if (atomic_dec_and_test(&ctl->count)) | |
299 | kfree(ctl); | |
300 | } | |
301 | ||
0f9dd46c JB |
302 | /* |
303 | * this is only called by cache_block_group, since we could have freed extents | |
304 | * we need to check the pinned_extents for any extents that can't be used yet | |
305 | * since their free space will be released as soon as the transaction commits. | |
306 | */ | |
817d52f8 | 307 | static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
0f9dd46c JB |
308 | struct btrfs_fs_info *info, u64 start, u64 end) |
309 | { | |
817d52f8 | 310 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
311 | int ret; |
312 | ||
313 | while (start < end) { | |
11833d66 | 314 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 315 | &extent_start, &extent_end, |
e6138876 JB |
316 | EXTENT_DIRTY | EXTENT_UPTODATE, |
317 | NULL); | |
0f9dd46c JB |
318 | if (ret) |
319 | break; | |
320 | ||
06b2331f | 321 | if (extent_start <= start) { |
0f9dd46c JB |
322 | start = extent_end + 1; |
323 | } else if (extent_start > start && extent_start < end) { | |
324 | size = extent_start - start; | |
817d52f8 | 325 | total_added += size; |
ea6a478e JB |
326 | ret = btrfs_add_free_space(block_group, start, |
327 | size); | |
79787eaa | 328 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
329 | start = extent_end + 1; |
330 | } else { | |
331 | break; | |
332 | } | |
333 | } | |
334 | ||
335 | if (start < end) { | |
336 | size = end - start; | |
817d52f8 | 337 | total_added += size; |
ea6a478e | 338 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 339 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
340 | } |
341 | ||
817d52f8 | 342 | return total_added; |
0f9dd46c JB |
343 | } |
344 | ||
bab39bf9 | 345 | static noinline void caching_thread(struct btrfs_work *work) |
e37c9e69 | 346 | { |
bab39bf9 JB |
347 | struct btrfs_block_group_cache *block_group; |
348 | struct btrfs_fs_info *fs_info; | |
349 | struct btrfs_caching_control *caching_ctl; | |
350 | struct btrfs_root *extent_root; | |
e37c9e69 | 351 | struct btrfs_path *path; |
5f39d397 | 352 | struct extent_buffer *leaf; |
11833d66 | 353 | struct btrfs_key key; |
817d52f8 | 354 | u64 total_found = 0; |
11833d66 YZ |
355 | u64 last = 0; |
356 | u32 nritems; | |
357 | int ret = 0; | |
f510cfec | 358 | |
bab39bf9 JB |
359 | caching_ctl = container_of(work, struct btrfs_caching_control, work); |
360 | block_group = caching_ctl->block_group; | |
361 | fs_info = block_group->fs_info; | |
362 | extent_root = fs_info->extent_root; | |
363 | ||
e37c9e69 CM |
364 | path = btrfs_alloc_path(); |
365 | if (!path) | |
bab39bf9 | 366 | goto out; |
7d7d6068 | 367 | |
817d52f8 | 368 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 369 | |
5cd57b2c | 370 | /* |
817d52f8 JB |
371 | * We don't want to deadlock with somebody trying to allocate a new |
372 | * extent for the extent root while also trying to search the extent | |
373 | * root to add free space. So we skip locking and search the commit | |
374 | * root, since its read-only | |
5cd57b2c CM |
375 | */ |
376 | path->skip_locking = 1; | |
817d52f8 | 377 | path->search_commit_root = 1; |
026fd317 | 378 | path->reada = 1; |
817d52f8 | 379 | |
e4404d6e | 380 | key.objectid = last; |
e37c9e69 | 381 | key.offset = 0; |
11833d66 | 382 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 383 | again: |
11833d66 | 384 | mutex_lock(&caching_ctl->mutex); |
013f1b12 CM |
385 | /* need to make sure the commit_root doesn't disappear */ |
386 | down_read(&fs_info->extent_commit_sem); | |
387 | ||
11833d66 | 388 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 389 | if (ret < 0) |
ef8bbdfe | 390 | goto err; |
a512bbf8 | 391 | |
11833d66 YZ |
392 | leaf = path->nodes[0]; |
393 | nritems = btrfs_header_nritems(leaf); | |
394 | ||
d397712b | 395 | while (1) { |
7841cb28 | 396 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 397 | last = (u64)-1; |
817d52f8 | 398 | break; |
f25784b3 | 399 | } |
817d52f8 | 400 | |
11833d66 YZ |
401 | if (path->slots[0] < nritems) { |
402 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
403 | } else { | |
404 | ret = find_next_key(path, 0, &key); | |
405 | if (ret) | |
e37c9e69 | 406 | break; |
817d52f8 | 407 | |
589d8ade JB |
408 | if (need_resched() || |
409 | btrfs_next_leaf(extent_root, path)) { | |
410 | caching_ctl->progress = last; | |
ff5714cc | 411 | btrfs_release_path(path); |
589d8ade JB |
412 | up_read(&fs_info->extent_commit_sem); |
413 | mutex_unlock(&caching_ctl->mutex); | |
11833d66 | 414 | cond_resched(); |
589d8ade JB |
415 | goto again; |
416 | } | |
417 | leaf = path->nodes[0]; | |
418 | nritems = btrfs_header_nritems(leaf); | |
419 | continue; | |
11833d66 | 420 | } |
817d52f8 | 421 | |
11833d66 YZ |
422 | if (key.objectid < block_group->key.objectid) { |
423 | path->slots[0]++; | |
817d52f8 | 424 | continue; |
e37c9e69 | 425 | } |
0f9dd46c | 426 | |
e37c9e69 | 427 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 428 | block_group->key.offset) |
e37c9e69 | 429 | break; |
7d7d6068 | 430 | |
11833d66 | 431 | if (key.type == BTRFS_EXTENT_ITEM_KEY) { |
817d52f8 JB |
432 | total_found += add_new_free_space(block_group, |
433 | fs_info, last, | |
434 | key.objectid); | |
7d7d6068 | 435 | last = key.objectid + key.offset; |
817d52f8 | 436 | |
11833d66 YZ |
437 | if (total_found > (1024 * 1024 * 2)) { |
438 | total_found = 0; | |
439 | wake_up(&caching_ctl->wait); | |
440 | } | |
817d52f8 | 441 | } |
e37c9e69 CM |
442 | path->slots[0]++; |
443 | } | |
817d52f8 | 444 | ret = 0; |
e37c9e69 | 445 | |
817d52f8 JB |
446 | total_found += add_new_free_space(block_group, fs_info, last, |
447 | block_group->key.objectid + | |
448 | block_group->key.offset); | |
11833d66 | 449 | caching_ctl->progress = (u64)-1; |
817d52f8 JB |
450 | |
451 | spin_lock(&block_group->lock); | |
11833d66 | 452 | block_group->caching_ctl = NULL; |
817d52f8 JB |
453 | block_group->cached = BTRFS_CACHE_FINISHED; |
454 | spin_unlock(&block_group->lock); | |
0f9dd46c | 455 | |
54aa1f4d | 456 | err: |
e37c9e69 | 457 | btrfs_free_path(path); |
276e680d | 458 | up_read(&fs_info->extent_commit_sem); |
817d52f8 | 459 | |
11833d66 YZ |
460 | free_excluded_extents(extent_root, block_group); |
461 | ||
462 | mutex_unlock(&caching_ctl->mutex); | |
bab39bf9 | 463 | out: |
11833d66 YZ |
464 | wake_up(&caching_ctl->wait); |
465 | ||
466 | put_caching_control(caching_ctl); | |
11dfe35a | 467 | btrfs_put_block_group(block_group); |
817d52f8 JB |
468 | } |
469 | ||
9d66e233 JB |
470 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
471 | struct btrfs_trans_handle *trans, | |
b8399dee | 472 | struct btrfs_root *root, |
9d66e233 | 473 | int load_cache_only) |
817d52f8 | 474 | { |
291c7d2f | 475 | DEFINE_WAIT(wait); |
11833d66 YZ |
476 | struct btrfs_fs_info *fs_info = cache->fs_info; |
477 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
478 | int ret = 0; |
479 | ||
291c7d2f | 480 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
481 | if (!caching_ctl) |
482 | return -ENOMEM; | |
291c7d2f JB |
483 | |
484 | INIT_LIST_HEAD(&caching_ctl->list); | |
485 | mutex_init(&caching_ctl->mutex); | |
486 | init_waitqueue_head(&caching_ctl->wait); | |
487 | caching_ctl->block_group = cache; | |
488 | caching_ctl->progress = cache->key.objectid; | |
489 | atomic_set(&caching_ctl->count, 1); | |
490 | caching_ctl->work.func = caching_thread; | |
491 | ||
492 | spin_lock(&cache->lock); | |
493 | /* | |
494 | * This should be a rare occasion, but this could happen I think in the | |
495 | * case where one thread starts to load the space cache info, and then | |
496 | * some other thread starts a transaction commit which tries to do an | |
497 | * allocation while the other thread is still loading the space cache | |
498 | * info. The previous loop should have kept us from choosing this block | |
499 | * group, but if we've moved to the state where we will wait on caching | |
500 | * block groups we need to first check if we're doing a fast load here, | |
501 | * so we can wait for it to finish, otherwise we could end up allocating | |
502 | * from a block group who's cache gets evicted for one reason or | |
503 | * another. | |
504 | */ | |
505 | while (cache->cached == BTRFS_CACHE_FAST) { | |
506 | struct btrfs_caching_control *ctl; | |
507 | ||
508 | ctl = cache->caching_ctl; | |
509 | atomic_inc(&ctl->count); | |
510 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); | |
511 | spin_unlock(&cache->lock); | |
512 | ||
513 | schedule(); | |
514 | ||
515 | finish_wait(&ctl->wait, &wait); | |
516 | put_caching_control(ctl); | |
517 | spin_lock(&cache->lock); | |
518 | } | |
519 | ||
520 | if (cache->cached != BTRFS_CACHE_NO) { | |
521 | spin_unlock(&cache->lock); | |
522 | kfree(caching_ctl); | |
11833d66 | 523 | return 0; |
291c7d2f JB |
524 | } |
525 | WARN_ON(cache->caching_ctl); | |
526 | cache->caching_ctl = caching_ctl; | |
527 | cache->cached = BTRFS_CACHE_FAST; | |
528 | spin_unlock(&cache->lock); | |
11833d66 | 529 | |
d53ba474 | 530 | if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) { |
9d66e233 JB |
531 | ret = load_free_space_cache(fs_info, cache); |
532 | ||
533 | spin_lock(&cache->lock); | |
534 | if (ret == 1) { | |
291c7d2f | 535 | cache->caching_ctl = NULL; |
9d66e233 JB |
536 | cache->cached = BTRFS_CACHE_FINISHED; |
537 | cache->last_byte_to_unpin = (u64)-1; | |
538 | } else { | |
291c7d2f JB |
539 | if (load_cache_only) { |
540 | cache->caching_ctl = NULL; | |
541 | cache->cached = BTRFS_CACHE_NO; | |
542 | } else { | |
543 | cache->cached = BTRFS_CACHE_STARTED; | |
544 | } | |
9d66e233 JB |
545 | } |
546 | spin_unlock(&cache->lock); | |
291c7d2f | 547 | wake_up(&caching_ctl->wait); |
3c14874a | 548 | if (ret == 1) { |
291c7d2f | 549 | put_caching_control(caching_ctl); |
3c14874a | 550 | free_excluded_extents(fs_info->extent_root, cache); |
9d66e233 | 551 | return 0; |
3c14874a | 552 | } |
291c7d2f JB |
553 | } else { |
554 | /* | |
555 | * We are not going to do the fast caching, set cached to the | |
556 | * appropriate value and wakeup any waiters. | |
557 | */ | |
558 | spin_lock(&cache->lock); | |
559 | if (load_cache_only) { | |
560 | cache->caching_ctl = NULL; | |
561 | cache->cached = BTRFS_CACHE_NO; | |
562 | } else { | |
563 | cache->cached = BTRFS_CACHE_STARTED; | |
564 | } | |
565 | spin_unlock(&cache->lock); | |
566 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
567 | } |
568 | ||
291c7d2f JB |
569 | if (load_cache_only) { |
570 | put_caching_control(caching_ctl); | |
11833d66 | 571 | return 0; |
817d52f8 | 572 | } |
817d52f8 | 573 | |
11833d66 | 574 | down_write(&fs_info->extent_commit_sem); |
291c7d2f | 575 | atomic_inc(&caching_ctl->count); |
11833d66 YZ |
576 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
577 | up_write(&fs_info->extent_commit_sem); | |
578 | ||
11dfe35a | 579 | btrfs_get_block_group(cache); |
11833d66 | 580 | |
bab39bf9 | 581 | btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 582 | |
ef8bbdfe | 583 | return ret; |
e37c9e69 CM |
584 | } |
585 | ||
0f9dd46c JB |
586 | /* |
587 | * return the block group that starts at or after bytenr | |
588 | */ | |
d397712b CM |
589 | static struct btrfs_block_group_cache * |
590 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 591 | { |
0f9dd46c | 592 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 593 | |
0f9dd46c | 594 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 595 | |
0f9dd46c | 596 | return cache; |
0ef3e66b CM |
597 | } |
598 | ||
0f9dd46c | 599 | /* |
9f55684c | 600 | * return the block group that contains the given bytenr |
0f9dd46c | 601 | */ |
d397712b CM |
602 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
603 | struct btrfs_fs_info *info, | |
604 | u64 bytenr) | |
be744175 | 605 | { |
0f9dd46c | 606 | struct btrfs_block_group_cache *cache; |
be744175 | 607 | |
0f9dd46c | 608 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 609 | |
0f9dd46c | 610 | return cache; |
be744175 | 611 | } |
0b86a832 | 612 | |
0f9dd46c JB |
613 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
614 | u64 flags) | |
6324fbf3 | 615 | { |
0f9dd46c | 616 | struct list_head *head = &info->space_info; |
0f9dd46c | 617 | struct btrfs_space_info *found; |
4184ea7f | 618 | |
52ba6929 | 619 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 620 | |
4184ea7f CM |
621 | rcu_read_lock(); |
622 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 623 | if (found->flags & flags) { |
4184ea7f | 624 | rcu_read_unlock(); |
0f9dd46c | 625 | return found; |
4184ea7f | 626 | } |
0f9dd46c | 627 | } |
4184ea7f | 628 | rcu_read_unlock(); |
0f9dd46c | 629 | return NULL; |
6324fbf3 CM |
630 | } |
631 | ||
4184ea7f CM |
632 | /* |
633 | * after adding space to the filesystem, we need to clear the full flags | |
634 | * on all the space infos. | |
635 | */ | |
636 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
637 | { | |
638 | struct list_head *head = &info->space_info; | |
639 | struct btrfs_space_info *found; | |
640 | ||
641 | rcu_read_lock(); | |
642 | list_for_each_entry_rcu(found, head, list) | |
643 | found->full = 0; | |
644 | rcu_read_unlock(); | |
645 | } | |
646 | ||
d2fb3437 YZ |
647 | u64 btrfs_find_block_group(struct btrfs_root *root, |
648 | u64 search_start, u64 search_hint, int owner) | |
cd1bc465 | 649 | { |
96b5179d | 650 | struct btrfs_block_group_cache *cache; |
cd1bc465 | 651 | u64 used; |
d2fb3437 YZ |
652 | u64 last = max(search_hint, search_start); |
653 | u64 group_start = 0; | |
31f3c99b | 654 | int full_search = 0; |
d2fb3437 | 655 | int factor = 9; |
0ef3e66b | 656 | int wrapped = 0; |
31f3c99b | 657 | again: |
e8569813 ZY |
658 | while (1) { |
659 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
0f9dd46c JB |
660 | if (!cache) |
661 | break; | |
96b5179d | 662 | |
c286ac48 | 663 | spin_lock(&cache->lock); |
96b5179d CM |
664 | last = cache->key.objectid + cache->key.offset; |
665 | used = btrfs_block_group_used(&cache->item); | |
666 | ||
d2fb3437 YZ |
667 | if ((full_search || !cache->ro) && |
668 | block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) { | |
e8569813 | 669 | if (used + cache->pinned + cache->reserved < |
d2fb3437 YZ |
670 | div_factor(cache->key.offset, factor)) { |
671 | group_start = cache->key.objectid; | |
c286ac48 | 672 | spin_unlock(&cache->lock); |
fa9c0d79 | 673 | btrfs_put_block_group(cache); |
8790d502 CM |
674 | goto found; |
675 | } | |
6324fbf3 | 676 | } |
c286ac48 | 677 | spin_unlock(&cache->lock); |
fa9c0d79 | 678 | btrfs_put_block_group(cache); |
de428b63 | 679 | cond_resched(); |
cd1bc465 | 680 | } |
0ef3e66b CM |
681 | if (!wrapped) { |
682 | last = search_start; | |
683 | wrapped = 1; | |
684 | goto again; | |
685 | } | |
686 | if (!full_search && factor < 10) { | |
be744175 | 687 | last = search_start; |
31f3c99b | 688 | full_search = 1; |
0ef3e66b | 689 | factor = 10; |
31f3c99b CM |
690 | goto again; |
691 | } | |
be744175 | 692 | found: |
d2fb3437 | 693 | return group_start; |
925baedd | 694 | } |
0f9dd46c | 695 | |
e02119d5 | 696 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 697 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
698 | { |
699 | int ret; | |
700 | struct btrfs_key key; | |
31840ae1 | 701 | struct btrfs_path *path; |
e02119d5 | 702 | |
31840ae1 | 703 | path = btrfs_alloc_path(); |
d8926bb3 MF |
704 | if (!path) |
705 | return -ENOMEM; | |
706 | ||
e02119d5 CM |
707 | key.objectid = start; |
708 | key.offset = len; | |
709 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); | |
710 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, | |
711 | 0, 0); | |
31840ae1 | 712 | btrfs_free_path(path); |
7bb86316 CM |
713 | return ret; |
714 | } | |
715 | ||
a22285a6 YZ |
716 | /* |
717 | * helper function to lookup reference count and flags of extent. | |
718 | * | |
719 | * the head node for delayed ref is used to store the sum of all the | |
720 | * reference count modifications queued up in the rbtree. the head | |
721 | * node may also store the extent flags to set. This way you can check | |
722 | * to see what the reference count and extent flags would be if all of | |
723 | * the delayed refs are not processed. | |
724 | */ | |
725 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
726 | struct btrfs_root *root, u64 bytenr, | |
727 | u64 num_bytes, u64 *refs, u64 *flags) | |
728 | { | |
729 | struct btrfs_delayed_ref_head *head; | |
730 | struct btrfs_delayed_ref_root *delayed_refs; | |
731 | struct btrfs_path *path; | |
732 | struct btrfs_extent_item *ei; | |
733 | struct extent_buffer *leaf; | |
734 | struct btrfs_key key; | |
735 | u32 item_size; | |
736 | u64 num_refs; | |
737 | u64 extent_flags; | |
738 | int ret; | |
739 | ||
740 | path = btrfs_alloc_path(); | |
741 | if (!path) | |
742 | return -ENOMEM; | |
743 | ||
744 | key.objectid = bytenr; | |
745 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
746 | key.offset = num_bytes; | |
747 | if (!trans) { | |
748 | path->skip_locking = 1; | |
749 | path->search_commit_root = 1; | |
750 | } | |
751 | again: | |
752 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, | |
753 | &key, path, 0, 0); | |
754 | if (ret < 0) | |
755 | goto out_free; | |
756 | ||
757 | if (ret == 0) { | |
758 | leaf = path->nodes[0]; | |
759 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
760 | if (item_size >= sizeof(*ei)) { | |
761 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
762 | struct btrfs_extent_item); | |
763 | num_refs = btrfs_extent_refs(leaf, ei); | |
764 | extent_flags = btrfs_extent_flags(leaf, ei); | |
765 | } else { | |
766 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
767 | struct btrfs_extent_item_v0 *ei0; | |
768 | BUG_ON(item_size != sizeof(*ei0)); | |
769 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
770 | struct btrfs_extent_item_v0); | |
771 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
772 | /* FIXME: this isn't correct for data */ | |
773 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
774 | #else | |
775 | BUG(); | |
776 | #endif | |
777 | } | |
778 | BUG_ON(num_refs == 0); | |
779 | } else { | |
780 | num_refs = 0; | |
781 | extent_flags = 0; | |
782 | ret = 0; | |
783 | } | |
784 | ||
785 | if (!trans) | |
786 | goto out; | |
787 | ||
788 | delayed_refs = &trans->transaction->delayed_refs; | |
789 | spin_lock(&delayed_refs->lock); | |
790 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
791 | if (head) { | |
792 | if (!mutex_trylock(&head->mutex)) { | |
793 | atomic_inc(&head->node.refs); | |
794 | spin_unlock(&delayed_refs->lock); | |
795 | ||
b3b4aa74 | 796 | btrfs_release_path(path); |
a22285a6 | 797 | |
8cc33e5c DS |
798 | /* |
799 | * Mutex was contended, block until it's released and try | |
800 | * again | |
801 | */ | |
a22285a6 YZ |
802 | mutex_lock(&head->mutex); |
803 | mutex_unlock(&head->mutex); | |
804 | btrfs_put_delayed_ref(&head->node); | |
805 | goto again; | |
806 | } | |
807 | if (head->extent_op && head->extent_op->update_flags) | |
808 | extent_flags |= head->extent_op->flags_to_set; | |
809 | else | |
810 | BUG_ON(num_refs == 0); | |
811 | ||
812 | num_refs += head->node.ref_mod; | |
813 | mutex_unlock(&head->mutex); | |
814 | } | |
815 | spin_unlock(&delayed_refs->lock); | |
816 | out: | |
817 | WARN_ON(num_refs == 0); | |
818 | if (refs) | |
819 | *refs = num_refs; | |
820 | if (flags) | |
821 | *flags = extent_flags; | |
822 | out_free: | |
823 | btrfs_free_path(path); | |
824 | return ret; | |
825 | } | |
826 | ||
d8d5f3e1 CM |
827 | /* |
828 | * Back reference rules. Back refs have three main goals: | |
829 | * | |
830 | * 1) differentiate between all holders of references to an extent so that | |
831 | * when a reference is dropped we can make sure it was a valid reference | |
832 | * before freeing the extent. | |
833 | * | |
834 | * 2) Provide enough information to quickly find the holders of an extent | |
835 | * if we notice a given block is corrupted or bad. | |
836 | * | |
837 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
838 | * maintenance. This is actually the same as #2, but with a slightly | |
839 | * different use case. | |
840 | * | |
5d4f98a2 YZ |
841 | * There are two kinds of back refs. The implicit back refs is optimized |
842 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
843 | * back refs of this kind provide information about the block's owner tree | |
844 | * and the pointer's key. These information allow us to find the block by | |
845 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
846 | * referenced by their owner trees. The location of tree block is recorded | |
847 | * in the back refs. Actually the full back refs is generic, and can be | |
848 | * used in all cases the implicit back refs is used. The major shortcoming | |
849 | * of the full back refs is its overhead. Every time a tree block gets | |
850 | * COWed, we have to update back refs entry for all pointers in it. | |
851 | * | |
852 | * For a newly allocated tree block, we use implicit back refs for | |
853 | * pointers in it. This means most tree related operations only involve | |
854 | * implicit back refs. For a tree block created in old transaction, the | |
855 | * only way to drop a reference to it is COW it. So we can detect the | |
856 | * event that tree block loses its owner tree's reference and do the | |
857 | * back refs conversion. | |
858 | * | |
859 | * When a tree block is COW'd through a tree, there are four cases: | |
860 | * | |
861 | * The reference count of the block is one and the tree is the block's | |
862 | * owner tree. Nothing to do in this case. | |
863 | * | |
864 | * The reference count of the block is one and the tree is not the | |
865 | * block's owner tree. In this case, full back refs is used for pointers | |
866 | * in the block. Remove these full back refs, add implicit back refs for | |
867 | * every pointers in the new block. | |
868 | * | |
869 | * The reference count of the block is greater than one and the tree is | |
870 | * the block's owner tree. In this case, implicit back refs is used for | |
871 | * pointers in the block. Add full back refs for every pointers in the | |
872 | * block, increase lower level extents' reference counts. The original | |
873 | * implicit back refs are entailed to the new block. | |
874 | * | |
875 | * The reference count of the block is greater than one and the tree is | |
876 | * not the block's owner tree. Add implicit back refs for every pointer in | |
877 | * the new block, increase lower level extents' reference count. | |
878 | * | |
879 | * Back Reference Key composing: | |
880 | * | |
881 | * The key objectid corresponds to the first byte in the extent, | |
882 | * The key type is used to differentiate between types of back refs. | |
883 | * There are different meanings of the key offset for different types | |
884 | * of back refs. | |
885 | * | |
d8d5f3e1 CM |
886 | * File extents can be referenced by: |
887 | * | |
888 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 889 | * - different files inside a single subvolume |
d8d5f3e1 CM |
890 | * - different offsets inside a file (bookend extents in file.c) |
891 | * | |
5d4f98a2 | 892 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
893 | * |
894 | * - Objectid of the subvolume root | |
d8d5f3e1 | 895 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
896 | * - original offset in the file |
897 | * - how many bookend extents | |
d8d5f3e1 | 898 | * |
5d4f98a2 YZ |
899 | * The key offset for the implicit back refs is hash of the first |
900 | * three fields. | |
d8d5f3e1 | 901 | * |
5d4f98a2 | 902 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 903 | * |
5d4f98a2 | 904 | * - number of pointers in the tree leaf |
d8d5f3e1 | 905 | * |
5d4f98a2 YZ |
906 | * The key offset for the implicit back refs is the first byte of |
907 | * the tree leaf | |
d8d5f3e1 | 908 | * |
5d4f98a2 YZ |
909 | * When a file extent is allocated, The implicit back refs is used. |
910 | * the fields are filled in: | |
d8d5f3e1 | 911 | * |
5d4f98a2 | 912 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 913 | * |
5d4f98a2 YZ |
914 | * When a file extent is removed file truncation, we find the |
915 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 916 | * |
5d4f98a2 | 917 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 918 | * |
5d4f98a2 | 919 | * Btree extents can be referenced by: |
d8d5f3e1 | 920 | * |
5d4f98a2 | 921 | * - Different subvolumes |
d8d5f3e1 | 922 | * |
5d4f98a2 YZ |
923 | * Both the implicit back refs and the full back refs for tree blocks |
924 | * only consist of key. The key offset for the implicit back refs is | |
925 | * objectid of block's owner tree. The key offset for the full back refs | |
926 | * is the first byte of parent block. | |
d8d5f3e1 | 927 | * |
5d4f98a2 YZ |
928 | * When implicit back refs is used, information about the lowest key and |
929 | * level of the tree block are required. These information are stored in | |
930 | * tree block info structure. | |
d8d5f3e1 | 931 | */ |
31840ae1 | 932 | |
5d4f98a2 YZ |
933 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
934 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
935 | struct btrfs_root *root, | |
936 | struct btrfs_path *path, | |
937 | u64 owner, u32 extra_size) | |
7bb86316 | 938 | { |
5d4f98a2 YZ |
939 | struct btrfs_extent_item *item; |
940 | struct btrfs_extent_item_v0 *ei0; | |
941 | struct btrfs_extent_ref_v0 *ref0; | |
942 | struct btrfs_tree_block_info *bi; | |
943 | struct extent_buffer *leaf; | |
7bb86316 | 944 | struct btrfs_key key; |
5d4f98a2 YZ |
945 | struct btrfs_key found_key; |
946 | u32 new_size = sizeof(*item); | |
947 | u64 refs; | |
948 | int ret; | |
949 | ||
950 | leaf = path->nodes[0]; | |
951 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
952 | ||
953 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
954 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
955 | struct btrfs_extent_item_v0); | |
956 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
957 | ||
958 | if (owner == (u64)-1) { | |
959 | while (1) { | |
960 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
961 | ret = btrfs_next_leaf(root, path); | |
962 | if (ret < 0) | |
963 | return ret; | |
79787eaa | 964 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
965 | leaf = path->nodes[0]; |
966 | } | |
967 | btrfs_item_key_to_cpu(leaf, &found_key, | |
968 | path->slots[0]); | |
969 | BUG_ON(key.objectid != found_key.objectid); | |
970 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
971 | path->slots[0]++; | |
972 | continue; | |
973 | } | |
974 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
975 | struct btrfs_extent_ref_v0); | |
976 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
977 | break; | |
978 | } | |
979 | } | |
b3b4aa74 | 980 | btrfs_release_path(path); |
5d4f98a2 YZ |
981 | |
982 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
983 | new_size += sizeof(*bi); | |
984 | ||
985 | new_size -= sizeof(*ei0); | |
986 | ret = btrfs_search_slot(trans, root, &key, path, | |
987 | new_size + extra_size, 1); | |
988 | if (ret < 0) | |
989 | return ret; | |
79787eaa | 990 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 991 | |
143bede5 | 992 | btrfs_extend_item(trans, root, path, new_size); |
5d4f98a2 YZ |
993 | |
994 | leaf = path->nodes[0]; | |
995 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
996 | btrfs_set_extent_refs(leaf, item, refs); | |
997 | /* FIXME: get real generation */ | |
998 | btrfs_set_extent_generation(leaf, item, 0); | |
999 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1000 | btrfs_set_extent_flags(leaf, item, | |
1001 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1002 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1003 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1004 | /* FIXME: get first key of the block */ | |
1005 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
1006 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
1007 | } else { | |
1008 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1009 | } | |
1010 | btrfs_mark_buffer_dirty(leaf); | |
1011 | return 0; | |
1012 | } | |
1013 | #endif | |
1014 | ||
1015 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
1016 | { | |
1017 | u32 high_crc = ~(u32)0; | |
1018 | u32 low_crc = ~(u32)0; | |
1019 | __le64 lenum; | |
1020 | ||
1021 | lenum = cpu_to_le64(root_objectid); | |
163e783e | 1022 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1023 | lenum = cpu_to_le64(owner); |
163e783e | 1024 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1025 | lenum = cpu_to_le64(offset); |
163e783e | 1026 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1027 | |
1028 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1029 | } | |
1030 | ||
1031 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1032 | struct btrfs_extent_data_ref *ref) | |
1033 | { | |
1034 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1035 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1036 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1037 | } | |
1038 | ||
1039 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1040 | struct btrfs_extent_data_ref *ref, | |
1041 | u64 root_objectid, u64 owner, u64 offset) | |
1042 | { | |
1043 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1044 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1045 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1046 | return 0; | |
1047 | return 1; | |
1048 | } | |
1049 | ||
1050 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
1051 | struct btrfs_root *root, | |
1052 | struct btrfs_path *path, | |
1053 | u64 bytenr, u64 parent, | |
1054 | u64 root_objectid, | |
1055 | u64 owner, u64 offset) | |
1056 | { | |
1057 | struct btrfs_key key; | |
1058 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1059 | struct extent_buffer *leaf; |
5d4f98a2 | 1060 | u32 nritems; |
74493f7a | 1061 | int ret; |
5d4f98a2 YZ |
1062 | int recow; |
1063 | int err = -ENOENT; | |
74493f7a | 1064 | |
31840ae1 | 1065 | key.objectid = bytenr; |
5d4f98a2 YZ |
1066 | if (parent) { |
1067 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1068 | key.offset = parent; | |
1069 | } else { | |
1070 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1071 | key.offset = hash_extent_data_ref(root_objectid, | |
1072 | owner, offset); | |
1073 | } | |
1074 | again: | |
1075 | recow = 0; | |
1076 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1077 | if (ret < 0) { | |
1078 | err = ret; | |
1079 | goto fail; | |
1080 | } | |
31840ae1 | 1081 | |
5d4f98a2 YZ |
1082 | if (parent) { |
1083 | if (!ret) | |
1084 | return 0; | |
1085 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1086 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1087 | btrfs_release_path(path); |
5d4f98a2 YZ |
1088 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1089 | if (ret < 0) { | |
1090 | err = ret; | |
1091 | goto fail; | |
1092 | } | |
1093 | if (!ret) | |
1094 | return 0; | |
1095 | #endif | |
1096 | goto fail; | |
31840ae1 ZY |
1097 | } |
1098 | ||
1099 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1100 | nritems = btrfs_header_nritems(leaf); |
1101 | while (1) { | |
1102 | if (path->slots[0] >= nritems) { | |
1103 | ret = btrfs_next_leaf(root, path); | |
1104 | if (ret < 0) | |
1105 | err = ret; | |
1106 | if (ret) | |
1107 | goto fail; | |
1108 | ||
1109 | leaf = path->nodes[0]; | |
1110 | nritems = btrfs_header_nritems(leaf); | |
1111 | recow = 1; | |
1112 | } | |
1113 | ||
1114 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1115 | if (key.objectid != bytenr || | |
1116 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1117 | goto fail; | |
1118 | ||
1119 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1120 | struct btrfs_extent_data_ref); | |
1121 | ||
1122 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1123 | owner, offset)) { | |
1124 | if (recow) { | |
b3b4aa74 | 1125 | btrfs_release_path(path); |
5d4f98a2 YZ |
1126 | goto again; |
1127 | } | |
1128 | err = 0; | |
1129 | break; | |
1130 | } | |
1131 | path->slots[0]++; | |
31840ae1 | 1132 | } |
5d4f98a2 YZ |
1133 | fail: |
1134 | return err; | |
31840ae1 ZY |
1135 | } |
1136 | ||
5d4f98a2 YZ |
1137 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
1138 | struct btrfs_root *root, | |
1139 | struct btrfs_path *path, | |
1140 | u64 bytenr, u64 parent, | |
1141 | u64 root_objectid, u64 owner, | |
1142 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
1143 | { |
1144 | struct btrfs_key key; | |
1145 | struct extent_buffer *leaf; | |
5d4f98a2 | 1146 | u32 size; |
31840ae1 ZY |
1147 | u32 num_refs; |
1148 | int ret; | |
74493f7a | 1149 | |
74493f7a | 1150 | key.objectid = bytenr; |
5d4f98a2 YZ |
1151 | if (parent) { |
1152 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1153 | key.offset = parent; | |
1154 | size = sizeof(struct btrfs_shared_data_ref); | |
1155 | } else { | |
1156 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1157 | key.offset = hash_extent_data_ref(root_objectid, | |
1158 | owner, offset); | |
1159 | size = sizeof(struct btrfs_extent_data_ref); | |
1160 | } | |
74493f7a | 1161 | |
5d4f98a2 YZ |
1162 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1163 | if (ret && ret != -EEXIST) | |
1164 | goto fail; | |
1165 | ||
1166 | leaf = path->nodes[0]; | |
1167 | if (parent) { | |
1168 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1169 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1170 | struct btrfs_shared_data_ref); |
1171 | if (ret == 0) { | |
1172 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1173 | } else { | |
1174 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1175 | num_refs += refs_to_add; | |
1176 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1177 | } |
5d4f98a2 YZ |
1178 | } else { |
1179 | struct btrfs_extent_data_ref *ref; | |
1180 | while (ret == -EEXIST) { | |
1181 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1182 | struct btrfs_extent_data_ref); | |
1183 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1184 | owner, offset)) | |
1185 | break; | |
b3b4aa74 | 1186 | btrfs_release_path(path); |
5d4f98a2 YZ |
1187 | key.offset++; |
1188 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1189 | size); | |
1190 | if (ret && ret != -EEXIST) | |
1191 | goto fail; | |
31840ae1 | 1192 | |
5d4f98a2 YZ |
1193 | leaf = path->nodes[0]; |
1194 | } | |
1195 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1196 | struct btrfs_extent_data_ref); | |
1197 | if (ret == 0) { | |
1198 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1199 | root_objectid); | |
1200 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1201 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1202 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1203 | } else { | |
1204 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1205 | num_refs += refs_to_add; | |
1206 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1207 | } |
31840ae1 | 1208 | } |
5d4f98a2 YZ |
1209 | btrfs_mark_buffer_dirty(leaf); |
1210 | ret = 0; | |
1211 | fail: | |
b3b4aa74 | 1212 | btrfs_release_path(path); |
7bb86316 | 1213 | return ret; |
74493f7a CM |
1214 | } |
1215 | ||
5d4f98a2 YZ |
1216 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
1217 | struct btrfs_root *root, | |
1218 | struct btrfs_path *path, | |
1219 | int refs_to_drop) | |
31840ae1 | 1220 | { |
5d4f98a2 YZ |
1221 | struct btrfs_key key; |
1222 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1223 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1224 | struct extent_buffer *leaf; |
5d4f98a2 | 1225 | u32 num_refs = 0; |
31840ae1 ZY |
1226 | int ret = 0; |
1227 | ||
1228 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1229 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1230 | ||
1231 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1232 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1233 | struct btrfs_extent_data_ref); | |
1234 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1235 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1236 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1237 | struct btrfs_shared_data_ref); | |
1238 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1239 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1240 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1241 | struct btrfs_extent_ref_v0 *ref0; | |
1242 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1243 | struct btrfs_extent_ref_v0); | |
1244 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1245 | #endif | |
1246 | } else { | |
1247 | BUG(); | |
1248 | } | |
1249 | ||
56bec294 CM |
1250 | BUG_ON(num_refs < refs_to_drop); |
1251 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1252 | |
31840ae1 ZY |
1253 | if (num_refs == 0) { |
1254 | ret = btrfs_del_item(trans, root, path); | |
1255 | } else { | |
5d4f98a2 YZ |
1256 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1257 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1258 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1259 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1260 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1261 | else { | |
1262 | struct btrfs_extent_ref_v0 *ref0; | |
1263 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1264 | struct btrfs_extent_ref_v0); | |
1265 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1266 | } | |
1267 | #endif | |
31840ae1 ZY |
1268 | btrfs_mark_buffer_dirty(leaf); |
1269 | } | |
31840ae1 ZY |
1270 | return ret; |
1271 | } | |
1272 | ||
5d4f98a2 YZ |
1273 | static noinline u32 extent_data_ref_count(struct btrfs_root *root, |
1274 | struct btrfs_path *path, | |
1275 | struct btrfs_extent_inline_ref *iref) | |
15916de8 | 1276 | { |
5d4f98a2 YZ |
1277 | struct btrfs_key key; |
1278 | struct extent_buffer *leaf; | |
1279 | struct btrfs_extent_data_ref *ref1; | |
1280 | struct btrfs_shared_data_ref *ref2; | |
1281 | u32 num_refs = 0; | |
1282 | ||
1283 | leaf = path->nodes[0]; | |
1284 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1285 | if (iref) { | |
1286 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1287 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1288 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1289 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1290 | } else { | |
1291 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1292 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1293 | } | |
1294 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1295 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1296 | struct btrfs_extent_data_ref); | |
1297 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1298 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1299 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1300 | struct btrfs_shared_data_ref); | |
1301 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1302 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1303 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1304 | struct btrfs_extent_ref_v0 *ref0; | |
1305 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1306 | struct btrfs_extent_ref_v0); | |
1307 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1308 | #endif |
5d4f98a2 YZ |
1309 | } else { |
1310 | WARN_ON(1); | |
1311 | } | |
1312 | return num_refs; | |
1313 | } | |
15916de8 | 1314 | |
5d4f98a2 YZ |
1315 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1316 | struct btrfs_root *root, | |
1317 | struct btrfs_path *path, | |
1318 | u64 bytenr, u64 parent, | |
1319 | u64 root_objectid) | |
1f3c79a2 | 1320 | { |
5d4f98a2 | 1321 | struct btrfs_key key; |
1f3c79a2 | 1322 | int ret; |
1f3c79a2 | 1323 | |
5d4f98a2 YZ |
1324 | key.objectid = bytenr; |
1325 | if (parent) { | |
1326 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1327 | key.offset = parent; | |
1328 | } else { | |
1329 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1330 | key.offset = root_objectid; | |
1f3c79a2 LH |
1331 | } |
1332 | ||
5d4f98a2 YZ |
1333 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1334 | if (ret > 0) | |
1335 | ret = -ENOENT; | |
1336 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1337 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1338 | btrfs_release_path(path); |
5d4f98a2 YZ |
1339 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1340 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1341 | if (ret > 0) | |
1342 | ret = -ENOENT; | |
1343 | } | |
1f3c79a2 | 1344 | #endif |
5d4f98a2 | 1345 | return ret; |
1f3c79a2 LH |
1346 | } |
1347 | ||
5d4f98a2 YZ |
1348 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1349 | struct btrfs_root *root, | |
1350 | struct btrfs_path *path, | |
1351 | u64 bytenr, u64 parent, | |
1352 | u64 root_objectid) | |
31840ae1 | 1353 | { |
5d4f98a2 | 1354 | struct btrfs_key key; |
31840ae1 | 1355 | int ret; |
31840ae1 | 1356 | |
5d4f98a2 YZ |
1357 | key.objectid = bytenr; |
1358 | if (parent) { | |
1359 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1360 | key.offset = parent; | |
1361 | } else { | |
1362 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1363 | key.offset = root_objectid; | |
1364 | } | |
1365 | ||
1366 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
b3b4aa74 | 1367 | btrfs_release_path(path); |
31840ae1 ZY |
1368 | return ret; |
1369 | } | |
1370 | ||
5d4f98a2 | 1371 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1372 | { |
5d4f98a2 YZ |
1373 | int type; |
1374 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1375 | if (parent > 0) | |
1376 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1377 | else | |
1378 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1379 | } else { | |
1380 | if (parent > 0) | |
1381 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1382 | else | |
1383 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1384 | } | |
1385 | return type; | |
31840ae1 | 1386 | } |
56bec294 | 1387 | |
2c47e605 YZ |
1388 | static int find_next_key(struct btrfs_path *path, int level, |
1389 | struct btrfs_key *key) | |
56bec294 | 1390 | |
02217ed2 | 1391 | { |
2c47e605 | 1392 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1393 | if (!path->nodes[level]) |
1394 | break; | |
5d4f98a2 YZ |
1395 | if (path->slots[level] + 1 >= |
1396 | btrfs_header_nritems(path->nodes[level])) | |
1397 | continue; | |
1398 | if (level == 0) | |
1399 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1400 | path->slots[level] + 1); | |
1401 | else | |
1402 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1403 | path->slots[level] + 1); | |
1404 | return 0; | |
1405 | } | |
1406 | return 1; | |
1407 | } | |
037e6390 | 1408 | |
5d4f98a2 YZ |
1409 | /* |
1410 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1411 | * to the address of inline back ref, and 0 is returned. | |
1412 | * | |
1413 | * if back ref isn't found, *ref_ret is set to the address where it | |
1414 | * should be inserted, and -ENOENT is returned. | |
1415 | * | |
1416 | * if insert is true and there are too many inline back refs, the path | |
1417 | * points to the extent item, and -EAGAIN is returned. | |
1418 | * | |
1419 | * NOTE: inline back refs are ordered in the same way that back ref | |
1420 | * items in the tree are ordered. | |
1421 | */ | |
1422 | static noinline_for_stack | |
1423 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1424 | struct btrfs_root *root, | |
1425 | struct btrfs_path *path, | |
1426 | struct btrfs_extent_inline_ref **ref_ret, | |
1427 | u64 bytenr, u64 num_bytes, | |
1428 | u64 parent, u64 root_objectid, | |
1429 | u64 owner, u64 offset, int insert) | |
1430 | { | |
1431 | struct btrfs_key key; | |
1432 | struct extent_buffer *leaf; | |
1433 | struct btrfs_extent_item *ei; | |
1434 | struct btrfs_extent_inline_ref *iref; | |
1435 | u64 flags; | |
1436 | u64 item_size; | |
1437 | unsigned long ptr; | |
1438 | unsigned long end; | |
1439 | int extra_size; | |
1440 | int type; | |
1441 | int want; | |
1442 | int ret; | |
1443 | int err = 0; | |
26b8003f | 1444 | |
db94535d | 1445 | key.objectid = bytenr; |
31840ae1 | 1446 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1447 | key.offset = num_bytes; |
31840ae1 | 1448 | |
5d4f98a2 YZ |
1449 | want = extent_ref_type(parent, owner); |
1450 | if (insert) { | |
1451 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1452 | path->keep_locks = 1; |
5d4f98a2 YZ |
1453 | } else |
1454 | extra_size = -1; | |
1455 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); | |
b9473439 | 1456 | if (ret < 0) { |
5d4f98a2 YZ |
1457 | err = ret; |
1458 | goto out; | |
1459 | } | |
79787eaa JM |
1460 | if (ret && !insert) { |
1461 | err = -ENOENT; | |
1462 | goto out; | |
1463 | } | |
1464 | BUG_ON(ret); /* Corruption */ | |
5d4f98a2 YZ |
1465 | |
1466 | leaf = path->nodes[0]; | |
1467 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1468 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1469 | if (item_size < sizeof(*ei)) { | |
1470 | if (!insert) { | |
1471 | err = -ENOENT; | |
1472 | goto out; | |
1473 | } | |
1474 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1475 | extra_size); | |
1476 | if (ret < 0) { | |
1477 | err = ret; | |
1478 | goto out; | |
1479 | } | |
1480 | leaf = path->nodes[0]; | |
1481 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1482 | } | |
1483 | #endif | |
1484 | BUG_ON(item_size < sizeof(*ei)); | |
1485 | ||
5d4f98a2 YZ |
1486 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1487 | flags = btrfs_extent_flags(leaf, ei); | |
1488 | ||
1489 | ptr = (unsigned long)(ei + 1); | |
1490 | end = (unsigned long)ei + item_size; | |
1491 | ||
1492 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { | |
1493 | ptr += sizeof(struct btrfs_tree_block_info); | |
1494 | BUG_ON(ptr > end); | |
1495 | } else { | |
1496 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA)); | |
1497 | } | |
1498 | ||
1499 | err = -ENOENT; | |
1500 | while (1) { | |
1501 | if (ptr >= end) { | |
1502 | WARN_ON(ptr > end); | |
1503 | break; | |
1504 | } | |
1505 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1506 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1507 | if (want < type) | |
1508 | break; | |
1509 | if (want > type) { | |
1510 | ptr += btrfs_extent_inline_ref_size(type); | |
1511 | continue; | |
1512 | } | |
1513 | ||
1514 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1515 | struct btrfs_extent_data_ref *dref; | |
1516 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1517 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1518 | owner, offset)) { | |
1519 | err = 0; | |
1520 | break; | |
1521 | } | |
1522 | if (hash_extent_data_ref_item(leaf, dref) < | |
1523 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1524 | break; | |
1525 | } else { | |
1526 | u64 ref_offset; | |
1527 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1528 | if (parent > 0) { | |
1529 | if (parent == ref_offset) { | |
1530 | err = 0; | |
1531 | break; | |
1532 | } | |
1533 | if (ref_offset < parent) | |
1534 | break; | |
1535 | } else { | |
1536 | if (root_objectid == ref_offset) { | |
1537 | err = 0; | |
1538 | break; | |
1539 | } | |
1540 | if (ref_offset < root_objectid) | |
1541 | break; | |
1542 | } | |
1543 | } | |
1544 | ptr += btrfs_extent_inline_ref_size(type); | |
1545 | } | |
1546 | if (err == -ENOENT && insert) { | |
1547 | if (item_size + extra_size >= | |
1548 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1549 | err = -EAGAIN; | |
1550 | goto out; | |
1551 | } | |
1552 | /* | |
1553 | * To add new inline back ref, we have to make sure | |
1554 | * there is no corresponding back ref item. | |
1555 | * For simplicity, we just do not add new inline back | |
1556 | * ref if there is any kind of item for this block | |
1557 | */ | |
2c47e605 YZ |
1558 | if (find_next_key(path, 0, &key) == 0 && |
1559 | key.objectid == bytenr && | |
85d4198e | 1560 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1561 | err = -EAGAIN; |
1562 | goto out; | |
1563 | } | |
1564 | } | |
1565 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1566 | out: | |
85d4198e | 1567 | if (insert) { |
5d4f98a2 YZ |
1568 | path->keep_locks = 0; |
1569 | btrfs_unlock_up_safe(path, 1); | |
1570 | } | |
1571 | return err; | |
1572 | } | |
1573 | ||
1574 | /* | |
1575 | * helper to add new inline back ref | |
1576 | */ | |
1577 | static noinline_for_stack | |
143bede5 JM |
1578 | void setup_inline_extent_backref(struct btrfs_trans_handle *trans, |
1579 | struct btrfs_root *root, | |
1580 | struct btrfs_path *path, | |
1581 | struct btrfs_extent_inline_ref *iref, | |
1582 | u64 parent, u64 root_objectid, | |
1583 | u64 owner, u64 offset, int refs_to_add, | |
1584 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1585 | { |
1586 | struct extent_buffer *leaf; | |
1587 | struct btrfs_extent_item *ei; | |
1588 | unsigned long ptr; | |
1589 | unsigned long end; | |
1590 | unsigned long item_offset; | |
1591 | u64 refs; | |
1592 | int size; | |
1593 | int type; | |
5d4f98a2 YZ |
1594 | |
1595 | leaf = path->nodes[0]; | |
1596 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1597 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1598 | ||
1599 | type = extent_ref_type(parent, owner); | |
1600 | size = btrfs_extent_inline_ref_size(type); | |
1601 | ||
143bede5 | 1602 | btrfs_extend_item(trans, root, path, size); |
5d4f98a2 YZ |
1603 | |
1604 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1605 | refs = btrfs_extent_refs(leaf, ei); | |
1606 | refs += refs_to_add; | |
1607 | btrfs_set_extent_refs(leaf, ei, refs); | |
1608 | if (extent_op) | |
1609 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1610 | ||
1611 | ptr = (unsigned long)ei + item_offset; | |
1612 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1613 | if (ptr < end - size) | |
1614 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1615 | end - size - ptr); | |
1616 | ||
1617 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1618 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1619 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1620 | struct btrfs_extent_data_ref *dref; | |
1621 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1622 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1623 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1624 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1625 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1626 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1627 | struct btrfs_shared_data_ref *sref; | |
1628 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1629 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1630 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1631 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1632 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1633 | } else { | |
1634 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1635 | } | |
1636 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1637 | } |
1638 | ||
1639 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1640 | struct btrfs_root *root, | |
1641 | struct btrfs_path *path, | |
1642 | struct btrfs_extent_inline_ref **ref_ret, | |
1643 | u64 bytenr, u64 num_bytes, u64 parent, | |
1644 | u64 root_objectid, u64 owner, u64 offset) | |
1645 | { | |
1646 | int ret; | |
1647 | ||
1648 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1649 | bytenr, num_bytes, parent, | |
1650 | root_objectid, owner, offset, 0); | |
1651 | if (ret != -ENOENT) | |
54aa1f4d | 1652 | return ret; |
5d4f98a2 | 1653 | |
b3b4aa74 | 1654 | btrfs_release_path(path); |
5d4f98a2 YZ |
1655 | *ref_ret = NULL; |
1656 | ||
1657 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1658 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1659 | root_objectid); | |
1660 | } else { | |
1661 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1662 | root_objectid, owner, offset); | |
b9473439 | 1663 | } |
5d4f98a2 YZ |
1664 | return ret; |
1665 | } | |
31840ae1 | 1666 | |
5d4f98a2 YZ |
1667 | /* |
1668 | * helper to update/remove inline back ref | |
1669 | */ | |
1670 | static noinline_for_stack | |
143bede5 JM |
1671 | void update_inline_extent_backref(struct btrfs_trans_handle *trans, |
1672 | struct btrfs_root *root, | |
1673 | struct btrfs_path *path, | |
1674 | struct btrfs_extent_inline_ref *iref, | |
1675 | int refs_to_mod, | |
1676 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1677 | { |
1678 | struct extent_buffer *leaf; | |
1679 | struct btrfs_extent_item *ei; | |
1680 | struct btrfs_extent_data_ref *dref = NULL; | |
1681 | struct btrfs_shared_data_ref *sref = NULL; | |
1682 | unsigned long ptr; | |
1683 | unsigned long end; | |
1684 | u32 item_size; | |
1685 | int size; | |
1686 | int type; | |
5d4f98a2 YZ |
1687 | u64 refs; |
1688 | ||
1689 | leaf = path->nodes[0]; | |
1690 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1691 | refs = btrfs_extent_refs(leaf, ei); | |
1692 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1693 | refs += refs_to_mod; | |
1694 | btrfs_set_extent_refs(leaf, ei, refs); | |
1695 | if (extent_op) | |
1696 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1697 | ||
1698 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1699 | ||
1700 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1701 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1702 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1703 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1704 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1705 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1706 | } else { | |
1707 | refs = 1; | |
1708 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1709 | } |
31840ae1 | 1710 | |
5d4f98a2 YZ |
1711 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1712 | refs += refs_to_mod; | |
1713 | ||
1714 | if (refs > 0) { | |
1715 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1716 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1717 | else | |
1718 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1719 | } else { | |
1720 | size = btrfs_extent_inline_ref_size(type); | |
1721 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1722 | ptr = (unsigned long)iref; | |
1723 | end = (unsigned long)ei + item_size; | |
1724 | if (ptr + size < end) | |
1725 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1726 | end - ptr - size); | |
1727 | item_size -= size; | |
143bede5 | 1728 | btrfs_truncate_item(trans, root, path, item_size, 1); |
5d4f98a2 YZ |
1729 | } |
1730 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1731 | } |
1732 | ||
1733 | static noinline_for_stack | |
1734 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1735 | struct btrfs_root *root, | |
1736 | struct btrfs_path *path, | |
1737 | u64 bytenr, u64 num_bytes, u64 parent, | |
1738 | u64 root_objectid, u64 owner, | |
1739 | u64 offset, int refs_to_add, | |
1740 | struct btrfs_delayed_extent_op *extent_op) | |
1741 | { | |
1742 | struct btrfs_extent_inline_ref *iref; | |
1743 | int ret; | |
1744 | ||
1745 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1746 | bytenr, num_bytes, parent, | |
1747 | root_objectid, owner, offset, 1); | |
1748 | if (ret == 0) { | |
1749 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
143bede5 JM |
1750 | update_inline_extent_backref(trans, root, path, iref, |
1751 | refs_to_add, extent_op); | |
5d4f98a2 | 1752 | } else if (ret == -ENOENT) { |
143bede5 JM |
1753 | setup_inline_extent_backref(trans, root, path, iref, parent, |
1754 | root_objectid, owner, offset, | |
1755 | refs_to_add, extent_op); | |
1756 | ret = 0; | |
771ed689 | 1757 | } |
5d4f98a2 YZ |
1758 | return ret; |
1759 | } | |
31840ae1 | 1760 | |
5d4f98a2 YZ |
1761 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1762 | struct btrfs_root *root, | |
1763 | struct btrfs_path *path, | |
1764 | u64 bytenr, u64 parent, u64 root_objectid, | |
1765 | u64 owner, u64 offset, int refs_to_add) | |
1766 | { | |
1767 | int ret; | |
1768 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1769 | BUG_ON(refs_to_add != 1); | |
1770 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1771 | parent, root_objectid); | |
1772 | } else { | |
1773 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1774 | parent, root_objectid, | |
1775 | owner, offset, refs_to_add); | |
1776 | } | |
1777 | return ret; | |
1778 | } | |
56bec294 | 1779 | |
5d4f98a2 YZ |
1780 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1781 | struct btrfs_root *root, | |
1782 | struct btrfs_path *path, | |
1783 | struct btrfs_extent_inline_ref *iref, | |
1784 | int refs_to_drop, int is_data) | |
1785 | { | |
143bede5 | 1786 | int ret = 0; |
b9473439 | 1787 | |
5d4f98a2 YZ |
1788 | BUG_ON(!is_data && refs_to_drop != 1); |
1789 | if (iref) { | |
143bede5 JM |
1790 | update_inline_extent_backref(trans, root, path, iref, |
1791 | -refs_to_drop, NULL); | |
5d4f98a2 YZ |
1792 | } else if (is_data) { |
1793 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop); | |
1794 | } else { | |
1795 | ret = btrfs_del_item(trans, root, path); | |
1796 | } | |
1797 | return ret; | |
1798 | } | |
1799 | ||
5378e607 | 1800 | static int btrfs_issue_discard(struct block_device *bdev, |
5d4f98a2 YZ |
1801 | u64 start, u64 len) |
1802 | { | |
5378e607 | 1803 | return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0); |
5d4f98a2 | 1804 | } |
5d4f98a2 YZ |
1805 | |
1806 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 1807 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1808 | { |
5d4f98a2 | 1809 | int ret; |
5378e607 | 1810 | u64 discarded_bytes = 0; |
a1d3c478 | 1811 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1812 | |
e244a0ae | 1813 | |
5d4f98a2 | 1814 | /* Tell the block device(s) that the sectors can be discarded */ |
3ec706c8 | 1815 | ret = btrfs_map_block(root->fs_info, REQ_DISCARD, |
a1d3c478 | 1816 | bytenr, &num_bytes, &bbio, 0); |
79787eaa | 1817 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 1818 | if (!ret) { |
a1d3c478 | 1819 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
1820 | int i; |
1821 | ||
5d4f98a2 | 1822 | |
a1d3c478 | 1823 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d5e2003c JB |
1824 | if (!stripe->dev->can_discard) |
1825 | continue; | |
1826 | ||
5378e607 LD |
1827 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1828 | stripe->physical, | |
1829 | stripe->length); | |
1830 | if (!ret) | |
1831 | discarded_bytes += stripe->length; | |
1832 | else if (ret != -EOPNOTSUPP) | |
79787eaa | 1833 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
1834 | |
1835 | /* | |
1836 | * Just in case we get back EOPNOTSUPP for some reason, | |
1837 | * just ignore the return value so we don't screw up | |
1838 | * people calling discard_extent. | |
1839 | */ | |
1840 | ret = 0; | |
5d4f98a2 | 1841 | } |
a1d3c478 | 1842 | kfree(bbio); |
5d4f98a2 | 1843 | } |
5378e607 LD |
1844 | |
1845 | if (actual_bytes) | |
1846 | *actual_bytes = discarded_bytes; | |
1847 | ||
5d4f98a2 YZ |
1848 | |
1849 | return ret; | |
5d4f98a2 YZ |
1850 | } |
1851 | ||
79787eaa | 1852 | /* Can return -ENOMEM */ |
5d4f98a2 YZ |
1853 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
1854 | struct btrfs_root *root, | |
1855 | u64 bytenr, u64 num_bytes, u64 parent, | |
66d7e7f0 | 1856 | u64 root_objectid, u64 owner, u64 offset, int for_cow) |
5d4f98a2 YZ |
1857 | { |
1858 | int ret; | |
66d7e7f0 AJ |
1859 | struct btrfs_fs_info *fs_info = root->fs_info; |
1860 | ||
5d4f98a2 YZ |
1861 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
1862 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
1863 | ||
1864 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
66d7e7f0 AJ |
1865 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
1866 | num_bytes, | |
5d4f98a2 | 1867 | parent, root_objectid, (int)owner, |
66d7e7f0 | 1868 | BTRFS_ADD_DELAYED_REF, NULL, for_cow); |
5d4f98a2 | 1869 | } else { |
66d7e7f0 AJ |
1870 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
1871 | num_bytes, | |
5d4f98a2 | 1872 | parent, root_objectid, owner, offset, |
66d7e7f0 | 1873 | BTRFS_ADD_DELAYED_REF, NULL, for_cow); |
5d4f98a2 YZ |
1874 | } |
1875 | return ret; | |
1876 | } | |
1877 | ||
1878 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1879 | struct btrfs_root *root, | |
1880 | u64 bytenr, u64 num_bytes, | |
1881 | u64 parent, u64 root_objectid, | |
1882 | u64 owner, u64 offset, int refs_to_add, | |
1883 | struct btrfs_delayed_extent_op *extent_op) | |
1884 | { | |
1885 | struct btrfs_path *path; | |
1886 | struct extent_buffer *leaf; | |
1887 | struct btrfs_extent_item *item; | |
1888 | u64 refs; | |
1889 | int ret; | |
1890 | int err = 0; | |
1891 | ||
1892 | path = btrfs_alloc_path(); | |
1893 | if (!path) | |
1894 | return -ENOMEM; | |
1895 | ||
1896 | path->reada = 1; | |
1897 | path->leave_spinning = 1; | |
1898 | /* this will setup the path even if it fails to insert the back ref */ | |
1899 | ret = insert_inline_extent_backref(trans, root->fs_info->extent_root, | |
1900 | path, bytenr, num_bytes, parent, | |
1901 | root_objectid, owner, offset, | |
1902 | refs_to_add, extent_op); | |
1903 | if (ret == 0) | |
1904 | goto out; | |
1905 | ||
1906 | if (ret != -EAGAIN) { | |
1907 | err = ret; | |
1908 | goto out; | |
1909 | } | |
1910 | ||
1911 | leaf = path->nodes[0]; | |
1912 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1913 | refs = btrfs_extent_refs(leaf, item); | |
1914 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
1915 | if (extent_op) | |
1916 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 1917 | |
5d4f98a2 | 1918 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 1919 | btrfs_release_path(path); |
56bec294 CM |
1920 | |
1921 | path->reada = 1; | |
b9473439 CM |
1922 | path->leave_spinning = 1; |
1923 | ||
56bec294 CM |
1924 | /* now insert the actual backref */ |
1925 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
1926 | path, bytenr, parent, root_objectid, |
1927 | owner, offset, refs_to_add); | |
79787eaa JM |
1928 | if (ret) |
1929 | btrfs_abort_transaction(trans, root, ret); | |
5d4f98a2 | 1930 | out: |
56bec294 | 1931 | btrfs_free_path(path); |
5d4f98a2 | 1932 | return err; |
56bec294 CM |
1933 | } |
1934 | ||
5d4f98a2 YZ |
1935 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
1936 | struct btrfs_root *root, | |
1937 | struct btrfs_delayed_ref_node *node, | |
1938 | struct btrfs_delayed_extent_op *extent_op, | |
1939 | int insert_reserved) | |
56bec294 | 1940 | { |
5d4f98a2 YZ |
1941 | int ret = 0; |
1942 | struct btrfs_delayed_data_ref *ref; | |
1943 | struct btrfs_key ins; | |
1944 | u64 parent = 0; | |
1945 | u64 ref_root = 0; | |
1946 | u64 flags = 0; | |
1947 | ||
1948 | ins.objectid = node->bytenr; | |
1949 | ins.offset = node->num_bytes; | |
1950 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
1951 | ||
1952 | ref = btrfs_delayed_node_to_data_ref(node); | |
1953 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) | |
1954 | parent = ref->parent; | |
1955 | else | |
1956 | ref_root = ref->root; | |
1957 | ||
1958 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
1959 | if (extent_op) { | |
1960 | BUG_ON(extent_op->update_key); | |
1961 | flags |= extent_op->flags_to_set; | |
1962 | } | |
1963 | ret = alloc_reserved_file_extent(trans, root, | |
1964 | parent, ref_root, flags, | |
1965 | ref->objectid, ref->offset, | |
1966 | &ins, node->ref_mod); | |
5d4f98a2 YZ |
1967 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
1968 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
1969 | node->num_bytes, parent, | |
1970 | ref_root, ref->objectid, | |
1971 | ref->offset, node->ref_mod, | |
1972 | extent_op); | |
1973 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
1974 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
1975 | node->num_bytes, parent, | |
1976 | ref_root, ref->objectid, | |
1977 | ref->offset, node->ref_mod, | |
1978 | extent_op); | |
1979 | } else { | |
1980 | BUG(); | |
1981 | } | |
1982 | return ret; | |
1983 | } | |
1984 | ||
1985 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
1986 | struct extent_buffer *leaf, | |
1987 | struct btrfs_extent_item *ei) | |
1988 | { | |
1989 | u64 flags = btrfs_extent_flags(leaf, ei); | |
1990 | if (extent_op->update_flags) { | |
1991 | flags |= extent_op->flags_to_set; | |
1992 | btrfs_set_extent_flags(leaf, ei, flags); | |
1993 | } | |
1994 | ||
1995 | if (extent_op->update_key) { | |
1996 | struct btrfs_tree_block_info *bi; | |
1997 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
1998 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
1999 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2000 | } | |
2001 | } | |
2002 | ||
2003 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2004 | struct btrfs_root *root, | |
2005 | struct btrfs_delayed_ref_node *node, | |
2006 | struct btrfs_delayed_extent_op *extent_op) | |
2007 | { | |
2008 | struct btrfs_key key; | |
2009 | struct btrfs_path *path; | |
2010 | struct btrfs_extent_item *ei; | |
2011 | struct extent_buffer *leaf; | |
2012 | u32 item_size; | |
56bec294 | 2013 | int ret; |
5d4f98a2 YZ |
2014 | int err = 0; |
2015 | ||
79787eaa JM |
2016 | if (trans->aborted) |
2017 | return 0; | |
2018 | ||
5d4f98a2 YZ |
2019 | path = btrfs_alloc_path(); |
2020 | if (!path) | |
2021 | return -ENOMEM; | |
2022 | ||
2023 | key.objectid = node->bytenr; | |
2024 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2025 | key.offset = node->num_bytes; | |
2026 | ||
2027 | path->reada = 1; | |
2028 | path->leave_spinning = 1; | |
2029 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
2030 | path, 0, 1); | |
2031 | if (ret < 0) { | |
2032 | err = ret; | |
2033 | goto out; | |
2034 | } | |
2035 | if (ret > 0) { | |
2036 | err = -EIO; | |
2037 | goto out; | |
2038 | } | |
2039 | ||
2040 | leaf = path->nodes[0]; | |
2041 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2042 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2043 | if (item_size < sizeof(*ei)) { | |
2044 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
2045 | path, (u64)-1, 0); | |
2046 | if (ret < 0) { | |
2047 | err = ret; | |
2048 | goto out; | |
2049 | } | |
2050 | leaf = path->nodes[0]; | |
2051 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2052 | } | |
2053 | #endif | |
2054 | BUG_ON(item_size < sizeof(*ei)); | |
2055 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2056 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2057 | |
5d4f98a2 YZ |
2058 | btrfs_mark_buffer_dirty(leaf); |
2059 | out: | |
2060 | btrfs_free_path(path); | |
2061 | return err; | |
56bec294 CM |
2062 | } |
2063 | ||
5d4f98a2 YZ |
2064 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2065 | struct btrfs_root *root, | |
2066 | struct btrfs_delayed_ref_node *node, | |
2067 | struct btrfs_delayed_extent_op *extent_op, | |
2068 | int insert_reserved) | |
56bec294 CM |
2069 | { |
2070 | int ret = 0; | |
5d4f98a2 YZ |
2071 | struct btrfs_delayed_tree_ref *ref; |
2072 | struct btrfs_key ins; | |
2073 | u64 parent = 0; | |
2074 | u64 ref_root = 0; | |
56bec294 | 2075 | |
5d4f98a2 YZ |
2076 | ins.objectid = node->bytenr; |
2077 | ins.offset = node->num_bytes; | |
2078 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
56bec294 | 2079 | |
5d4f98a2 YZ |
2080 | ref = btrfs_delayed_node_to_tree_ref(node); |
2081 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2082 | parent = ref->parent; | |
2083 | else | |
2084 | ref_root = ref->root; | |
2085 | ||
2086 | BUG_ON(node->ref_mod != 1); | |
2087 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
2088 | BUG_ON(!extent_op || !extent_op->update_flags || | |
2089 | !extent_op->update_key); | |
2090 | ret = alloc_reserved_tree_block(trans, root, | |
2091 | parent, ref_root, | |
2092 | extent_op->flags_to_set, | |
2093 | &extent_op->key, | |
2094 | ref->level, &ins); | |
5d4f98a2 YZ |
2095 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2096 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
2097 | node->num_bytes, parent, ref_root, | |
2098 | ref->level, 0, 1, extent_op); | |
2099 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
2100 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
2101 | node->num_bytes, parent, ref_root, | |
2102 | ref->level, 0, 1, extent_op); | |
2103 | } else { | |
2104 | BUG(); | |
2105 | } | |
56bec294 CM |
2106 | return ret; |
2107 | } | |
2108 | ||
2109 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 YZ |
2110 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2111 | struct btrfs_root *root, | |
2112 | struct btrfs_delayed_ref_node *node, | |
2113 | struct btrfs_delayed_extent_op *extent_op, | |
2114 | int insert_reserved) | |
56bec294 | 2115 | { |
79787eaa JM |
2116 | int ret = 0; |
2117 | ||
2118 | if (trans->aborted) | |
2119 | return 0; | |
2120 | ||
5d4f98a2 | 2121 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2122 | struct btrfs_delayed_ref_head *head; |
2123 | /* | |
2124 | * we've hit the end of the chain and we were supposed | |
2125 | * to insert this extent into the tree. But, it got | |
2126 | * deleted before we ever needed to insert it, so all | |
2127 | * we have to do is clean up the accounting | |
2128 | */ | |
5d4f98a2 YZ |
2129 | BUG_ON(extent_op); |
2130 | head = btrfs_delayed_node_to_head(node); | |
56bec294 | 2131 | if (insert_reserved) { |
f0486c68 YZ |
2132 | btrfs_pin_extent(root, node->bytenr, |
2133 | node->num_bytes, 1); | |
5d4f98a2 YZ |
2134 | if (head->is_data) { |
2135 | ret = btrfs_del_csums(trans, root, | |
2136 | node->bytenr, | |
2137 | node->num_bytes); | |
5d4f98a2 | 2138 | } |
56bec294 | 2139 | } |
79787eaa | 2140 | return ret; |
56bec294 CM |
2141 | } |
2142 | ||
5d4f98a2 YZ |
2143 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2144 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2145 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2146 | insert_reserved); | |
2147 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2148 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2149 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2150 | insert_reserved); | |
2151 | else | |
2152 | BUG(); | |
2153 | return ret; | |
56bec294 CM |
2154 | } |
2155 | ||
2156 | static noinline struct btrfs_delayed_ref_node * | |
2157 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
2158 | { | |
2159 | struct rb_node *node; | |
2160 | struct btrfs_delayed_ref_node *ref; | |
2161 | int action = BTRFS_ADD_DELAYED_REF; | |
2162 | again: | |
2163 | /* | |
2164 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2165 | * this prevents ref count from going down to zero when | |
2166 | * there still are pending delayed ref. | |
2167 | */ | |
2168 | node = rb_prev(&head->node.rb_node); | |
2169 | while (1) { | |
2170 | if (!node) | |
2171 | break; | |
2172 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2173 | rb_node); | |
2174 | if (ref->bytenr != head->node.bytenr) | |
2175 | break; | |
5d4f98a2 | 2176 | if (ref->action == action) |
56bec294 CM |
2177 | return ref; |
2178 | node = rb_prev(node); | |
2179 | } | |
2180 | if (action == BTRFS_ADD_DELAYED_REF) { | |
2181 | action = BTRFS_DROP_DELAYED_REF; | |
2182 | goto again; | |
2183 | } | |
2184 | return NULL; | |
2185 | } | |
2186 | ||
79787eaa JM |
2187 | /* |
2188 | * Returns 0 on success or if called with an already aborted transaction. | |
2189 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2190 | */ | |
c3e69d58 CM |
2191 | static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, |
2192 | struct btrfs_root *root, | |
2193 | struct list_head *cluster) | |
56bec294 | 2194 | { |
56bec294 CM |
2195 | struct btrfs_delayed_ref_root *delayed_refs; |
2196 | struct btrfs_delayed_ref_node *ref; | |
2197 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2198 | struct btrfs_delayed_extent_op *extent_op; |
097b8a7c | 2199 | struct btrfs_fs_info *fs_info = root->fs_info; |
56bec294 | 2200 | int ret; |
c3e69d58 | 2201 | int count = 0; |
56bec294 | 2202 | int must_insert_reserved = 0; |
56bec294 CM |
2203 | |
2204 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2205 | while (1) { |
2206 | if (!locked_ref) { | |
c3e69d58 CM |
2207 | /* pick a new head ref from the cluster list */ |
2208 | if (list_empty(cluster)) | |
56bec294 | 2209 | break; |
56bec294 | 2210 | |
c3e69d58 CM |
2211 | locked_ref = list_entry(cluster->next, |
2212 | struct btrfs_delayed_ref_head, cluster); | |
2213 | ||
2214 | /* grab the lock that says we are going to process | |
2215 | * all the refs for this head */ | |
2216 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
2217 | ||
2218 | /* | |
2219 | * we may have dropped the spin lock to get the head | |
2220 | * mutex lock, and that might have given someone else | |
2221 | * time to free the head. If that's true, it has been | |
2222 | * removed from our list and we can move on. | |
2223 | */ | |
2224 | if (ret == -EAGAIN) { | |
2225 | locked_ref = NULL; | |
2226 | count++; | |
2227 | continue; | |
56bec294 CM |
2228 | } |
2229 | } | |
a28ec197 | 2230 | |
ae1e206b JB |
2231 | /* |
2232 | * We need to try and merge add/drops of the same ref since we | |
2233 | * can run into issues with relocate dropping the implicit ref | |
2234 | * and then it being added back again before the drop can | |
2235 | * finish. If we merged anything we need to re-loop so we can | |
2236 | * get a good ref. | |
2237 | */ | |
2238 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, | |
2239 | locked_ref); | |
2240 | ||
d1270cd9 AJ |
2241 | /* |
2242 | * locked_ref is the head node, so we have to go one | |
2243 | * node back for any delayed ref updates | |
2244 | */ | |
2245 | ref = select_delayed_ref(locked_ref); | |
2246 | ||
2247 | if (ref && ref->seq && | |
097b8a7c | 2248 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d1270cd9 AJ |
2249 | /* |
2250 | * there are still refs with lower seq numbers in the | |
2251 | * process of being added. Don't run this ref yet. | |
2252 | */ | |
2253 | list_del_init(&locked_ref->cluster); | |
093486c4 | 2254 | btrfs_delayed_ref_unlock(locked_ref); |
d1270cd9 AJ |
2255 | locked_ref = NULL; |
2256 | delayed_refs->num_heads_ready++; | |
2257 | spin_unlock(&delayed_refs->lock); | |
2258 | cond_resched(); | |
2259 | spin_lock(&delayed_refs->lock); | |
2260 | continue; | |
2261 | } | |
2262 | ||
56bec294 CM |
2263 | /* |
2264 | * record the must insert reserved flag before we | |
2265 | * drop the spin lock. | |
2266 | */ | |
2267 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2268 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2269 | |
5d4f98a2 YZ |
2270 | extent_op = locked_ref->extent_op; |
2271 | locked_ref->extent_op = NULL; | |
2272 | ||
56bec294 CM |
2273 | if (!ref) { |
2274 | /* All delayed refs have been processed, Go ahead | |
2275 | * and send the head node to run_one_delayed_ref, | |
2276 | * so that any accounting fixes can happen | |
2277 | */ | |
2278 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2279 | |
2280 | if (extent_op && must_insert_reserved) { | |
78a6184a | 2281 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2282 | extent_op = NULL; |
2283 | } | |
2284 | ||
2285 | if (extent_op) { | |
2286 | spin_unlock(&delayed_refs->lock); | |
2287 | ||
2288 | ret = run_delayed_extent_op(trans, root, | |
2289 | ref, extent_op); | |
78a6184a | 2290 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 | 2291 | |
79787eaa | 2292 | if (ret) { |
093486c4 MX |
2293 | printk(KERN_DEBUG |
2294 | "btrfs: run_delayed_extent_op " | |
2295 | "returned %d\n", ret); | |
253beebd | 2296 | spin_lock(&delayed_refs->lock); |
093486c4 | 2297 | btrfs_delayed_ref_unlock(locked_ref); |
79787eaa JM |
2298 | return ret; |
2299 | } | |
2300 | ||
203bf287 | 2301 | goto next; |
5d4f98a2 | 2302 | } |
56bec294 | 2303 | } |
02217ed2 | 2304 | |
56bec294 CM |
2305 | ref->in_tree = 0; |
2306 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
2307 | delayed_refs->num_entries--; | |
093486c4 | 2308 | if (!btrfs_delayed_ref_is_head(ref)) { |
22cd2e7d AJ |
2309 | /* |
2310 | * when we play the delayed ref, also correct the | |
2311 | * ref_mod on head | |
2312 | */ | |
2313 | switch (ref->action) { | |
2314 | case BTRFS_ADD_DELAYED_REF: | |
2315 | case BTRFS_ADD_DELAYED_EXTENT: | |
2316 | locked_ref->node.ref_mod -= ref->ref_mod; | |
2317 | break; | |
2318 | case BTRFS_DROP_DELAYED_REF: | |
2319 | locked_ref->node.ref_mod += ref->ref_mod; | |
2320 | break; | |
2321 | default: | |
2322 | WARN_ON(1); | |
2323 | } | |
2324 | } | |
56bec294 | 2325 | spin_unlock(&delayed_refs->lock); |
925baedd | 2326 | |
5d4f98a2 | 2327 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 | 2328 | must_insert_reserved); |
eb099670 | 2329 | |
78a6184a | 2330 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2331 | if (ret) { |
093486c4 MX |
2332 | btrfs_delayed_ref_unlock(locked_ref); |
2333 | btrfs_put_delayed_ref(ref); | |
2334 | printk(KERN_DEBUG | |
2335 | "btrfs: run_one_delayed_ref returned %d\n", ret); | |
253beebd | 2336 | spin_lock(&delayed_refs->lock); |
79787eaa JM |
2337 | return ret; |
2338 | } | |
2339 | ||
093486c4 MX |
2340 | /* |
2341 | * If this node is a head, that means all the refs in this head | |
2342 | * have been dealt with, and we will pick the next head to deal | |
2343 | * with, so we must unlock the head and drop it from the cluster | |
2344 | * list before we release it. | |
2345 | */ | |
2346 | if (btrfs_delayed_ref_is_head(ref)) { | |
2347 | list_del_init(&locked_ref->cluster); | |
2348 | btrfs_delayed_ref_unlock(locked_ref); | |
2349 | locked_ref = NULL; | |
2350 | } | |
2351 | btrfs_put_delayed_ref(ref); | |
2352 | count++; | |
203bf287 | 2353 | next: |
c3e69d58 CM |
2354 | cond_resched(); |
2355 | spin_lock(&delayed_refs->lock); | |
2356 | } | |
2357 | return count; | |
2358 | } | |
2359 | ||
709c0486 AJ |
2360 | #ifdef SCRAMBLE_DELAYED_REFS |
2361 | /* | |
2362 | * Normally delayed refs get processed in ascending bytenr order. This | |
2363 | * correlates in most cases to the order added. To expose dependencies on this | |
2364 | * order, we start to process the tree in the middle instead of the beginning | |
2365 | */ | |
2366 | static u64 find_middle(struct rb_root *root) | |
2367 | { | |
2368 | struct rb_node *n = root->rb_node; | |
2369 | struct btrfs_delayed_ref_node *entry; | |
2370 | int alt = 1; | |
2371 | u64 middle; | |
2372 | u64 first = 0, last = 0; | |
2373 | ||
2374 | n = rb_first(root); | |
2375 | if (n) { | |
2376 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2377 | first = entry->bytenr; | |
2378 | } | |
2379 | n = rb_last(root); | |
2380 | if (n) { | |
2381 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2382 | last = entry->bytenr; | |
2383 | } | |
2384 | n = root->rb_node; | |
2385 | ||
2386 | while (n) { | |
2387 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2388 | WARN_ON(!entry->in_tree); | |
2389 | ||
2390 | middle = entry->bytenr; | |
2391 | ||
2392 | if (alt) | |
2393 | n = n->rb_left; | |
2394 | else | |
2395 | n = n->rb_right; | |
2396 | ||
2397 | alt = 1 - alt; | |
2398 | } | |
2399 | return middle; | |
2400 | } | |
2401 | #endif | |
2402 | ||
bed92eae AJ |
2403 | int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans, |
2404 | struct btrfs_fs_info *fs_info) | |
2405 | { | |
2406 | struct qgroup_update *qgroup_update; | |
2407 | int ret = 0; | |
2408 | ||
2409 | if (list_empty(&trans->qgroup_ref_list) != | |
2410 | !trans->delayed_ref_elem.seq) { | |
2411 | /* list without seq or seq without list */ | |
2412 | printk(KERN_ERR "btrfs: qgroup accounting update error, list is%s empty, seq is %llu\n", | |
2413 | list_empty(&trans->qgroup_ref_list) ? "" : " not", | |
2414 | trans->delayed_ref_elem.seq); | |
2415 | BUG(); | |
2416 | } | |
2417 | ||
2418 | if (!trans->delayed_ref_elem.seq) | |
2419 | return 0; | |
2420 | ||
2421 | while (!list_empty(&trans->qgroup_ref_list)) { | |
2422 | qgroup_update = list_first_entry(&trans->qgroup_ref_list, | |
2423 | struct qgroup_update, list); | |
2424 | list_del(&qgroup_update->list); | |
2425 | if (!ret) | |
2426 | ret = btrfs_qgroup_account_ref( | |
2427 | trans, fs_info, qgroup_update->node, | |
2428 | qgroup_update->extent_op); | |
2429 | kfree(qgroup_update); | |
2430 | } | |
2431 | ||
2432 | btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem); | |
2433 | ||
2434 | return ret; | |
2435 | } | |
2436 | ||
c3e69d58 CM |
2437 | /* |
2438 | * this starts processing the delayed reference count updates and | |
2439 | * extent insertions we have queued up so far. count can be | |
2440 | * 0, which means to process everything in the tree at the start | |
2441 | * of the run (but not newly added entries), or it can be some target | |
2442 | * number you'd like to process. | |
79787eaa JM |
2443 | * |
2444 | * Returns 0 on success or if called with an aborted transaction | |
2445 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2446 | */ |
2447 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2448 | struct btrfs_root *root, unsigned long count) | |
2449 | { | |
2450 | struct rb_node *node; | |
2451 | struct btrfs_delayed_ref_root *delayed_refs; | |
2452 | struct btrfs_delayed_ref_node *ref; | |
2453 | struct list_head cluster; | |
2454 | int ret; | |
a168650c | 2455 | u64 delayed_start; |
c3e69d58 CM |
2456 | int run_all = count == (unsigned long)-1; |
2457 | int run_most = 0; | |
1fa11e26 | 2458 | int loops; |
c3e69d58 | 2459 | |
79787eaa JM |
2460 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2461 | if (trans->aborted) | |
2462 | return 0; | |
2463 | ||
c3e69d58 CM |
2464 | if (root == root->fs_info->extent_root) |
2465 | root = root->fs_info->tree_root; | |
2466 | ||
edf39272 JS |
2467 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); |
2468 | ||
c3e69d58 CM |
2469 | delayed_refs = &trans->transaction->delayed_refs; |
2470 | INIT_LIST_HEAD(&cluster); | |
2471 | again: | |
1fa11e26 | 2472 | loops = 0; |
c3e69d58 | 2473 | spin_lock(&delayed_refs->lock); |
097b8a7c | 2474 | |
709c0486 AJ |
2475 | #ifdef SCRAMBLE_DELAYED_REFS |
2476 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2477 | #endif | |
2478 | ||
c3e69d58 CM |
2479 | if (count == 0) { |
2480 | count = delayed_refs->num_entries * 2; | |
2481 | run_most = 1; | |
2482 | } | |
2483 | while (1) { | |
2484 | if (!(run_all || run_most) && | |
2485 | delayed_refs->num_heads_ready < 64) | |
2486 | break; | |
eb099670 | 2487 | |
56bec294 | 2488 | /* |
c3e69d58 CM |
2489 | * go find something we can process in the rbtree. We start at |
2490 | * the beginning of the tree, and then build a cluster | |
2491 | * of refs to process starting at the first one we are able to | |
2492 | * lock | |
56bec294 | 2493 | */ |
a168650c | 2494 | delayed_start = delayed_refs->run_delayed_start; |
c3e69d58 CM |
2495 | ret = btrfs_find_ref_cluster(trans, &cluster, |
2496 | delayed_refs->run_delayed_start); | |
2497 | if (ret) | |
56bec294 CM |
2498 | break; |
2499 | ||
c3e69d58 | 2500 | ret = run_clustered_refs(trans, root, &cluster); |
79787eaa | 2501 | if (ret < 0) { |
093486c4 | 2502 | btrfs_release_ref_cluster(&cluster); |
79787eaa JM |
2503 | spin_unlock(&delayed_refs->lock); |
2504 | btrfs_abort_transaction(trans, root, ret); | |
2505 | return ret; | |
2506 | } | |
c3e69d58 CM |
2507 | |
2508 | count -= min_t(unsigned long, ret, count); | |
2509 | ||
2510 | if (count == 0) | |
2511 | break; | |
a168650c | 2512 | |
1fa11e26 AJ |
2513 | if (delayed_start >= delayed_refs->run_delayed_start) { |
2514 | if (loops == 0) { | |
2515 | /* | |
2516 | * btrfs_find_ref_cluster looped. let's do one | |
2517 | * more cycle. if we don't run any delayed ref | |
2518 | * during that cycle (because we can't because | |
2519 | * all of them are blocked), bail out. | |
2520 | */ | |
2521 | loops = 1; | |
2522 | } else { | |
2523 | /* | |
2524 | * no runnable refs left, stop trying | |
2525 | */ | |
2526 | BUG_ON(run_all); | |
2527 | break; | |
2528 | } | |
2529 | } | |
2530 | if (ret) { | |
a168650c | 2531 | /* refs were run, let's reset staleness detection */ |
1fa11e26 | 2532 | loops = 0; |
a168650c | 2533 | } |
eb099670 | 2534 | } |
c3e69d58 | 2535 | |
56bec294 | 2536 | if (run_all) { |
ea658bad JB |
2537 | if (!list_empty(&trans->new_bgs)) { |
2538 | spin_unlock(&delayed_refs->lock); | |
2539 | btrfs_create_pending_block_groups(trans, root); | |
2540 | spin_lock(&delayed_refs->lock); | |
2541 | } | |
2542 | ||
56bec294 | 2543 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 2544 | if (!node) |
56bec294 | 2545 | goto out; |
c3e69d58 | 2546 | count = (unsigned long)-1; |
e9d0b13b | 2547 | |
56bec294 CM |
2548 | while (node) { |
2549 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2550 | rb_node); | |
2551 | if (btrfs_delayed_ref_is_head(ref)) { | |
2552 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 2553 | |
56bec294 CM |
2554 | head = btrfs_delayed_node_to_head(ref); |
2555 | atomic_inc(&ref->refs); | |
2556 | ||
2557 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2558 | /* |
2559 | * Mutex was contended, block until it's | |
2560 | * released and try again | |
2561 | */ | |
56bec294 CM |
2562 | mutex_lock(&head->mutex); |
2563 | mutex_unlock(&head->mutex); | |
2564 | ||
2565 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2566 | cond_resched(); |
56bec294 CM |
2567 | goto again; |
2568 | } | |
2569 | node = rb_next(node); | |
2570 | } | |
2571 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
2572 | schedule_timeout(1); |
2573 | goto again; | |
5f39d397 | 2574 | } |
54aa1f4d | 2575 | out: |
c3e69d58 | 2576 | spin_unlock(&delayed_refs->lock); |
edf39272 | 2577 | assert_qgroups_uptodate(trans); |
a28ec197 CM |
2578 | return 0; |
2579 | } | |
2580 | ||
5d4f98a2 YZ |
2581 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2582 | struct btrfs_root *root, | |
2583 | u64 bytenr, u64 num_bytes, u64 flags, | |
2584 | int is_data) | |
2585 | { | |
2586 | struct btrfs_delayed_extent_op *extent_op; | |
2587 | int ret; | |
2588 | ||
78a6184a | 2589 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2590 | if (!extent_op) |
2591 | return -ENOMEM; | |
2592 | ||
2593 | extent_op->flags_to_set = flags; | |
2594 | extent_op->update_flags = 1; | |
2595 | extent_op->update_key = 0; | |
2596 | extent_op->is_data = is_data ? 1 : 0; | |
2597 | ||
66d7e7f0 AJ |
2598 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, |
2599 | num_bytes, extent_op); | |
5d4f98a2 | 2600 | if (ret) |
78a6184a | 2601 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2602 | return ret; |
2603 | } | |
2604 | ||
2605 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2606 | struct btrfs_root *root, | |
2607 | struct btrfs_path *path, | |
2608 | u64 objectid, u64 offset, u64 bytenr) | |
2609 | { | |
2610 | struct btrfs_delayed_ref_head *head; | |
2611 | struct btrfs_delayed_ref_node *ref; | |
2612 | struct btrfs_delayed_data_ref *data_ref; | |
2613 | struct btrfs_delayed_ref_root *delayed_refs; | |
2614 | struct rb_node *node; | |
2615 | int ret = 0; | |
2616 | ||
2617 | ret = -ENOENT; | |
2618 | delayed_refs = &trans->transaction->delayed_refs; | |
2619 | spin_lock(&delayed_refs->lock); | |
2620 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2621 | if (!head) | |
2622 | goto out; | |
2623 | ||
2624 | if (!mutex_trylock(&head->mutex)) { | |
2625 | atomic_inc(&head->node.refs); | |
2626 | spin_unlock(&delayed_refs->lock); | |
2627 | ||
b3b4aa74 | 2628 | btrfs_release_path(path); |
5d4f98a2 | 2629 | |
8cc33e5c DS |
2630 | /* |
2631 | * Mutex was contended, block until it's released and let | |
2632 | * caller try again | |
2633 | */ | |
5d4f98a2 YZ |
2634 | mutex_lock(&head->mutex); |
2635 | mutex_unlock(&head->mutex); | |
2636 | btrfs_put_delayed_ref(&head->node); | |
2637 | return -EAGAIN; | |
2638 | } | |
2639 | ||
2640 | node = rb_prev(&head->node.rb_node); | |
2641 | if (!node) | |
2642 | goto out_unlock; | |
2643 | ||
2644 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2645 | ||
2646 | if (ref->bytenr != bytenr) | |
2647 | goto out_unlock; | |
2648 | ||
2649 | ret = 1; | |
2650 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2651 | goto out_unlock; | |
2652 | ||
2653 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2654 | ||
2655 | node = rb_prev(node); | |
2656 | if (node) { | |
df57dbe6 LB |
2657 | int seq = ref->seq; |
2658 | ||
5d4f98a2 | 2659 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); |
df57dbe6 | 2660 | if (ref->bytenr == bytenr && ref->seq == seq) |
5d4f98a2 YZ |
2661 | goto out_unlock; |
2662 | } | |
2663 | ||
2664 | if (data_ref->root != root->root_key.objectid || | |
2665 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2666 | goto out_unlock; | |
2667 | ||
2668 | ret = 0; | |
2669 | out_unlock: | |
2670 | mutex_unlock(&head->mutex); | |
2671 | out: | |
2672 | spin_unlock(&delayed_refs->lock); | |
2673 | return ret; | |
2674 | } | |
2675 | ||
2676 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2677 | struct btrfs_root *root, | |
2678 | struct btrfs_path *path, | |
2679 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2680 | { |
2681 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2682 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2683 | struct btrfs_extent_data_ref *ref; |
2684 | struct btrfs_extent_inline_ref *iref; | |
2685 | struct btrfs_extent_item *ei; | |
f321e491 | 2686 | struct btrfs_key key; |
5d4f98a2 | 2687 | u32 item_size; |
be20aa9d | 2688 | int ret; |
925baedd | 2689 | |
be20aa9d | 2690 | key.objectid = bytenr; |
31840ae1 | 2691 | key.offset = (u64)-1; |
f321e491 | 2692 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2693 | |
be20aa9d CM |
2694 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2695 | if (ret < 0) | |
2696 | goto out; | |
79787eaa | 2697 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2698 | |
2699 | ret = -ENOENT; | |
2700 | if (path->slots[0] == 0) | |
31840ae1 | 2701 | goto out; |
be20aa9d | 2702 | |
31840ae1 | 2703 | path->slots[0]--; |
f321e491 | 2704 | leaf = path->nodes[0]; |
5d4f98a2 | 2705 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2706 | |
5d4f98a2 | 2707 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2708 | goto out; |
f321e491 | 2709 | |
5d4f98a2 YZ |
2710 | ret = 1; |
2711 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2712 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2713 | if (item_size < sizeof(*ei)) { | |
2714 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2715 | goto out; | |
2716 | } | |
2717 | #endif | |
2718 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2719 | |
5d4f98a2 YZ |
2720 | if (item_size != sizeof(*ei) + |
2721 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2722 | goto out; | |
be20aa9d | 2723 | |
5d4f98a2 YZ |
2724 | if (btrfs_extent_generation(leaf, ei) <= |
2725 | btrfs_root_last_snapshot(&root->root_item)) | |
2726 | goto out; | |
2727 | ||
2728 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2729 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2730 | BTRFS_EXTENT_DATA_REF_KEY) | |
2731 | goto out; | |
2732 | ||
2733 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2734 | if (btrfs_extent_refs(leaf, ei) != | |
2735 | btrfs_extent_data_ref_count(leaf, ref) || | |
2736 | btrfs_extent_data_ref_root(leaf, ref) != | |
2737 | root->root_key.objectid || | |
2738 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2739 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2740 | goto out; | |
2741 | ||
2742 | ret = 0; | |
2743 | out: | |
2744 | return ret; | |
2745 | } | |
2746 | ||
2747 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2748 | struct btrfs_root *root, | |
2749 | u64 objectid, u64 offset, u64 bytenr) | |
2750 | { | |
2751 | struct btrfs_path *path; | |
2752 | int ret; | |
2753 | int ret2; | |
2754 | ||
2755 | path = btrfs_alloc_path(); | |
2756 | if (!path) | |
2757 | return -ENOENT; | |
2758 | ||
2759 | do { | |
2760 | ret = check_committed_ref(trans, root, path, objectid, | |
2761 | offset, bytenr); | |
2762 | if (ret && ret != -ENOENT) | |
f321e491 | 2763 | goto out; |
80ff3856 | 2764 | |
5d4f98a2 YZ |
2765 | ret2 = check_delayed_ref(trans, root, path, objectid, |
2766 | offset, bytenr); | |
2767 | } while (ret2 == -EAGAIN); | |
2768 | ||
2769 | if (ret2 && ret2 != -ENOENT) { | |
2770 | ret = ret2; | |
2771 | goto out; | |
f321e491 | 2772 | } |
5d4f98a2 YZ |
2773 | |
2774 | if (ret != -ENOENT || ret2 != -ENOENT) | |
2775 | ret = 0; | |
be20aa9d | 2776 | out: |
80ff3856 | 2777 | btrfs_free_path(path); |
f0486c68 YZ |
2778 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2779 | WARN_ON(ret > 0); | |
f321e491 | 2780 | return ret; |
be20aa9d | 2781 | } |
c5739bba | 2782 | |
5d4f98a2 | 2783 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2784 | struct btrfs_root *root, |
5d4f98a2 | 2785 | struct extent_buffer *buf, |
66d7e7f0 | 2786 | int full_backref, int inc, int for_cow) |
31840ae1 ZY |
2787 | { |
2788 | u64 bytenr; | |
5d4f98a2 YZ |
2789 | u64 num_bytes; |
2790 | u64 parent; | |
31840ae1 | 2791 | u64 ref_root; |
31840ae1 | 2792 | u32 nritems; |
31840ae1 ZY |
2793 | struct btrfs_key key; |
2794 | struct btrfs_file_extent_item *fi; | |
2795 | int i; | |
2796 | int level; | |
2797 | int ret = 0; | |
31840ae1 | 2798 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
66d7e7f0 | 2799 | u64, u64, u64, u64, u64, u64, int); |
31840ae1 ZY |
2800 | |
2801 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
2802 | nritems = btrfs_header_nritems(buf); |
2803 | level = btrfs_header_level(buf); | |
2804 | ||
5d4f98a2 YZ |
2805 | if (!root->ref_cows && level == 0) |
2806 | return 0; | |
31840ae1 | 2807 | |
5d4f98a2 YZ |
2808 | if (inc) |
2809 | process_func = btrfs_inc_extent_ref; | |
2810 | else | |
2811 | process_func = btrfs_free_extent; | |
31840ae1 | 2812 | |
5d4f98a2 YZ |
2813 | if (full_backref) |
2814 | parent = buf->start; | |
2815 | else | |
2816 | parent = 0; | |
2817 | ||
2818 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 2819 | if (level == 0) { |
5d4f98a2 | 2820 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
2821 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
2822 | continue; | |
5d4f98a2 | 2823 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
2824 | struct btrfs_file_extent_item); |
2825 | if (btrfs_file_extent_type(buf, fi) == | |
2826 | BTRFS_FILE_EXTENT_INLINE) | |
2827 | continue; | |
2828 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
2829 | if (bytenr == 0) | |
2830 | continue; | |
5d4f98a2 YZ |
2831 | |
2832 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
2833 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
2834 | ret = process_func(trans, root, bytenr, num_bytes, | |
2835 | parent, ref_root, key.objectid, | |
66d7e7f0 | 2836 | key.offset, for_cow); |
31840ae1 ZY |
2837 | if (ret) |
2838 | goto fail; | |
2839 | } else { | |
5d4f98a2 YZ |
2840 | bytenr = btrfs_node_blockptr(buf, i); |
2841 | num_bytes = btrfs_level_size(root, level - 1); | |
2842 | ret = process_func(trans, root, bytenr, num_bytes, | |
66d7e7f0 AJ |
2843 | parent, ref_root, level - 1, 0, |
2844 | for_cow); | |
31840ae1 ZY |
2845 | if (ret) |
2846 | goto fail; | |
2847 | } | |
2848 | } | |
2849 | return 0; | |
2850 | fail: | |
5d4f98a2 YZ |
2851 | return ret; |
2852 | } | |
2853 | ||
2854 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 2855 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 2856 | { |
66d7e7f0 | 2857 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow); |
5d4f98a2 YZ |
2858 | } |
2859 | ||
2860 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 2861 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 2862 | { |
66d7e7f0 | 2863 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow); |
31840ae1 ZY |
2864 | } |
2865 | ||
9078a3e1 CM |
2866 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2867 | struct btrfs_root *root, | |
2868 | struct btrfs_path *path, | |
2869 | struct btrfs_block_group_cache *cache) | |
2870 | { | |
2871 | int ret; | |
9078a3e1 | 2872 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
2873 | unsigned long bi; |
2874 | struct extent_buffer *leaf; | |
9078a3e1 | 2875 | |
9078a3e1 | 2876 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
2877 | if (ret < 0) |
2878 | goto fail; | |
79787eaa | 2879 | BUG_ON(ret); /* Corruption */ |
5f39d397 CM |
2880 | |
2881 | leaf = path->nodes[0]; | |
2882 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
2883 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
2884 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 2885 | btrfs_release_path(path); |
54aa1f4d | 2886 | fail: |
79787eaa JM |
2887 | if (ret) { |
2888 | btrfs_abort_transaction(trans, root, ret); | |
9078a3e1 | 2889 | return ret; |
79787eaa | 2890 | } |
9078a3e1 CM |
2891 | return 0; |
2892 | ||
2893 | } | |
2894 | ||
4a8c9a62 YZ |
2895 | static struct btrfs_block_group_cache * |
2896 | next_block_group(struct btrfs_root *root, | |
2897 | struct btrfs_block_group_cache *cache) | |
2898 | { | |
2899 | struct rb_node *node; | |
2900 | spin_lock(&root->fs_info->block_group_cache_lock); | |
2901 | node = rb_next(&cache->cache_node); | |
2902 | btrfs_put_block_group(cache); | |
2903 | if (node) { | |
2904 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
2905 | cache_node); | |
11dfe35a | 2906 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
2907 | } else |
2908 | cache = NULL; | |
2909 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
2910 | return cache; | |
2911 | } | |
2912 | ||
0af3d00b JB |
2913 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
2914 | struct btrfs_trans_handle *trans, | |
2915 | struct btrfs_path *path) | |
2916 | { | |
2917 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
2918 | struct inode *inode = NULL; | |
2919 | u64 alloc_hint = 0; | |
2b20982e | 2920 | int dcs = BTRFS_DC_ERROR; |
0af3d00b JB |
2921 | int num_pages = 0; |
2922 | int retries = 0; | |
2923 | int ret = 0; | |
2924 | ||
2925 | /* | |
2926 | * If this block group is smaller than 100 megs don't bother caching the | |
2927 | * block group. | |
2928 | */ | |
2929 | if (block_group->key.offset < (100 * 1024 * 1024)) { | |
2930 | spin_lock(&block_group->lock); | |
2931 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
2932 | spin_unlock(&block_group->lock); | |
2933 | return 0; | |
2934 | } | |
2935 | ||
2936 | again: | |
2937 | inode = lookup_free_space_inode(root, block_group, path); | |
2938 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
2939 | ret = PTR_ERR(inode); | |
b3b4aa74 | 2940 | btrfs_release_path(path); |
0af3d00b JB |
2941 | goto out; |
2942 | } | |
2943 | ||
2944 | if (IS_ERR(inode)) { | |
2945 | BUG_ON(retries); | |
2946 | retries++; | |
2947 | ||
2948 | if (block_group->ro) | |
2949 | goto out_free; | |
2950 | ||
2951 | ret = create_free_space_inode(root, trans, block_group, path); | |
2952 | if (ret) | |
2953 | goto out_free; | |
2954 | goto again; | |
2955 | } | |
2956 | ||
5b0e95bf JB |
2957 | /* We've already setup this transaction, go ahead and exit */ |
2958 | if (block_group->cache_generation == trans->transid && | |
2959 | i_size_read(inode)) { | |
2960 | dcs = BTRFS_DC_SETUP; | |
2961 | goto out_put; | |
2962 | } | |
2963 | ||
0af3d00b JB |
2964 | /* |
2965 | * We want to set the generation to 0, that way if anything goes wrong | |
2966 | * from here on out we know not to trust this cache when we load up next | |
2967 | * time. | |
2968 | */ | |
2969 | BTRFS_I(inode)->generation = 0; | |
2970 | ret = btrfs_update_inode(trans, root, inode); | |
2971 | WARN_ON(ret); | |
2972 | ||
2973 | if (i_size_read(inode) > 0) { | |
2974 | ret = btrfs_truncate_free_space_cache(root, trans, path, | |
2975 | inode); | |
2976 | if (ret) | |
2977 | goto out_put; | |
2978 | } | |
2979 | ||
2980 | spin_lock(&block_group->lock); | |
cf7c1ef6 LB |
2981 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
2982 | !btrfs_test_opt(root, SPACE_CACHE)) { | |
2983 | /* | |
2984 | * don't bother trying to write stuff out _if_ | |
2985 | * a) we're not cached, | |
2986 | * b) we're with nospace_cache mount option. | |
2987 | */ | |
2b20982e | 2988 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
2989 | spin_unlock(&block_group->lock); |
2990 | goto out_put; | |
2991 | } | |
2992 | spin_unlock(&block_group->lock); | |
2993 | ||
6fc823b1 JB |
2994 | /* |
2995 | * Try to preallocate enough space based on how big the block group is. | |
2996 | * Keep in mind this has to include any pinned space which could end up | |
2997 | * taking up quite a bit since it's not folded into the other space | |
2998 | * cache. | |
2999 | */ | |
3000 | num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024); | |
0af3d00b JB |
3001 | if (!num_pages) |
3002 | num_pages = 1; | |
3003 | ||
0af3d00b JB |
3004 | num_pages *= 16; |
3005 | num_pages *= PAGE_CACHE_SIZE; | |
3006 | ||
3007 | ret = btrfs_check_data_free_space(inode, num_pages); | |
3008 | if (ret) | |
3009 | goto out_put; | |
3010 | ||
3011 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3012 | num_pages, num_pages, | |
3013 | &alloc_hint); | |
2b20982e JB |
3014 | if (!ret) |
3015 | dcs = BTRFS_DC_SETUP; | |
0af3d00b | 3016 | btrfs_free_reserved_data_space(inode, num_pages); |
c09544e0 | 3017 | |
0af3d00b JB |
3018 | out_put: |
3019 | iput(inode); | |
3020 | out_free: | |
b3b4aa74 | 3021 | btrfs_release_path(path); |
0af3d00b JB |
3022 | out: |
3023 | spin_lock(&block_group->lock); | |
e65cbb94 | 3024 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3025 | block_group->cache_generation = trans->transid; |
2b20982e | 3026 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3027 | spin_unlock(&block_group->lock); |
3028 | ||
3029 | return ret; | |
3030 | } | |
3031 | ||
96b5179d CM |
3032 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
3033 | struct btrfs_root *root) | |
9078a3e1 | 3034 | { |
4a8c9a62 | 3035 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 3036 | int err = 0; |
9078a3e1 | 3037 | struct btrfs_path *path; |
96b5179d | 3038 | u64 last = 0; |
9078a3e1 CM |
3039 | |
3040 | path = btrfs_alloc_path(); | |
3041 | if (!path) | |
3042 | return -ENOMEM; | |
3043 | ||
0af3d00b JB |
3044 | again: |
3045 | while (1) { | |
3046 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
3047 | while (cache) { | |
3048 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3049 | break; | |
3050 | cache = next_block_group(root, cache); | |
3051 | } | |
3052 | if (!cache) { | |
3053 | if (last == 0) | |
3054 | break; | |
3055 | last = 0; | |
3056 | continue; | |
3057 | } | |
3058 | err = cache_save_setup(cache, trans, path); | |
3059 | last = cache->key.objectid + cache->key.offset; | |
3060 | btrfs_put_block_group(cache); | |
3061 | } | |
3062 | ||
d397712b | 3063 | while (1) { |
4a8c9a62 YZ |
3064 | if (last == 0) { |
3065 | err = btrfs_run_delayed_refs(trans, root, | |
3066 | (unsigned long)-1); | |
79787eaa JM |
3067 | if (err) /* File system offline */ |
3068 | goto out; | |
0f9dd46c | 3069 | } |
54aa1f4d | 3070 | |
4a8c9a62 YZ |
3071 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
3072 | while (cache) { | |
0af3d00b JB |
3073 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) { |
3074 | btrfs_put_block_group(cache); | |
3075 | goto again; | |
3076 | } | |
3077 | ||
4a8c9a62 YZ |
3078 | if (cache->dirty) |
3079 | break; | |
3080 | cache = next_block_group(root, cache); | |
3081 | } | |
3082 | if (!cache) { | |
3083 | if (last == 0) | |
3084 | break; | |
3085 | last = 0; | |
3086 | continue; | |
3087 | } | |
0f9dd46c | 3088 | |
0cb59c99 JB |
3089 | if (cache->disk_cache_state == BTRFS_DC_SETUP) |
3090 | cache->disk_cache_state = BTRFS_DC_NEED_WRITE; | |
e8569813 | 3091 | cache->dirty = 0; |
4a8c9a62 | 3092 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 3093 | |
4a8c9a62 | 3094 | err = write_one_cache_group(trans, root, path, cache); |
79787eaa JM |
3095 | if (err) /* File system offline */ |
3096 | goto out; | |
3097 | ||
4a8c9a62 | 3098 | btrfs_put_block_group(cache); |
9078a3e1 | 3099 | } |
4a8c9a62 | 3100 | |
0cb59c99 JB |
3101 | while (1) { |
3102 | /* | |
3103 | * I don't think this is needed since we're just marking our | |
3104 | * preallocated extent as written, but just in case it can't | |
3105 | * hurt. | |
3106 | */ | |
3107 | if (last == 0) { | |
3108 | err = btrfs_run_delayed_refs(trans, root, | |
3109 | (unsigned long)-1); | |
79787eaa JM |
3110 | if (err) /* File system offline */ |
3111 | goto out; | |
0cb59c99 JB |
3112 | } |
3113 | ||
3114 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
3115 | while (cache) { | |
3116 | /* | |
3117 | * Really this shouldn't happen, but it could if we | |
3118 | * couldn't write the entire preallocated extent and | |
3119 | * splitting the extent resulted in a new block. | |
3120 | */ | |
3121 | if (cache->dirty) { | |
3122 | btrfs_put_block_group(cache); | |
3123 | goto again; | |
3124 | } | |
3125 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
3126 | break; | |
3127 | cache = next_block_group(root, cache); | |
3128 | } | |
3129 | if (!cache) { | |
3130 | if (last == 0) | |
3131 | break; | |
3132 | last = 0; | |
3133 | continue; | |
3134 | } | |
3135 | ||
79787eaa | 3136 | err = btrfs_write_out_cache(root, trans, cache, path); |
0cb59c99 JB |
3137 | |
3138 | /* | |
3139 | * If we didn't have an error then the cache state is still | |
3140 | * NEED_WRITE, so we can set it to WRITTEN. | |
3141 | */ | |
79787eaa | 3142 | if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE) |
0cb59c99 JB |
3143 | cache->disk_cache_state = BTRFS_DC_WRITTEN; |
3144 | last = cache->key.objectid + cache->key.offset; | |
3145 | btrfs_put_block_group(cache); | |
3146 | } | |
79787eaa | 3147 | out: |
0cb59c99 | 3148 | |
9078a3e1 | 3149 | btrfs_free_path(path); |
79787eaa | 3150 | return err; |
9078a3e1 CM |
3151 | } |
3152 | ||
d2fb3437 YZ |
3153 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
3154 | { | |
3155 | struct btrfs_block_group_cache *block_group; | |
3156 | int readonly = 0; | |
3157 | ||
3158 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
3159 | if (!block_group || block_group->ro) | |
3160 | readonly = 1; | |
3161 | if (block_group) | |
fa9c0d79 | 3162 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3163 | return readonly; |
3164 | } | |
3165 | ||
593060d7 CM |
3166 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3167 | u64 total_bytes, u64 bytes_used, | |
3168 | struct btrfs_space_info **space_info) | |
3169 | { | |
3170 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3171 | int i; |
3172 | int factor; | |
3173 | ||
3174 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3175 | BTRFS_BLOCK_GROUP_RAID10)) | |
3176 | factor = 2; | |
3177 | else | |
3178 | factor = 1; | |
593060d7 CM |
3179 | |
3180 | found = __find_space_info(info, flags); | |
3181 | if (found) { | |
25179201 | 3182 | spin_lock(&found->lock); |
593060d7 | 3183 | found->total_bytes += total_bytes; |
89a55897 | 3184 | found->disk_total += total_bytes * factor; |
593060d7 | 3185 | found->bytes_used += bytes_used; |
b742bb82 | 3186 | found->disk_used += bytes_used * factor; |
8f18cf13 | 3187 | found->full = 0; |
25179201 | 3188 | spin_unlock(&found->lock); |
593060d7 CM |
3189 | *space_info = found; |
3190 | return 0; | |
3191 | } | |
c146afad | 3192 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3193 | if (!found) |
3194 | return -ENOMEM; | |
3195 | ||
b742bb82 YZ |
3196 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
3197 | INIT_LIST_HEAD(&found->block_groups[i]); | |
80eb234a | 3198 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3199 | spin_lock_init(&found->lock); |
52ba6929 | 3200 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
593060d7 | 3201 | found->total_bytes = total_bytes; |
89a55897 | 3202 | found->disk_total = total_bytes * factor; |
593060d7 | 3203 | found->bytes_used = bytes_used; |
b742bb82 | 3204 | found->disk_used = bytes_used * factor; |
593060d7 | 3205 | found->bytes_pinned = 0; |
e8569813 | 3206 | found->bytes_reserved = 0; |
c146afad | 3207 | found->bytes_readonly = 0; |
f0486c68 | 3208 | found->bytes_may_use = 0; |
593060d7 | 3209 | found->full = 0; |
0e4f8f88 | 3210 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3211 | found->chunk_alloc = 0; |
fdb5effd JB |
3212 | found->flush = 0; |
3213 | init_waitqueue_head(&found->wait); | |
593060d7 | 3214 | *space_info = found; |
4184ea7f | 3215 | list_add_rcu(&found->list, &info->space_info); |
b4d7c3c9 LZ |
3216 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3217 | info->data_sinfo = found; | |
593060d7 CM |
3218 | return 0; |
3219 | } | |
3220 | ||
8790d502 CM |
3221 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3222 | { | |
899c81ea ID |
3223 | u64 extra_flags = chunk_to_extended(flags) & |
3224 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 ID |
3225 | |
3226 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3227 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3228 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3229 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3230 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3231 | fs_info->avail_system_alloc_bits |= extra_flags; | |
8790d502 | 3232 | } |
593060d7 | 3233 | |
fc67c450 ID |
3234 | /* |
3235 | * returns target flags in extended format or 0 if restripe for this | |
3236 | * chunk_type is not in progress | |
c6664b42 ID |
3237 | * |
3238 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
3239 | */ |
3240 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
3241 | { | |
3242 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3243 | u64 target = 0; | |
3244 | ||
fc67c450 ID |
3245 | if (!bctl) |
3246 | return 0; | |
3247 | ||
3248 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
3249 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3250 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3251 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
3252 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3253 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3254 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
3255 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3256 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3257 | } | |
3258 | ||
3259 | return target; | |
3260 | } | |
3261 | ||
a46d11a8 ID |
3262 | /* |
3263 | * @flags: available profiles in extended format (see ctree.h) | |
3264 | * | |
e4d8ec0f ID |
3265 | * Returns reduced profile in chunk format. If profile changing is in |
3266 | * progress (either running or paused) picks the target profile (if it's | |
3267 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 3268 | */ |
2b82032c | 3269 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 3270 | { |
cd02dca5 CM |
3271 | /* |
3272 | * we add in the count of missing devices because we want | |
3273 | * to make sure that any RAID levels on a degraded FS | |
3274 | * continue to be honored. | |
3275 | */ | |
3276 | u64 num_devices = root->fs_info->fs_devices->rw_devices + | |
3277 | root->fs_info->fs_devices->missing_devices; | |
fc67c450 | 3278 | u64 target; |
a061fc8d | 3279 | |
fc67c450 ID |
3280 | /* |
3281 | * see if restripe for this chunk_type is in progress, if so | |
3282 | * try to reduce to the target profile | |
3283 | */ | |
e4d8ec0f | 3284 | spin_lock(&root->fs_info->balance_lock); |
fc67c450 ID |
3285 | target = get_restripe_target(root->fs_info, flags); |
3286 | if (target) { | |
3287 | /* pick target profile only if it's already available */ | |
3288 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
e4d8ec0f | 3289 | spin_unlock(&root->fs_info->balance_lock); |
fc67c450 | 3290 | return extended_to_chunk(target); |
e4d8ec0f ID |
3291 | } |
3292 | } | |
3293 | spin_unlock(&root->fs_info->balance_lock); | |
3294 | ||
a061fc8d CM |
3295 | if (num_devices == 1) |
3296 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); | |
3297 | if (num_devices < 4) | |
3298 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
3299 | ||
ec44a35c CM |
3300 | if ((flags & BTRFS_BLOCK_GROUP_DUP) && |
3301 | (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
a061fc8d | 3302 | BTRFS_BLOCK_GROUP_RAID10))) { |
ec44a35c | 3303 | flags &= ~BTRFS_BLOCK_GROUP_DUP; |
a061fc8d | 3304 | } |
ec44a35c CM |
3305 | |
3306 | if ((flags & BTRFS_BLOCK_GROUP_RAID1) && | |
a061fc8d | 3307 | (flags & BTRFS_BLOCK_GROUP_RAID10)) { |
ec44a35c | 3308 | flags &= ~BTRFS_BLOCK_GROUP_RAID1; |
a061fc8d | 3309 | } |
ec44a35c CM |
3310 | |
3311 | if ((flags & BTRFS_BLOCK_GROUP_RAID0) && | |
3312 | ((flags & BTRFS_BLOCK_GROUP_RAID1) | | |
3313 | (flags & BTRFS_BLOCK_GROUP_RAID10) | | |
a46d11a8 | 3314 | (flags & BTRFS_BLOCK_GROUP_DUP))) { |
ec44a35c | 3315 | flags &= ~BTRFS_BLOCK_GROUP_RAID0; |
a46d11a8 ID |
3316 | } |
3317 | ||
899c81ea | 3318 | return extended_to_chunk(flags); |
ec44a35c CM |
3319 | } |
3320 | ||
b742bb82 | 3321 | static u64 get_alloc_profile(struct btrfs_root *root, u64 flags) |
6a63209f | 3322 | { |
b742bb82 | 3323 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
6fef8df1 | 3324 | flags |= root->fs_info->avail_data_alloc_bits; |
b742bb82 | 3325 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
6fef8df1 | 3326 | flags |= root->fs_info->avail_system_alloc_bits; |
b742bb82 | 3327 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
6fef8df1 ID |
3328 | flags |= root->fs_info->avail_metadata_alloc_bits; |
3329 | ||
b742bb82 | 3330 | return btrfs_reduce_alloc_profile(root, flags); |
6a63209f JB |
3331 | } |
3332 | ||
6d07bcec | 3333 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 3334 | { |
b742bb82 | 3335 | u64 flags; |
9ed74f2d | 3336 | |
b742bb82 YZ |
3337 | if (data) |
3338 | flags = BTRFS_BLOCK_GROUP_DATA; | |
3339 | else if (root == root->fs_info->chunk_root) | |
3340 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 3341 | else |
b742bb82 | 3342 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3343 | |
b742bb82 | 3344 | return get_alloc_profile(root, flags); |
6a63209f | 3345 | } |
9ed74f2d | 3346 | |
6a63209f | 3347 | /* |
6a63209f JB |
3348 | * This will check the space that the inode allocates from to make sure we have |
3349 | * enough space for bytes. | |
6a63209f | 3350 | */ |
0ca1f7ce | 3351 | int btrfs_check_data_free_space(struct inode *inode, u64 bytes) |
6a63209f | 3352 | { |
6a63209f | 3353 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 3354 | struct btrfs_root *root = BTRFS_I(inode)->root; |
b4d7c3c9 | 3355 | struct btrfs_fs_info *fs_info = root->fs_info; |
ab6e2410 | 3356 | u64 used; |
0af3d00b | 3357 | int ret = 0, committed = 0, alloc_chunk = 1; |
6a63209f | 3358 | |
6a63209f JB |
3359 | /* make sure bytes are sectorsize aligned */ |
3360 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
6a63209f | 3361 | |
82d5902d LZ |
3362 | if (root == root->fs_info->tree_root || |
3363 | BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) { | |
0af3d00b JB |
3364 | alloc_chunk = 0; |
3365 | committed = 1; | |
3366 | } | |
3367 | ||
b4d7c3c9 | 3368 | data_sinfo = fs_info->data_sinfo; |
33b4d47f CM |
3369 | if (!data_sinfo) |
3370 | goto alloc; | |
9ed74f2d | 3371 | |
6a63209f JB |
3372 | again: |
3373 | /* make sure we have enough space to handle the data first */ | |
3374 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
3375 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
3376 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
3377 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
3378 | |
3379 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 3380 | struct btrfs_trans_handle *trans; |
9ed74f2d | 3381 | |
6a63209f JB |
3382 | /* |
3383 | * if we don't have enough free bytes in this space then we need | |
3384 | * to alloc a new chunk. | |
3385 | */ | |
0af3d00b | 3386 | if (!data_sinfo->full && alloc_chunk) { |
6a63209f | 3387 | u64 alloc_target; |
9ed74f2d | 3388 | |
0e4f8f88 | 3389 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 3390 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 3391 | alloc: |
6a63209f | 3392 | alloc_target = btrfs_get_alloc_profile(root, 1); |
7a7eaa40 | 3393 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3394 | if (IS_ERR(trans)) |
3395 | return PTR_ERR(trans); | |
9ed74f2d | 3396 | |
6a63209f | 3397 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
3398 | alloc_target, |
3399 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 3400 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
3401 | if (ret < 0) { |
3402 | if (ret != -ENOSPC) | |
3403 | return ret; | |
3404 | else | |
3405 | goto commit_trans; | |
3406 | } | |
9ed74f2d | 3407 | |
b4d7c3c9 LZ |
3408 | if (!data_sinfo) |
3409 | data_sinfo = fs_info->data_sinfo; | |
3410 | ||
6a63209f JB |
3411 | goto again; |
3412 | } | |
f2bb8f5c JB |
3413 | |
3414 | /* | |
3415 | * If we have less pinned bytes than we want to allocate then | |
3416 | * don't bother committing the transaction, it won't help us. | |
3417 | */ | |
3418 | if (data_sinfo->bytes_pinned < bytes) | |
3419 | committed = 1; | |
6a63209f | 3420 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3421 | |
4e06bdd6 | 3422 | /* commit the current transaction and try again */ |
d52a5b5f | 3423 | commit_trans: |
a4abeea4 JB |
3424 | if (!committed && |
3425 | !atomic_read(&root->fs_info->open_ioctl_trans)) { | |
4e06bdd6 | 3426 | committed = 1; |
7a7eaa40 | 3427 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3428 | if (IS_ERR(trans)) |
3429 | return PTR_ERR(trans); | |
4e06bdd6 JB |
3430 | ret = btrfs_commit_transaction(trans, root); |
3431 | if (ret) | |
3432 | return ret; | |
3433 | goto again; | |
3434 | } | |
9ed74f2d | 3435 | |
6a63209f JB |
3436 | return -ENOSPC; |
3437 | } | |
3438 | data_sinfo->bytes_may_use += bytes; | |
8c2a3ca2 | 3439 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3440 | data_sinfo->flags, bytes, 1); |
6a63209f | 3441 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3442 | |
9ed74f2d | 3443 | return 0; |
9ed74f2d | 3444 | } |
6a63209f | 3445 | |
6a63209f | 3446 | /* |
fb25e914 | 3447 | * Called if we need to clear a data reservation for this inode. |
6a63209f | 3448 | */ |
0ca1f7ce | 3449 | void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes) |
e3ccfa98 | 3450 | { |
0ca1f7ce | 3451 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6a63209f | 3452 | struct btrfs_space_info *data_sinfo; |
e3ccfa98 | 3453 | |
6a63209f JB |
3454 | /* make sure bytes are sectorsize aligned */ |
3455 | bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); | |
e3ccfa98 | 3456 | |
b4d7c3c9 | 3457 | data_sinfo = root->fs_info->data_sinfo; |
6a63209f JB |
3458 | spin_lock(&data_sinfo->lock); |
3459 | data_sinfo->bytes_may_use -= bytes; | |
8c2a3ca2 | 3460 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3461 | data_sinfo->flags, bytes, 0); |
6a63209f | 3462 | spin_unlock(&data_sinfo->lock); |
e3ccfa98 JB |
3463 | } |
3464 | ||
97e728d4 | 3465 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 3466 | { |
97e728d4 JB |
3467 | struct list_head *head = &info->space_info; |
3468 | struct btrfs_space_info *found; | |
e3ccfa98 | 3469 | |
97e728d4 JB |
3470 | rcu_read_lock(); |
3471 | list_for_each_entry_rcu(found, head, list) { | |
3472 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 3473 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 3474 | } |
97e728d4 | 3475 | rcu_read_unlock(); |
e3ccfa98 JB |
3476 | } |
3477 | ||
e5bc2458 | 3478 | static int should_alloc_chunk(struct btrfs_root *root, |
698d0082 | 3479 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 3480 | { |
fb25e914 | 3481 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
424499db | 3482 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 3483 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 3484 | u64 thresh; |
e3ccfa98 | 3485 | |
0e4f8f88 CM |
3486 | if (force == CHUNK_ALLOC_FORCE) |
3487 | return 1; | |
3488 | ||
fb25e914 JB |
3489 | /* |
3490 | * We need to take into account the global rsv because for all intents | |
3491 | * and purposes it's used space. Don't worry about locking the | |
3492 | * global_rsv, it doesn't change except when the transaction commits. | |
3493 | */ | |
54338b5c JB |
3494 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
3495 | num_allocated += global_rsv->size; | |
fb25e914 | 3496 | |
0e4f8f88 CM |
3497 | /* |
3498 | * in limited mode, we want to have some free space up to | |
3499 | * about 1% of the FS size. | |
3500 | */ | |
3501 | if (force == CHUNK_ALLOC_LIMITED) { | |
6c41761f | 3502 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 CM |
3503 | thresh = max_t(u64, 64 * 1024 * 1024, |
3504 | div_factor_fine(thresh, 1)); | |
3505 | ||
3506 | if (num_bytes - num_allocated < thresh) | |
3507 | return 1; | |
3508 | } | |
0e4f8f88 | 3509 | |
698d0082 | 3510 | if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8)) |
14ed0ca6 | 3511 | return 0; |
424499db | 3512 | return 1; |
32c00aff JB |
3513 | } |
3514 | ||
15d1ff81 LB |
3515 | static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type) |
3516 | { | |
3517 | u64 num_dev; | |
3518 | ||
3519 | if (type & BTRFS_BLOCK_GROUP_RAID10 || | |
3520 | type & BTRFS_BLOCK_GROUP_RAID0) | |
3521 | num_dev = root->fs_info->fs_devices->rw_devices; | |
3522 | else if (type & BTRFS_BLOCK_GROUP_RAID1) | |
3523 | num_dev = 2; | |
3524 | else | |
3525 | num_dev = 1; /* DUP or single */ | |
3526 | ||
3527 | /* metadata for updaing devices and chunk tree */ | |
3528 | return btrfs_calc_trans_metadata_size(root, num_dev + 1); | |
3529 | } | |
3530 | ||
3531 | static void check_system_chunk(struct btrfs_trans_handle *trans, | |
3532 | struct btrfs_root *root, u64 type) | |
3533 | { | |
3534 | struct btrfs_space_info *info; | |
3535 | u64 left; | |
3536 | u64 thresh; | |
3537 | ||
3538 | info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
3539 | spin_lock(&info->lock); | |
3540 | left = info->total_bytes - info->bytes_used - info->bytes_pinned - | |
3541 | info->bytes_reserved - info->bytes_readonly; | |
3542 | spin_unlock(&info->lock); | |
3543 | ||
3544 | thresh = get_system_chunk_thresh(root, type); | |
3545 | if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
3546 | printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n", | |
3547 | left, thresh, type); | |
3548 | dump_space_info(info, 0, 0); | |
3549 | } | |
3550 | ||
3551 | if (left < thresh) { | |
3552 | u64 flags; | |
3553 | ||
3554 | flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0); | |
3555 | btrfs_alloc_chunk(trans, root, flags); | |
3556 | } | |
3557 | } | |
3558 | ||
6324fbf3 | 3559 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 | 3560 | struct btrfs_root *extent_root, u64 flags, int force) |
9ed74f2d | 3561 | { |
6324fbf3 | 3562 | struct btrfs_space_info *space_info; |
97e728d4 | 3563 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 3564 | int wait_for_alloc = 0; |
9ed74f2d | 3565 | int ret = 0; |
9ed74f2d | 3566 | |
c6b305a8 JB |
3567 | /* Don't re-enter if we're already allocating a chunk */ |
3568 | if (trans->allocating_chunk) | |
3569 | return -ENOSPC; | |
3570 | ||
6324fbf3 | 3571 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
3572 | if (!space_info) { |
3573 | ret = update_space_info(extent_root->fs_info, flags, | |
3574 | 0, 0, &space_info); | |
79787eaa | 3575 | BUG_ON(ret); /* -ENOMEM */ |
9ed74f2d | 3576 | } |
79787eaa | 3577 | BUG_ON(!space_info); /* Logic error */ |
9ed74f2d | 3578 | |
6d74119f | 3579 | again: |
25179201 | 3580 | spin_lock(&space_info->lock); |
9e622d6b | 3581 | if (force < space_info->force_alloc) |
0e4f8f88 | 3582 | force = space_info->force_alloc; |
25179201 JB |
3583 | if (space_info->full) { |
3584 | spin_unlock(&space_info->lock); | |
6d74119f | 3585 | return 0; |
9ed74f2d JB |
3586 | } |
3587 | ||
698d0082 | 3588 | if (!should_alloc_chunk(extent_root, space_info, force)) { |
25179201 | 3589 | spin_unlock(&space_info->lock); |
6d74119f JB |
3590 | return 0; |
3591 | } else if (space_info->chunk_alloc) { | |
3592 | wait_for_alloc = 1; | |
3593 | } else { | |
3594 | space_info->chunk_alloc = 1; | |
9ed74f2d | 3595 | } |
0e4f8f88 | 3596 | |
25179201 | 3597 | spin_unlock(&space_info->lock); |
9ed74f2d | 3598 | |
6d74119f JB |
3599 | mutex_lock(&fs_info->chunk_mutex); |
3600 | ||
3601 | /* | |
3602 | * The chunk_mutex is held throughout the entirety of a chunk | |
3603 | * allocation, so once we've acquired the chunk_mutex we know that the | |
3604 | * other guy is done and we need to recheck and see if we should | |
3605 | * allocate. | |
3606 | */ | |
3607 | if (wait_for_alloc) { | |
3608 | mutex_unlock(&fs_info->chunk_mutex); | |
3609 | wait_for_alloc = 0; | |
3610 | goto again; | |
3611 | } | |
3612 | ||
c6b305a8 JB |
3613 | trans->allocating_chunk = true; |
3614 | ||
67377734 JB |
3615 | /* |
3616 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3617 | * allocating mixed chunks instead of individual chunks. | |
3618 | */ | |
3619 | if (btrfs_mixed_space_info(space_info)) | |
3620 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3621 | ||
97e728d4 JB |
3622 | /* |
3623 | * if we're doing a data chunk, go ahead and make sure that | |
3624 | * we keep a reasonable number of metadata chunks allocated in the | |
3625 | * FS as well. | |
3626 | */ | |
9ed74f2d | 3627 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
3628 | fs_info->data_chunk_allocations++; |
3629 | if (!(fs_info->data_chunk_allocations % | |
3630 | fs_info->metadata_ratio)) | |
3631 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
3632 | } |
3633 | ||
15d1ff81 LB |
3634 | /* |
3635 | * Check if we have enough space in SYSTEM chunk because we may need | |
3636 | * to update devices. | |
3637 | */ | |
3638 | check_system_chunk(trans, extent_root, flags); | |
3639 | ||
2b82032c | 3640 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
c6b305a8 | 3641 | trans->allocating_chunk = false; |
92b8e897 MF |
3642 | if (ret < 0 && ret != -ENOSPC) |
3643 | goto out; | |
3644 | ||
9ed74f2d | 3645 | spin_lock(&space_info->lock); |
9ed74f2d | 3646 | if (ret) |
6324fbf3 | 3647 | space_info->full = 1; |
424499db YZ |
3648 | else |
3649 | ret = 1; | |
6d74119f | 3650 | |
0e4f8f88 | 3651 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3652 | space_info->chunk_alloc = 0; |
9ed74f2d | 3653 | spin_unlock(&space_info->lock); |
92b8e897 | 3654 | out: |
a25c75d5 | 3655 | mutex_unlock(&fs_info->chunk_mutex); |
0f9dd46c | 3656 | return ret; |
6324fbf3 | 3657 | } |
9ed74f2d | 3658 | |
a80c8dcf JB |
3659 | static int can_overcommit(struct btrfs_root *root, |
3660 | struct btrfs_space_info *space_info, u64 bytes, | |
08e007d2 | 3661 | enum btrfs_reserve_flush_enum flush) |
a80c8dcf JB |
3662 | { |
3663 | u64 profile = btrfs_get_alloc_profile(root, 0); | |
3664 | u64 avail; | |
3665 | u64 used; | |
3666 | ||
3667 | used = space_info->bytes_used + space_info->bytes_reserved + | |
3668 | space_info->bytes_pinned + space_info->bytes_readonly + | |
3669 | space_info->bytes_may_use; | |
3670 | ||
3671 | spin_lock(&root->fs_info->free_chunk_lock); | |
3672 | avail = root->fs_info->free_chunk_space; | |
3673 | spin_unlock(&root->fs_info->free_chunk_lock); | |
3674 | ||
3675 | /* | |
3676 | * If we have dup, raid1 or raid10 then only half of the free | |
3677 | * space is actually useable. | |
3678 | */ | |
3679 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
3680 | BTRFS_BLOCK_GROUP_RAID1 | | |
3681 | BTRFS_BLOCK_GROUP_RAID10)) | |
3682 | avail >>= 1; | |
3683 | ||
3684 | /* | |
561c294d MX |
3685 | * If we aren't flushing all things, let us overcommit up to |
3686 | * 1/2th of the space. If we can flush, don't let us overcommit | |
3687 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 3688 | */ |
08e007d2 | 3689 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
a80c8dcf JB |
3690 | avail >>= 3; |
3691 | else | |
3692 | avail >>= 1; | |
3693 | ||
3694 | if (used + bytes < space_info->total_bytes + avail) | |
3695 | return 1; | |
3696 | return 0; | |
3697 | } | |
3698 | ||
da633a42 MX |
3699 | static inline int writeback_inodes_sb_nr_if_idle_safe(struct super_block *sb, |
3700 | unsigned long nr_pages, | |
3701 | enum wb_reason reason) | |
1135d6df | 3702 | { |
da633a42 MX |
3703 | /* the flusher is dealing with the dirty inodes now. */ |
3704 | if (writeback_in_progress(sb->s_bdi)) | |
3705 | return 1; | |
3706 | ||
3707 | if (down_read_trylock(&sb->s_umount)) { | |
1135d6df JB |
3708 | writeback_inodes_sb_nr(sb, nr_pages, reason); |
3709 | up_read(&sb->s_umount); | |
3710 | return 1; | |
3711 | } | |
3712 | ||
3713 | return 0; | |
3714 | } | |
3715 | ||
da633a42 MX |
3716 | void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root, |
3717 | unsigned long nr_pages) | |
3718 | { | |
3719 | struct super_block *sb = root->fs_info->sb; | |
3720 | int started; | |
3721 | ||
3722 | /* If we can not start writeback, just sync all the delalloc file. */ | |
3723 | started = writeback_inodes_sb_nr_if_idle_safe(sb, nr_pages, | |
3724 | WB_REASON_FS_FREE_SPACE); | |
3725 | if (!started) { | |
3726 | /* | |
3727 | * We needn't worry the filesystem going from r/w to r/o though | |
3728 | * we don't acquire ->s_umount mutex, because the filesystem | |
3729 | * should guarantee the delalloc inodes list be empty after | |
3730 | * the filesystem is readonly(all dirty pages are written to | |
3731 | * the disk). | |
3732 | */ | |
3733 | btrfs_start_delalloc_inodes(root, 0); | |
3734 | btrfs_wait_ordered_extents(root, 0); | |
3735 | } | |
3736 | } | |
3737 | ||
9ed74f2d | 3738 | /* |
5da9d01b | 3739 | * shrink metadata reservation for delalloc |
9ed74f2d | 3740 | */ |
f4c738c2 JB |
3741 | static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig, |
3742 | bool wait_ordered) | |
5da9d01b | 3743 | { |
0ca1f7ce | 3744 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 3745 | struct btrfs_space_info *space_info; |
663350ac | 3746 | struct btrfs_trans_handle *trans; |
f4c738c2 | 3747 | u64 delalloc_bytes; |
5da9d01b | 3748 | u64 max_reclaim; |
b1953bce | 3749 | long time_left; |
877da174 | 3750 | unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT; |
b1953bce | 3751 | int loops = 0; |
08e007d2 | 3752 | enum btrfs_reserve_flush_enum flush; |
5da9d01b | 3753 | |
663350ac | 3754 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0ca1f7ce | 3755 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 3756 | space_info = block_rsv->space_info; |
bf9022e0 CM |
3757 | |
3758 | smp_mb(); | |
f4c738c2 JB |
3759 | delalloc_bytes = root->fs_info->delalloc_bytes; |
3760 | if (delalloc_bytes == 0) { | |
fdb5effd | 3761 | if (trans) |
f4c738c2 | 3762 | return; |
6bbe3a9c | 3763 | btrfs_wait_ordered_extents(root, 0); |
f4c738c2 | 3764 | return; |
fdb5effd JB |
3765 | } |
3766 | ||
f4c738c2 JB |
3767 | while (delalloc_bytes && loops < 3) { |
3768 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
3769 | nr_pages = max_reclaim >> PAGE_CACHE_SHIFT; | |
da633a42 | 3770 | btrfs_writeback_inodes_sb_nr(root, nr_pages); |
dea31f52 JB |
3771 | /* |
3772 | * We need to wait for the async pages to actually start before | |
3773 | * we do anything. | |
3774 | */ | |
3775 | wait_event(root->fs_info->async_submit_wait, | |
3776 | !atomic_read(&root->fs_info->async_delalloc_pages)); | |
3777 | ||
08e007d2 MX |
3778 | if (!trans) |
3779 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
3780 | else | |
3781 | flush = BTRFS_RESERVE_NO_FLUSH; | |
0019f10d | 3782 | spin_lock(&space_info->lock); |
08e007d2 | 3783 | if (can_overcommit(root, space_info, orig, flush)) { |
f4c738c2 JB |
3784 | spin_unlock(&space_info->lock); |
3785 | break; | |
3786 | } | |
0019f10d | 3787 | spin_unlock(&space_info->lock); |
5da9d01b | 3788 | |
36e39c40 | 3789 | loops++; |
f104d044 | 3790 | if (wait_ordered && !trans) { |
6bbe3a9c | 3791 | btrfs_wait_ordered_extents(root, 0); |
f104d044 | 3792 | } else { |
f4c738c2 | 3793 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
3794 | if (time_left) |
3795 | break; | |
3796 | } | |
f4c738c2 JB |
3797 | smp_mb(); |
3798 | delalloc_bytes = root->fs_info->delalloc_bytes; | |
5da9d01b | 3799 | } |
5da9d01b YZ |
3800 | } |
3801 | ||
663350ac JB |
3802 | /** |
3803 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
3804 | * @root - the root we're allocating for | |
3805 | * @bytes - the number of bytes we want to reserve | |
3806 | * @force - force the commit | |
8bb8ab2e | 3807 | * |
663350ac JB |
3808 | * This will check to make sure that committing the transaction will actually |
3809 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
3810 | * will return -ENOSPC. | |
8bb8ab2e | 3811 | */ |
663350ac JB |
3812 | static int may_commit_transaction(struct btrfs_root *root, |
3813 | struct btrfs_space_info *space_info, | |
3814 | u64 bytes, int force) | |
3815 | { | |
3816 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; | |
3817 | struct btrfs_trans_handle *trans; | |
3818 | ||
3819 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
3820 | if (trans) | |
3821 | return -EAGAIN; | |
3822 | ||
3823 | if (force) | |
3824 | goto commit; | |
3825 | ||
3826 | /* See if there is enough pinned space to make this reservation */ | |
3827 | spin_lock(&space_info->lock); | |
3828 | if (space_info->bytes_pinned >= bytes) { | |
3829 | spin_unlock(&space_info->lock); | |
3830 | goto commit; | |
3831 | } | |
3832 | spin_unlock(&space_info->lock); | |
3833 | ||
3834 | /* | |
3835 | * See if there is some space in the delayed insertion reservation for | |
3836 | * this reservation. | |
3837 | */ | |
3838 | if (space_info != delayed_rsv->space_info) | |
3839 | return -ENOSPC; | |
3840 | ||
d9b0218f | 3841 | spin_lock(&space_info->lock); |
663350ac | 3842 | spin_lock(&delayed_rsv->lock); |
d9b0218f | 3843 | if (space_info->bytes_pinned + delayed_rsv->size < bytes) { |
663350ac | 3844 | spin_unlock(&delayed_rsv->lock); |
d9b0218f | 3845 | spin_unlock(&space_info->lock); |
663350ac JB |
3846 | return -ENOSPC; |
3847 | } | |
3848 | spin_unlock(&delayed_rsv->lock); | |
d9b0218f | 3849 | spin_unlock(&space_info->lock); |
663350ac JB |
3850 | |
3851 | commit: | |
3852 | trans = btrfs_join_transaction(root); | |
3853 | if (IS_ERR(trans)) | |
3854 | return -ENOSPC; | |
3855 | ||
3856 | return btrfs_commit_transaction(trans, root); | |
3857 | } | |
3858 | ||
96c3f433 | 3859 | enum flush_state { |
67b0fd63 JB |
3860 | FLUSH_DELAYED_ITEMS_NR = 1, |
3861 | FLUSH_DELAYED_ITEMS = 2, | |
3862 | FLUSH_DELALLOC = 3, | |
3863 | FLUSH_DELALLOC_WAIT = 4, | |
ea658bad JB |
3864 | ALLOC_CHUNK = 5, |
3865 | COMMIT_TRANS = 6, | |
96c3f433 JB |
3866 | }; |
3867 | ||
3868 | static int flush_space(struct btrfs_root *root, | |
3869 | struct btrfs_space_info *space_info, u64 num_bytes, | |
3870 | u64 orig_bytes, int state) | |
3871 | { | |
3872 | struct btrfs_trans_handle *trans; | |
3873 | int nr; | |
f4c738c2 | 3874 | int ret = 0; |
96c3f433 JB |
3875 | |
3876 | switch (state) { | |
96c3f433 JB |
3877 | case FLUSH_DELAYED_ITEMS_NR: |
3878 | case FLUSH_DELAYED_ITEMS: | |
3879 | if (state == FLUSH_DELAYED_ITEMS_NR) { | |
3880 | u64 bytes = btrfs_calc_trans_metadata_size(root, 1); | |
3881 | ||
3882 | nr = (int)div64_u64(num_bytes, bytes); | |
3883 | if (!nr) | |
3884 | nr = 1; | |
3885 | nr *= 2; | |
3886 | } else { | |
3887 | nr = -1; | |
3888 | } | |
3889 | trans = btrfs_join_transaction(root); | |
3890 | if (IS_ERR(trans)) { | |
3891 | ret = PTR_ERR(trans); | |
3892 | break; | |
3893 | } | |
3894 | ret = btrfs_run_delayed_items_nr(trans, root, nr); | |
3895 | btrfs_end_transaction(trans, root); | |
3896 | break; | |
67b0fd63 JB |
3897 | case FLUSH_DELALLOC: |
3898 | case FLUSH_DELALLOC_WAIT: | |
3899 | shrink_delalloc(root, num_bytes, orig_bytes, | |
3900 | state == FLUSH_DELALLOC_WAIT); | |
3901 | break; | |
ea658bad JB |
3902 | case ALLOC_CHUNK: |
3903 | trans = btrfs_join_transaction(root); | |
3904 | if (IS_ERR(trans)) { | |
3905 | ret = PTR_ERR(trans); | |
3906 | break; | |
3907 | } | |
3908 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
ea658bad JB |
3909 | btrfs_get_alloc_profile(root, 0), |
3910 | CHUNK_ALLOC_NO_FORCE); | |
3911 | btrfs_end_transaction(trans, root); | |
3912 | if (ret == -ENOSPC) | |
3913 | ret = 0; | |
3914 | break; | |
96c3f433 JB |
3915 | case COMMIT_TRANS: |
3916 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); | |
3917 | break; | |
3918 | default: | |
3919 | ret = -ENOSPC; | |
3920 | break; | |
3921 | } | |
3922 | ||
3923 | return ret; | |
3924 | } | |
4a92b1b8 JB |
3925 | /** |
3926 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
3927 | * @root - the root we're allocating for | |
3928 | * @block_rsv - the block_rsv we're allocating for | |
3929 | * @orig_bytes - the number of bytes we want | |
3930 | * @flush - wether or not we can flush to make our reservation | |
8bb8ab2e | 3931 | * |
4a92b1b8 JB |
3932 | * This will reserve orgi_bytes number of bytes from the space info associated |
3933 | * with the block_rsv. If there is not enough space it will make an attempt to | |
3934 | * flush out space to make room. It will do this by flushing delalloc if | |
3935 | * possible or committing the transaction. If flush is 0 then no attempts to | |
3936 | * regain reservations will be made and this will fail if there is not enough | |
3937 | * space already. | |
8bb8ab2e | 3938 | */ |
4a92b1b8 | 3939 | static int reserve_metadata_bytes(struct btrfs_root *root, |
8bb8ab2e | 3940 | struct btrfs_block_rsv *block_rsv, |
08e007d2 MX |
3941 | u64 orig_bytes, |
3942 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 3943 | { |
f0486c68 | 3944 | struct btrfs_space_info *space_info = block_rsv->space_info; |
2bf64758 | 3945 | u64 used; |
8bb8ab2e | 3946 | u64 num_bytes = orig_bytes; |
67b0fd63 | 3947 | int flush_state = FLUSH_DELAYED_ITEMS_NR; |
8bb8ab2e | 3948 | int ret = 0; |
fdb5effd | 3949 | bool flushing = false; |
9ed74f2d | 3950 | |
8bb8ab2e | 3951 | again: |
fdb5effd | 3952 | ret = 0; |
8bb8ab2e | 3953 | spin_lock(&space_info->lock); |
fdb5effd | 3954 | /* |
08e007d2 MX |
3955 | * We only want to wait if somebody other than us is flushing and we |
3956 | * are actually allowed to flush all things. | |
fdb5effd | 3957 | */ |
08e007d2 MX |
3958 | while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing && |
3959 | space_info->flush) { | |
fdb5effd JB |
3960 | spin_unlock(&space_info->lock); |
3961 | /* | |
3962 | * If we have a trans handle we can't wait because the flusher | |
3963 | * may have to commit the transaction, which would mean we would | |
3964 | * deadlock since we are waiting for the flusher to finish, but | |
3965 | * hold the current transaction open. | |
3966 | */ | |
663350ac | 3967 | if (current->journal_info) |
fdb5effd | 3968 | return -EAGAIN; |
b9688bb8 AJ |
3969 | ret = wait_event_killable(space_info->wait, !space_info->flush); |
3970 | /* Must have been killed, return */ | |
3971 | if (ret) | |
fdb5effd JB |
3972 | return -EINTR; |
3973 | ||
3974 | spin_lock(&space_info->lock); | |
3975 | } | |
3976 | ||
3977 | ret = -ENOSPC; | |
2bf64758 JB |
3978 | used = space_info->bytes_used + space_info->bytes_reserved + |
3979 | space_info->bytes_pinned + space_info->bytes_readonly + | |
3980 | space_info->bytes_may_use; | |
9ed74f2d | 3981 | |
8bb8ab2e JB |
3982 | /* |
3983 | * The idea here is that we've not already over-reserved the block group | |
3984 | * then we can go ahead and save our reservation first and then start | |
3985 | * flushing if we need to. Otherwise if we've already overcommitted | |
3986 | * lets start flushing stuff first and then come back and try to make | |
3987 | * our reservation. | |
3988 | */ | |
2bf64758 JB |
3989 | if (used <= space_info->total_bytes) { |
3990 | if (used + orig_bytes <= space_info->total_bytes) { | |
fb25e914 | 3991 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 3992 | trace_btrfs_space_reservation(root->fs_info, |
2bcc0328 | 3993 | "space_info", space_info->flags, orig_bytes, 1); |
8bb8ab2e JB |
3994 | ret = 0; |
3995 | } else { | |
3996 | /* | |
3997 | * Ok set num_bytes to orig_bytes since we aren't | |
3998 | * overocmmitted, this way we only try and reclaim what | |
3999 | * we need. | |
4000 | */ | |
4001 | num_bytes = orig_bytes; | |
4002 | } | |
4003 | } else { | |
4004 | /* | |
4005 | * Ok we're over committed, set num_bytes to the overcommitted | |
4006 | * amount plus the amount of bytes that we need for this | |
4007 | * reservation. | |
4008 | */ | |
2bf64758 | 4009 | num_bytes = used - space_info->total_bytes + |
96c3f433 | 4010 | (orig_bytes * 2); |
8bb8ab2e | 4011 | } |
9ed74f2d | 4012 | |
44734ed1 JB |
4013 | if (ret && can_overcommit(root, space_info, orig_bytes, flush)) { |
4014 | space_info->bytes_may_use += orig_bytes; | |
4015 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
4016 | space_info->flags, orig_bytes, | |
4017 | 1); | |
4018 | ret = 0; | |
2bf64758 JB |
4019 | } |
4020 | ||
8bb8ab2e JB |
4021 | /* |
4022 | * Couldn't make our reservation, save our place so while we're trying | |
4023 | * to reclaim space we can actually use it instead of somebody else | |
4024 | * stealing it from us. | |
08e007d2 MX |
4025 | * |
4026 | * We make the other tasks wait for the flush only when we can flush | |
4027 | * all things. | |
8bb8ab2e | 4028 | */ |
72bcd99d | 4029 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
fdb5effd JB |
4030 | flushing = true; |
4031 | space_info->flush = 1; | |
8bb8ab2e | 4032 | } |
9ed74f2d | 4033 | |
f0486c68 | 4034 | spin_unlock(&space_info->lock); |
9ed74f2d | 4035 | |
08e007d2 | 4036 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
8bb8ab2e | 4037 | goto out; |
f0486c68 | 4038 | |
96c3f433 JB |
4039 | ret = flush_space(root, space_info, num_bytes, orig_bytes, |
4040 | flush_state); | |
4041 | flush_state++; | |
08e007d2 MX |
4042 | |
4043 | /* | |
4044 | * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock | |
4045 | * would happen. So skip delalloc flush. | |
4046 | */ | |
4047 | if (flush == BTRFS_RESERVE_FLUSH_LIMIT && | |
4048 | (flush_state == FLUSH_DELALLOC || | |
4049 | flush_state == FLUSH_DELALLOC_WAIT)) | |
4050 | flush_state = ALLOC_CHUNK; | |
4051 | ||
96c3f433 | 4052 | if (!ret) |
8bb8ab2e | 4053 | goto again; |
08e007d2 MX |
4054 | else if (flush == BTRFS_RESERVE_FLUSH_LIMIT && |
4055 | flush_state < COMMIT_TRANS) | |
4056 | goto again; | |
4057 | else if (flush == BTRFS_RESERVE_FLUSH_ALL && | |
4058 | flush_state <= COMMIT_TRANS) | |
8bb8ab2e JB |
4059 | goto again; |
4060 | ||
4061 | out: | |
fdb5effd | 4062 | if (flushing) { |
8bb8ab2e | 4063 | spin_lock(&space_info->lock); |
fdb5effd JB |
4064 | space_info->flush = 0; |
4065 | wake_up_all(&space_info->wait); | |
8bb8ab2e | 4066 | spin_unlock(&space_info->lock); |
f0486c68 | 4067 | } |
f0486c68 YZ |
4068 | return ret; |
4069 | } | |
4070 | ||
79787eaa JM |
4071 | static struct btrfs_block_rsv *get_block_rsv( |
4072 | const struct btrfs_trans_handle *trans, | |
4073 | const struct btrfs_root *root) | |
f0486c68 | 4074 | { |
4c13d758 JB |
4075 | struct btrfs_block_rsv *block_rsv = NULL; |
4076 | ||
0e721106 JB |
4077 | if (root->ref_cows) |
4078 | block_rsv = trans->block_rsv; | |
4079 | ||
4080 | if (root == root->fs_info->csum_root && trans->adding_csums) | |
f0486c68 | 4081 | block_rsv = trans->block_rsv; |
4c13d758 JB |
4082 | |
4083 | if (!block_rsv) | |
f0486c68 YZ |
4084 | block_rsv = root->block_rsv; |
4085 | ||
4086 | if (!block_rsv) | |
4087 | block_rsv = &root->fs_info->empty_block_rsv; | |
4088 | ||
4089 | return block_rsv; | |
4090 | } | |
4091 | ||
4092 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
4093 | u64 num_bytes) | |
4094 | { | |
4095 | int ret = -ENOSPC; | |
4096 | spin_lock(&block_rsv->lock); | |
4097 | if (block_rsv->reserved >= num_bytes) { | |
4098 | block_rsv->reserved -= num_bytes; | |
4099 | if (block_rsv->reserved < block_rsv->size) | |
4100 | block_rsv->full = 0; | |
4101 | ret = 0; | |
4102 | } | |
4103 | spin_unlock(&block_rsv->lock); | |
4104 | return ret; | |
4105 | } | |
4106 | ||
4107 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
4108 | u64 num_bytes, int update_size) | |
4109 | { | |
4110 | spin_lock(&block_rsv->lock); | |
4111 | block_rsv->reserved += num_bytes; | |
4112 | if (update_size) | |
4113 | block_rsv->size += num_bytes; | |
4114 | else if (block_rsv->reserved >= block_rsv->size) | |
4115 | block_rsv->full = 1; | |
4116 | spin_unlock(&block_rsv->lock); | |
4117 | } | |
4118 | ||
8c2a3ca2 JB |
4119 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
4120 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 4121 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
4122 | { |
4123 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
4124 | ||
4125 | spin_lock(&block_rsv->lock); | |
4126 | if (num_bytes == (u64)-1) | |
4127 | num_bytes = block_rsv->size; | |
4128 | block_rsv->size -= num_bytes; | |
4129 | if (block_rsv->reserved >= block_rsv->size) { | |
4130 | num_bytes = block_rsv->reserved - block_rsv->size; | |
4131 | block_rsv->reserved = block_rsv->size; | |
4132 | block_rsv->full = 1; | |
4133 | } else { | |
4134 | num_bytes = 0; | |
4135 | } | |
4136 | spin_unlock(&block_rsv->lock); | |
4137 | ||
4138 | if (num_bytes > 0) { | |
4139 | if (dest) { | |
e9e22899 JB |
4140 | spin_lock(&dest->lock); |
4141 | if (!dest->full) { | |
4142 | u64 bytes_to_add; | |
4143 | ||
4144 | bytes_to_add = dest->size - dest->reserved; | |
4145 | bytes_to_add = min(num_bytes, bytes_to_add); | |
4146 | dest->reserved += bytes_to_add; | |
4147 | if (dest->reserved >= dest->size) | |
4148 | dest->full = 1; | |
4149 | num_bytes -= bytes_to_add; | |
4150 | } | |
4151 | spin_unlock(&dest->lock); | |
4152 | } | |
4153 | if (num_bytes) { | |
f0486c68 | 4154 | spin_lock(&space_info->lock); |
fb25e914 | 4155 | space_info->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4156 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4157 | space_info->flags, num_bytes, 0); |
36e39c40 | 4158 | space_info->reservation_progress++; |
f0486c68 | 4159 | spin_unlock(&space_info->lock); |
4e06bdd6 | 4160 | } |
9ed74f2d | 4161 | } |
f0486c68 | 4162 | } |
4e06bdd6 | 4163 | |
f0486c68 YZ |
4164 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
4165 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
4166 | { | |
4167 | int ret; | |
9ed74f2d | 4168 | |
f0486c68 YZ |
4169 | ret = block_rsv_use_bytes(src, num_bytes); |
4170 | if (ret) | |
4171 | return ret; | |
9ed74f2d | 4172 | |
f0486c68 | 4173 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
4174 | return 0; |
4175 | } | |
4176 | ||
66d8f3dd | 4177 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 4178 | { |
f0486c68 YZ |
4179 | memset(rsv, 0, sizeof(*rsv)); |
4180 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 4181 | rsv->type = type; |
f0486c68 YZ |
4182 | } |
4183 | ||
66d8f3dd MX |
4184 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root, |
4185 | unsigned short type) | |
f0486c68 YZ |
4186 | { |
4187 | struct btrfs_block_rsv *block_rsv; | |
4188 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 4189 | |
f0486c68 YZ |
4190 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
4191 | if (!block_rsv) | |
4192 | return NULL; | |
9ed74f2d | 4193 | |
66d8f3dd | 4194 | btrfs_init_block_rsv(block_rsv, type); |
f0486c68 YZ |
4195 | block_rsv->space_info = __find_space_info(fs_info, |
4196 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
4197 | return block_rsv; |
4198 | } | |
9ed74f2d | 4199 | |
f0486c68 YZ |
4200 | void btrfs_free_block_rsv(struct btrfs_root *root, |
4201 | struct btrfs_block_rsv *rsv) | |
4202 | { | |
2aaa6655 JB |
4203 | if (!rsv) |
4204 | return; | |
dabdb640 JB |
4205 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
4206 | kfree(rsv); | |
9ed74f2d JB |
4207 | } |
4208 | ||
08e007d2 MX |
4209 | int btrfs_block_rsv_add(struct btrfs_root *root, |
4210 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
4211 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 4212 | { |
f0486c68 | 4213 | int ret; |
9ed74f2d | 4214 | |
f0486c68 YZ |
4215 | if (num_bytes == 0) |
4216 | return 0; | |
8bb8ab2e | 4217 | |
61b520a9 | 4218 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
4219 | if (!ret) { |
4220 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
4221 | return 0; | |
4222 | } | |
9ed74f2d | 4223 | |
f0486c68 | 4224 | return ret; |
f0486c68 | 4225 | } |
9ed74f2d | 4226 | |
4a92b1b8 | 4227 | int btrfs_block_rsv_check(struct btrfs_root *root, |
36ba022a | 4228 | struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
4229 | { |
4230 | u64 num_bytes = 0; | |
f0486c68 | 4231 | int ret = -ENOSPC; |
9ed74f2d | 4232 | |
f0486c68 YZ |
4233 | if (!block_rsv) |
4234 | return 0; | |
9ed74f2d | 4235 | |
f0486c68 | 4236 | spin_lock(&block_rsv->lock); |
36ba022a JB |
4237 | num_bytes = div_factor(block_rsv->size, min_factor); |
4238 | if (block_rsv->reserved >= num_bytes) | |
4239 | ret = 0; | |
4240 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 4241 | |
36ba022a JB |
4242 | return ret; |
4243 | } | |
4244 | ||
08e007d2 MX |
4245 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
4246 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
4247 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
4248 | { |
4249 | u64 num_bytes = 0; | |
4250 | int ret = -ENOSPC; | |
4251 | ||
4252 | if (!block_rsv) | |
4253 | return 0; | |
4254 | ||
4255 | spin_lock(&block_rsv->lock); | |
4256 | num_bytes = min_reserved; | |
13553e52 | 4257 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 4258 | ret = 0; |
13553e52 | 4259 | else |
f0486c68 | 4260 | num_bytes -= block_rsv->reserved; |
f0486c68 | 4261 | spin_unlock(&block_rsv->lock); |
13553e52 | 4262 | |
f0486c68 YZ |
4263 | if (!ret) |
4264 | return 0; | |
4265 | ||
aa38a711 | 4266 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
4267 | if (!ret) { |
4268 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 4269 | return 0; |
6a63209f | 4270 | } |
9ed74f2d | 4271 | |
13553e52 | 4272 | return ret; |
f0486c68 YZ |
4273 | } |
4274 | ||
4275 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, | |
4276 | struct btrfs_block_rsv *dst_rsv, | |
4277 | u64 num_bytes) | |
4278 | { | |
4279 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
4280 | } | |
4281 | ||
4282 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
4283 | struct btrfs_block_rsv *block_rsv, | |
4284 | u64 num_bytes) | |
4285 | { | |
4286 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
4287 | if (global_rsv->full || global_rsv == block_rsv || | |
4288 | block_rsv->space_info != global_rsv->space_info) | |
4289 | global_rsv = NULL; | |
8c2a3ca2 JB |
4290 | block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv, |
4291 | num_bytes); | |
6a63209f JB |
4292 | } |
4293 | ||
4294 | /* | |
8929ecfa YZ |
4295 | * helper to calculate size of global block reservation. |
4296 | * the desired value is sum of space used by extent tree, | |
4297 | * checksum tree and root tree | |
6a63209f | 4298 | */ |
8929ecfa | 4299 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 4300 | { |
8929ecfa YZ |
4301 | struct btrfs_space_info *sinfo; |
4302 | u64 num_bytes; | |
4303 | u64 meta_used; | |
4304 | u64 data_used; | |
6c41761f | 4305 | int csum_size = btrfs_super_csum_size(fs_info->super_copy); |
6a63209f | 4306 | |
8929ecfa YZ |
4307 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
4308 | spin_lock(&sinfo->lock); | |
4309 | data_used = sinfo->bytes_used; | |
4310 | spin_unlock(&sinfo->lock); | |
33b4d47f | 4311 | |
8929ecfa YZ |
4312 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
4313 | spin_lock(&sinfo->lock); | |
6d48755d JB |
4314 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
4315 | data_used = 0; | |
8929ecfa YZ |
4316 | meta_used = sinfo->bytes_used; |
4317 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 4318 | |
8929ecfa YZ |
4319 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
4320 | csum_size * 2; | |
4321 | num_bytes += div64_u64(data_used + meta_used, 50); | |
4e06bdd6 | 4322 | |
8929ecfa | 4323 | if (num_bytes * 3 > meta_used) |
8e62c2de | 4324 | num_bytes = div64_u64(meta_used, 3); |
ab6e2410 | 4325 | |
8929ecfa YZ |
4326 | return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10); |
4327 | } | |
6a63209f | 4328 | |
8929ecfa YZ |
4329 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
4330 | { | |
4331 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
4332 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
4333 | u64 num_bytes; | |
6a63209f | 4334 | |
8929ecfa | 4335 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 4336 | |
8929ecfa | 4337 | spin_lock(&sinfo->lock); |
1f699d38 | 4338 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 4339 | |
8929ecfa | 4340 | block_rsv->size = num_bytes; |
4e06bdd6 | 4341 | |
8929ecfa | 4342 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
6d48755d JB |
4343 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
4344 | sinfo->bytes_may_use; | |
8929ecfa YZ |
4345 | |
4346 | if (sinfo->total_bytes > num_bytes) { | |
4347 | num_bytes = sinfo->total_bytes - num_bytes; | |
4348 | block_rsv->reserved += num_bytes; | |
fb25e914 | 4349 | sinfo->bytes_may_use += num_bytes; |
8c2a3ca2 | 4350 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4351 | sinfo->flags, num_bytes, 1); |
6a63209f | 4352 | } |
6a63209f | 4353 | |
8929ecfa YZ |
4354 | if (block_rsv->reserved >= block_rsv->size) { |
4355 | num_bytes = block_rsv->reserved - block_rsv->size; | |
fb25e914 | 4356 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4357 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4358 | sinfo->flags, num_bytes, 0); |
36e39c40 | 4359 | sinfo->reservation_progress++; |
8929ecfa YZ |
4360 | block_rsv->reserved = block_rsv->size; |
4361 | block_rsv->full = 1; | |
4362 | } | |
182608c8 | 4363 | |
8929ecfa | 4364 | spin_unlock(&block_rsv->lock); |
1f699d38 | 4365 | spin_unlock(&sinfo->lock); |
6a63209f JB |
4366 | } |
4367 | ||
f0486c68 | 4368 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4369 | { |
f0486c68 | 4370 | struct btrfs_space_info *space_info; |
6a63209f | 4371 | |
f0486c68 YZ |
4372 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
4373 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 4374 | |
f0486c68 | 4375 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 4376 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 4377 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
4378 | fs_info->trans_block_rsv.space_info = space_info; |
4379 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 4380 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 4381 | |
8929ecfa YZ |
4382 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
4383 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
4384 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
4385 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 4386 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 4387 | |
8929ecfa | 4388 | update_global_block_rsv(fs_info); |
6a63209f JB |
4389 | } |
4390 | ||
8929ecfa | 4391 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4392 | { |
8c2a3ca2 JB |
4393 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
4394 | (u64)-1); | |
8929ecfa YZ |
4395 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
4396 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
4397 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
4398 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
4399 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
4400 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
4401 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
4402 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
4403 | } |
4404 | ||
a22285a6 YZ |
4405 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
4406 | struct btrfs_root *root) | |
6a63209f | 4407 | { |
0e721106 JB |
4408 | if (!trans->block_rsv) |
4409 | return; | |
4410 | ||
a22285a6 YZ |
4411 | if (!trans->bytes_reserved) |
4412 | return; | |
6a63209f | 4413 | |
e77266e4 | 4414 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 4415 | trans->transid, trans->bytes_reserved, 0); |
b24e03db | 4416 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); |
a22285a6 YZ |
4417 | trans->bytes_reserved = 0; |
4418 | } | |
6a63209f | 4419 | |
79787eaa | 4420 | /* Can only return 0 or -ENOSPC */ |
d68fc57b YZ |
4421 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
4422 | struct inode *inode) | |
4423 | { | |
4424 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4425 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4426 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
4427 | ||
4428 | /* | |
fcb80c2a JB |
4429 | * We need to hold space in order to delete our orphan item once we've |
4430 | * added it, so this takes the reservation so we can release it later | |
4431 | * when we are truly done with the orphan item. | |
d68fc57b | 4432 | */ |
ff5714cc | 4433 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4434 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4435 | btrfs_ino(inode), num_bytes, 1); | |
d68fc57b | 4436 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
6a63209f JB |
4437 | } |
4438 | ||
d68fc57b | 4439 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 4440 | { |
d68fc57b | 4441 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 4442 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4443 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4444 | btrfs_ino(inode), num_bytes, 0); | |
d68fc57b YZ |
4445 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
4446 | } | |
97e728d4 | 4447 | |
a22285a6 YZ |
4448 | int btrfs_snap_reserve_metadata(struct btrfs_trans_handle *trans, |
4449 | struct btrfs_pending_snapshot *pending) | |
4450 | { | |
4451 | struct btrfs_root *root = pending->root; | |
4452 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4453 | struct btrfs_block_rsv *dst_rsv = &pending->block_rsv; | |
4454 | /* | |
48c03c4b MX |
4455 | * two for root back/forward refs, two for directory entries, |
4456 | * one for root of the snapshot and one for parent inode. | |
a22285a6 | 4457 | */ |
48c03c4b | 4458 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 6); |
a22285a6 YZ |
4459 | dst_rsv->space_info = src_rsv->space_info; |
4460 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
97e728d4 JB |
4461 | } |
4462 | ||
7709cde3 JB |
4463 | /** |
4464 | * drop_outstanding_extent - drop an outstanding extent | |
4465 | * @inode: the inode we're dropping the extent for | |
4466 | * | |
4467 | * This is called when we are freeing up an outstanding extent, either called | |
4468 | * after an error or after an extent is written. This will return the number of | |
4469 | * reserved extents that need to be freed. This must be called with | |
4470 | * BTRFS_I(inode)->lock held. | |
4471 | */ | |
9e0baf60 JB |
4472 | static unsigned drop_outstanding_extent(struct inode *inode) |
4473 | { | |
7fd2ae21 | 4474 | unsigned drop_inode_space = 0; |
9e0baf60 JB |
4475 | unsigned dropped_extents = 0; |
4476 | ||
9e0baf60 JB |
4477 | BUG_ON(!BTRFS_I(inode)->outstanding_extents); |
4478 | BTRFS_I(inode)->outstanding_extents--; | |
4479 | ||
7fd2ae21 | 4480 | if (BTRFS_I(inode)->outstanding_extents == 0 && |
72ac3c0d JB |
4481 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4482 | &BTRFS_I(inode)->runtime_flags)) | |
7fd2ae21 | 4483 | drop_inode_space = 1; |
7fd2ae21 | 4484 | |
9e0baf60 JB |
4485 | /* |
4486 | * If we have more or the same amount of outsanding extents than we have | |
4487 | * reserved then we need to leave the reserved extents count alone. | |
4488 | */ | |
4489 | if (BTRFS_I(inode)->outstanding_extents >= | |
4490 | BTRFS_I(inode)->reserved_extents) | |
7fd2ae21 | 4491 | return drop_inode_space; |
9e0baf60 JB |
4492 | |
4493 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
4494 | BTRFS_I(inode)->outstanding_extents; | |
4495 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
7fd2ae21 | 4496 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
4497 | } |
4498 | ||
7709cde3 JB |
4499 | /** |
4500 | * calc_csum_metadata_size - return the amount of metada space that must be | |
4501 | * reserved/free'd for the given bytes. | |
4502 | * @inode: the inode we're manipulating | |
4503 | * @num_bytes: the number of bytes in question | |
4504 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
4505 | * | |
4506 | * This adjusts the number of csum_bytes in the inode and then returns the | |
4507 | * correct amount of metadata that must either be reserved or freed. We | |
4508 | * calculate how many checksums we can fit into one leaf and then divide the | |
4509 | * number of bytes that will need to be checksumed by this value to figure out | |
4510 | * how many checksums will be required. If we are adding bytes then the number | |
4511 | * may go up and we will return the number of additional bytes that must be | |
4512 | * reserved. If it is going down we will return the number of bytes that must | |
4513 | * be freed. | |
4514 | * | |
4515 | * This must be called with BTRFS_I(inode)->lock held. | |
4516 | */ | |
4517 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, | |
4518 | int reserve) | |
6324fbf3 | 4519 | { |
7709cde3 JB |
4520 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4521 | u64 csum_size; | |
4522 | int num_csums_per_leaf; | |
4523 | int num_csums; | |
4524 | int old_csums; | |
4525 | ||
4526 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && | |
4527 | BTRFS_I(inode)->csum_bytes == 0) | |
4528 | return 0; | |
4529 | ||
4530 | old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4531 | if (reserve) | |
4532 | BTRFS_I(inode)->csum_bytes += num_bytes; | |
4533 | else | |
4534 | BTRFS_I(inode)->csum_bytes -= num_bytes; | |
4535 | csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item); | |
4536 | num_csums_per_leaf = (int)div64_u64(csum_size, | |
4537 | sizeof(struct btrfs_csum_item) + | |
4538 | sizeof(struct btrfs_disk_key)); | |
4539 | num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4540 | num_csums = num_csums + num_csums_per_leaf - 1; | |
4541 | num_csums = num_csums / num_csums_per_leaf; | |
4542 | ||
4543 | old_csums = old_csums + num_csums_per_leaf - 1; | |
4544 | old_csums = old_csums / num_csums_per_leaf; | |
4545 | ||
4546 | /* No change, no need to reserve more */ | |
4547 | if (old_csums == num_csums) | |
4548 | return 0; | |
4549 | ||
4550 | if (reserve) | |
4551 | return btrfs_calc_trans_metadata_size(root, | |
4552 | num_csums - old_csums); | |
4553 | ||
4554 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); | |
0ca1f7ce | 4555 | } |
c146afad | 4556 | |
0ca1f7ce YZ |
4557 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
4558 | { | |
4559 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4560 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 | 4561 | u64 to_reserve = 0; |
660d3f6c | 4562 | u64 csum_bytes; |
9e0baf60 | 4563 | unsigned nr_extents = 0; |
660d3f6c | 4564 | int extra_reserve = 0; |
08e007d2 | 4565 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 4566 | int ret = 0; |
c64c2bd8 | 4567 | bool delalloc_lock = true; |
6324fbf3 | 4568 | |
c64c2bd8 JB |
4569 | /* If we are a free space inode we need to not flush since we will be in |
4570 | * the middle of a transaction commit. We also don't need the delalloc | |
4571 | * mutex since we won't race with anybody. We need this mostly to make | |
4572 | * lockdep shut its filthy mouth. | |
4573 | */ | |
4574 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 4575 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 JB |
4576 | delalloc_lock = false; |
4577 | } | |
c09544e0 | 4578 | |
08e007d2 MX |
4579 | if (flush != BTRFS_RESERVE_NO_FLUSH && |
4580 | btrfs_transaction_in_commit(root->fs_info)) | |
0ca1f7ce | 4581 | schedule_timeout(1); |
ec44a35c | 4582 | |
c64c2bd8 JB |
4583 | if (delalloc_lock) |
4584 | mutex_lock(&BTRFS_I(inode)->delalloc_mutex); | |
4585 | ||
0ca1f7ce | 4586 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 4587 | |
9e0baf60 JB |
4588 | spin_lock(&BTRFS_I(inode)->lock); |
4589 | BTRFS_I(inode)->outstanding_extents++; | |
4590 | ||
4591 | if (BTRFS_I(inode)->outstanding_extents > | |
660d3f6c | 4592 | BTRFS_I(inode)->reserved_extents) |
9e0baf60 JB |
4593 | nr_extents = BTRFS_I(inode)->outstanding_extents - |
4594 | BTRFS_I(inode)->reserved_extents; | |
57a45ced | 4595 | |
7fd2ae21 JB |
4596 | /* |
4597 | * Add an item to reserve for updating the inode when we complete the | |
4598 | * delalloc io. | |
4599 | */ | |
72ac3c0d JB |
4600 | if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4601 | &BTRFS_I(inode)->runtime_flags)) { | |
7fd2ae21 | 4602 | nr_extents++; |
660d3f6c | 4603 | extra_reserve = 1; |
593060d7 | 4604 | } |
7fd2ae21 JB |
4605 | |
4606 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); | |
7709cde3 | 4607 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
660d3f6c | 4608 | csum_bytes = BTRFS_I(inode)->csum_bytes; |
9e0baf60 | 4609 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 4610 | |
eb6b88d9 | 4611 | if (root->fs_info->quota_enabled) |
c5567237 AJ |
4612 | ret = btrfs_qgroup_reserve(root, num_bytes + |
4613 | nr_extents * root->leafsize); | |
c5567237 | 4614 | |
eb6b88d9 JS |
4615 | /* |
4616 | * ret != 0 here means the qgroup reservation failed, we go straight to | |
4617 | * the shared error handling then. | |
4618 | */ | |
4619 | if (ret == 0) | |
4620 | ret = reserve_metadata_bytes(root, block_rsv, | |
4621 | to_reserve, flush); | |
4622 | ||
9e0baf60 | 4623 | if (ret) { |
7ed49f18 | 4624 | u64 to_free = 0; |
9e0baf60 | 4625 | unsigned dropped; |
7ed49f18 | 4626 | |
7709cde3 | 4627 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 4628 | dropped = drop_outstanding_extent(inode); |
9e0baf60 | 4629 | /* |
660d3f6c JB |
4630 | * If the inodes csum_bytes is the same as the original |
4631 | * csum_bytes then we know we haven't raced with any free()ers | |
4632 | * so we can just reduce our inodes csum bytes and carry on. | |
4633 | * Otherwise we have to do the normal free thing to account for | |
4634 | * the case that the free side didn't free up its reserve | |
4635 | * because of this outstanding reservation. | |
9e0baf60 | 4636 | */ |
660d3f6c JB |
4637 | if (BTRFS_I(inode)->csum_bytes == csum_bytes) |
4638 | calc_csum_metadata_size(inode, num_bytes, 0); | |
4639 | else | |
4640 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
4641 | spin_unlock(&BTRFS_I(inode)->lock); | |
4642 | if (dropped) | |
4643 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
4644 | ||
8c2a3ca2 | 4645 | if (to_free) { |
7ed49f18 | 4646 | btrfs_block_rsv_release(root, block_rsv, to_free); |
8c2a3ca2 JB |
4647 | trace_btrfs_space_reservation(root->fs_info, |
4648 | "delalloc", | |
4649 | btrfs_ino(inode), | |
4650 | to_free, 0); | |
4651 | } | |
4b5829a8 MX |
4652 | if (root->fs_info->quota_enabled) { |
4653 | btrfs_qgroup_free(root, num_bytes + | |
4654 | nr_extents * root->leafsize); | |
4655 | } | |
c64c2bd8 JB |
4656 | if (delalloc_lock) |
4657 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
0ca1f7ce | 4658 | return ret; |
9e0baf60 | 4659 | } |
25179201 | 4660 | |
660d3f6c JB |
4661 | spin_lock(&BTRFS_I(inode)->lock); |
4662 | if (extra_reserve) { | |
72ac3c0d JB |
4663 | set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4664 | &BTRFS_I(inode)->runtime_flags); | |
660d3f6c JB |
4665 | nr_extents--; |
4666 | } | |
4667 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
4668 | spin_unlock(&BTRFS_I(inode)->lock); | |
c64c2bd8 JB |
4669 | |
4670 | if (delalloc_lock) | |
4671 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
660d3f6c | 4672 | |
8c2a3ca2 JB |
4673 | if (to_reserve) |
4674 | trace_btrfs_space_reservation(root->fs_info,"delalloc", | |
4675 | btrfs_ino(inode), to_reserve, 1); | |
0ca1f7ce YZ |
4676 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
4677 | ||
0ca1f7ce YZ |
4678 | return 0; |
4679 | } | |
4680 | ||
7709cde3 JB |
4681 | /** |
4682 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
4683 | * @inode: the inode to release the reservation for | |
4684 | * @num_bytes: the number of bytes we're releasing | |
4685 | * | |
4686 | * This will release the metadata reservation for an inode. This can be called | |
4687 | * once we complete IO for a given set of bytes to release their metadata | |
4688 | * reservations. | |
4689 | */ | |
0ca1f7ce YZ |
4690 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) |
4691 | { | |
4692 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
4693 | u64 to_free = 0; |
4694 | unsigned dropped; | |
0ca1f7ce YZ |
4695 | |
4696 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
7709cde3 | 4697 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 4698 | dropped = drop_outstanding_extent(inode); |
97e728d4 | 4699 | |
7709cde3 JB |
4700 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); |
4701 | spin_unlock(&BTRFS_I(inode)->lock); | |
9e0baf60 JB |
4702 | if (dropped > 0) |
4703 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce | 4704 | |
8c2a3ca2 JB |
4705 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
4706 | btrfs_ino(inode), to_free, 0); | |
c5567237 AJ |
4707 | if (root->fs_info->quota_enabled) { |
4708 | btrfs_qgroup_free(root, num_bytes + | |
4709 | dropped * root->leafsize); | |
4710 | } | |
4711 | ||
0ca1f7ce YZ |
4712 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, |
4713 | to_free); | |
4714 | } | |
4715 | ||
7709cde3 JB |
4716 | /** |
4717 | * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc | |
4718 | * @inode: inode we're writing to | |
4719 | * @num_bytes: the number of bytes we want to allocate | |
4720 | * | |
4721 | * This will do the following things | |
4722 | * | |
4723 | * o reserve space in the data space info for num_bytes | |
4724 | * o reserve space in the metadata space info based on number of outstanding | |
4725 | * extents and how much csums will be needed | |
4726 | * o add to the inodes ->delalloc_bytes | |
4727 | * o add it to the fs_info's delalloc inodes list. | |
4728 | * | |
4729 | * This will return 0 for success and -ENOSPC if there is no space left. | |
4730 | */ | |
0ca1f7ce YZ |
4731 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes) |
4732 | { | |
4733 | int ret; | |
4734 | ||
4735 | ret = btrfs_check_data_free_space(inode, num_bytes); | |
d397712b | 4736 | if (ret) |
0ca1f7ce YZ |
4737 | return ret; |
4738 | ||
4739 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes); | |
4740 | if (ret) { | |
4741 | btrfs_free_reserved_data_space(inode, num_bytes); | |
4742 | return ret; | |
4743 | } | |
4744 | ||
4745 | return 0; | |
4746 | } | |
4747 | ||
7709cde3 JB |
4748 | /** |
4749 | * btrfs_delalloc_release_space - release data and metadata space for delalloc | |
4750 | * @inode: inode we're releasing space for | |
4751 | * @num_bytes: the number of bytes we want to free up | |
4752 | * | |
4753 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
4754 | * called in the case that we don't need the metadata AND data reservations | |
4755 | * anymore. So if there is an error or we insert an inline extent. | |
4756 | * | |
4757 | * This function will release the metadata space that was not used and will | |
4758 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
4759 | * list if there are no delalloc bytes left. | |
4760 | */ | |
0ca1f7ce YZ |
4761 | void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes) |
4762 | { | |
4763 | btrfs_delalloc_release_metadata(inode, num_bytes); | |
4764 | btrfs_free_reserved_data_space(inode, num_bytes); | |
6324fbf3 CM |
4765 | } |
4766 | ||
9078a3e1 CM |
4767 | static int update_block_group(struct btrfs_trans_handle *trans, |
4768 | struct btrfs_root *root, | |
f0486c68 | 4769 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 4770 | { |
0af3d00b | 4771 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 4772 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 4773 | u64 total = num_bytes; |
9078a3e1 | 4774 | u64 old_val; |
db94535d | 4775 | u64 byte_in_group; |
0af3d00b | 4776 | int factor; |
3e1ad54f | 4777 | |
5d4f98a2 YZ |
4778 | /* block accounting for super block */ |
4779 | spin_lock(&info->delalloc_lock); | |
6c41761f | 4780 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
4781 | if (alloc) |
4782 | old_val += num_bytes; | |
4783 | else | |
4784 | old_val -= num_bytes; | |
6c41761f | 4785 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
5d4f98a2 YZ |
4786 | spin_unlock(&info->delalloc_lock); |
4787 | ||
d397712b | 4788 | while (total) { |
db94535d | 4789 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 4790 | if (!cache) |
79787eaa | 4791 | return -ENOENT; |
b742bb82 YZ |
4792 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
4793 | BTRFS_BLOCK_GROUP_RAID1 | | |
4794 | BTRFS_BLOCK_GROUP_RAID10)) | |
4795 | factor = 2; | |
4796 | else | |
4797 | factor = 1; | |
9d66e233 JB |
4798 | /* |
4799 | * If this block group has free space cache written out, we | |
4800 | * need to make sure to load it if we are removing space. This | |
4801 | * is because we need the unpinning stage to actually add the | |
4802 | * space back to the block group, otherwise we will leak space. | |
4803 | */ | |
4804 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
b8399dee | 4805 | cache_block_group(cache, trans, NULL, 1); |
0af3d00b | 4806 | |
db94535d CM |
4807 | byte_in_group = bytenr - cache->key.objectid; |
4808 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 4809 | |
25179201 | 4810 | spin_lock(&cache->space_info->lock); |
c286ac48 | 4811 | spin_lock(&cache->lock); |
0af3d00b | 4812 | |
73bc1876 | 4813 | if (btrfs_test_opt(root, SPACE_CACHE) && |
0af3d00b JB |
4814 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
4815 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
4816 | ||
0f9dd46c | 4817 | cache->dirty = 1; |
9078a3e1 | 4818 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 4819 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 4820 | if (alloc) { |
db94535d | 4821 | old_val += num_bytes; |
11833d66 YZ |
4822 | btrfs_set_block_group_used(&cache->item, old_val); |
4823 | cache->reserved -= num_bytes; | |
11833d66 | 4824 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
4825 | cache->space_info->bytes_used += num_bytes; |
4826 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 4827 | spin_unlock(&cache->lock); |
25179201 | 4828 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 4829 | } else { |
db94535d | 4830 | old_val -= num_bytes; |
c286ac48 | 4831 | btrfs_set_block_group_used(&cache->item, old_val); |
f0486c68 YZ |
4832 | cache->pinned += num_bytes; |
4833 | cache->space_info->bytes_pinned += num_bytes; | |
6324fbf3 | 4834 | cache->space_info->bytes_used -= num_bytes; |
b742bb82 | 4835 | cache->space_info->disk_used -= num_bytes * factor; |
c286ac48 | 4836 | spin_unlock(&cache->lock); |
25179201 | 4837 | spin_unlock(&cache->space_info->lock); |
1f3c79a2 | 4838 | |
f0486c68 YZ |
4839 | set_extent_dirty(info->pinned_extents, |
4840 | bytenr, bytenr + num_bytes - 1, | |
4841 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 4842 | } |
fa9c0d79 | 4843 | btrfs_put_block_group(cache); |
db94535d CM |
4844 | total -= num_bytes; |
4845 | bytenr += num_bytes; | |
9078a3e1 CM |
4846 | } |
4847 | return 0; | |
4848 | } | |
6324fbf3 | 4849 | |
a061fc8d CM |
4850 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
4851 | { | |
0f9dd46c | 4852 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 4853 | u64 bytenr; |
0f9dd46c JB |
4854 | |
4855 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); | |
4856 | if (!cache) | |
a061fc8d | 4857 | return 0; |
0f9dd46c | 4858 | |
d2fb3437 | 4859 | bytenr = cache->key.objectid; |
fa9c0d79 | 4860 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
4861 | |
4862 | return bytenr; | |
a061fc8d CM |
4863 | } |
4864 | ||
f0486c68 YZ |
4865 | static int pin_down_extent(struct btrfs_root *root, |
4866 | struct btrfs_block_group_cache *cache, | |
4867 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 4868 | { |
11833d66 YZ |
4869 | spin_lock(&cache->space_info->lock); |
4870 | spin_lock(&cache->lock); | |
4871 | cache->pinned += num_bytes; | |
4872 | cache->space_info->bytes_pinned += num_bytes; | |
4873 | if (reserved) { | |
4874 | cache->reserved -= num_bytes; | |
4875 | cache->space_info->bytes_reserved -= num_bytes; | |
4876 | } | |
4877 | spin_unlock(&cache->lock); | |
4878 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 4879 | |
f0486c68 YZ |
4880 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
4881 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
4882 | return 0; | |
4883 | } | |
68b38550 | 4884 | |
f0486c68 YZ |
4885 | /* |
4886 | * this function must be called within transaction | |
4887 | */ | |
4888 | int btrfs_pin_extent(struct btrfs_root *root, | |
4889 | u64 bytenr, u64 num_bytes, int reserved) | |
4890 | { | |
4891 | struct btrfs_block_group_cache *cache; | |
68b38550 | 4892 | |
f0486c68 | 4893 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
79787eaa | 4894 | BUG_ON(!cache); /* Logic error */ |
f0486c68 YZ |
4895 | |
4896 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
4897 | ||
4898 | btrfs_put_block_group(cache); | |
11833d66 YZ |
4899 | return 0; |
4900 | } | |
4901 | ||
f0486c68 | 4902 | /* |
e688b725 CM |
4903 | * this function must be called within transaction |
4904 | */ | |
4905 | int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans, | |
4906 | struct btrfs_root *root, | |
4907 | u64 bytenr, u64 num_bytes) | |
4908 | { | |
4909 | struct btrfs_block_group_cache *cache; | |
4910 | ||
4911 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
79787eaa | 4912 | BUG_ON(!cache); /* Logic error */ |
e688b725 CM |
4913 | |
4914 | /* | |
4915 | * pull in the free space cache (if any) so that our pin | |
4916 | * removes the free space from the cache. We have load_only set | |
4917 | * to one because the slow code to read in the free extents does check | |
4918 | * the pinned extents. | |
4919 | */ | |
4920 | cache_block_group(cache, trans, root, 1); | |
4921 | ||
4922 | pin_down_extent(root, cache, bytenr, num_bytes, 0); | |
4923 | ||
4924 | /* remove us from the free space cache (if we're there at all) */ | |
4925 | btrfs_remove_free_space(cache, bytenr, num_bytes); | |
4926 | btrfs_put_block_group(cache); | |
4927 | return 0; | |
4928 | } | |
4929 | ||
fb25e914 JB |
4930 | /** |
4931 | * btrfs_update_reserved_bytes - update the block_group and space info counters | |
4932 | * @cache: The cache we are manipulating | |
4933 | * @num_bytes: The number of bytes in question | |
4934 | * @reserve: One of the reservation enums | |
4935 | * | |
4936 | * This is called by the allocator when it reserves space, or by somebody who is | |
4937 | * freeing space that was never actually used on disk. For example if you | |
4938 | * reserve some space for a new leaf in transaction A and before transaction A | |
4939 | * commits you free that leaf, you call this with reserve set to 0 in order to | |
4940 | * clear the reservation. | |
4941 | * | |
4942 | * Metadata reservations should be called with RESERVE_ALLOC so we do the proper | |
4943 | * ENOSPC accounting. For data we handle the reservation through clearing the | |
4944 | * delalloc bits in the io_tree. We have to do this since we could end up | |
4945 | * allocating less disk space for the amount of data we have reserved in the | |
4946 | * case of compression. | |
4947 | * | |
4948 | * If this is a reservation and the block group has become read only we cannot | |
4949 | * make the reservation and return -EAGAIN, otherwise this function always | |
4950 | * succeeds. | |
f0486c68 | 4951 | */ |
fb25e914 JB |
4952 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
4953 | u64 num_bytes, int reserve) | |
11833d66 | 4954 | { |
fb25e914 | 4955 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 4956 | int ret = 0; |
79787eaa | 4957 | |
fb25e914 JB |
4958 | spin_lock(&space_info->lock); |
4959 | spin_lock(&cache->lock); | |
4960 | if (reserve != RESERVE_FREE) { | |
f0486c68 YZ |
4961 | if (cache->ro) { |
4962 | ret = -EAGAIN; | |
4963 | } else { | |
fb25e914 JB |
4964 | cache->reserved += num_bytes; |
4965 | space_info->bytes_reserved += num_bytes; | |
4966 | if (reserve == RESERVE_ALLOC) { | |
8c2a3ca2 | 4967 | trace_btrfs_space_reservation(cache->fs_info, |
2bcc0328 LB |
4968 | "space_info", space_info->flags, |
4969 | num_bytes, 0); | |
fb25e914 JB |
4970 | space_info->bytes_may_use -= num_bytes; |
4971 | } | |
f0486c68 | 4972 | } |
fb25e914 JB |
4973 | } else { |
4974 | if (cache->ro) | |
4975 | space_info->bytes_readonly += num_bytes; | |
4976 | cache->reserved -= num_bytes; | |
4977 | space_info->bytes_reserved -= num_bytes; | |
4978 | space_info->reservation_progress++; | |
324ae4df | 4979 | } |
fb25e914 JB |
4980 | spin_unlock(&cache->lock); |
4981 | spin_unlock(&space_info->lock); | |
f0486c68 | 4982 | return ret; |
324ae4df | 4983 | } |
9078a3e1 | 4984 | |
143bede5 | 4985 | void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
11833d66 | 4986 | struct btrfs_root *root) |
e8569813 | 4987 | { |
e8569813 | 4988 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
4989 | struct btrfs_caching_control *next; |
4990 | struct btrfs_caching_control *caching_ctl; | |
4991 | struct btrfs_block_group_cache *cache; | |
e8569813 | 4992 | |
11833d66 | 4993 | down_write(&fs_info->extent_commit_sem); |
25179201 | 4994 | |
11833d66 YZ |
4995 | list_for_each_entry_safe(caching_ctl, next, |
4996 | &fs_info->caching_block_groups, list) { | |
4997 | cache = caching_ctl->block_group; | |
4998 | if (block_group_cache_done(cache)) { | |
4999 | cache->last_byte_to_unpin = (u64)-1; | |
5000 | list_del_init(&caching_ctl->list); | |
5001 | put_caching_control(caching_ctl); | |
e8569813 | 5002 | } else { |
11833d66 | 5003 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 5004 | } |
e8569813 | 5005 | } |
11833d66 YZ |
5006 | |
5007 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
5008 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
5009 | else | |
5010 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
5011 | ||
5012 | up_write(&fs_info->extent_commit_sem); | |
8929ecfa YZ |
5013 | |
5014 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
5015 | } |
5016 | ||
11833d66 | 5017 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
ccd467d6 | 5018 | { |
11833d66 YZ |
5019 | struct btrfs_fs_info *fs_info = root->fs_info; |
5020 | struct btrfs_block_group_cache *cache = NULL; | |
7b398f8e JB |
5021 | struct btrfs_space_info *space_info; |
5022 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
11833d66 | 5023 | u64 len; |
7b398f8e | 5024 | bool readonly; |
ccd467d6 | 5025 | |
11833d66 | 5026 | while (start <= end) { |
7b398f8e | 5027 | readonly = false; |
11833d66 YZ |
5028 | if (!cache || |
5029 | start >= cache->key.objectid + cache->key.offset) { | |
5030 | if (cache) | |
5031 | btrfs_put_block_group(cache); | |
5032 | cache = btrfs_lookup_block_group(fs_info, start); | |
79787eaa | 5033 | BUG_ON(!cache); /* Logic error */ |
11833d66 YZ |
5034 | } |
5035 | ||
5036 | len = cache->key.objectid + cache->key.offset - start; | |
5037 | len = min(len, end + 1 - start); | |
5038 | ||
5039 | if (start < cache->last_byte_to_unpin) { | |
5040 | len = min(len, cache->last_byte_to_unpin - start); | |
5041 | btrfs_add_free_space(cache, start, len); | |
5042 | } | |
5043 | ||
f0486c68 | 5044 | start += len; |
7b398f8e | 5045 | space_info = cache->space_info; |
f0486c68 | 5046 | |
7b398f8e | 5047 | spin_lock(&space_info->lock); |
11833d66 YZ |
5048 | spin_lock(&cache->lock); |
5049 | cache->pinned -= len; | |
7b398f8e JB |
5050 | space_info->bytes_pinned -= len; |
5051 | if (cache->ro) { | |
5052 | space_info->bytes_readonly += len; | |
5053 | readonly = true; | |
5054 | } | |
11833d66 | 5055 | spin_unlock(&cache->lock); |
7b398f8e JB |
5056 | if (!readonly && global_rsv->space_info == space_info) { |
5057 | spin_lock(&global_rsv->lock); | |
5058 | if (!global_rsv->full) { | |
5059 | len = min(len, global_rsv->size - | |
5060 | global_rsv->reserved); | |
5061 | global_rsv->reserved += len; | |
5062 | space_info->bytes_may_use += len; | |
5063 | if (global_rsv->reserved >= global_rsv->size) | |
5064 | global_rsv->full = 1; | |
5065 | } | |
5066 | spin_unlock(&global_rsv->lock); | |
5067 | } | |
5068 | spin_unlock(&space_info->lock); | |
ccd467d6 | 5069 | } |
11833d66 YZ |
5070 | |
5071 | if (cache) | |
5072 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
5073 | return 0; |
5074 | } | |
5075 | ||
5076 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 5077 | struct btrfs_root *root) |
a28ec197 | 5078 | { |
11833d66 YZ |
5079 | struct btrfs_fs_info *fs_info = root->fs_info; |
5080 | struct extent_io_tree *unpin; | |
1a5bc167 CM |
5081 | u64 start; |
5082 | u64 end; | |
a28ec197 | 5083 | int ret; |
a28ec197 | 5084 | |
79787eaa JM |
5085 | if (trans->aborted) |
5086 | return 0; | |
5087 | ||
11833d66 YZ |
5088 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
5089 | unpin = &fs_info->freed_extents[1]; | |
5090 | else | |
5091 | unpin = &fs_info->freed_extents[0]; | |
5092 | ||
d397712b | 5093 | while (1) { |
1a5bc167 | 5094 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 5095 | EXTENT_DIRTY, NULL); |
1a5bc167 | 5096 | if (ret) |
a28ec197 | 5097 | break; |
1f3c79a2 | 5098 | |
5378e607 LD |
5099 | if (btrfs_test_opt(root, DISCARD)) |
5100 | ret = btrfs_discard_extent(root, start, | |
5101 | end + 1 - start, NULL); | |
1f3c79a2 | 5102 | |
1a5bc167 | 5103 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
11833d66 | 5104 | unpin_extent_range(root, start, end); |
b9473439 | 5105 | cond_resched(); |
a28ec197 | 5106 | } |
817d52f8 | 5107 | |
e20d96d6 CM |
5108 | return 0; |
5109 | } | |
5110 | ||
5d4f98a2 YZ |
5111 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
5112 | struct btrfs_root *root, | |
5113 | u64 bytenr, u64 num_bytes, u64 parent, | |
5114 | u64 root_objectid, u64 owner_objectid, | |
5115 | u64 owner_offset, int refs_to_drop, | |
5116 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 5117 | { |
e2fa7227 | 5118 | struct btrfs_key key; |
5d4f98a2 | 5119 | struct btrfs_path *path; |
1261ec42 CM |
5120 | struct btrfs_fs_info *info = root->fs_info; |
5121 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 5122 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
5123 | struct btrfs_extent_item *ei; |
5124 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 5125 | int ret; |
5d4f98a2 | 5126 | int is_data; |
952fccac CM |
5127 | int extent_slot = 0; |
5128 | int found_extent = 0; | |
5129 | int num_to_del = 1; | |
5d4f98a2 YZ |
5130 | u32 item_size; |
5131 | u64 refs; | |
037e6390 | 5132 | |
5caf2a00 | 5133 | path = btrfs_alloc_path(); |
54aa1f4d CM |
5134 | if (!path) |
5135 | return -ENOMEM; | |
5f26f772 | 5136 | |
3c12ac72 | 5137 | path->reada = 1; |
b9473439 | 5138 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5139 | |
5140 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
5141 | BUG_ON(!is_data && refs_to_drop != 1); | |
5142 | ||
5143 | ret = lookup_extent_backref(trans, extent_root, path, &iref, | |
5144 | bytenr, num_bytes, parent, | |
5145 | root_objectid, owner_objectid, | |
5146 | owner_offset); | |
7bb86316 | 5147 | if (ret == 0) { |
952fccac | 5148 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
5149 | while (extent_slot >= 0) { |
5150 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 5151 | extent_slot); |
5d4f98a2 | 5152 | if (key.objectid != bytenr) |
952fccac | 5153 | break; |
5d4f98a2 YZ |
5154 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
5155 | key.offset == num_bytes) { | |
952fccac CM |
5156 | found_extent = 1; |
5157 | break; | |
5158 | } | |
5159 | if (path->slots[0] - extent_slot > 5) | |
5160 | break; | |
5d4f98a2 | 5161 | extent_slot--; |
952fccac | 5162 | } |
5d4f98a2 YZ |
5163 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
5164 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
5165 | if (found_extent && item_size < sizeof(*ei)) | |
5166 | found_extent = 0; | |
5167 | #endif | |
31840ae1 | 5168 | if (!found_extent) { |
5d4f98a2 | 5169 | BUG_ON(iref); |
56bec294 | 5170 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
5171 | NULL, refs_to_drop, |
5172 | is_data); | |
005d6427 DS |
5173 | if (ret) { |
5174 | btrfs_abort_transaction(trans, extent_root, ret); | |
5175 | goto out; | |
5176 | } | |
b3b4aa74 | 5177 | btrfs_release_path(path); |
b9473439 | 5178 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5179 | |
5180 | key.objectid = bytenr; | |
5181 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5182 | key.offset = num_bytes; | |
5183 | ||
31840ae1 ZY |
5184 | ret = btrfs_search_slot(trans, extent_root, |
5185 | &key, path, -1, 1); | |
f3465ca4 JB |
5186 | if (ret) { |
5187 | printk(KERN_ERR "umm, got %d back from search" | |
d397712b CM |
5188 | ", was looking for %llu\n", ret, |
5189 | (unsigned long long)bytenr); | |
b783e62d JB |
5190 | if (ret > 0) |
5191 | btrfs_print_leaf(extent_root, | |
5192 | path->nodes[0]); | |
f3465ca4 | 5193 | } |
005d6427 DS |
5194 | if (ret < 0) { |
5195 | btrfs_abort_transaction(trans, extent_root, ret); | |
5196 | goto out; | |
5197 | } | |
31840ae1 ZY |
5198 | extent_slot = path->slots[0]; |
5199 | } | |
79787eaa | 5200 | } else if (ret == -ENOENT) { |
7bb86316 CM |
5201 | btrfs_print_leaf(extent_root, path->nodes[0]); |
5202 | WARN_ON(1); | |
d397712b | 5203 | printk(KERN_ERR "btrfs unable to find ref byte nr %llu " |
5d4f98a2 | 5204 | "parent %llu root %llu owner %llu offset %llu\n", |
d397712b | 5205 | (unsigned long long)bytenr, |
56bec294 | 5206 | (unsigned long long)parent, |
d397712b | 5207 | (unsigned long long)root_objectid, |
5d4f98a2 YZ |
5208 | (unsigned long long)owner_objectid, |
5209 | (unsigned long long)owner_offset); | |
79787eaa | 5210 | } else { |
005d6427 DS |
5211 | btrfs_abort_transaction(trans, extent_root, ret); |
5212 | goto out; | |
7bb86316 | 5213 | } |
5f39d397 CM |
5214 | |
5215 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
5216 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
5217 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
5218 | if (item_size < sizeof(*ei)) { | |
5219 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
5220 | ret = convert_extent_item_v0(trans, extent_root, path, | |
5221 | owner_objectid, 0); | |
005d6427 DS |
5222 | if (ret < 0) { |
5223 | btrfs_abort_transaction(trans, extent_root, ret); | |
5224 | goto out; | |
5225 | } | |
5d4f98a2 | 5226 | |
b3b4aa74 | 5227 | btrfs_release_path(path); |
5d4f98a2 YZ |
5228 | path->leave_spinning = 1; |
5229 | ||
5230 | key.objectid = bytenr; | |
5231 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5232 | key.offset = num_bytes; | |
5233 | ||
5234 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
5235 | -1, 1); | |
5236 | if (ret) { | |
5237 | printk(KERN_ERR "umm, got %d back from search" | |
5238 | ", was looking for %llu\n", ret, | |
5239 | (unsigned long long)bytenr); | |
5240 | btrfs_print_leaf(extent_root, path->nodes[0]); | |
5241 | } | |
005d6427 DS |
5242 | if (ret < 0) { |
5243 | btrfs_abort_transaction(trans, extent_root, ret); | |
5244 | goto out; | |
5245 | } | |
5246 | ||
5d4f98a2 YZ |
5247 | extent_slot = path->slots[0]; |
5248 | leaf = path->nodes[0]; | |
5249 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
5250 | } | |
5251 | #endif | |
5252 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 5253 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 5254 | struct btrfs_extent_item); |
5d4f98a2 YZ |
5255 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { |
5256 | struct btrfs_tree_block_info *bi; | |
5257 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
5258 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
5259 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
5260 | } | |
56bec294 | 5261 | |
5d4f98a2 | 5262 | refs = btrfs_extent_refs(leaf, ei); |
56bec294 CM |
5263 | BUG_ON(refs < refs_to_drop); |
5264 | refs -= refs_to_drop; | |
5f39d397 | 5265 | |
5d4f98a2 YZ |
5266 | if (refs > 0) { |
5267 | if (extent_op) | |
5268 | __run_delayed_extent_op(extent_op, leaf, ei); | |
5269 | /* | |
5270 | * In the case of inline back ref, reference count will | |
5271 | * be updated by remove_extent_backref | |
952fccac | 5272 | */ |
5d4f98a2 YZ |
5273 | if (iref) { |
5274 | BUG_ON(!found_extent); | |
5275 | } else { | |
5276 | btrfs_set_extent_refs(leaf, ei, refs); | |
5277 | btrfs_mark_buffer_dirty(leaf); | |
5278 | } | |
5279 | if (found_extent) { | |
5280 | ret = remove_extent_backref(trans, extent_root, path, | |
5281 | iref, refs_to_drop, | |
5282 | is_data); | |
005d6427 DS |
5283 | if (ret) { |
5284 | btrfs_abort_transaction(trans, extent_root, ret); | |
5285 | goto out; | |
5286 | } | |
952fccac | 5287 | } |
5d4f98a2 | 5288 | } else { |
5d4f98a2 YZ |
5289 | if (found_extent) { |
5290 | BUG_ON(is_data && refs_to_drop != | |
5291 | extent_data_ref_count(root, path, iref)); | |
5292 | if (iref) { | |
5293 | BUG_ON(path->slots[0] != extent_slot); | |
5294 | } else { | |
5295 | BUG_ON(path->slots[0] != extent_slot + 1); | |
5296 | path->slots[0] = extent_slot; | |
5297 | num_to_del = 2; | |
5298 | } | |
78fae27e | 5299 | } |
b9473439 | 5300 | |
952fccac CM |
5301 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
5302 | num_to_del); | |
005d6427 DS |
5303 | if (ret) { |
5304 | btrfs_abort_transaction(trans, extent_root, ret); | |
5305 | goto out; | |
5306 | } | |
b3b4aa74 | 5307 | btrfs_release_path(path); |
21af804c | 5308 | |
5d4f98a2 | 5309 | if (is_data) { |
459931ec | 5310 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
005d6427 DS |
5311 | if (ret) { |
5312 | btrfs_abort_transaction(trans, extent_root, ret); | |
5313 | goto out; | |
5314 | } | |
459931ec CM |
5315 | } |
5316 | ||
f0486c68 | 5317 | ret = update_block_group(trans, root, bytenr, num_bytes, 0); |
005d6427 DS |
5318 | if (ret) { |
5319 | btrfs_abort_transaction(trans, extent_root, ret); | |
5320 | goto out; | |
5321 | } | |
a28ec197 | 5322 | } |
79787eaa | 5323 | out: |
5caf2a00 | 5324 | btrfs_free_path(path); |
a28ec197 CM |
5325 | return ret; |
5326 | } | |
5327 | ||
1887be66 | 5328 | /* |
f0486c68 | 5329 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
5330 | * delayed ref for that extent as well. This searches the delayed ref tree for |
5331 | * a given extent, and if there are no other delayed refs to be processed, it | |
5332 | * removes it from the tree. | |
5333 | */ | |
5334 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
5335 | struct btrfs_root *root, u64 bytenr) | |
5336 | { | |
5337 | struct btrfs_delayed_ref_head *head; | |
5338 | struct btrfs_delayed_ref_root *delayed_refs; | |
5339 | struct btrfs_delayed_ref_node *ref; | |
5340 | struct rb_node *node; | |
f0486c68 | 5341 | int ret = 0; |
1887be66 CM |
5342 | |
5343 | delayed_refs = &trans->transaction->delayed_refs; | |
5344 | spin_lock(&delayed_refs->lock); | |
5345 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
5346 | if (!head) | |
5347 | goto out; | |
5348 | ||
5349 | node = rb_prev(&head->node.rb_node); | |
5350 | if (!node) | |
5351 | goto out; | |
5352 | ||
5353 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
5354 | ||
5355 | /* there are still entries for this ref, we can't drop it */ | |
5356 | if (ref->bytenr == bytenr) | |
5357 | goto out; | |
5358 | ||
5d4f98a2 YZ |
5359 | if (head->extent_op) { |
5360 | if (!head->must_insert_reserved) | |
5361 | goto out; | |
78a6184a | 5362 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
5363 | head->extent_op = NULL; |
5364 | } | |
5365 | ||
1887be66 CM |
5366 | /* |
5367 | * waiting for the lock here would deadlock. If someone else has it | |
5368 | * locked they are already in the process of dropping it anyway | |
5369 | */ | |
5370 | if (!mutex_trylock(&head->mutex)) | |
5371 | goto out; | |
5372 | ||
5373 | /* | |
5374 | * at this point we have a head with no other entries. Go | |
5375 | * ahead and process it. | |
5376 | */ | |
5377 | head->node.in_tree = 0; | |
5378 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 5379 | |
1887be66 CM |
5380 | delayed_refs->num_entries--; |
5381 | ||
5382 | /* | |
5383 | * we don't take a ref on the node because we're removing it from the | |
5384 | * tree, so we just steal the ref the tree was holding. | |
5385 | */ | |
c3e69d58 CM |
5386 | delayed_refs->num_heads--; |
5387 | if (list_empty(&head->cluster)) | |
5388 | delayed_refs->num_heads_ready--; | |
5389 | ||
5390 | list_del_init(&head->cluster); | |
1887be66 CM |
5391 | spin_unlock(&delayed_refs->lock); |
5392 | ||
f0486c68 YZ |
5393 | BUG_ON(head->extent_op); |
5394 | if (head->must_insert_reserved) | |
5395 | ret = 1; | |
5396 | ||
5397 | mutex_unlock(&head->mutex); | |
1887be66 | 5398 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 5399 | return ret; |
1887be66 CM |
5400 | out: |
5401 | spin_unlock(&delayed_refs->lock); | |
5402 | return 0; | |
5403 | } | |
5404 | ||
f0486c68 YZ |
5405 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
5406 | struct btrfs_root *root, | |
5407 | struct extent_buffer *buf, | |
5581a51a | 5408 | u64 parent, int last_ref) |
f0486c68 | 5409 | { |
f0486c68 YZ |
5410 | struct btrfs_block_group_cache *cache = NULL; |
5411 | int ret; | |
5412 | ||
5413 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
66d7e7f0 AJ |
5414 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
5415 | buf->start, buf->len, | |
5416 | parent, root->root_key.objectid, | |
5417 | btrfs_header_level(buf), | |
5581a51a | 5418 | BTRFS_DROP_DELAYED_REF, NULL, 0); |
79787eaa | 5419 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
5420 | } |
5421 | ||
5422 | if (!last_ref) | |
5423 | return; | |
5424 | ||
f0486c68 | 5425 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); |
f0486c68 YZ |
5426 | |
5427 | if (btrfs_header_generation(buf) == trans->transid) { | |
5428 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
5429 | ret = check_ref_cleanup(trans, root, buf->start); | |
5430 | if (!ret) | |
37be25bc | 5431 | goto out; |
f0486c68 YZ |
5432 | } |
5433 | ||
5434 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
5435 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
37be25bc | 5436 | goto out; |
f0486c68 YZ |
5437 | } |
5438 | ||
5439 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
5440 | ||
5441 | btrfs_add_free_space(cache, buf->start, buf->len); | |
fb25e914 | 5442 | btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE); |
f0486c68 YZ |
5443 | } |
5444 | out: | |
a826d6dc JB |
5445 | /* |
5446 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
5447 | * anymore. | |
5448 | */ | |
5449 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
5450 | btrfs_put_block_group(cache); |
5451 | } | |
5452 | ||
79787eaa | 5453 | /* Can return -ENOMEM */ |
66d7e7f0 AJ |
5454 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
5455 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, | |
5456 | u64 owner, u64 offset, int for_cow) | |
925baedd CM |
5457 | { |
5458 | int ret; | |
66d7e7f0 | 5459 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 5460 | |
56bec294 CM |
5461 | /* |
5462 | * tree log blocks never actually go into the extent allocation | |
5463 | * tree, just update pinning info and exit early. | |
56bec294 | 5464 | */ |
5d4f98a2 YZ |
5465 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
5466 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 5467 | /* unlocks the pinned mutex */ |
11833d66 | 5468 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 5469 | ret = 0; |
5d4f98a2 | 5470 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
5471 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
5472 | num_bytes, | |
5d4f98a2 | 5473 | parent, root_objectid, (int)owner, |
66d7e7f0 | 5474 | BTRFS_DROP_DELAYED_REF, NULL, for_cow); |
5d4f98a2 | 5475 | } else { |
66d7e7f0 AJ |
5476 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
5477 | num_bytes, | |
5478 | parent, root_objectid, owner, | |
5479 | offset, BTRFS_DROP_DELAYED_REF, | |
5480 | NULL, for_cow); | |
56bec294 | 5481 | } |
925baedd CM |
5482 | return ret; |
5483 | } | |
5484 | ||
87ee04eb CM |
5485 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
5486 | { | |
5487 | u64 mask = ((u64)root->stripesize - 1); | |
5488 | u64 ret = (val + mask) & ~mask; | |
5489 | return ret; | |
5490 | } | |
5491 | ||
817d52f8 JB |
5492 | /* |
5493 | * when we wait for progress in the block group caching, its because | |
5494 | * our allocation attempt failed at least once. So, we must sleep | |
5495 | * and let some progress happen before we try again. | |
5496 | * | |
5497 | * This function will sleep at least once waiting for new free space to | |
5498 | * show up, and then it will check the block group free space numbers | |
5499 | * for our min num_bytes. Another option is to have it go ahead | |
5500 | * and look in the rbtree for a free extent of a given size, but this | |
5501 | * is a good start. | |
5502 | */ | |
5503 | static noinline int | |
5504 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
5505 | u64 num_bytes) | |
5506 | { | |
11833d66 | 5507 | struct btrfs_caching_control *caching_ctl; |
817d52f8 JB |
5508 | DEFINE_WAIT(wait); |
5509 | ||
11833d66 YZ |
5510 | caching_ctl = get_caching_control(cache); |
5511 | if (!caching_ctl) | |
817d52f8 | 5512 | return 0; |
817d52f8 | 5513 | |
11833d66 | 5514 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 5515 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
5516 | |
5517 | put_caching_control(caching_ctl); | |
5518 | return 0; | |
5519 | } | |
5520 | ||
5521 | static noinline int | |
5522 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
5523 | { | |
5524 | struct btrfs_caching_control *caching_ctl; | |
5525 | DEFINE_WAIT(wait); | |
5526 | ||
5527 | caching_ctl = get_caching_control(cache); | |
5528 | if (!caching_ctl) | |
5529 | return 0; | |
5530 | ||
5531 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
5532 | ||
5533 | put_caching_control(caching_ctl); | |
817d52f8 JB |
5534 | return 0; |
5535 | } | |
5536 | ||
31e50229 | 5537 | int __get_raid_index(u64 flags) |
b742bb82 YZ |
5538 | { |
5539 | int index; | |
7738a53a ID |
5540 | |
5541 | if (flags & BTRFS_BLOCK_GROUP_RAID10) | |
b742bb82 | 5542 | index = 0; |
7738a53a | 5543 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
b742bb82 | 5544 | index = 1; |
7738a53a | 5545 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
b742bb82 | 5546 | index = 2; |
7738a53a | 5547 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
b742bb82 YZ |
5548 | index = 3; |
5549 | else | |
5550 | index = 4; | |
7738a53a | 5551 | |
b742bb82 YZ |
5552 | return index; |
5553 | } | |
5554 | ||
7738a53a ID |
5555 | static int get_block_group_index(struct btrfs_block_group_cache *cache) |
5556 | { | |
31e50229 | 5557 | return __get_raid_index(cache->flags); |
7738a53a ID |
5558 | } |
5559 | ||
817d52f8 | 5560 | enum btrfs_loop_type { |
285ff5af JB |
5561 | LOOP_CACHING_NOWAIT = 0, |
5562 | LOOP_CACHING_WAIT = 1, | |
5563 | LOOP_ALLOC_CHUNK = 2, | |
5564 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
5565 | }; |
5566 | ||
fec577fb CM |
5567 | /* |
5568 | * walks the btree of allocated extents and find a hole of a given size. | |
5569 | * The key ins is changed to record the hole: | |
5570 | * ins->objectid == block start | |
62e2749e | 5571 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
5572 | * ins->offset == number of blocks |
5573 | * Any available blocks before search_start are skipped. | |
5574 | */ | |
d397712b | 5575 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
5576 | struct btrfs_root *orig_root, |
5577 | u64 num_bytes, u64 empty_size, | |
98ed5174 | 5578 | u64 hint_byte, struct btrfs_key *ins, |
e0f54067 | 5579 | u64 data) |
fec577fb | 5580 | { |
80eb234a | 5581 | int ret = 0; |
d397712b | 5582 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 5583 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 5584 | struct btrfs_block_group_cache *block_group = NULL; |
274bd4fb | 5585 | struct btrfs_block_group_cache *used_block_group; |
81c9ad23 | 5586 | u64 search_start = 0; |
239b14b3 | 5587 | int empty_cluster = 2 * 1024 * 1024; |
80eb234a | 5588 | struct btrfs_space_info *space_info; |
fa9c0d79 | 5589 | int loop = 0; |
ac5c9300 | 5590 | int index = __get_raid_index(data); |
fb25e914 JB |
5591 | int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ? |
5592 | RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC; | |
817d52f8 | 5593 | bool found_uncached_bg = false; |
0a24325e | 5594 | bool failed_cluster_refill = false; |
1cdda9b8 | 5595 | bool failed_alloc = false; |
67377734 | 5596 | bool use_cluster = true; |
60d2adbb | 5597 | bool have_caching_bg = false; |
fec577fb | 5598 | |
db94535d | 5599 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 5600 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
5601 | ins->objectid = 0; |
5602 | ins->offset = 0; | |
b1a4d965 | 5603 | |
3f7de037 JB |
5604 | trace_find_free_extent(orig_root, num_bytes, empty_size, data); |
5605 | ||
2552d17e | 5606 | space_info = __find_space_info(root->fs_info, data); |
1b1d1f66 | 5607 | if (!space_info) { |
e0f54067 | 5608 | printk(KERN_ERR "No space info for %llu\n", data); |
1b1d1f66 JB |
5609 | return -ENOSPC; |
5610 | } | |
2552d17e | 5611 | |
67377734 JB |
5612 | /* |
5613 | * If the space info is for both data and metadata it means we have a | |
5614 | * small filesystem and we can't use the clustering stuff. | |
5615 | */ | |
5616 | if (btrfs_mixed_space_info(space_info)) | |
5617 | use_cluster = false; | |
5618 | ||
67377734 | 5619 | if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) { |
fa9c0d79 | 5620 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
5621 | if (!btrfs_test_opt(root, SSD)) |
5622 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
5623 | } |
5624 | ||
67377734 JB |
5625 | if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster && |
5626 | btrfs_test_opt(root, SSD)) { | |
fa9c0d79 CM |
5627 | last_ptr = &root->fs_info->data_alloc_cluster; |
5628 | } | |
0f9dd46c | 5629 | |
239b14b3 | 5630 | if (last_ptr) { |
fa9c0d79 CM |
5631 | spin_lock(&last_ptr->lock); |
5632 | if (last_ptr->block_group) | |
5633 | hint_byte = last_ptr->window_start; | |
5634 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 5635 | } |
fa9c0d79 | 5636 | |
a061fc8d | 5637 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 5638 | search_start = max(search_start, hint_byte); |
0b86a832 | 5639 | |
817d52f8 | 5640 | if (!last_ptr) |
fa9c0d79 | 5641 | empty_cluster = 0; |
fa9c0d79 | 5642 | |
2552d17e | 5643 | if (search_start == hint_byte) { |
2552d17e JB |
5644 | block_group = btrfs_lookup_block_group(root->fs_info, |
5645 | search_start); | |
274bd4fb | 5646 | used_block_group = block_group; |
817d52f8 JB |
5647 | /* |
5648 | * we don't want to use the block group if it doesn't match our | |
5649 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
5650 | * |
5651 | * However if we are re-searching with an ideal block group | |
5652 | * picked out then we don't care that the block group is cached. | |
817d52f8 JB |
5653 | */ |
5654 | if (block_group && block_group_bits(block_group, data) && | |
285ff5af | 5655 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 5656 | down_read(&space_info->groups_sem); |
44fb5511 CM |
5657 | if (list_empty(&block_group->list) || |
5658 | block_group->ro) { | |
5659 | /* | |
5660 | * someone is removing this block group, | |
5661 | * we can't jump into the have_block_group | |
5662 | * target because our list pointers are not | |
5663 | * valid | |
5664 | */ | |
5665 | btrfs_put_block_group(block_group); | |
5666 | up_read(&space_info->groups_sem); | |
ccf0e725 | 5667 | } else { |
b742bb82 | 5668 | index = get_block_group_index(block_group); |
44fb5511 | 5669 | goto have_block_group; |
ccf0e725 | 5670 | } |
2552d17e | 5671 | } else if (block_group) { |
fa9c0d79 | 5672 | btrfs_put_block_group(block_group); |
2552d17e | 5673 | } |
42e70e7a | 5674 | } |
2552d17e | 5675 | search: |
60d2adbb | 5676 | have_caching_bg = false; |
80eb234a | 5677 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
5678 | list_for_each_entry(block_group, &space_info->block_groups[index], |
5679 | list) { | |
6226cb0a | 5680 | u64 offset; |
817d52f8 | 5681 | int cached; |
8a1413a2 | 5682 | |
274bd4fb | 5683 | used_block_group = block_group; |
11dfe35a | 5684 | btrfs_get_block_group(block_group); |
2552d17e | 5685 | search_start = block_group->key.objectid; |
42e70e7a | 5686 | |
83a50de9 CM |
5687 | /* |
5688 | * this can happen if we end up cycling through all the | |
5689 | * raid types, but we want to make sure we only allocate | |
5690 | * for the proper type. | |
5691 | */ | |
5692 | if (!block_group_bits(block_group, data)) { | |
5693 | u64 extra = BTRFS_BLOCK_GROUP_DUP | | |
5694 | BTRFS_BLOCK_GROUP_RAID1 | | |
5695 | BTRFS_BLOCK_GROUP_RAID10; | |
5696 | ||
5697 | /* | |
5698 | * if they asked for extra copies and this block group | |
5699 | * doesn't provide them, bail. This does allow us to | |
5700 | * fill raid0 from raid1. | |
5701 | */ | |
5702 | if ((data & extra) && !(block_group->flags & extra)) | |
5703 | goto loop; | |
5704 | } | |
5705 | ||
2552d17e | 5706 | have_block_group: |
291c7d2f JB |
5707 | cached = block_group_cache_done(block_group); |
5708 | if (unlikely(!cached)) { | |
291c7d2f | 5709 | found_uncached_bg = true; |
b8399dee | 5710 | ret = cache_block_group(block_group, trans, |
285ff5af | 5711 | orig_root, 0); |
1d4284bd CM |
5712 | BUG_ON(ret < 0); |
5713 | ret = 0; | |
817d52f8 JB |
5714 | } |
5715 | ||
ea6a478e | 5716 | if (unlikely(block_group->ro)) |
2552d17e | 5717 | goto loop; |
0f9dd46c | 5718 | |
0a24325e | 5719 | /* |
062c05c4 AO |
5720 | * Ok we want to try and use the cluster allocator, so |
5721 | * lets look there | |
0a24325e | 5722 | */ |
062c05c4 | 5723 | if (last_ptr) { |
fa9c0d79 CM |
5724 | /* |
5725 | * the refill lock keeps out other | |
5726 | * people trying to start a new cluster | |
5727 | */ | |
5728 | spin_lock(&last_ptr->refill_lock); | |
274bd4fb AO |
5729 | used_block_group = last_ptr->block_group; |
5730 | if (used_block_group != block_group && | |
5731 | (!used_block_group || | |
5732 | used_block_group->ro || | |
5733 | !block_group_bits(used_block_group, data))) { | |
5734 | used_block_group = block_group; | |
44fb5511 | 5735 | goto refill_cluster; |
274bd4fb AO |
5736 | } |
5737 | ||
5738 | if (used_block_group != block_group) | |
5739 | btrfs_get_block_group(used_block_group); | |
44fb5511 | 5740 | |
274bd4fb AO |
5741 | offset = btrfs_alloc_from_cluster(used_block_group, |
5742 | last_ptr, num_bytes, used_block_group->key.objectid); | |
fa9c0d79 CM |
5743 | if (offset) { |
5744 | /* we have a block, we're done */ | |
5745 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
5746 | trace_btrfs_reserve_extent_cluster(root, |
5747 | block_group, search_start, num_bytes); | |
fa9c0d79 CM |
5748 | goto checks; |
5749 | } | |
5750 | ||
274bd4fb AO |
5751 | WARN_ON(last_ptr->block_group != used_block_group); |
5752 | if (used_block_group != block_group) { | |
5753 | btrfs_put_block_group(used_block_group); | |
5754 | used_block_group = block_group; | |
fa9c0d79 | 5755 | } |
44fb5511 | 5756 | refill_cluster: |
274bd4fb | 5757 | BUG_ON(used_block_group != block_group); |
062c05c4 AO |
5758 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
5759 | * set up a new clusters, so lets just skip it | |
5760 | * and let the allocator find whatever block | |
5761 | * it can find. If we reach this point, we | |
5762 | * will have tried the cluster allocator | |
5763 | * plenty of times and not have found | |
5764 | * anything, so we are likely way too | |
5765 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
5766 | * anything. |
5767 | * | |
5768 | * However, if the cluster is taken from the | |
5769 | * current block group, release the cluster | |
5770 | * first, so that we stand a better chance of | |
5771 | * succeeding in the unclustered | |
5772 | * allocation. */ | |
5773 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
5774 | last_ptr->block_group != block_group) { | |
062c05c4 AO |
5775 | spin_unlock(&last_ptr->refill_lock); |
5776 | goto unclustered_alloc; | |
5777 | } | |
5778 | ||
fa9c0d79 CM |
5779 | /* |
5780 | * this cluster didn't work out, free it and | |
5781 | * start over | |
5782 | */ | |
5783 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
5784 | ||
a5f6f719 AO |
5785 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
5786 | spin_unlock(&last_ptr->refill_lock); | |
5787 | goto unclustered_alloc; | |
5788 | } | |
5789 | ||
fa9c0d79 | 5790 | /* allocate a cluster in this block group */ |
451d7585 | 5791 | ret = btrfs_find_space_cluster(trans, root, |
fa9c0d79 | 5792 | block_group, last_ptr, |
1b22bad7 | 5793 | search_start, num_bytes, |
fa9c0d79 CM |
5794 | empty_cluster + empty_size); |
5795 | if (ret == 0) { | |
5796 | /* | |
5797 | * now pull our allocation out of this | |
5798 | * cluster | |
5799 | */ | |
5800 | offset = btrfs_alloc_from_cluster(block_group, | |
5801 | last_ptr, num_bytes, | |
5802 | search_start); | |
5803 | if (offset) { | |
5804 | /* we found one, proceed */ | |
5805 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
5806 | trace_btrfs_reserve_extent_cluster(root, |
5807 | block_group, search_start, | |
5808 | num_bytes); | |
fa9c0d79 CM |
5809 | goto checks; |
5810 | } | |
0a24325e JB |
5811 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
5812 | && !failed_cluster_refill) { | |
817d52f8 JB |
5813 | spin_unlock(&last_ptr->refill_lock); |
5814 | ||
0a24325e | 5815 | failed_cluster_refill = true; |
817d52f8 JB |
5816 | wait_block_group_cache_progress(block_group, |
5817 | num_bytes + empty_cluster + empty_size); | |
5818 | goto have_block_group; | |
fa9c0d79 | 5819 | } |
817d52f8 | 5820 | |
fa9c0d79 CM |
5821 | /* |
5822 | * at this point we either didn't find a cluster | |
5823 | * or we weren't able to allocate a block from our | |
5824 | * cluster. Free the cluster we've been trying | |
5825 | * to use, and go to the next block group | |
5826 | */ | |
0a24325e | 5827 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 5828 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 5829 | goto loop; |
fa9c0d79 CM |
5830 | } |
5831 | ||
062c05c4 | 5832 | unclustered_alloc: |
a5f6f719 AO |
5833 | spin_lock(&block_group->free_space_ctl->tree_lock); |
5834 | if (cached && | |
5835 | block_group->free_space_ctl->free_space < | |
5836 | num_bytes + empty_cluster + empty_size) { | |
5837 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
5838 | goto loop; | |
5839 | } | |
5840 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
5841 | ||
6226cb0a JB |
5842 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
5843 | num_bytes, empty_size); | |
1cdda9b8 JB |
5844 | /* |
5845 | * If we didn't find a chunk, and we haven't failed on this | |
5846 | * block group before, and this block group is in the middle of | |
5847 | * caching and we are ok with waiting, then go ahead and wait | |
5848 | * for progress to be made, and set failed_alloc to true. | |
5849 | * | |
5850 | * If failed_alloc is true then we've already waited on this | |
5851 | * block group once and should move on to the next block group. | |
5852 | */ | |
5853 | if (!offset && !failed_alloc && !cached && | |
5854 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 5855 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
5856 | num_bytes + empty_size); |
5857 | failed_alloc = true; | |
817d52f8 | 5858 | goto have_block_group; |
1cdda9b8 | 5859 | } else if (!offset) { |
60d2adbb MX |
5860 | if (!cached) |
5861 | have_caching_bg = true; | |
1cdda9b8 | 5862 | goto loop; |
817d52f8 | 5863 | } |
fa9c0d79 | 5864 | checks: |
6226cb0a | 5865 | search_start = stripe_align(root, offset); |
25179201 | 5866 | |
2552d17e JB |
5867 | /* move on to the next group */ |
5868 | if (search_start + num_bytes > | |
274bd4fb AO |
5869 | used_block_group->key.objectid + used_block_group->key.offset) { |
5870 | btrfs_add_free_space(used_block_group, offset, num_bytes); | |
2552d17e | 5871 | goto loop; |
6226cb0a | 5872 | } |
f5a31e16 | 5873 | |
f0486c68 | 5874 | if (offset < search_start) |
274bd4fb | 5875 | btrfs_add_free_space(used_block_group, offset, |
f0486c68 YZ |
5876 | search_start - offset); |
5877 | BUG_ON(offset > search_start); | |
2552d17e | 5878 | |
274bd4fb | 5879 | ret = btrfs_update_reserved_bytes(used_block_group, num_bytes, |
fb25e914 | 5880 | alloc_type); |
f0486c68 | 5881 | if (ret == -EAGAIN) { |
274bd4fb | 5882 | btrfs_add_free_space(used_block_group, offset, num_bytes); |
2552d17e | 5883 | goto loop; |
0f9dd46c | 5884 | } |
0b86a832 | 5885 | |
f0486c68 | 5886 | /* we are all good, lets return */ |
2552d17e JB |
5887 | ins->objectid = search_start; |
5888 | ins->offset = num_bytes; | |
d2fb3437 | 5889 | |
3f7de037 JB |
5890 | trace_btrfs_reserve_extent(orig_root, block_group, |
5891 | search_start, num_bytes); | |
274bd4fb AO |
5892 | if (used_block_group != block_group) |
5893 | btrfs_put_block_group(used_block_group); | |
d82a6f1d | 5894 | btrfs_put_block_group(block_group); |
2552d17e JB |
5895 | break; |
5896 | loop: | |
0a24325e | 5897 | failed_cluster_refill = false; |
1cdda9b8 | 5898 | failed_alloc = false; |
b742bb82 | 5899 | BUG_ON(index != get_block_group_index(block_group)); |
274bd4fb AO |
5900 | if (used_block_group != block_group) |
5901 | btrfs_put_block_group(used_block_group); | |
fa9c0d79 | 5902 | btrfs_put_block_group(block_group); |
2552d17e JB |
5903 | } |
5904 | up_read(&space_info->groups_sem); | |
5905 | ||
60d2adbb MX |
5906 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
5907 | goto search; | |
5908 | ||
b742bb82 YZ |
5909 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
5910 | goto search; | |
5911 | ||
285ff5af | 5912 | /* |
ccf0e725 JB |
5913 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
5914 | * caching kthreads as we move along | |
817d52f8 JB |
5915 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
5916 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
5917 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
5918 | * again | |
fa9c0d79 | 5919 | */ |
723bda20 | 5920 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 5921 | index = 0; |
723bda20 | 5922 | loop++; |
817d52f8 | 5923 | if (loop == LOOP_ALLOC_CHUNK) { |
698d0082 | 5924 | ret = do_chunk_alloc(trans, root, data, |
ea658bad JB |
5925 | CHUNK_ALLOC_FORCE); |
5926 | /* | |
5927 | * Do not bail out on ENOSPC since we | |
5928 | * can do more things. | |
5929 | */ | |
5930 | if (ret < 0 && ret != -ENOSPC) { | |
5931 | btrfs_abort_transaction(trans, | |
5932 | root, ret); | |
5933 | goto out; | |
723bda20 | 5934 | } |
2552d17e JB |
5935 | } |
5936 | ||
723bda20 JB |
5937 | if (loop == LOOP_NO_EMPTY_SIZE) { |
5938 | empty_size = 0; | |
5939 | empty_cluster = 0; | |
fa9c0d79 | 5940 | } |
723bda20 JB |
5941 | |
5942 | goto search; | |
2552d17e JB |
5943 | } else if (!ins->objectid) { |
5944 | ret = -ENOSPC; | |
d82a6f1d | 5945 | } else if (ins->objectid) { |
80eb234a | 5946 | ret = 0; |
be744175 | 5947 | } |
79787eaa | 5948 | out: |
be744175 | 5949 | |
0f70abe2 | 5950 | return ret; |
fec577fb | 5951 | } |
ec44a35c | 5952 | |
9ed74f2d JB |
5953 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
5954 | int dump_block_groups) | |
0f9dd46c JB |
5955 | { |
5956 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 5957 | int index = 0; |
0f9dd46c | 5958 | |
9ed74f2d | 5959 | spin_lock(&info->lock); |
fb25e914 JB |
5960 | printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n", |
5961 | (unsigned long long)info->flags, | |
d397712b | 5962 | (unsigned long long)(info->total_bytes - info->bytes_used - |
9ed74f2d | 5963 | info->bytes_pinned - info->bytes_reserved - |
8929ecfa | 5964 | info->bytes_readonly), |
d397712b | 5965 | (info->full) ? "" : "not "); |
8929ecfa YZ |
5966 | printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, " |
5967 | "reserved=%llu, may_use=%llu, readonly=%llu\n", | |
21380931 | 5968 | (unsigned long long)info->total_bytes, |
8929ecfa | 5969 | (unsigned long long)info->bytes_used, |
21380931 | 5970 | (unsigned long long)info->bytes_pinned, |
8929ecfa | 5971 | (unsigned long long)info->bytes_reserved, |
21380931 | 5972 | (unsigned long long)info->bytes_may_use, |
8929ecfa | 5973 | (unsigned long long)info->bytes_readonly); |
9ed74f2d JB |
5974 | spin_unlock(&info->lock); |
5975 | ||
5976 | if (!dump_block_groups) | |
5977 | return; | |
0f9dd46c | 5978 | |
80eb234a | 5979 | down_read(&info->groups_sem); |
b742bb82 YZ |
5980 | again: |
5981 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 5982 | spin_lock(&cache->lock); |
799ffc3c | 5983 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n", |
d397712b CM |
5984 | (unsigned long long)cache->key.objectid, |
5985 | (unsigned long long)cache->key.offset, | |
5986 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
5987 | (unsigned long long)cache->pinned, | |
799ffc3c LB |
5988 | (unsigned long long)cache->reserved, |
5989 | cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
5990 | btrfs_dump_free_space(cache, bytes); |
5991 | spin_unlock(&cache->lock); | |
5992 | } | |
b742bb82 YZ |
5993 | if (++index < BTRFS_NR_RAID_TYPES) |
5994 | goto again; | |
80eb234a | 5995 | up_read(&info->groups_sem); |
0f9dd46c | 5996 | } |
e8569813 | 5997 | |
11833d66 YZ |
5998 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
5999 | struct btrfs_root *root, | |
6000 | u64 num_bytes, u64 min_alloc_size, | |
6001 | u64 empty_size, u64 hint_byte, | |
81c9ad23 | 6002 | struct btrfs_key *ins, u64 data) |
fec577fb | 6003 | { |
9e622d6b | 6004 | bool final_tried = false; |
fec577fb | 6005 | int ret; |
925baedd | 6006 | |
6a63209f | 6007 | data = btrfs_get_alloc_profile(root, data); |
98d20f67 | 6008 | again: |
db94535d CM |
6009 | WARN_ON(num_bytes < root->sectorsize); |
6010 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
81c9ad23 | 6011 | hint_byte, ins, data); |
3b951516 | 6012 | |
9e622d6b MX |
6013 | if (ret == -ENOSPC) { |
6014 | if (!final_tried) { | |
6015 | num_bytes = num_bytes >> 1; | |
6016 | num_bytes = num_bytes & ~(root->sectorsize - 1); | |
6017 | num_bytes = max(num_bytes, min_alloc_size); | |
9e622d6b MX |
6018 | if (num_bytes == min_alloc_size) |
6019 | final_tried = true; | |
6020 | goto again; | |
6021 | } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
6022 | struct btrfs_space_info *sinfo; | |
6023 | ||
6024 | sinfo = __find_space_info(root->fs_info, data); | |
6025 | printk(KERN_ERR "btrfs allocation failed flags %llu, " | |
6026 | "wanted %llu\n", (unsigned long long)data, | |
6027 | (unsigned long long)num_bytes); | |
53804280 JM |
6028 | if (sinfo) |
6029 | dump_space_info(sinfo, num_bytes, 1); | |
9e622d6b | 6030 | } |
925baedd | 6031 | } |
0f9dd46c | 6032 | |
1abe9b8a | 6033 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
6034 | ||
0f9dd46c | 6035 | return ret; |
e6dcd2dc CM |
6036 | } |
6037 | ||
e688b725 CM |
6038 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, |
6039 | u64 start, u64 len, int pin) | |
65b51a00 | 6040 | { |
0f9dd46c | 6041 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 6042 | int ret = 0; |
0f9dd46c | 6043 | |
0f9dd46c JB |
6044 | cache = btrfs_lookup_block_group(root->fs_info, start); |
6045 | if (!cache) { | |
d397712b CM |
6046 | printk(KERN_ERR "Unable to find block group for %llu\n", |
6047 | (unsigned long long)start); | |
0f9dd46c JB |
6048 | return -ENOSPC; |
6049 | } | |
1f3c79a2 | 6050 | |
5378e607 LD |
6051 | if (btrfs_test_opt(root, DISCARD)) |
6052 | ret = btrfs_discard_extent(root, start, len, NULL); | |
1f3c79a2 | 6053 | |
e688b725 CM |
6054 | if (pin) |
6055 | pin_down_extent(root, cache, start, len, 1); | |
6056 | else { | |
6057 | btrfs_add_free_space(cache, start, len); | |
6058 | btrfs_update_reserved_bytes(cache, len, RESERVE_FREE); | |
6059 | } | |
fa9c0d79 | 6060 | btrfs_put_block_group(cache); |
817d52f8 | 6061 | |
1abe9b8a | 6062 | trace_btrfs_reserved_extent_free(root, start, len); |
6063 | ||
e6dcd2dc CM |
6064 | return ret; |
6065 | } | |
6066 | ||
e688b725 CM |
6067 | int btrfs_free_reserved_extent(struct btrfs_root *root, |
6068 | u64 start, u64 len) | |
6069 | { | |
6070 | return __btrfs_free_reserved_extent(root, start, len, 0); | |
6071 | } | |
6072 | ||
6073 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, | |
6074 | u64 start, u64 len) | |
6075 | { | |
6076 | return __btrfs_free_reserved_extent(root, start, len, 1); | |
6077 | } | |
6078 | ||
5d4f98a2 YZ |
6079 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
6080 | struct btrfs_root *root, | |
6081 | u64 parent, u64 root_objectid, | |
6082 | u64 flags, u64 owner, u64 offset, | |
6083 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
6084 | { |
6085 | int ret; | |
5d4f98a2 | 6086 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 6087 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 6088 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 6089 | struct btrfs_path *path; |
5d4f98a2 YZ |
6090 | struct extent_buffer *leaf; |
6091 | int type; | |
6092 | u32 size; | |
26b8003f | 6093 | |
5d4f98a2 YZ |
6094 | if (parent > 0) |
6095 | type = BTRFS_SHARED_DATA_REF_KEY; | |
6096 | else | |
6097 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 6098 | |
5d4f98a2 | 6099 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
6100 | |
6101 | path = btrfs_alloc_path(); | |
db5b493a TI |
6102 | if (!path) |
6103 | return -ENOMEM; | |
47e4bb98 | 6104 | |
b9473439 | 6105 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6106 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
6107 | ins, size); | |
79787eaa JM |
6108 | if (ret) { |
6109 | btrfs_free_path(path); | |
6110 | return ret; | |
6111 | } | |
0f9dd46c | 6112 | |
5d4f98a2 YZ |
6113 | leaf = path->nodes[0]; |
6114 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 6115 | struct btrfs_extent_item); |
5d4f98a2 YZ |
6116 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
6117 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
6118 | btrfs_set_extent_flags(leaf, extent_item, | |
6119 | flags | BTRFS_EXTENT_FLAG_DATA); | |
6120 | ||
6121 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
6122 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
6123 | if (parent > 0) { | |
6124 | struct btrfs_shared_data_ref *ref; | |
6125 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
6126 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
6127 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
6128 | } else { | |
6129 | struct btrfs_extent_data_ref *ref; | |
6130 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
6131 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
6132 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
6133 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
6134 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
6135 | } | |
47e4bb98 CM |
6136 | |
6137 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 6138 | btrfs_free_path(path); |
f510cfec | 6139 | |
f0486c68 | 6140 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
79787eaa | 6141 | if (ret) { /* -ENOENT, logic error */ |
d397712b CM |
6142 | printk(KERN_ERR "btrfs update block group failed for %llu " |
6143 | "%llu\n", (unsigned long long)ins->objectid, | |
6144 | (unsigned long long)ins->offset); | |
f5947066 CM |
6145 | BUG(); |
6146 | } | |
e6dcd2dc CM |
6147 | return ret; |
6148 | } | |
6149 | ||
5d4f98a2 YZ |
6150 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
6151 | struct btrfs_root *root, | |
6152 | u64 parent, u64 root_objectid, | |
6153 | u64 flags, struct btrfs_disk_key *key, | |
6154 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
6155 | { |
6156 | int ret; | |
5d4f98a2 YZ |
6157 | struct btrfs_fs_info *fs_info = root->fs_info; |
6158 | struct btrfs_extent_item *extent_item; | |
6159 | struct btrfs_tree_block_info *block_info; | |
6160 | struct btrfs_extent_inline_ref *iref; | |
6161 | struct btrfs_path *path; | |
6162 | struct extent_buffer *leaf; | |
6163 | u32 size = sizeof(*extent_item) + sizeof(*block_info) + sizeof(*iref); | |
1c2308f8 | 6164 | |
5d4f98a2 | 6165 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6166 | if (!path) |
6167 | return -ENOMEM; | |
56bec294 | 6168 | |
5d4f98a2 YZ |
6169 | path->leave_spinning = 1; |
6170 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
6171 | ins, size); | |
79787eaa JM |
6172 | if (ret) { |
6173 | btrfs_free_path(path); | |
6174 | return ret; | |
6175 | } | |
5d4f98a2 YZ |
6176 | |
6177 | leaf = path->nodes[0]; | |
6178 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
6179 | struct btrfs_extent_item); | |
6180 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
6181 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
6182 | btrfs_set_extent_flags(leaf, extent_item, | |
6183 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
6184 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
6185 | ||
6186 | btrfs_set_tree_block_key(leaf, block_info, key); | |
6187 | btrfs_set_tree_block_level(leaf, block_info, level); | |
6188 | ||
6189 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
6190 | if (parent > 0) { | |
6191 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
6192 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6193 | BTRFS_SHARED_BLOCK_REF_KEY); | |
6194 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
6195 | } else { | |
6196 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6197 | BTRFS_TREE_BLOCK_REF_KEY); | |
6198 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
6199 | } | |
6200 | ||
6201 | btrfs_mark_buffer_dirty(leaf); | |
6202 | btrfs_free_path(path); | |
6203 | ||
f0486c68 | 6204 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
79787eaa | 6205 | if (ret) { /* -ENOENT, logic error */ |
5d4f98a2 YZ |
6206 | printk(KERN_ERR "btrfs update block group failed for %llu " |
6207 | "%llu\n", (unsigned long long)ins->objectid, | |
6208 | (unsigned long long)ins->offset); | |
6209 | BUG(); | |
6210 | } | |
6211 | return ret; | |
6212 | } | |
6213 | ||
6214 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
6215 | struct btrfs_root *root, | |
6216 | u64 root_objectid, u64 owner, | |
6217 | u64 offset, struct btrfs_key *ins) | |
6218 | { | |
6219 | int ret; | |
6220 | ||
6221 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
6222 | ||
66d7e7f0 AJ |
6223 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, |
6224 | ins->offset, 0, | |
6225 | root_objectid, owner, offset, | |
6226 | BTRFS_ADD_DELAYED_EXTENT, NULL, 0); | |
e6dcd2dc CM |
6227 | return ret; |
6228 | } | |
e02119d5 CM |
6229 | |
6230 | /* | |
6231 | * this is used by the tree logging recovery code. It records that | |
6232 | * an extent has been allocated and makes sure to clear the free | |
6233 | * space cache bits as well | |
6234 | */ | |
5d4f98a2 YZ |
6235 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
6236 | struct btrfs_root *root, | |
6237 | u64 root_objectid, u64 owner, u64 offset, | |
6238 | struct btrfs_key *ins) | |
e02119d5 CM |
6239 | { |
6240 | int ret; | |
6241 | struct btrfs_block_group_cache *block_group; | |
11833d66 YZ |
6242 | struct btrfs_caching_control *caching_ctl; |
6243 | u64 start = ins->objectid; | |
6244 | u64 num_bytes = ins->offset; | |
e02119d5 | 6245 | |
e02119d5 | 6246 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
b8399dee | 6247 | cache_block_group(block_group, trans, NULL, 0); |
11833d66 | 6248 | caching_ctl = get_caching_control(block_group); |
e02119d5 | 6249 | |
11833d66 YZ |
6250 | if (!caching_ctl) { |
6251 | BUG_ON(!block_group_cache_done(block_group)); | |
6252 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
79787eaa | 6253 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6254 | } else { |
6255 | mutex_lock(&caching_ctl->mutex); | |
6256 | ||
6257 | if (start >= caching_ctl->progress) { | |
6258 | ret = add_excluded_extent(root, start, num_bytes); | |
79787eaa | 6259 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6260 | } else if (start + num_bytes <= caching_ctl->progress) { |
6261 | ret = btrfs_remove_free_space(block_group, | |
6262 | start, num_bytes); | |
79787eaa | 6263 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6264 | } else { |
6265 | num_bytes = caching_ctl->progress - start; | |
6266 | ret = btrfs_remove_free_space(block_group, | |
6267 | start, num_bytes); | |
79787eaa | 6268 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6269 | |
6270 | start = caching_ctl->progress; | |
6271 | num_bytes = ins->objectid + ins->offset - | |
6272 | caching_ctl->progress; | |
6273 | ret = add_excluded_extent(root, start, num_bytes); | |
79787eaa | 6274 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6275 | } |
6276 | ||
6277 | mutex_unlock(&caching_ctl->mutex); | |
6278 | put_caching_control(caching_ctl); | |
6279 | } | |
6280 | ||
fb25e914 JB |
6281 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, |
6282 | RESERVE_ALLOC_NO_ACCOUNT); | |
79787eaa | 6283 | BUG_ON(ret); /* logic error */ |
fa9c0d79 | 6284 | btrfs_put_block_group(block_group); |
5d4f98a2 YZ |
6285 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
6286 | 0, owner, offset, ins, 1); | |
e02119d5 CM |
6287 | return ret; |
6288 | } | |
6289 | ||
65b51a00 CM |
6290 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, |
6291 | struct btrfs_root *root, | |
4008c04a CM |
6292 | u64 bytenr, u32 blocksize, |
6293 | int level) | |
65b51a00 CM |
6294 | { |
6295 | struct extent_buffer *buf; | |
6296 | ||
6297 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
6298 | if (!buf) | |
6299 | return ERR_PTR(-ENOMEM); | |
6300 | btrfs_set_header_generation(buf, trans->transid); | |
85d4e461 | 6301 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 CM |
6302 | btrfs_tree_lock(buf); |
6303 | clean_tree_block(trans, root, buf); | |
3083ee2e | 6304 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
6305 | |
6306 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 6307 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 6308 | |
d0c803c4 | 6309 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
8cef4e16 YZ |
6310 | /* |
6311 | * we allow two log transactions at a time, use different | |
6312 | * EXENT bit to differentiate dirty pages. | |
6313 | */ | |
6314 | if (root->log_transid % 2 == 0) | |
6315 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
6316 | buf->start + buf->len - 1, GFP_NOFS); | |
6317 | else | |
6318 | set_extent_new(&root->dirty_log_pages, buf->start, | |
6319 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 CM |
6320 | } else { |
6321 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 6322 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 6323 | } |
65b51a00 | 6324 | trans->blocks_used++; |
b4ce94de | 6325 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
6326 | return buf; |
6327 | } | |
6328 | ||
f0486c68 YZ |
6329 | static struct btrfs_block_rsv * |
6330 | use_block_rsv(struct btrfs_trans_handle *trans, | |
6331 | struct btrfs_root *root, u32 blocksize) | |
6332 | { | |
6333 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 6334 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 YZ |
6335 | int ret; |
6336 | ||
6337 | block_rsv = get_block_rsv(trans, root); | |
6338 | ||
6339 | if (block_rsv->size == 0) { | |
08e007d2 MX |
6340 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, |
6341 | BTRFS_RESERVE_NO_FLUSH); | |
68a82277 JB |
6342 | /* |
6343 | * If we couldn't reserve metadata bytes try and use some from | |
6344 | * the global reserve. | |
6345 | */ | |
6346 | if (ret && block_rsv != global_rsv) { | |
6347 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
6348 | if (!ret) | |
6349 | return global_rsv; | |
f0486c68 | 6350 | return ERR_PTR(ret); |
68a82277 | 6351 | } else if (ret) { |
f0486c68 | 6352 | return ERR_PTR(ret); |
68a82277 | 6353 | } |
f0486c68 YZ |
6354 | return block_rsv; |
6355 | } | |
6356 | ||
6357 | ret = block_rsv_use_bytes(block_rsv, blocksize); | |
6358 | if (!ret) | |
6359 | return block_rsv; | |
ca7e70f5 | 6360 | if (ret && !block_rsv->failfast) { |
dff51cd1 DS |
6361 | static DEFINE_RATELIMIT_STATE(_rs, |
6362 | DEFAULT_RATELIMIT_INTERVAL, | |
6363 | /*DEFAULT_RATELIMIT_BURST*/ 2); | |
31b1a2bd JL |
6364 | if (__ratelimit(&_rs)) |
6365 | WARN(1, KERN_DEBUG "btrfs: block rsv returned %d\n", | |
6366 | ret); | |
08e007d2 MX |
6367 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, |
6368 | BTRFS_RESERVE_NO_FLUSH); | |
68a82277 | 6369 | if (!ret) { |
68a82277 JB |
6370 | return block_rsv; |
6371 | } else if (ret && block_rsv != global_rsv) { | |
6372 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
6373 | if (!ret) | |
6374 | return global_rsv; | |
6375 | } | |
6376 | } | |
f0486c68 | 6377 | |
f0486c68 YZ |
6378 | return ERR_PTR(-ENOSPC); |
6379 | } | |
6380 | ||
8c2a3ca2 JB |
6381 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
6382 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
6383 | { |
6384 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 6385 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
6386 | } |
6387 | ||
fec577fb | 6388 | /* |
f0486c68 YZ |
6389 | * finds a free extent and does all the dirty work required for allocation |
6390 | * returns the key for the extent through ins, and a tree buffer for | |
6391 | * the first block of the extent through buf. | |
6392 | * | |
fec577fb CM |
6393 | * returns the tree buffer or NULL. |
6394 | */ | |
5f39d397 | 6395 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
6396 | struct btrfs_root *root, u32 blocksize, |
6397 | u64 parent, u64 root_objectid, | |
6398 | struct btrfs_disk_key *key, int level, | |
5581a51a | 6399 | u64 hint, u64 empty_size) |
fec577fb | 6400 | { |
e2fa7227 | 6401 | struct btrfs_key ins; |
f0486c68 | 6402 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 6403 | struct extent_buffer *buf; |
f0486c68 YZ |
6404 | u64 flags = 0; |
6405 | int ret; | |
6406 | ||
fec577fb | 6407 | |
f0486c68 YZ |
6408 | block_rsv = use_block_rsv(trans, root, blocksize); |
6409 | if (IS_ERR(block_rsv)) | |
6410 | return ERR_CAST(block_rsv); | |
6411 | ||
6412 | ret = btrfs_reserve_extent(trans, root, blocksize, blocksize, | |
81c9ad23 | 6413 | empty_size, hint, &ins, 0); |
fec577fb | 6414 | if (ret) { |
8c2a3ca2 | 6415 | unuse_block_rsv(root->fs_info, block_rsv, blocksize); |
54aa1f4d | 6416 | return ERR_PTR(ret); |
fec577fb | 6417 | } |
55c69072 | 6418 | |
4008c04a CM |
6419 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
6420 | blocksize, level); | |
79787eaa | 6421 | BUG_ON(IS_ERR(buf)); /* -ENOMEM */ |
f0486c68 YZ |
6422 | |
6423 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
6424 | if (parent == 0) | |
6425 | parent = ins.objectid; | |
6426 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6427 | } else | |
6428 | BUG_ON(parent > 0); | |
6429 | ||
6430 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
6431 | struct btrfs_delayed_extent_op *extent_op; | |
78a6184a | 6432 | extent_op = btrfs_alloc_delayed_extent_op(); |
79787eaa | 6433 | BUG_ON(!extent_op); /* -ENOMEM */ |
f0486c68 YZ |
6434 | if (key) |
6435 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
6436 | else | |
6437 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
6438 | extent_op->flags_to_set = flags; | |
6439 | extent_op->update_key = 1; | |
6440 | extent_op->update_flags = 1; | |
6441 | extent_op->is_data = 0; | |
6442 | ||
66d7e7f0 AJ |
6443 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
6444 | ins.objectid, | |
f0486c68 YZ |
6445 | ins.offset, parent, root_objectid, |
6446 | level, BTRFS_ADD_DELAYED_EXTENT, | |
5581a51a | 6447 | extent_op, 0); |
79787eaa | 6448 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 | 6449 | } |
fec577fb CM |
6450 | return buf; |
6451 | } | |
a28ec197 | 6452 | |
2c47e605 YZ |
6453 | struct walk_control { |
6454 | u64 refs[BTRFS_MAX_LEVEL]; | |
6455 | u64 flags[BTRFS_MAX_LEVEL]; | |
6456 | struct btrfs_key update_progress; | |
6457 | int stage; | |
6458 | int level; | |
6459 | int shared_level; | |
6460 | int update_ref; | |
6461 | int keep_locks; | |
1c4850e2 YZ |
6462 | int reada_slot; |
6463 | int reada_count; | |
66d7e7f0 | 6464 | int for_reloc; |
2c47e605 YZ |
6465 | }; |
6466 | ||
6467 | #define DROP_REFERENCE 1 | |
6468 | #define UPDATE_BACKREF 2 | |
6469 | ||
1c4850e2 YZ |
6470 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
6471 | struct btrfs_root *root, | |
6472 | struct walk_control *wc, | |
6473 | struct btrfs_path *path) | |
6407bf6d | 6474 | { |
1c4850e2 YZ |
6475 | u64 bytenr; |
6476 | u64 generation; | |
6477 | u64 refs; | |
94fcca9f | 6478 | u64 flags; |
5d4f98a2 | 6479 | u32 nritems; |
1c4850e2 YZ |
6480 | u32 blocksize; |
6481 | struct btrfs_key key; | |
6482 | struct extent_buffer *eb; | |
6407bf6d | 6483 | int ret; |
1c4850e2 YZ |
6484 | int slot; |
6485 | int nread = 0; | |
6407bf6d | 6486 | |
1c4850e2 YZ |
6487 | if (path->slots[wc->level] < wc->reada_slot) { |
6488 | wc->reada_count = wc->reada_count * 2 / 3; | |
6489 | wc->reada_count = max(wc->reada_count, 2); | |
6490 | } else { | |
6491 | wc->reada_count = wc->reada_count * 3 / 2; | |
6492 | wc->reada_count = min_t(int, wc->reada_count, | |
6493 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
6494 | } | |
7bb86316 | 6495 | |
1c4850e2 YZ |
6496 | eb = path->nodes[wc->level]; |
6497 | nritems = btrfs_header_nritems(eb); | |
6498 | blocksize = btrfs_level_size(root, wc->level - 1); | |
bd56b302 | 6499 | |
1c4850e2 YZ |
6500 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
6501 | if (nread >= wc->reada_count) | |
6502 | break; | |
bd56b302 | 6503 | |
2dd3e67b | 6504 | cond_resched(); |
1c4850e2 YZ |
6505 | bytenr = btrfs_node_blockptr(eb, slot); |
6506 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 6507 | |
1c4850e2 YZ |
6508 | if (slot == path->slots[wc->level]) |
6509 | goto reada; | |
5d4f98a2 | 6510 | |
1c4850e2 YZ |
6511 | if (wc->stage == UPDATE_BACKREF && |
6512 | generation <= root->root_key.offset) | |
bd56b302 CM |
6513 | continue; |
6514 | ||
94fcca9f YZ |
6515 | /* We don't lock the tree block, it's OK to be racy here */ |
6516 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, | |
6517 | &refs, &flags); | |
79787eaa JM |
6518 | /* We don't care about errors in readahead. */ |
6519 | if (ret < 0) | |
6520 | continue; | |
94fcca9f YZ |
6521 | BUG_ON(refs == 0); |
6522 | ||
1c4850e2 | 6523 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
6524 | if (refs == 1) |
6525 | goto reada; | |
bd56b302 | 6526 | |
94fcca9f YZ |
6527 | if (wc->level == 1 && |
6528 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6529 | continue; | |
1c4850e2 YZ |
6530 | if (!wc->update_ref || |
6531 | generation <= root->root_key.offset) | |
6532 | continue; | |
6533 | btrfs_node_key_to_cpu(eb, &key, slot); | |
6534 | ret = btrfs_comp_cpu_keys(&key, | |
6535 | &wc->update_progress); | |
6536 | if (ret < 0) | |
6537 | continue; | |
94fcca9f YZ |
6538 | } else { |
6539 | if (wc->level == 1 && | |
6540 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6541 | continue; | |
6407bf6d | 6542 | } |
1c4850e2 YZ |
6543 | reada: |
6544 | ret = readahead_tree_block(root, bytenr, blocksize, | |
6545 | generation); | |
6546 | if (ret) | |
bd56b302 | 6547 | break; |
1c4850e2 | 6548 | nread++; |
20524f02 | 6549 | } |
1c4850e2 | 6550 | wc->reada_slot = slot; |
20524f02 | 6551 | } |
2c47e605 | 6552 | |
f82d02d9 | 6553 | /* |
2c47e605 YZ |
6554 | * hepler to process tree block while walking down the tree. |
6555 | * | |
2c47e605 YZ |
6556 | * when wc->stage == UPDATE_BACKREF, this function updates |
6557 | * back refs for pointers in the block. | |
6558 | * | |
6559 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 6560 | */ |
2c47e605 | 6561 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 6562 | struct btrfs_root *root, |
2c47e605 | 6563 | struct btrfs_path *path, |
94fcca9f | 6564 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 6565 | { |
2c47e605 YZ |
6566 | int level = wc->level; |
6567 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 6568 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
6569 | int ret; |
6570 | ||
2c47e605 YZ |
6571 | if (wc->stage == UPDATE_BACKREF && |
6572 | btrfs_header_owner(eb) != root->root_key.objectid) | |
6573 | return 1; | |
f82d02d9 | 6574 | |
2c47e605 YZ |
6575 | /* |
6576 | * when reference count of tree block is 1, it won't increase | |
6577 | * again. once full backref flag is set, we never clear it. | |
6578 | */ | |
94fcca9f YZ |
6579 | if (lookup_info && |
6580 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
6581 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
6582 | BUG_ON(!path->locks[level]); |
6583 | ret = btrfs_lookup_extent_info(trans, root, | |
6584 | eb->start, eb->len, | |
6585 | &wc->refs[level], | |
6586 | &wc->flags[level]); | |
79787eaa JM |
6587 | BUG_ON(ret == -ENOMEM); |
6588 | if (ret) | |
6589 | return ret; | |
2c47e605 YZ |
6590 | BUG_ON(wc->refs[level] == 0); |
6591 | } | |
5d4f98a2 | 6592 | |
2c47e605 YZ |
6593 | if (wc->stage == DROP_REFERENCE) { |
6594 | if (wc->refs[level] > 1) | |
6595 | return 1; | |
f82d02d9 | 6596 | |
2c47e605 | 6597 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 6598 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6599 | path->locks[level] = 0; |
6600 | } | |
6601 | return 0; | |
6602 | } | |
f82d02d9 | 6603 | |
2c47e605 YZ |
6604 | /* wc->stage == UPDATE_BACKREF */ |
6605 | if (!(wc->flags[level] & flag)) { | |
6606 | BUG_ON(!path->locks[level]); | |
66d7e7f0 | 6607 | ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc); |
79787eaa | 6608 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 6609 | ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc); |
79787eaa | 6610 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6611 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, |
6612 | eb->len, flag, 0); | |
79787eaa | 6613 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6614 | wc->flags[level] |= flag; |
6615 | } | |
6616 | ||
6617 | /* | |
6618 | * the block is shared by multiple trees, so it's not good to | |
6619 | * keep the tree lock | |
6620 | */ | |
6621 | if (path->locks[level] && level > 0) { | |
bd681513 | 6622 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6623 | path->locks[level] = 0; |
6624 | } | |
6625 | return 0; | |
6626 | } | |
6627 | ||
1c4850e2 YZ |
6628 | /* |
6629 | * hepler to process tree block pointer. | |
6630 | * | |
6631 | * when wc->stage == DROP_REFERENCE, this function checks | |
6632 | * reference count of the block pointed to. if the block | |
6633 | * is shared and we need update back refs for the subtree | |
6634 | * rooted at the block, this function changes wc->stage to | |
6635 | * UPDATE_BACKREF. if the block is shared and there is no | |
6636 | * need to update back, this function drops the reference | |
6637 | * to the block. | |
6638 | * | |
6639 | * NOTE: return value 1 means we should stop walking down. | |
6640 | */ | |
6641 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
6642 | struct btrfs_root *root, | |
6643 | struct btrfs_path *path, | |
94fcca9f | 6644 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
6645 | { |
6646 | u64 bytenr; | |
6647 | u64 generation; | |
6648 | u64 parent; | |
6649 | u32 blocksize; | |
6650 | struct btrfs_key key; | |
6651 | struct extent_buffer *next; | |
6652 | int level = wc->level; | |
6653 | int reada = 0; | |
6654 | int ret = 0; | |
6655 | ||
6656 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
6657 | path->slots[level]); | |
6658 | /* | |
6659 | * if the lower level block was created before the snapshot | |
6660 | * was created, we know there is no need to update back refs | |
6661 | * for the subtree | |
6662 | */ | |
6663 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
6664 | generation <= root->root_key.offset) { |
6665 | *lookup_info = 1; | |
1c4850e2 | 6666 | return 1; |
94fcca9f | 6667 | } |
1c4850e2 YZ |
6668 | |
6669 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
6670 | blocksize = btrfs_level_size(root, level - 1); | |
6671 | ||
6672 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
6673 | if (!next) { | |
6674 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
90d2c51d MX |
6675 | if (!next) |
6676 | return -ENOMEM; | |
1c4850e2 YZ |
6677 | reada = 1; |
6678 | } | |
6679 | btrfs_tree_lock(next); | |
6680 | btrfs_set_lock_blocking(next); | |
6681 | ||
94fcca9f YZ |
6682 | ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize, |
6683 | &wc->refs[level - 1], | |
6684 | &wc->flags[level - 1]); | |
79787eaa JM |
6685 | if (ret < 0) { |
6686 | btrfs_tree_unlock(next); | |
6687 | return ret; | |
6688 | } | |
6689 | ||
94fcca9f YZ |
6690 | BUG_ON(wc->refs[level - 1] == 0); |
6691 | *lookup_info = 0; | |
1c4850e2 | 6692 | |
94fcca9f | 6693 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 6694 | if (wc->refs[level - 1] > 1) { |
94fcca9f YZ |
6695 | if (level == 1 && |
6696 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6697 | goto skip; | |
6698 | ||
1c4850e2 YZ |
6699 | if (!wc->update_ref || |
6700 | generation <= root->root_key.offset) | |
6701 | goto skip; | |
6702 | ||
6703 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
6704 | path->slots[level]); | |
6705 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
6706 | if (ret < 0) | |
6707 | goto skip; | |
6708 | ||
6709 | wc->stage = UPDATE_BACKREF; | |
6710 | wc->shared_level = level - 1; | |
6711 | } | |
94fcca9f YZ |
6712 | } else { |
6713 | if (level == 1 && | |
6714 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6715 | goto skip; | |
1c4850e2 YZ |
6716 | } |
6717 | ||
b9fab919 | 6718 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
6719 | btrfs_tree_unlock(next); |
6720 | free_extent_buffer(next); | |
6721 | next = NULL; | |
94fcca9f | 6722 | *lookup_info = 1; |
1c4850e2 YZ |
6723 | } |
6724 | ||
6725 | if (!next) { | |
6726 | if (reada && level == 1) | |
6727 | reada_walk_down(trans, root, wc, path); | |
6728 | next = read_tree_block(root, bytenr, blocksize, generation); | |
97d9a8a4 TI |
6729 | if (!next) |
6730 | return -EIO; | |
1c4850e2 YZ |
6731 | btrfs_tree_lock(next); |
6732 | btrfs_set_lock_blocking(next); | |
6733 | } | |
6734 | ||
6735 | level--; | |
6736 | BUG_ON(level != btrfs_header_level(next)); | |
6737 | path->nodes[level] = next; | |
6738 | path->slots[level] = 0; | |
bd681513 | 6739 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
6740 | wc->level = level; |
6741 | if (wc->level == 1) | |
6742 | wc->reada_slot = 0; | |
6743 | return 0; | |
6744 | skip: | |
6745 | wc->refs[level - 1] = 0; | |
6746 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
6747 | if (wc->stage == DROP_REFERENCE) { |
6748 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
6749 | parent = path->nodes[level]->start; | |
6750 | } else { | |
6751 | BUG_ON(root->root_key.objectid != | |
6752 | btrfs_header_owner(path->nodes[level])); | |
6753 | parent = 0; | |
6754 | } | |
1c4850e2 | 6755 | |
94fcca9f | 6756 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
66d7e7f0 | 6757 | root->root_key.objectid, level - 1, 0, 0); |
79787eaa | 6758 | BUG_ON(ret); /* -ENOMEM */ |
1c4850e2 | 6759 | } |
1c4850e2 YZ |
6760 | btrfs_tree_unlock(next); |
6761 | free_extent_buffer(next); | |
94fcca9f | 6762 | *lookup_info = 1; |
1c4850e2 YZ |
6763 | return 1; |
6764 | } | |
6765 | ||
2c47e605 YZ |
6766 | /* |
6767 | * hepler to process tree block while walking up the tree. | |
6768 | * | |
6769 | * when wc->stage == DROP_REFERENCE, this function drops | |
6770 | * reference count on the block. | |
6771 | * | |
6772 | * when wc->stage == UPDATE_BACKREF, this function changes | |
6773 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
6774 | * to UPDATE_BACKREF previously while processing the block. | |
6775 | * | |
6776 | * NOTE: return value 1 means we should stop walking up. | |
6777 | */ | |
6778 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
6779 | struct btrfs_root *root, | |
6780 | struct btrfs_path *path, | |
6781 | struct walk_control *wc) | |
6782 | { | |
f0486c68 | 6783 | int ret; |
2c47e605 YZ |
6784 | int level = wc->level; |
6785 | struct extent_buffer *eb = path->nodes[level]; | |
6786 | u64 parent = 0; | |
6787 | ||
6788 | if (wc->stage == UPDATE_BACKREF) { | |
6789 | BUG_ON(wc->shared_level < level); | |
6790 | if (level < wc->shared_level) | |
6791 | goto out; | |
6792 | ||
2c47e605 YZ |
6793 | ret = find_next_key(path, level + 1, &wc->update_progress); |
6794 | if (ret > 0) | |
6795 | wc->update_ref = 0; | |
6796 | ||
6797 | wc->stage = DROP_REFERENCE; | |
6798 | wc->shared_level = -1; | |
6799 | path->slots[level] = 0; | |
6800 | ||
6801 | /* | |
6802 | * check reference count again if the block isn't locked. | |
6803 | * we should start walking down the tree again if reference | |
6804 | * count is one. | |
6805 | */ | |
6806 | if (!path->locks[level]) { | |
6807 | BUG_ON(level == 0); | |
6808 | btrfs_tree_lock(eb); | |
6809 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6810 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6811 | |
6812 | ret = btrfs_lookup_extent_info(trans, root, | |
6813 | eb->start, eb->len, | |
6814 | &wc->refs[level], | |
6815 | &wc->flags[level]); | |
79787eaa JM |
6816 | if (ret < 0) { |
6817 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 6818 | path->locks[level] = 0; |
79787eaa JM |
6819 | return ret; |
6820 | } | |
2c47e605 YZ |
6821 | BUG_ON(wc->refs[level] == 0); |
6822 | if (wc->refs[level] == 1) { | |
bd681513 | 6823 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 6824 | path->locks[level] = 0; |
2c47e605 YZ |
6825 | return 1; |
6826 | } | |
f82d02d9 | 6827 | } |
2c47e605 | 6828 | } |
f82d02d9 | 6829 | |
2c47e605 YZ |
6830 | /* wc->stage == DROP_REFERENCE */ |
6831 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 6832 | |
2c47e605 YZ |
6833 | if (wc->refs[level] == 1) { |
6834 | if (level == 0) { | |
6835 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
66d7e7f0 AJ |
6836 | ret = btrfs_dec_ref(trans, root, eb, 1, |
6837 | wc->for_reloc); | |
2c47e605 | 6838 | else |
66d7e7f0 AJ |
6839 | ret = btrfs_dec_ref(trans, root, eb, 0, |
6840 | wc->for_reloc); | |
79787eaa | 6841 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6842 | } |
6843 | /* make block locked assertion in clean_tree_block happy */ | |
6844 | if (!path->locks[level] && | |
6845 | btrfs_header_generation(eb) == trans->transid) { | |
6846 | btrfs_tree_lock(eb); | |
6847 | btrfs_set_lock_blocking(eb); | |
bd681513 | 6848 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6849 | } |
6850 | clean_tree_block(trans, root, eb); | |
6851 | } | |
6852 | ||
6853 | if (eb == root->node) { | |
6854 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6855 | parent = eb->start; | |
6856 | else | |
6857 | BUG_ON(root->root_key.objectid != | |
6858 | btrfs_header_owner(eb)); | |
6859 | } else { | |
6860 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
6861 | parent = path->nodes[level + 1]->start; | |
6862 | else | |
6863 | BUG_ON(root->root_key.objectid != | |
6864 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 6865 | } |
f82d02d9 | 6866 | |
5581a51a | 6867 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
6868 | out: |
6869 | wc->refs[level] = 0; | |
6870 | wc->flags[level] = 0; | |
f0486c68 | 6871 | return 0; |
2c47e605 YZ |
6872 | } |
6873 | ||
6874 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
6875 | struct btrfs_root *root, | |
6876 | struct btrfs_path *path, | |
6877 | struct walk_control *wc) | |
6878 | { | |
2c47e605 | 6879 | int level = wc->level; |
94fcca9f | 6880 | int lookup_info = 1; |
2c47e605 YZ |
6881 | int ret; |
6882 | ||
6883 | while (level >= 0) { | |
94fcca9f | 6884 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
6885 | if (ret > 0) |
6886 | break; | |
6887 | ||
6888 | if (level == 0) | |
6889 | break; | |
6890 | ||
7a7965f8 YZ |
6891 | if (path->slots[level] >= |
6892 | btrfs_header_nritems(path->nodes[level])) | |
6893 | break; | |
6894 | ||
94fcca9f | 6895 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
6896 | if (ret > 0) { |
6897 | path->slots[level]++; | |
6898 | continue; | |
90d2c51d MX |
6899 | } else if (ret < 0) |
6900 | return ret; | |
1c4850e2 | 6901 | level = wc->level; |
f82d02d9 | 6902 | } |
f82d02d9 YZ |
6903 | return 0; |
6904 | } | |
6905 | ||
d397712b | 6906 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 6907 | struct btrfs_root *root, |
f82d02d9 | 6908 | struct btrfs_path *path, |
2c47e605 | 6909 | struct walk_control *wc, int max_level) |
20524f02 | 6910 | { |
2c47e605 | 6911 | int level = wc->level; |
20524f02 | 6912 | int ret; |
9f3a7427 | 6913 | |
2c47e605 YZ |
6914 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
6915 | while (level < max_level && path->nodes[level]) { | |
6916 | wc->level = level; | |
6917 | if (path->slots[level] + 1 < | |
6918 | btrfs_header_nritems(path->nodes[level])) { | |
6919 | path->slots[level]++; | |
20524f02 CM |
6920 | return 0; |
6921 | } else { | |
2c47e605 YZ |
6922 | ret = walk_up_proc(trans, root, path, wc); |
6923 | if (ret > 0) | |
6924 | return 0; | |
bd56b302 | 6925 | |
2c47e605 | 6926 | if (path->locks[level]) { |
bd681513 CM |
6927 | btrfs_tree_unlock_rw(path->nodes[level], |
6928 | path->locks[level]); | |
2c47e605 | 6929 | path->locks[level] = 0; |
f82d02d9 | 6930 | } |
2c47e605 YZ |
6931 | free_extent_buffer(path->nodes[level]); |
6932 | path->nodes[level] = NULL; | |
6933 | level++; | |
20524f02 CM |
6934 | } |
6935 | } | |
6936 | return 1; | |
6937 | } | |
6938 | ||
9aca1d51 | 6939 | /* |
2c47e605 YZ |
6940 | * drop a subvolume tree. |
6941 | * | |
6942 | * this function traverses the tree freeing any blocks that only | |
6943 | * referenced by the tree. | |
6944 | * | |
6945 | * when a shared tree block is found. this function decreases its | |
6946 | * reference count by one. if update_ref is true, this function | |
6947 | * also make sure backrefs for the shared block and all lower level | |
6948 | * blocks are properly updated. | |
9aca1d51 | 6949 | */ |
2c536799 | 6950 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
6951 | struct btrfs_block_rsv *block_rsv, int update_ref, |
6952 | int for_reloc) | |
20524f02 | 6953 | { |
5caf2a00 | 6954 | struct btrfs_path *path; |
2c47e605 YZ |
6955 | struct btrfs_trans_handle *trans; |
6956 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 6957 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
6958 | struct walk_control *wc; |
6959 | struct btrfs_key key; | |
6960 | int err = 0; | |
6961 | int ret; | |
6962 | int level; | |
20524f02 | 6963 | |
5caf2a00 | 6964 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
6965 | if (!path) { |
6966 | err = -ENOMEM; | |
6967 | goto out; | |
6968 | } | |
20524f02 | 6969 | |
2c47e605 | 6970 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
6971 | if (!wc) { |
6972 | btrfs_free_path(path); | |
cb1b69f4 TI |
6973 | err = -ENOMEM; |
6974 | goto out; | |
38a1a919 | 6975 | } |
2c47e605 | 6976 | |
a22285a6 | 6977 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
6978 | if (IS_ERR(trans)) { |
6979 | err = PTR_ERR(trans); | |
6980 | goto out_free; | |
6981 | } | |
98d5dc13 | 6982 | |
3fd0a558 YZ |
6983 | if (block_rsv) |
6984 | trans->block_rsv = block_rsv; | |
2c47e605 | 6985 | |
9f3a7427 | 6986 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 6987 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
6988 | path->nodes[level] = btrfs_lock_root_node(root); |
6989 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 6990 | path->slots[level] = 0; |
bd681513 | 6991 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
6992 | memset(&wc->update_progress, 0, |
6993 | sizeof(wc->update_progress)); | |
9f3a7427 | 6994 | } else { |
9f3a7427 | 6995 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
6996 | memcpy(&wc->update_progress, &key, |
6997 | sizeof(wc->update_progress)); | |
6998 | ||
6702ed49 | 6999 | level = root_item->drop_level; |
2c47e605 | 7000 | BUG_ON(level == 0); |
6702ed49 | 7001 | path->lowest_level = level; |
2c47e605 YZ |
7002 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
7003 | path->lowest_level = 0; | |
7004 | if (ret < 0) { | |
7005 | err = ret; | |
79787eaa | 7006 | goto out_end_trans; |
9f3a7427 | 7007 | } |
1c4850e2 | 7008 | WARN_ON(ret > 0); |
2c47e605 | 7009 | |
7d9eb12c CM |
7010 | /* |
7011 | * unlock our path, this is safe because only this | |
7012 | * function is allowed to delete this snapshot | |
7013 | */ | |
5d4f98a2 | 7014 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
7015 | |
7016 | level = btrfs_header_level(root->node); | |
7017 | while (1) { | |
7018 | btrfs_tree_lock(path->nodes[level]); | |
7019 | btrfs_set_lock_blocking(path->nodes[level]); | |
7020 | ||
7021 | ret = btrfs_lookup_extent_info(trans, root, | |
7022 | path->nodes[level]->start, | |
7023 | path->nodes[level]->len, | |
7024 | &wc->refs[level], | |
7025 | &wc->flags[level]); | |
79787eaa JM |
7026 | if (ret < 0) { |
7027 | err = ret; | |
7028 | goto out_end_trans; | |
7029 | } | |
2c47e605 YZ |
7030 | BUG_ON(wc->refs[level] == 0); |
7031 | ||
7032 | if (level == root_item->drop_level) | |
7033 | break; | |
7034 | ||
7035 | btrfs_tree_unlock(path->nodes[level]); | |
7036 | WARN_ON(wc->refs[level] != 1); | |
7037 | level--; | |
7038 | } | |
9f3a7427 | 7039 | } |
2c47e605 YZ |
7040 | |
7041 | wc->level = level; | |
7042 | wc->shared_level = -1; | |
7043 | wc->stage = DROP_REFERENCE; | |
7044 | wc->update_ref = update_ref; | |
7045 | wc->keep_locks = 0; | |
66d7e7f0 | 7046 | wc->for_reloc = for_reloc; |
1c4850e2 | 7047 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 7048 | |
d397712b | 7049 | while (1) { |
2c47e605 YZ |
7050 | ret = walk_down_tree(trans, root, path, wc); |
7051 | if (ret < 0) { | |
7052 | err = ret; | |
20524f02 | 7053 | break; |
2c47e605 | 7054 | } |
9aca1d51 | 7055 | |
2c47e605 YZ |
7056 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
7057 | if (ret < 0) { | |
7058 | err = ret; | |
20524f02 | 7059 | break; |
2c47e605 YZ |
7060 | } |
7061 | ||
7062 | if (ret > 0) { | |
7063 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
7064 | break; |
7065 | } | |
2c47e605 YZ |
7066 | |
7067 | if (wc->stage == DROP_REFERENCE) { | |
7068 | level = wc->level; | |
7069 | btrfs_node_key(path->nodes[level], | |
7070 | &root_item->drop_progress, | |
7071 | path->slots[level]); | |
7072 | root_item->drop_level = level; | |
7073 | } | |
7074 | ||
7075 | BUG_ON(wc->level == 0); | |
3fd0a558 | 7076 | if (btrfs_should_end_transaction(trans, tree_root)) { |
2c47e605 YZ |
7077 | ret = btrfs_update_root(trans, tree_root, |
7078 | &root->root_key, | |
7079 | root_item); | |
79787eaa JM |
7080 | if (ret) { |
7081 | btrfs_abort_transaction(trans, tree_root, ret); | |
7082 | err = ret; | |
7083 | goto out_end_trans; | |
7084 | } | |
2c47e605 | 7085 | |
3fd0a558 | 7086 | btrfs_end_transaction_throttle(trans, tree_root); |
a22285a6 | 7087 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
7088 | if (IS_ERR(trans)) { |
7089 | err = PTR_ERR(trans); | |
7090 | goto out_free; | |
7091 | } | |
3fd0a558 YZ |
7092 | if (block_rsv) |
7093 | trans->block_rsv = block_rsv; | |
c3e69d58 | 7094 | } |
20524f02 | 7095 | } |
b3b4aa74 | 7096 | btrfs_release_path(path); |
79787eaa JM |
7097 | if (err) |
7098 | goto out_end_trans; | |
2c47e605 YZ |
7099 | |
7100 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
79787eaa JM |
7101 | if (ret) { |
7102 | btrfs_abort_transaction(trans, tree_root, ret); | |
7103 | goto out_end_trans; | |
7104 | } | |
2c47e605 | 7105 | |
76dda93c YZ |
7106 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
7107 | ret = btrfs_find_last_root(tree_root, root->root_key.objectid, | |
7108 | NULL, NULL); | |
79787eaa JM |
7109 | if (ret < 0) { |
7110 | btrfs_abort_transaction(trans, tree_root, ret); | |
7111 | err = ret; | |
7112 | goto out_end_trans; | |
7113 | } else if (ret > 0) { | |
84cd948c JB |
7114 | /* if we fail to delete the orphan item this time |
7115 | * around, it'll get picked up the next time. | |
7116 | * | |
7117 | * The most common failure here is just -ENOENT. | |
7118 | */ | |
7119 | btrfs_del_orphan_item(trans, tree_root, | |
7120 | root->root_key.objectid); | |
76dda93c YZ |
7121 | } |
7122 | } | |
7123 | ||
7124 | if (root->in_radix) { | |
7125 | btrfs_free_fs_root(tree_root->fs_info, root); | |
7126 | } else { | |
7127 | free_extent_buffer(root->node); | |
7128 | free_extent_buffer(root->commit_root); | |
7129 | kfree(root); | |
7130 | } | |
79787eaa | 7131 | out_end_trans: |
3fd0a558 | 7132 | btrfs_end_transaction_throttle(trans, tree_root); |
79787eaa | 7133 | out_free: |
2c47e605 | 7134 | kfree(wc); |
5caf2a00 | 7135 | btrfs_free_path(path); |
cb1b69f4 TI |
7136 | out: |
7137 | if (err) | |
7138 | btrfs_std_error(root->fs_info, err); | |
2c536799 | 7139 | return err; |
20524f02 | 7140 | } |
9078a3e1 | 7141 | |
2c47e605 YZ |
7142 | /* |
7143 | * drop subtree rooted at tree block 'node'. | |
7144 | * | |
7145 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 7146 | * only used by relocation code |
2c47e605 | 7147 | */ |
f82d02d9 YZ |
7148 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
7149 | struct btrfs_root *root, | |
7150 | struct extent_buffer *node, | |
7151 | struct extent_buffer *parent) | |
7152 | { | |
7153 | struct btrfs_path *path; | |
2c47e605 | 7154 | struct walk_control *wc; |
f82d02d9 YZ |
7155 | int level; |
7156 | int parent_level; | |
7157 | int ret = 0; | |
7158 | int wret; | |
7159 | ||
2c47e605 YZ |
7160 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
7161 | ||
f82d02d9 | 7162 | path = btrfs_alloc_path(); |
db5b493a TI |
7163 | if (!path) |
7164 | return -ENOMEM; | |
f82d02d9 | 7165 | |
2c47e605 | 7166 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
7167 | if (!wc) { |
7168 | btrfs_free_path(path); | |
7169 | return -ENOMEM; | |
7170 | } | |
2c47e605 | 7171 | |
b9447ef8 | 7172 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
7173 | parent_level = btrfs_header_level(parent); |
7174 | extent_buffer_get(parent); | |
7175 | path->nodes[parent_level] = parent; | |
7176 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
7177 | ||
b9447ef8 | 7178 | btrfs_assert_tree_locked(node); |
f82d02d9 | 7179 | level = btrfs_header_level(node); |
f82d02d9 YZ |
7180 | path->nodes[level] = node; |
7181 | path->slots[level] = 0; | |
bd681513 | 7182 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7183 | |
7184 | wc->refs[parent_level] = 1; | |
7185 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
7186 | wc->level = level; | |
7187 | wc->shared_level = -1; | |
7188 | wc->stage = DROP_REFERENCE; | |
7189 | wc->update_ref = 0; | |
7190 | wc->keep_locks = 1; | |
66d7e7f0 | 7191 | wc->for_reloc = 1; |
1c4850e2 | 7192 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
7193 | |
7194 | while (1) { | |
2c47e605 YZ |
7195 | wret = walk_down_tree(trans, root, path, wc); |
7196 | if (wret < 0) { | |
f82d02d9 | 7197 | ret = wret; |
f82d02d9 | 7198 | break; |
2c47e605 | 7199 | } |
f82d02d9 | 7200 | |
2c47e605 | 7201 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
7202 | if (wret < 0) |
7203 | ret = wret; | |
7204 | if (wret != 0) | |
7205 | break; | |
7206 | } | |
7207 | ||
2c47e605 | 7208 | kfree(wc); |
f82d02d9 YZ |
7209 | btrfs_free_path(path); |
7210 | return ret; | |
7211 | } | |
7212 | ||
ec44a35c CM |
7213 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
7214 | { | |
7215 | u64 num_devices; | |
fc67c450 | 7216 | u64 stripped; |
e4d8ec0f | 7217 | |
fc67c450 ID |
7218 | /* |
7219 | * if restripe for this chunk_type is on pick target profile and | |
7220 | * return, otherwise do the usual balance | |
7221 | */ | |
7222 | stripped = get_restripe_target(root->fs_info, flags); | |
7223 | if (stripped) | |
7224 | return extended_to_chunk(stripped); | |
e4d8ec0f | 7225 | |
cd02dca5 CM |
7226 | /* |
7227 | * we add in the count of missing devices because we want | |
7228 | * to make sure that any RAID levels on a degraded FS | |
7229 | * continue to be honored. | |
7230 | */ | |
7231 | num_devices = root->fs_info->fs_devices->rw_devices + | |
7232 | root->fs_info->fs_devices->missing_devices; | |
7233 | ||
fc67c450 ID |
7234 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
7235 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; | |
7236 | ||
ec44a35c CM |
7237 | if (num_devices == 1) { |
7238 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7239 | stripped = flags & ~stripped; | |
7240 | ||
7241 | /* turn raid0 into single device chunks */ | |
7242 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
7243 | return stripped; | |
7244 | ||
7245 | /* turn mirroring into duplication */ | |
7246 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7247 | BTRFS_BLOCK_GROUP_RAID10)) | |
7248 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
7249 | } else { |
7250 | /* they already had raid on here, just return */ | |
ec44a35c CM |
7251 | if (flags & stripped) |
7252 | return flags; | |
7253 | ||
7254 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7255 | stripped = flags & ~stripped; | |
7256 | ||
7257 | /* switch duplicated blocks with raid1 */ | |
7258 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
7259 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
7260 | ||
e3176ca2 | 7261 | /* this is drive concat, leave it alone */ |
ec44a35c | 7262 | } |
e3176ca2 | 7263 | |
ec44a35c CM |
7264 | return flags; |
7265 | } | |
7266 | ||
199c36ea | 7267 | static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 7268 | { |
f0486c68 YZ |
7269 | struct btrfs_space_info *sinfo = cache->space_info; |
7270 | u64 num_bytes; | |
199c36ea | 7271 | u64 min_allocable_bytes; |
f0486c68 | 7272 | int ret = -ENOSPC; |
0ef3e66b | 7273 | |
c286ac48 | 7274 | |
199c36ea MX |
7275 | /* |
7276 | * We need some metadata space and system metadata space for | |
7277 | * allocating chunks in some corner cases until we force to set | |
7278 | * it to be readonly. | |
7279 | */ | |
7280 | if ((sinfo->flags & | |
7281 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
7282 | !force) | |
7283 | min_allocable_bytes = 1 * 1024 * 1024; | |
7284 | else | |
7285 | min_allocable_bytes = 0; | |
7286 | ||
f0486c68 YZ |
7287 | spin_lock(&sinfo->lock); |
7288 | spin_lock(&cache->lock); | |
61cfea9b W |
7289 | |
7290 | if (cache->ro) { | |
7291 | ret = 0; | |
7292 | goto out; | |
7293 | } | |
7294 | ||
f0486c68 YZ |
7295 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
7296 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7297 | ||
7298 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
37be25bc JB |
7299 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + |
7300 | min_allocable_bytes <= sinfo->total_bytes) { | |
f0486c68 | 7301 | sinfo->bytes_readonly += num_bytes; |
f0486c68 YZ |
7302 | cache->ro = 1; |
7303 | ret = 0; | |
7304 | } | |
61cfea9b | 7305 | out: |
f0486c68 YZ |
7306 | spin_unlock(&cache->lock); |
7307 | spin_unlock(&sinfo->lock); | |
7308 | return ret; | |
7309 | } | |
7d9eb12c | 7310 | |
f0486c68 YZ |
7311 | int btrfs_set_block_group_ro(struct btrfs_root *root, |
7312 | struct btrfs_block_group_cache *cache) | |
c286ac48 | 7313 | |
f0486c68 YZ |
7314 | { |
7315 | struct btrfs_trans_handle *trans; | |
7316 | u64 alloc_flags; | |
7317 | int ret; | |
7d9eb12c | 7318 | |
f0486c68 | 7319 | BUG_ON(cache->ro); |
0ef3e66b | 7320 | |
ff5714cc | 7321 | trans = btrfs_join_transaction(root); |
79787eaa JM |
7322 | if (IS_ERR(trans)) |
7323 | return PTR_ERR(trans); | |
5d4f98a2 | 7324 | |
f0486c68 | 7325 | alloc_flags = update_block_group_flags(root, cache->flags); |
79787eaa | 7326 | if (alloc_flags != cache->flags) { |
698d0082 | 7327 | ret = do_chunk_alloc(trans, root, alloc_flags, |
79787eaa JM |
7328 | CHUNK_ALLOC_FORCE); |
7329 | if (ret < 0) | |
7330 | goto out; | |
7331 | } | |
5d4f98a2 | 7332 | |
199c36ea | 7333 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7334 | if (!ret) |
7335 | goto out; | |
7336 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
698d0082 | 7337 | ret = do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 7338 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
7339 | if (ret < 0) |
7340 | goto out; | |
199c36ea | 7341 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7342 | out: |
7343 | btrfs_end_transaction(trans, root); | |
7344 | return ret; | |
7345 | } | |
5d4f98a2 | 7346 | |
c87f08ca CM |
7347 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
7348 | struct btrfs_root *root, u64 type) | |
7349 | { | |
7350 | u64 alloc_flags = get_alloc_profile(root, type); | |
698d0082 | 7351 | return do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 7352 | CHUNK_ALLOC_FORCE); |
c87f08ca CM |
7353 | } |
7354 | ||
6d07bcec MX |
7355 | /* |
7356 | * helper to account the unused space of all the readonly block group in the | |
7357 | * list. takes mirrors into account. | |
7358 | */ | |
7359 | static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list) | |
7360 | { | |
7361 | struct btrfs_block_group_cache *block_group; | |
7362 | u64 free_bytes = 0; | |
7363 | int factor; | |
7364 | ||
7365 | list_for_each_entry(block_group, groups_list, list) { | |
7366 | spin_lock(&block_group->lock); | |
7367 | ||
7368 | if (!block_group->ro) { | |
7369 | spin_unlock(&block_group->lock); | |
7370 | continue; | |
7371 | } | |
7372 | ||
7373 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7374 | BTRFS_BLOCK_GROUP_RAID10 | | |
7375 | BTRFS_BLOCK_GROUP_DUP)) | |
7376 | factor = 2; | |
7377 | else | |
7378 | factor = 1; | |
7379 | ||
7380 | free_bytes += (block_group->key.offset - | |
7381 | btrfs_block_group_used(&block_group->item)) * | |
7382 | factor; | |
7383 | ||
7384 | spin_unlock(&block_group->lock); | |
7385 | } | |
7386 | ||
7387 | return free_bytes; | |
7388 | } | |
7389 | ||
7390 | /* | |
7391 | * helper to account the unused space of all the readonly block group in the | |
7392 | * space_info. takes mirrors into account. | |
7393 | */ | |
7394 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) | |
7395 | { | |
7396 | int i; | |
7397 | u64 free_bytes = 0; | |
7398 | ||
7399 | spin_lock(&sinfo->lock); | |
7400 | ||
7401 | for(i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
7402 | if (!list_empty(&sinfo->block_groups[i])) | |
7403 | free_bytes += __btrfs_get_ro_block_group_free_space( | |
7404 | &sinfo->block_groups[i]); | |
7405 | ||
7406 | spin_unlock(&sinfo->lock); | |
7407 | ||
7408 | return free_bytes; | |
7409 | } | |
7410 | ||
143bede5 | 7411 | void btrfs_set_block_group_rw(struct btrfs_root *root, |
f0486c68 | 7412 | struct btrfs_block_group_cache *cache) |
5d4f98a2 | 7413 | { |
f0486c68 YZ |
7414 | struct btrfs_space_info *sinfo = cache->space_info; |
7415 | u64 num_bytes; | |
7416 | ||
7417 | BUG_ON(!cache->ro); | |
7418 | ||
7419 | spin_lock(&sinfo->lock); | |
7420 | spin_lock(&cache->lock); | |
7421 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
7422 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7423 | sinfo->bytes_readonly -= num_bytes; | |
7424 | cache->ro = 0; | |
7425 | spin_unlock(&cache->lock); | |
7426 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
7427 | } |
7428 | ||
ba1bf481 JB |
7429 | /* |
7430 | * checks to see if its even possible to relocate this block group. | |
7431 | * | |
7432 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
7433 | * ok to go ahead and try. | |
7434 | */ | |
7435 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 7436 | { |
ba1bf481 JB |
7437 | struct btrfs_block_group_cache *block_group; |
7438 | struct btrfs_space_info *space_info; | |
7439 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
7440 | struct btrfs_device *device; | |
cdcb725c | 7441 | u64 min_free; |
6719db6a JB |
7442 | u64 dev_min = 1; |
7443 | u64 dev_nr = 0; | |
4a5e98f5 | 7444 | u64 target; |
cdcb725c | 7445 | int index; |
ba1bf481 JB |
7446 | int full = 0; |
7447 | int ret = 0; | |
1a40e23b | 7448 | |
ba1bf481 | 7449 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 7450 | |
ba1bf481 JB |
7451 | /* odd, couldn't find the block group, leave it alone */ |
7452 | if (!block_group) | |
7453 | return -1; | |
1a40e23b | 7454 | |
cdcb725c | 7455 | min_free = btrfs_block_group_used(&block_group->item); |
7456 | ||
ba1bf481 | 7457 | /* no bytes used, we're good */ |
cdcb725c | 7458 | if (!min_free) |
1a40e23b ZY |
7459 | goto out; |
7460 | ||
ba1bf481 JB |
7461 | space_info = block_group->space_info; |
7462 | spin_lock(&space_info->lock); | |
17d217fe | 7463 | |
ba1bf481 | 7464 | full = space_info->full; |
17d217fe | 7465 | |
ba1bf481 JB |
7466 | /* |
7467 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
7468 | * relocate it unless we're able to allocate a new chunk below. |
7469 | * | |
7470 | * Otherwise, we need to make sure we have room in the space to handle | |
7471 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 7472 | */ |
7ce618db | 7473 | if ((space_info->total_bytes != block_group->key.offset) && |
cdcb725c | 7474 | (space_info->bytes_used + space_info->bytes_reserved + |
7475 | space_info->bytes_pinned + space_info->bytes_readonly + | |
7476 | min_free < space_info->total_bytes)) { | |
ba1bf481 JB |
7477 | spin_unlock(&space_info->lock); |
7478 | goto out; | |
17d217fe | 7479 | } |
ba1bf481 | 7480 | spin_unlock(&space_info->lock); |
ea8c2819 | 7481 | |
ba1bf481 JB |
7482 | /* |
7483 | * ok we don't have enough space, but maybe we have free space on our | |
7484 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
7485 | * alloc devices and guess if we have enough space. if this block |
7486 | * group is going to be restriped, run checks against the target | |
7487 | * profile instead of the current one. | |
ba1bf481 JB |
7488 | */ |
7489 | ret = -1; | |
ea8c2819 | 7490 | |
cdcb725c | 7491 | /* |
7492 | * index: | |
7493 | * 0: raid10 | |
7494 | * 1: raid1 | |
7495 | * 2: dup | |
7496 | * 3: raid0 | |
7497 | * 4: single | |
7498 | */ | |
4a5e98f5 ID |
7499 | target = get_restripe_target(root->fs_info, block_group->flags); |
7500 | if (target) { | |
31e50229 | 7501 | index = __get_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
7502 | } else { |
7503 | /* | |
7504 | * this is just a balance, so if we were marked as full | |
7505 | * we know there is no space for a new chunk | |
7506 | */ | |
7507 | if (full) | |
7508 | goto out; | |
7509 | ||
7510 | index = get_block_group_index(block_group); | |
7511 | } | |
7512 | ||
cdcb725c | 7513 | if (index == 0) { |
7514 | dev_min = 4; | |
6719db6a JB |
7515 | /* Divide by 2 */ |
7516 | min_free >>= 1; | |
cdcb725c | 7517 | } else if (index == 1) { |
7518 | dev_min = 2; | |
7519 | } else if (index == 2) { | |
6719db6a JB |
7520 | /* Multiply by 2 */ |
7521 | min_free <<= 1; | |
cdcb725c | 7522 | } else if (index == 3) { |
7523 | dev_min = fs_devices->rw_devices; | |
6719db6a | 7524 | do_div(min_free, dev_min); |
cdcb725c | 7525 | } |
7526 | ||
ba1bf481 JB |
7527 | mutex_lock(&root->fs_info->chunk_mutex); |
7528 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7bfc837d | 7529 | u64 dev_offset; |
56bec294 | 7530 | |
ba1bf481 JB |
7531 | /* |
7532 | * check to make sure we can actually find a chunk with enough | |
7533 | * space to fit our block group in. | |
7534 | */ | |
63a212ab SB |
7535 | if (device->total_bytes > device->bytes_used + min_free && |
7536 | !device->is_tgtdev_for_dev_replace) { | |
125ccb0a | 7537 | ret = find_free_dev_extent(device, min_free, |
7bfc837d | 7538 | &dev_offset, NULL); |
ba1bf481 | 7539 | if (!ret) |
cdcb725c | 7540 | dev_nr++; |
7541 | ||
7542 | if (dev_nr >= dev_min) | |
73e48b27 | 7543 | break; |
cdcb725c | 7544 | |
ba1bf481 | 7545 | ret = -1; |
725c8463 | 7546 | } |
edbd8d4e | 7547 | } |
ba1bf481 | 7548 | mutex_unlock(&root->fs_info->chunk_mutex); |
edbd8d4e | 7549 | out: |
ba1bf481 | 7550 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
7551 | return ret; |
7552 | } | |
7553 | ||
b2950863 CH |
7554 | static int find_first_block_group(struct btrfs_root *root, |
7555 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 7556 | { |
925baedd | 7557 | int ret = 0; |
0b86a832 CM |
7558 | struct btrfs_key found_key; |
7559 | struct extent_buffer *leaf; | |
7560 | int slot; | |
edbd8d4e | 7561 | |
0b86a832 CM |
7562 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
7563 | if (ret < 0) | |
925baedd CM |
7564 | goto out; |
7565 | ||
d397712b | 7566 | while (1) { |
0b86a832 | 7567 | slot = path->slots[0]; |
edbd8d4e | 7568 | leaf = path->nodes[0]; |
0b86a832 CM |
7569 | if (slot >= btrfs_header_nritems(leaf)) { |
7570 | ret = btrfs_next_leaf(root, path); | |
7571 | if (ret == 0) | |
7572 | continue; | |
7573 | if (ret < 0) | |
925baedd | 7574 | goto out; |
0b86a832 | 7575 | break; |
edbd8d4e | 7576 | } |
0b86a832 | 7577 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 7578 | |
0b86a832 | 7579 | if (found_key.objectid >= key->objectid && |
925baedd CM |
7580 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
7581 | ret = 0; | |
7582 | goto out; | |
7583 | } | |
0b86a832 | 7584 | path->slots[0]++; |
edbd8d4e | 7585 | } |
925baedd | 7586 | out: |
0b86a832 | 7587 | return ret; |
edbd8d4e CM |
7588 | } |
7589 | ||
0af3d00b JB |
7590 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
7591 | { | |
7592 | struct btrfs_block_group_cache *block_group; | |
7593 | u64 last = 0; | |
7594 | ||
7595 | while (1) { | |
7596 | struct inode *inode; | |
7597 | ||
7598 | block_group = btrfs_lookup_first_block_group(info, last); | |
7599 | while (block_group) { | |
7600 | spin_lock(&block_group->lock); | |
7601 | if (block_group->iref) | |
7602 | break; | |
7603 | spin_unlock(&block_group->lock); | |
7604 | block_group = next_block_group(info->tree_root, | |
7605 | block_group); | |
7606 | } | |
7607 | if (!block_group) { | |
7608 | if (last == 0) | |
7609 | break; | |
7610 | last = 0; | |
7611 | continue; | |
7612 | } | |
7613 | ||
7614 | inode = block_group->inode; | |
7615 | block_group->iref = 0; | |
7616 | block_group->inode = NULL; | |
7617 | spin_unlock(&block_group->lock); | |
7618 | iput(inode); | |
7619 | last = block_group->key.objectid + block_group->key.offset; | |
7620 | btrfs_put_block_group(block_group); | |
7621 | } | |
7622 | } | |
7623 | ||
1a40e23b ZY |
7624 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
7625 | { | |
7626 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 7627 | struct btrfs_space_info *space_info; |
11833d66 | 7628 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
7629 | struct rb_node *n; |
7630 | ||
11833d66 YZ |
7631 | down_write(&info->extent_commit_sem); |
7632 | while (!list_empty(&info->caching_block_groups)) { | |
7633 | caching_ctl = list_entry(info->caching_block_groups.next, | |
7634 | struct btrfs_caching_control, list); | |
7635 | list_del(&caching_ctl->list); | |
7636 | put_caching_control(caching_ctl); | |
7637 | } | |
7638 | up_write(&info->extent_commit_sem); | |
7639 | ||
1a40e23b ZY |
7640 | spin_lock(&info->block_group_cache_lock); |
7641 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
7642 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
7643 | cache_node); | |
1a40e23b ZY |
7644 | rb_erase(&block_group->cache_node, |
7645 | &info->block_group_cache_tree); | |
d899e052 YZ |
7646 | spin_unlock(&info->block_group_cache_lock); |
7647 | ||
80eb234a | 7648 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 7649 | list_del(&block_group->list); |
80eb234a | 7650 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 7651 | |
817d52f8 | 7652 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 7653 | wait_block_group_cache_done(block_group); |
817d52f8 | 7654 | |
3c14874a JB |
7655 | /* |
7656 | * We haven't cached this block group, which means we could | |
7657 | * possibly have excluded extents on this block group. | |
7658 | */ | |
7659 | if (block_group->cached == BTRFS_CACHE_NO) | |
7660 | free_excluded_extents(info->extent_root, block_group); | |
7661 | ||
817d52f8 | 7662 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 7663 | btrfs_put_block_group(block_group); |
d899e052 YZ |
7664 | |
7665 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
7666 | } |
7667 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
7668 | |
7669 | /* now that all the block groups are freed, go through and | |
7670 | * free all the space_info structs. This is only called during | |
7671 | * the final stages of unmount, and so we know nobody is | |
7672 | * using them. We call synchronize_rcu() once before we start, | |
7673 | * just to be on the safe side. | |
7674 | */ | |
7675 | synchronize_rcu(); | |
7676 | ||
8929ecfa YZ |
7677 | release_global_block_rsv(info); |
7678 | ||
4184ea7f CM |
7679 | while(!list_empty(&info->space_info)) { |
7680 | space_info = list_entry(info->space_info.next, | |
7681 | struct btrfs_space_info, | |
7682 | list); | |
f0486c68 | 7683 | if (space_info->bytes_pinned > 0 || |
fb25e914 JB |
7684 | space_info->bytes_reserved > 0 || |
7685 | space_info->bytes_may_use > 0) { | |
f0486c68 YZ |
7686 | WARN_ON(1); |
7687 | dump_space_info(space_info, 0, 0); | |
7688 | } | |
4184ea7f CM |
7689 | list_del(&space_info->list); |
7690 | kfree(space_info); | |
7691 | } | |
1a40e23b ZY |
7692 | return 0; |
7693 | } | |
7694 | ||
b742bb82 YZ |
7695 | static void __link_block_group(struct btrfs_space_info *space_info, |
7696 | struct btrfs_block_group_cache *cache) | |
7697 | { | |
7698 | int index = get_block_group_index(cache); | |
7699 | ||
7700 | down_write(&space_info->groups_sem); | |
7701 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
7702 | up_write(&space_info->groups_sem); | |
7703 | } | |
7704 | ||
9078a3e1 CM |
7705 | int btrfs_read_block_groups(struct btrfs_root *root) |
7706 | { | |
7707 | struct btrfs_path *path; | |
7708 | int ret; | |
9078a3e1 | 7709 | struct btrfs_block_group_cache *cache; |
be744175 | 7710 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 7711 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
7712 | struct btrfs_key key; |
7713 | struct btrfs_key found_key; | |
5f39d397 | 7714 | struct extent_buffer *leaf; |
0af3d00b JB |
7715 | int need_clear = 0; |
7716 | u64 cache_gen; | |
96b5179d | 7717 | |
be744175 | 7718 | root = info->extent_root; |
9078a3e1 | 7719 | key.objectid = 0; |
0b86a832 | 7720 | key.offset = 0; |
9078a3e1 | 7721 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
7722 | path = btrfs_alloc_path(); |
7723 | if (!path) | |
7724 | return -ENOMEM; | |
026fd317 | 7725 | path->reada = 1; |
9078a3e1 | 7726 | |
6c41761f | 7727 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); |
73bc1876 | 7728 | if (btrfs_test_opt(root, SPACE_CACHE) && |
6c41761f | 7729 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) |
0af3d00b | 7730 | need_clear = 1; |
88c2ba3b JB |
7731 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
7732 | need_clear = 1; | |
0af3d00b | 7733 | |
d397712b | 7734 | while (1) { |
0b86a832 | 7735 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
7736 | if (ret > 0) |
7737 | break; | |
0b86a832 CM |
7738 | if (ret != 0) |
7739 | goto error; | |
5f39d397 CM |
7740 | leaf = path->nodes[0]; |
7741 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 7742 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 7743 | if (!cache) { |
0b86a832 | 7744 | ret = -ENOMEM; |
f0486c68 | 7745 | goto error; |
9078a3e1 | 7746 | } |
34d52cb6 LZ |
7747 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
7748 | GFP_NOFS); | |
7749 | if (!cache->free_space_ctl) { | |
7750 | kfree(cache); | |
7751 | ret = -ENOMEM; | |
7752 | goto error; | |
7753 | } | |
3e1ad54f | 7754 | |
d2fb3437 | 7755 | atomic_set(&cache->count, 1); |
c286ac48 | 7756 | spin_lock_init(&cache->lock); |
817d52f8 | 7757 | cache->fs_info = info; |
0f9dd46c | 7758 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7759 | INIT_LIST_HEAD(&cache->cluster_list); |
96303081 | 7760 | |
cf7c1ef6 LB |
7761 | if (need_clear) { |
7762 | /* | |
7763 | * When we mount with old space cache, we need to | |
7764 | * set BTRFS_DC_CLEAR and set dirty flag. | |
7765 | * | |
7766 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
7767 | * truncate the old free space cache inode and | |
7768 | * setup a new one. | |
7769 | * b) Setting 'dirty flag' makes sure that we flush | |
7770 | * the new space cache info onto disk. | |
7771 | */ | |
0af3d00b | 7772 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 LB |
7773 | if (btrfs_test_opt(root, SPACE_CACHE)) |
7774 | cache->dirty = 1; | |
7775 | } | |
0af3d00b | 7776 | |
5f39d397 CM |
7777 | read_extent_buffer(leaf, &cache->item, |
7778 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
7779 | sizeof(cache->item)); | |
9078a3e1 | 7780 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 7781 | |
9078a3e1 | 7782 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 7783 | btrfs_release_path(path); |
0b86a832 | 7784 | cache->flags = btrfs_block_group_flags(&cache->item); |
817d52f8 JB |
7785 | cache->sectorsize = root->sectorsize; |
7786 | ||
34d52cb6 LZ |
7787 | btrfs_init_free_space_ctl(cache); |
7788 | ||
3c14874a JB |
7789 | /* |
7790 | * We need to exclude the super stripes now so that the space | |
7791 | * info has super bytes accounted for, otherwise we'll think | |
7792 | * we have more space than we actually do. | |
7793 | */ | |
7794 | exclude_super_stripes(root, cache); | |
7795 | ||
817d52f8 JB |
7796 | /* |
7797 | * check for two cases, either we are full, and therefore | |
7798 | * don't need to bother with the caching work since we won't | |
7799 | * find any space, or we are empty, and we can just add all | |
7800 | * the space in and be done with it. This saves us _alot_ of | |
7801 | * time, particularly in the full case. | |
7802 | */ | |
7803 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 7804 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7805 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 7806 | free_excluded_extents(root, cache); |
817d52f8 | 7807 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 7808 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
7809 | cache->cached = BTRFS_CACHE_FINISHED; |
7810 | add_new_free_space(cache, root->fs_info, | |
7811 | found_key.objectid, | |
7812 | found_key.objectid + | |
7813 | found_key.offset); | |
11833d66 | 7814 | free_excluded_extents(root, cache); |
817d52f8 | 7815 | } |
96b5179d | 7816 | |
6324fbf3 CM |
7817 | ret = update_space_info(info, cache->flags, found_key.offset, |
7818 | btrfs_block_group_used(&cache->item), | |
7819 | &space_info); | |
79787eaa | 7820 | BUG_ON(ret); /* -ENOMEM */ |
6324fbf3 | 7821 | cache->space_info = space_info; |
1b2da372 | 7822 | spin_lock(&cache->space_info->lock); |
f0486c68 | 7823 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7824 | spin_unlock(&cache->space_info->lock); |
7825 | ||
b742bb82 | 7826 | __link_block_group(space_info, cache); |
0f9dd46c JB |
7827 | |
7828 | ret = btrfs_add_block_group_cache(root->fs_info, cache); | |
79787eaa | 7829 | BUG_ON(ret); /* Logic error */ |
75ccf47d CM |
7830 | |
7831 | set_avail_alloc_bits(root->fs_info, cache->flags); | |
2b82032c | 7832 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
199c36ea | 7833 | set_block_group_ro(cache, 1); |
9078a3e1 | 7834 | } |
b742bb82 YZ |
7835 | |
7836 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
7837 | if (!(get_alloc_profile(root, space_info->flags) & | |
7838 | (BTRFS_BLOCK_GROUP_RAID10 | | |
7839 | BTRFS_BLOCK_GROUP_RAID1 | | |
7840 | BTRFS_BLOCK_GROUP_DUP))) | |
7841 | continue; | |
7842 | /* | |
7843 | * avoid allocating from un-mirrored block group if there are | |
7844 | * mirrored block groups. | |
7845 | */ | |
7846 | list_for_each_entry(cache, &space_info->block_groups[3], list) | |
199c36ea | 7847 | set_block_group_ro(cache, 1); |
b742bb82 | 7848 | list_for_each_entry(cache, &space_info->block_groups[4], list) |
199c36ea | 7849 | set_block_group_ro(cache, 1); |
9078a3e1 | 7850 | } |
f0486c68 YZ |
7851 | |
7852 | init_global_block_rsv(info); | |
0b86a832 CM |
7853 | ret = 0; |
7854 | error: | |
9078a3e1 | 7855 | btrfs_free_path(path); |
0b86a832 | 7856 | return ret; |
9078a3e1 | 7857 | } |
6324fbf3 | 7858 | |
ea658bad JB |
7859 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans, |
7860 | struct btrfs_root *root) | |
7861 | { | |
7862 | struct btrfs_block_group_cache *block_group, *tmp; | |
7863 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
7864 | struct btrfs_block_group_item item; | |
7865 | struct btrfs_key key; | |
7866 | int ret = 0; | |
7867 | ||
7868 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, | |
7869 | new_bg_list) { | |
7870 | list_del_init(&block_group->new_bg_list); | |
7871 | ||
7872 | if (ret) | |
7873 | continue; | |
7874 | ||
7875 | spin_lock(&block_group->lock); | |
7876 | memcpy(&item, &block_group->item, sizeof(item)); | |
7877 | memcpy(&key, &block_group->key, sizeof(key)); | |
7878 | spin_unlock(&block_group->lock); | |
7879 | ||
7880 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
7881 | sizeof(item)); | |
7882 | if (ret) | |
7883 | btrfs_abort_transaction(trans, extent_root, ret); | |
7884 | } | |
7885 | } | |
7886 | ||
6324fbf3 CM |
7887 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
7888 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 7889 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
7890 | u64 size) |
7891 | { | |
7892 | int ret; | |
6324fbf3 CM |
7893 | struct btrfs_root *extent_root; |
7894 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
7895 | |
7896 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 7897 | |
12fcfd22 | 7898 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 7899 | |
8f18cf13 | 7900 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
7901 | if (!cache) |
7902 | return -ENOMEM; | |
34d52cb6 LZ |
7903 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
7904 | GFP_NOFS); | |
7905 | if (!cache->free_space_ctl) { | |
7906 | kfree(cache); | |
7907 | return -ENOMEM; | |
7908 | } | |
0f9dd46c | 7909 | |
e17cade2 | 7910 | cache->key.objectid = chunk_offset; |
6324fbf3 | 7911 | cache->key.offset = size; |
d2fb3437 | 7912 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
96303081 | 7913 | cache->sectorsize = root->sectorsize; |
0af3d00b | 7914 | cache->fs_info = root->fs_info; |
96303081 | 7915 | |
d2fb3437 | 7916 | atomic_set(&cache->count, 1); |
c286ac48 | 7917 | spin_lock_init(&cache->lock); |
0f9dd46c | 7918 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 7919 | INIT_LIST_HEAD(&cache->cluster_list); |
ea658bad | 7920 | INIT_LIST_HEAD(&cache->new_bg_list); |
0ef3e66b | 7921 | |
34d52cb6 LZ |
7922 | btrfs_init_free_space_ctl(cache); |
7923 | ||
6324fbf3 | 7924 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
7925 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
7926 | cache->flags = type; | |
7927 | btrfs_set_block_group_flags(&cache->item, type); | |
7928 | ||
11833d66 | 7929 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 7930 | cache->cached = BTRFS_CACHE_FINISHED; |
11833d66 | 7931 | exclude_super_stripes(root, cache); |
96303081 | 7932 | |
817d52f8 JB |
7933 | add_new_free_space(cache, root->fs_info, chunk_offset, |
7934 | chunk_offset + size); | |
7935 | ||
11833d66 YZ |
7936 | free_excluded_extents(root, cache); |
7937 | ||
6324fbf3 CM |
7938 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
7939 | &cache->space_info); | |
79787eaa | 7940 | BUG_ON(ret); /* -ENOMEM */ |
c7c144db | 7941 | update_global_block_rsv(root->fs_info); |
1b2da372 JB |
7942 | |
7943 | spin_lock(&cache->space_info->lock); | |
f0486c68 | 7944 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
7945 | spin_unlock(&cache->space_info->lock); |
7946 | ||
b742bb82 | 7947 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 7948 | |
0f9dd46c | 7949 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
79787eaa | 7950 | BUG_ON(ret); /* Logic error */ |
c286ac48 | 7951 | |
ea658bad | 7952 | list_add_tail(&cache->new_bg_list, &trans->new_bgs); |
6324fbf3 | 7953 | |
d18a2c44 | 7954 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 7955 | |
6324fbf3 CM |
7956 | return 0; |
7957 | } | |
1a40e23b | 7958 | |
10ea00f5 ID |
7959 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
7960 | { | |
899c81ea ID |
7961 | u64 extra_flags = chunk_to_extended(flags) & |
7962 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 ID |
7963 | |
7964 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
7965 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
7966 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
7967 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
7968 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
7969 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
7970 | } | |
7971 | ||
1a40e23b ZY |
7972 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
7973 | struct btrfs_root *root, u64 group_start) | |
7974 | { | |
7975 | struct btrfs_path *path; | |
7976 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 7977 | struct btrfs_free_cluster *cluster; |
0af3d00b | 7978 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 7979 | struct btrfs_key key; |
0af3d00b | 7980 | struct inode *inode; |
1a40e23b | 7981 | int ret; |
10ea00f5 | 7982 | int index; |
89a55897 | 7983 | int factor; |
1a40e23b | 7984 | |
1a40e23b ZY |
7985 | root = root->fs_info->extent_root; |
7986 | ||
7987 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
7988 | BUG_ON(!block_group); | |
c146afad | 7989 | BUG_ON(!block_group->ro); |
1a40e23b | 7990 | |
9f7c43c9 | 7991 | /* |
7992 | * Free the reserved super bytes from this block group before | |
7993 | * remove it. | |
7994 | */ | |
7995 | free_excluded_extents(root, block_group); | |
7996 | ||
1a40e23b | 7997 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 7998 | index = get_block_group_index(block_group); |
89a55897 JB |
7999 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
8000 | BTRFS_BLOCK_GROUP_RAID1 | | |
8001 | BTRFS_BLOCK_GROUP_RAID10)) | |
8002 | factor = 2; | |
8003 | else | |
8004 | factor = 1; | |
1a40e23b | 8005 | |
44fb5511 CM |
8006 | /* make sure this block group isn't part of an allocation cluster */ |
8007 | cluster = &root->fs_info->data_alloc_cluster; | |
8008 | spin_lock(&cluster->refill_lock); | |
8009 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
8010 | spin_unlock(&cluster->refill_lock); | |
8011 | ||
8012 | /* | |
8013 | * make sure this block group isn't part of a metadata | |
8014 | * allocation cluster | |
8015 | */ | |
8016 | cluster = &root->fs_info->meta_alloc_cluster; | |
8017 | spin_lock(&cluster->refill_lock); | |
8018 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
8019 | spin_unlock(&cluster->refill_lock); | |
8020 | ||
1a40e23b | 8021 | path = btrfs_alloc_path(); |
d8926bb3 MF |
8022 | if (!path) { |
8023 | ret = -ENOMEM; | |
8024 | goto out; | |
8025 | } | |
1a40e23b | 8026 | |
10b2f34d | 8027 | inode = lookup_free_space_inode(tree_root, block_group, path); |
0af3d00b | 8028 | if (!IS_ERR(inode)) { |
b532402e | 8029 | ret = btrfs_orphan_add(trans, inode); |
79787eaa JM |
8030 | if (ret) { |
8031 | btrfs_add_delayed_iput(inode); | |
8032 | goto out; | |
8033 | } | |
0af3d00b JB |
8034 | clear_nlink(inode); |
8035 | /* One for the block groups ref */ | |
8036 | spin_lock(&block_group->lock); | |
8037 | if (block_group->iref) { | |
8038 | block_group->iref = 0; | |
8039 | block_group->inode = NULL; | |
8040 | spin_unlock(&block_group->lock); | |
8041 | iput(inode); | |
8042 | } else { | |
8043 | spin_unlock(&block_group->lock); | |
8044 | } | |
8045 | /* One for our lookup ref */ | |
455757c3 | 8046 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
8047 | } |
8048 | ||
8049 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
8050 | key.offset = block_group->key.objectid; | |
8051 | key.type = 0; | |
8052 | ||
8053 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
8054 | if (ret < 0) | |
8055 | goto out; | |
8056 | if (ret > 0) | |
b3b4aa74 | 8057 | btrfs_release_path(path); |
0af3d00b JB |
8058 | if (ret == 0) { |
8059 | ret = btrfs_del_item(trans, tree_root, path); | |
8060 | if (ret) | |
8061 | goto out; | |
b3b4aa74 | 8062 | btrfs_release_path(path); |
0af3d00b JB |
8063 | } |
8064 | ||
3dfdb934 | 8065 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
8066 | rb_erase(&block_group->cache_node, |
8067 | &root->fs_info->block_group_cache_tree); | |
3dfdb934 | 8068 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 8069 | |
80eb234a | 8070 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
8071 | /* |
8072 | * we must use list_del_init so people can check to see if they | |
8073 | * are still on the list after taking the semaphore | |
8074 | */ | |
8075 | list_del_init(&block_group->list); | |
10ea00f5 ID |
8076 | if (list_empty(&block_group->space_info->block_groups[index])) |
8077 | clear_avail_alloc_bits(root->fs_info, block_group->flags); | |
80eb234a | 8078 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 8079 | |
817d52f8 | 8080 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 8081 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
8082 | |
8083 | btrfs_remove_free_space_cache(block_group); | |
8084 | ||
c146afad YZ |
8085 | spin_lock(&block_group->space_info->lock); |
8086 | block_group->space_info->total_bytes -= block_group->key.offset; | |
8087 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 8088 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
c146afad | 8089 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 8090 | |
0af3d00b JB |
8091 | memcpy(&key, &block_group->key, sizeof(key)); |
8092 | ||
283bb197 | 8093 | btrfs_clear_space_info_full(root->fs_info); |
c146afad | 8094 | |
fa9c0d79 CM |
8095 | btrfs_put_block_group(block_group); |
8096 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
8097 | |
8098 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
8099 | if (ret > 0) | |
8100 | ret = -EIO; | |
8101 | if (ret < 0) | |
8102 | goto out; | |
8103 | ||
8104 | ret = btrfs_del_item(trans, root, path); | |
8105 | out: | |
8106 | btrfs_free_path(path); | |
8107 | return ret; | |
8108 | } | |
acce952b | 8109 | |
c59021f8 | 8110 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
8111 | { | |
8112 | struct btrfs_space_info *space_info; | |
1aba86d6 | 8113 | struct btrfs_super_block *disk_super; |
8114 | u64 features; | |
8115 | u64 flags; | |
8116 | int mixed = 0; | |
c59021f8 | 8117 | int ret; |
8118 | ||
6c41761f | 8119 | disk_super = fs_info->super_copy; |
1aba86d6 | 8120 | if (!btrfs_super_root(disk_super)) |
8121 | return 1; | |
c59021f8 | 8122 | |
1aba86d6 | 8123 | features = btrfs_super_incompat_flags(disk_super); |
8124 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
8125 | mixed = 1; | |
c59021f8 | 8126 | |
1aba86d6 | 8127 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
8128 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
c59021f8 | 8129 | if (ret) |
1aba86d6 | 8130 | goto out; |
c59021f8 | 8131 | |
1aba86d6 | 8132 | if (mixed) { |
8133 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
8134 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8135 | } else { | |
8136 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
8137 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8138 | if (ret) | |
8139 | goto out; | |
8140 | ||
8141 | flags = BTRFS_BLOCK_GROUP_DATA; | |
8142 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8143 | } | |
8144 | out: | |
c59021f8 | 8145 | return ret; |
8146 | } | |
8147 | ||
acce952b | 8148 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
8149 | { | |
8150 | return unpin_extent_range(root, start, end); | |
8151 | } | |
8152 | ||
8153 | int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 8154 | u64 num_bytes, u64 *actual_bytes) |
acce952b | 8155 | { |
5378e607 | 8156 | return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes); |
acce952b | 8157 | } |
f7039b1d LD |
8158 | |
8159 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) | |
8160 | { | |
8161 | struct btrfs_fs_info *fs_info = root->fs_info; | |
8162 | struct btrfs_block_group_cache *cache = NULL; | |
8163 | u64 group_trimmed; | |
8164 | u64 start; | |
8165 | u64 end; | |
8166 | u64 trimmed = 0; | |
2cac13e4 | 8167 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
8168 | int ret = 0; |
8169 | ||
2cac13e4 LB |
8170 | /* |
8171 | * try to trim all FS space, our block group may start from non-zero. | |
8172 | */ | |
8173 | if (range->len == total_bytes) | |
8174 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
8175 | else | |
8176 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
8177 | |
8178 | while (cache) { | |
8179 | if (cache->key.objectid >= (range->start + range->len)) { | |
8180 | btrfs_put_block_group(cache); | |
8181 | break; | |
8182 | } | |
8183 | ||
8184 | start = max(range->start, cache->key.objectid); | |
8185 | end = min(range->start + range->len, | |
8186 | cache->key.objectid + cache->key.offset); | |
8187 | ||
8188 | if (end - start >= range->minlen) { | |
8189 | if (!block_group_cache_done(cache)) { | |
8190 | ret = cache_block_group(cache, NULL, root, 0); | |
8191 | if (!ret) | |
8192 | wait_block_group_cache_done(cache); | |
8193 | } | |
8194 | ret = btrfs_trim_block_group(cache, | |
8195 | &group_trimmed, | |
8196 | start, | |
8197 | end, | |
8198 | range->minlen); | |
8199 | ||
8200 | trimmed += group_trimmed; | |
8201 | if (ret) { | |
8202 | btrfs_put_block_group(cache); | |
8203 | break; | |
8204 | } | |
8205 | } | |
8206 | ||
8207 | cache = next_block_group(fs_info->tree_root, cache); | |
8208 | } | |
8209 | ||
8210 | range->len = trimmed; | |
8211 | return ret; | |
8212 | } |