]>
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" |
53b381b3 | 34 | #include "raid56.h" |
925baedd | 35 | #include "locking.h" |
fa9c0d79 | 36 | #include "free-space-cache.h" |
3fed40cc | 37 | #include "math.h" |
fec577fb | 38 | |
709c0486 AJ |
39 | #undef SCRAMBLE_DELAYED_REFS |
40 | ||
9e622d6b MX |
41 | /* |
42 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
43 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
44 | * if we really need one. | |
45 | * | |
0e4f8f88 CM |
46 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
47 | * if we have very few chunks already allocated. This is | |
48 | * used as part of the clustering code to help make sure | |
49 | * we have a good pool of storage to cluster in, without | |
50 | * filling the FS with empty chunks | |
51 | * | |
9e622d6b MX |
52 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
53 | * | |
0e4f8f88 CM |
54 | */ |
55 | enum { | |
56 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
57 | CHUNK_ALLOC_LIMITED = 1, |
58 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
59 | }; |
60 | ||
fb25e914 JB |
61 | /* |
62 | * Control how reservations are dealt with. | |
63 | * | |
64 | * RESERVE_FREE - freeing a reservation. | |
65 | * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for | |
66 | * ENOSPC accounting | |
67 | * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update | |
68 | * bytes_may_use as the ENOSPC accounting is done elsewhere | |
69 | */ | |
70 | enum { | |
71 | RESERVE_FREE = 0, | |
72 | RESERVE_ALLOC = 1, | |
73 | RESERVE_ALLOC_NO_ACCOUNT = 2, | |
74 | }; | |
75 | ||
c53d613e | 76 | static int update_block_group(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); | |
5d80366e JB |
106 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
107 | u64 num_bytes); | |
6a63209f | 108 | |
817d52f8 JB |
109 | static noinline int |
110 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
111 | { | |
112 | smp_mb(); | |
113 | return cache->cached == BTRFS_CACHE_FINISHED; | |
114 | } | |
115 | ||
0f9dd46c JB |
116 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
117 | { | |
118 | return (cache->flags & bits) == bits; | |
119 | } | |
120 | ||
62a45b60 | 121 | static void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
122 | { |
123 | atomic_inc(&cache->count); | |
124 | } | |
125 | ||
126 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
127 | { | |
f0486c68 YZ |
128 | if (atomic_dec_and_test(&cache->count)) { |
129 | WARN_ON(cache->pinned > 0); | |
130 | WARN_ON(cache->reserved > 0); | |
34d52cb6 | 131 | kfree(cache->free_space_ctl); |
11dfe35a | 132 | kfree(cache); |
f0486c68 | 133 | } |
11dfe35a JB |
134 | } |
135 | ||
0f9dd46c JB |
136 | /* |
137 | * this adds the block group to the fs_info rb tree for the block group | |
138 | * cache | |
139 | */ | |
b2950863 | 140 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
141 | struct btrfs_block_group_cache *block_group) |
142 | { | |
143 | struct rb_node **p; | |
144 | struct rb_node *parent = NULL; | |
145 | struct btrfs_block_group_cache *cache; | |
146 | ||
147 | spin_lock(&info->block_group_cache_lock); | |
148 | p = &info->block_group_cache_tree.rb_node; | |
149 | ||
150 | while (*p) { | |
151 | parent = *p; | |
152 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
153 | cache_node); | |
154 | if (block_group->key.objectid < cache->key.objectid) { | |
155 | p = &(*p)->rb_left; | |
156 | } else if (block_group->key.objectid > cache->key.objectid) { | |
157 | p = &(*p)->rb_right; | |
158 | } else { | |
159 | spin_unlock(&info->block_group_cache_lock); | |
160 | return -EEXIST; | |
161 | } | |
162 | } | |
163 | ||
164 | rb_link_node(&block_group->cache_node, parent, p); | |
165 | rb_insert_color(&block_group->cache_node, | |
166 | &info->block_group_cache_tree); | |
a1897fdd LB |
167 | |
168 | if (info->first_logical_byte > block_group->key.objectid) | |
169 | info->first_logical_byte = block_group->key.objectid; | |
170 | ||
0f9dd46c JB |
171 | spin_unlock(&info->block_group_cache_lock); |
172 | ||
173 | return 0; | |
174 | } | |
175 | ||
176 | /* | |
177 | * This will return the block group at or after bytenr if contains is 0, else | |
178 | * it will return the block group that contains the bytenr | |
179 | */ | |
180 | static struct btrfs_block_group_cache * | |
181 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
182 | int contains) | |
183 | { | |
184 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
185 | struct rb_node *n; | |
186 | u64 end, start; | |
187 | ||
188 | spin_lock(&info->block_group_cache_lock); | |
189 | n = info->block_group_cache_tree.rb_node; | |
190 | ||
191 | while (n) { | |
192 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
193 | cache_node); | |
194 | end = cache->key.objectid + cache->key.offset - 1; | |
195 | start = cache->key.objectid; | |
196 | ||
197 | if (bytenr < start) { | |
198 | if (!contains && (!ret || start < ret->key.objectid)) | |
199 | ret = cache; | |
200 | n = n->rb_left; | |
201 | } else if (bytenr > start) { | |
202 | if (contains && bytenr <= end) { | |
203 | ret = cache; | |
204 | break; | |
205 | } | |
206 | n = n->rb_right; | |
207 | } else { | |
208 | ret = cache; | |
209 | break; | |
210 | } | |
211 | } | |
a1897fdd | 212 | if (ret) { |
11dfe35a | 213 | btrfs_get_block_group(ret); |
a1897fdd LB |
214 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
215 | info->first_logical_byte = ret->key.objectid; | |
216 | } | |
0f9dd46c JB |
217 | spin_unlock(&info->block_group_cache_lock); |
218 | ||
219 | return ret; | |
220 | } | |
221 | ||
11833d66 YZ |
222 | static int add_excluded_extent(struct btrfs_root *root, |
223 | u64 start, u64 num_bytes) | |
817d52f8 | 224 | { |
11833d66 YZ |
225 | u64 end = start + num_bytes - 1; |
226 | set_extent_bits(&root->fs_info->freed_extents[0], | |
227 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
228 | set_extent_bits(&root->fs_info->freed_extents[1], | |
229 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
230 | return 0; | |
231 | } | |
817d52f8 | 232 | |
11833d66 YZ |
233 | static void free_excluded_extents(struct btrfs_root *root, |
234 | struct btrfs_block_group_cache *cache) | |
235 | { | |
236 | u64 start, end; | |
817d52f8 | 237 | |
11833d66 YZ |
238 | start = cache->key.objectid; |
239 | end = start + cache->key.offset - 1; | |
240 | ||
241 | clear_extent_bits(&root->fs_info->freed_extents[0], | |
242 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
243 | clear_extent_bits(&root->fs_info->freed_extents[1], | |
244 | start, end, EXTENT_UPTODATE, GFP_NOFS); | |
817d52f8 JB |
245 | } |
246 | ||
11833d66 YZ |
247 | static int exclude_super_stripes(struct btrfs_root *root, |
248 | struct btrfs_block_group_cache *cache) | |
817d52f8 | 249 | { |
817d52f8 JB |
250 | u64 bytenr; |
251 | u64 *logical; | |
252 | int stripe_len; | |
253 | int i, nr, ret; | |
254 | ||
06b2331f YZ |
255 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
256 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
257 | cache->bytes_super += stripe_len; | |
258 | ret = add_excluded_extent(root, cache->key.objectid, | |
259 | stripe_len); | |
835d974f JB |
260 | if (ret) |
261 | return ret; | |
06b2331f YZ |
262 | } |
263 | ||
817d52f8 JB |
264 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
265 | bytenr = btrfs_sb_offset(i); | |
266 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, | |
267 | cache->key.objectid, bytenr, | |
268 | 0, &logical, &nr, &stripe_len); | |
835d974f JB |
269 | if (ret) |
270 | return ret; | |
11833d66 | 271 | |
817d52f8 | 272 | while (nr--) { |
51bf5f0b JB |
273 | u64 start, len; |
274 | ||
275 | if (logical[nr] > cache->key.objectid + | |
276 | cache->key.offset) | |
277 | continue; | |
278 | ||
279 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
280 | continue; | |
281 | ||
282 | start = logical[nr]; | |
283 | if (start < cache->key.objectid) { | |
284 | start = cache->key.objectid; | |
285 | len = (logical[nr] + stripe_len) - start; | |
286 | } else { | |
287 | len = min_t(u64, stripe_len, | |
288 | cache->key.objectid + | |
289 | cache->key.offset - start); | |
290 | } | |
291 | ||
292 | cache->bytes_super += len; | |
293 | ret = add_excluded_extent(root, start, len); | |
835d974f JB |
294 | if (ret) { |
295 | kfree(logical); | |
296 | return ret; | |
297 | } | |
817d52f8 | 298 | } |
11833d66 | 299 | |
817d52f8 JB |
300 | kfree(logical); |
301 | } | |
817d52f8 JB |
302 | return 0; |
303 | } | |
304 | ||
11833d66 YZ |
305 | static struct btrfs_caching_control * |
306 | get_caching_control(struct btrfs_block_group_cache *cache) | |
307 | { | |
308 | struct btrfs_caching_control *ctl; | |
309 | ||
310 | spin_lock(&cache->lock); | |
311 | if (cache->cached != BTRFS_CACHE_STARTED) { | |
312 | spin_unlock(&cache->lock); | |
313 | return NULL; | |
314 | } | |
315 | ||
dde5abee JB |
316 | /* We're loading it the fast way, so we don't have a caching_ctl. */ |
317 | if (!cache->caching_ctl) { | |
318 | spin_unlock(&cache->lock); | |
11833d66 YZ |
319 | return NULL; |
320 | } | |
321 | ||
322 | ctl = cache->caching_ctl; | |
323 | atomic_inc(&ctl->count); | |
324 | spin_unlock(&cache->lock); | |
325 | return ctl; | |
326 | } | |
327 | ||
328 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
329 | { | |
330 | if (atomic_dec_and_test(&ctl->count)) | |
331 | kfree(ctl); | |
332 | } | |
333 | ||
0f9dd46c JB |
334 | /* |
335 | * this is only called by cache_block_group, since we could have freed extents | |
336 | * we need to check the pinned_extents for any extents that can't be used yet | |
337 | * since their free space will be released as soon as the transaction commits. | |
338 | */ | |
817d52f8 | 339 | static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
0f9dd46c JB |
340 | struct btrfs_fs_info *info, u64 start, u64 end) |
341 | { | |
817d52f8 | 342 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
343 | int ret; |
344 | ||
345 | while (start < end) { | |
11833d66 | 346 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 347 | &extent_start, &extent_end, |
e6138876 JB |
348 | EXTENT_DIRTY | EXTENT_UPTODATE, |
349 | NULL); | |
0f9dd46c JB |
350 | if (ret) |
351 | break; | |
352 | ||
06b2331f | 353 | if (extent_start <= start) { |
0f9dd46c JB |
354 | start = extent_end + 1; |
355 | } else if (extent_start > start && extent_start < end) { | |
356 | size = extent_start - start; | |
817d52f8 | 357 | total_added += size; |
ea6a478e JB |
358 | ret = btrfs_add_free_space(block_group, start, |
359 | size); | |
79787eaa | 360 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
361 | start = extent_end + 1; |
362 | } else { | |
363 | break; | |
364 | } | |
365 | } | |
366 | ||
367 | if (start < end) { | |
368 | size = end - start; | |
817d52f8 | 369 | total_added += size; |
ea6a478e | 370 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 371 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
372 | } |
373 | ||
817d52f8 | 374 | return total_added; |
0f9dd46c JB |
375 | } |
376 | ||
bab39bf9 | 377 | static noinline void caching_thread(struct btrfs_work *work) |
e37c9e69 | 378 | { |
bab39bf9 JB |
379 | struct btrfs_block_group_cache *block_group; |
380 | struct btrfs_fs_info *fs_info; | |
381 | struct btrfs_caching_control *caching_ctl; | |
382 | struct btrfs_root *extent_root; | |
e37c9e69 | 383 | struct btrfs_path *path; |
5f39d397 | 384 | struct extent_buffer *leaf; |
11833d66 | 385 | struct btrfs_key key; |
817d52f8 | 386 | u64 total_found = 0; |
11833d66 YZ |
387 | u64 last = 0; |
388 | u32 nritems; | |
389 | int ret = 0; | |
f510cfec | 390 | |
bab39bf9 JB |
391 | caching_ctl = container_of(work, struct btrfs_caching_control, work); |
392 | block_group = caching_ctl->block_group; | |
393 | fs_info = block_group->fs_info; | |
394 | extent_root = fs_info->extent_root; | |
395 | ||
e37c9e69 CM |
396 | path = btrfs_alloc_path(); |
397 | if (!path) | |
bab39bf9 | 398 | goto out; |
7d7d6068 | 399 | |
817d52f8 | 400 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 401 | |
5cd57b2c | 402 | /* |
817d52f8 JB |
403 | * We don't want to deadlock with somebody trying to allocate a new |
404 | * extent for the extent root while also trying to search the extent | |
405 | * root to add free space. So we skip locking and search the commit | |
406 | * root, since its read-only | |
5cd57b2c CM |
407 | */ |
408 | path->skip_locking = 1; | |
817d52f8 | 409 | path->search_commit_root = 1; |
026fd317 | 410 | path->reada = 1; |
817d52f8 | 411 | |
e4404d6e | 412 | key.objectid = last; |
e37c9e69 | 413 | key.offset = 0; |
11833d66 | 414 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 415 | again: |
11833d66 | 416 | mutex_lock(&caching_ctl->mutex); |
013f1b12 CM |
417 | /* need to make sure the commit_root doesn't disappear */ |
418 | down_read(&fs_info->extent_commit_sem); | |
419 | ||
11833d66 | 420 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 421 | if (ret < 0) |
ef8bbdfe | 422 | goto err; |
a512bbf8 | 423 | |
11833d66 YZ |
424 | leaf = path->nodes[0]; |
425 | nritems = btrfs_header_nritems(leaf); | |
426 | ||
d397712b | 427 | while (1) { |
7841cb28 | 428 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 429 | last = (u64)-1; |
817d52f8 | 430 | break; |
f25784b3 | 431 | } |
817d52f8 | 432 | |
11833d66 YZ |
433 | if (path->slots[0] < nritems) { |
434 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
435 | } else { | |
436 | ret = find_next_key(path, 0, &key); | |
437 | if (ret) | |
e37c9e69 | 438 | break; |
817d52f8 | 439 | |
0a3896d0 | 440 | if (need_resched()) { |
589d8ade | 441 | caching_ctl->progress = last; |
ff5714cc | 442 | btrfs_release_path(path); |
589d8ade JB |
443 | up_read(&fs_info->extent_commit_sem); |
444 | mutex_unlock(&caching_ctl->mutex); | |
11833d66 | 445 | cond_resched(); |
589d8ade JB |
446 | goto again; |
447 | } | |
0a3896d0 JB |
448 | |
449 | ret = btrfs_next_leaf(extent_root, path); | |
450 | if (ret < 0) | |
451 | goto err; | |
452 | if (ret) | |
453 | break; | |
589d8ade JB |
454 | leaf = path->nodes[0]; |
455 | nritems = btrfs_header_nritems(leaf); | |
456 | continue; | |
11833d66 | 457 | } |
817d52f8 | 458 | |
11833d66 YZ |
459 | if (key.objectid < block_group->key.objectid) { |
460 | path->slots[0]++; | |
817d52f8 | 461 | continue; |
e37c9e69 | 462 | } |
0f9dd46c | 463 | |
e37c9e69 | 464 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 465 | block_group->key.offset) |
e37c9e69 | 466 | break; |
7d7d6068 | 467 | |
3173a18f JB |
468 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
469 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
817d52f8 JB |
470 | total_found += add_new_free_space(block_group, |
471 | fs_info, last, | |
472 | key.objectid); | |
3173a18f JB |
473 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
474 | last = key.objectid + | |
475 | fs_info->tree_root->leafsize; | |
476 | else | |
477 | last = key.objectid + key.offset; | |
817d52f8 | 478 | |
11833d66 YZ |
479 | if (total_found > (1024 * 1024 * 2)) { |
480 | total_found = 0; | |
481 | wake_up(&caching_ctl->wait); | |
482 | } | |
817d52f8 | 483 | } |
e37c9e69 CM |
484 | path->slots[0]++; |
485 | } | |
817d52f8 | 486 | ret = 0; |
e37c9e69 | 487 | |
817d52f8 JB |
488 | total_found += add_new_free_space(block_group, fs_info, last, |
489 | block_group->key.objectid + | |
490 | block_group->key.offset); | |
11833d66 | 491 | caching_ctl->progress = (u64)-1; |
817d52f8 JB |
492 | |
493 | spin_lock(&block_group->lock); | |
11833d66 | 494 | block_group->caching_ctl = NULL; |
817d52f8 JB |
495 | block_group->cached = BTRFS_CACHE_FINISHED; |
496 | spin_unlock(&block_group->lock); | |
0f9dd46c | 497 | |
54aa1f4d | 498 | err: |
e37c9e69 | 499 | btrfs_free_path(path); |
276e680d | 500 | up_read(&fs_info->extent_commit_sem); |
817d52f8 | 501 | |
11833d66 YZ |
502 | free_excluded_extents(extent_root, block_group); |
503 | ||
504 | mutex_unlock(&caching_ctl->mutex); | |
bab39bf9 | 505 | out: |
11833d66 YZ |
506 | wake_up(&caching_ctl->wait); |
507 | ||
508 | put_caching_control(caching_ctl); | |
11dfe35a | 509 | btrfs_put_block_group(block_group); |
817d52f8 JB |
510 | } |
511 | ||
9d66e233 | 512 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 513 | int load_cache_only) |
817d52f8 | 514 | { |
291c7d2f | 515 | DEFINE_WAIT(wait); |
11833d66 YZ |
516 | struct btrfs_fs_info *fs_info = cache->fs_info; |
517 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
518 | int ret = 0; |
519 | ||
291c7d2f | 520 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
521 | if (!caching_ctl) |
522 | return -ENOMEM; | |
291c7d2f JB |
523 | |
524 | INIT_LIST_HEAD(&caching_ctl->list); | |
525 | mutex_init(&caching_ctl->mutex); | |
526 | init_waitqueue_head(&caching_ctl->wait); | |
527 | caching_ctl->block_group = cache; | |
528 | caching_ctl->progress = cache->key.objectid; | |
529 | atomic_set(&caching_ctl->count, 1); | |
530 | caching_ctl->work.func = caching_thread; | |
531 | ||
532 | spin_lock(&cache->lock); | |
533 | /* | |
534 | * This should be a rare occasion, but this could happen I think in the | |
535 | * case where one thread starts to load the space cache info, and then | |
536 | * some other thread starts a transaction commit which tries to do an | |
537 | * allocation while the other thread is still loading the space cache | |
538 | * info. The previous loop should have kept us from choosing this block | |
539 | * group, but if we've moved to the state where we will wait on caching | |
540 | * block groups we need to first check if we're doing a fast load here, | |
541 | * so we can wait for it to finish, otherwise we could end up allocating | |
542 | * from a block group who's cache gets evicted for one reason or | |
543 | * another. | |
544 | */ | |
545 | while (cache->cached == BTRFS_CACHE_FAST) { | |
546 | struct btrfs_caching_control *ctl; | |
547 | ||
548 | ctl = cache->caching_ctl; | |
549 | atomic_inc(&ctl->count); | |
550 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); | |
551 | spin_unlock(&cache->lock); | |
552 | ||
553 | schedule(); | |
554 | ||
555 | finish_wait(&ctl->wait, &wait); | |
556 | put_caching_control(ctl); | |
557 | spin_lock(&cache->lock); | |
558 | } | |
559 | ||
560 | if (cache->cached != BTRFS_CACHE_NO) { | |
561 | spin_unlock(&cache->lock); | |
562 | kfree(caching_ctl); | |
11833d66 | 563 | return 0; |
291c7d2f JB |
564 | } |
565 | WARN_ON(cache->caching_ctl); | |
566 | cache->caching_ctl = caching_ctl; | |
567 | cache->cached = BTRFS_CACHE_FAST; | |
568 | spin_unlock(&cache->lock); | |
11833d66 | 569 | |
d53ba474 | 570 | if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) { |
9d66e233 JB |
571 | ret = load_free_space_cache(fs_info, cache); |
572 | ||
573 | spin_lock(&cache->lock); | |
574 | if (ret == 1) { | |
291c7d2f | 575 | cache->caching_ctl = NULL; |
9d66e233 JB |
576 | cache->cached = BTRFS_CACHE_FINISHED; |
577 | cache->last_byte_to_unpin = (u64)-1; | |
578 | } else { | |
291c7d2f JB |
579 | if (load_cache_only) { |
580 | cache->caching_ctl = NULL; | |
581 | cache->cached = BTRFS_CACHE_NO; | |
582 | } else { | |
583 | cache->cached = BTRFS_CACHE_STARTED; | |
584 | } | |
9d66e233 JB |
585 | } |
586 | spin_unlock(&cache->lock); | |
291c7d2f | 587 | wake_up(&caching_ctl->wait); |
3c14874a | 588 | if (ret == 1) { |
291c7d2f | 589 | put_caching_control(caching_ctl); |
3c14874a | 590 | free_excluded_extents(fs_info->extent_root, cache); |
9d66e233 | 591 | return 0; |
3c14874a | 592 | } |
291c7d2f JB |
593 | } else { |
594 | /* | |
595 | * We are not going to do the fast caching, set cached to the | |
596 | * appropriate value and wakeup any waiters. | |
597 | */ | |
598 | spin_lock(&cache->lock); | |
599 | if (load_cache_only) { | |
600 | cache->caching_ctl = NULL; | |
601 | cache->cached = BTRFS_CACHE_NO; | |
602 | } else { | |
603 | cache->cached = BTRFS_CACHE_STARTED; | |
604 | } | |
605 | spin_unlock(&cache->lock); | |
606 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
607 | } |
608 | ||
291c7d2f JB |
609 | if (load_cache_only) { |
610 | put_caching_control(caching_ctl); | |
11833d66 | 611 | return 0; |
817d52f8 | 612 | } |
817d52f8 | 613 | |
11833d66 | 614 | down_write(&fs_info->extent_commit_sem); |
291c7d2f | 615 | atomic_inc(&caching_ctl->count); |
11833d66 YZ |
616 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
617 | up_write(&fs_info->extent_commit_sem); | |
618 | ||
11dfe35a | 619 | btrfs_get_block_group(cache); |
11833d66 | 620 | |
bab39bf9 | 621 | btrfs_queue_worker(&fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 622 | |
ef8bbdfe | 623 | return ret; |
e37c9e69 CM |
624 | } |
625 | ||
0f9dd46c JB |
626 | /* |
627 | * return the block group that starts at or after bytenr | |
628 | */ | |
d397712b CM |
629 | static struct btrfs_block_group_cache * |
630 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 631 | { |
0f9dd46c | 632 | struct btrfs_block_group_cache *cache; |
0ef3e66b | 633 | |
0f9dd46c | 634 | cache = block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b | 635 | |
0f9dd46c | 636 | return cache; |
0ef3e66b CM |
637 | } |
638 | ||
0f9dd46c | 639 | /* |
9f55684c | 640 | * return the block group that contains the given bytenr |
0f9dd46c | 641 | */ |
d397712b CM |
642 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
643 | struct btrfs_fs_info *info, | |
644 | u64 bytenr) | |
be744175 | 645 | { |
0f9dd46c | 646 | struct btrfs_block_group_cache *cache; |
be744175 | 647 | |
0f9dd46c | 648 | cache = block_group_cache_tree_search(info, bytenr, 1); |
96b5179d | 649 | |
0f9dd46c | 650 | return cache; |
be744175 | 651 | } |
0b86a832 | 652 | |
0f9dd46c JB |
653 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
654 | u64 flags) | |
6324fbf3 | 655 | { |
0f9dd46c | 656 | struct list_head *head = &info->space_info; |
0f9dd46c | 657 | struct btrfs_space_info *found; |
4184ea7f | 658 | |
52ba6929 | 659 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 660 | |
4184ea7f CM |
661 | rcu_read_lock(); |
662 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 663 | if (found->flags & flags) { |
4184ea7f | 664 | rcu_read_unlock(); |
0f9dd46c | 665 | return found; |
4184ea7f | 666 | } |
0f9dd46c | 667 | } |
4184ea7f | 668 | rcu_read_unlock(); |
0f9dd46c | 669 | return NULL; |
6324fbf3 CM |
670 | } |
671 | ||
4184ea7f CM |
672 | /* |
673 | * after adding space to the filesystem, we need to clear the full flags | |
674 | * on all the space infos. | |
675 | */ | |
676 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
677 | { | |
678 | struct list_head *head = &info->space_info; | |
679 | struct btrfs_space_info *found; | |
680 | ||
681 | rcu_read_lock(); | |
682 | list_for_each_entry_rcu(found, head, list) | |
683 | found->full = 0; | |
684 | rcu_read_unlock(); | |
685 | } | |
686 | ||
d2fb3437 YZ |
687 | u64 btrfs_find_block_group(struct btrfs_root *root, |
688 | u64 search_start, u64 search_hint, int owner) | |
cd1bc465 | 689 | { |
96b5179d | 690 | struct btrfs_block_group_cache *cache; |
cd1bc465 | 691 | u64 used; |
d2fb3437 YZ |
692 | u64 last = max(search_hint, search_start); |
693 | u64 group_start = 0; | |
31f3c99b | 694 | int full_search = 0; |
d2fb3437 | 695 | int factor = 9; |
0ef3e66b | 696 | int wrapped = 0; |
31f3c99b | 697 | again: |
e8569813 ZY |
698 | while (1) { |
699 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
0f9dd46c JB |
700 | if (!cache) |
701 | break; | |
96b5179d | 702 | |
c286ac48 | 703 | spin_lock(&cache->lock); |
96b5179d CM |
704 | last = cache->key.objectid + cache->key.offset; |
705 | used = btrfs_block_group_used(&cache->item); | |
706 | ||
d2fb3437 YZ |
707 | if ((full_search || !cache->ro) && |
708 | block_group_bits(cache, BTRFS_BLOCK_GROUP_METADATA)) { | |
e8569813 | 709 | if (used + cache->pinned + cache->reserved < |
d2fb3437 YZ |
710 | div_factor(cache->key.offset, factor)) { |
711 | group_start = cache->key.objectid; | |
c286ac48 | 712 | spin_unlock(&cache->lock); |
fa9c0d79 | 713 | btrfs_put_block_group(cache); |
8790d502 CM |
714 | goto found; |
715 | } | |
6324fbf3 | 716 | } |
c286ac48 | 717 | spin_unlock(&cache->lock); |
fa9c0d79 | 718 | btrfs_put_block_group(cache); |
de428b63 | 719 | cond_resched(); |
cd1bc465 | 720 | } |
0ef3e66b CM |
721 | if (!wrapped) { |
722 | last = search_start; | |
723 | wrapped = 1; | |
724 | goto again; | |
725 | } | |
726 | if (!full_search && factor < 10) { | |
be744175 | 727 | last = search_start; |
31f3c99b | 728 | full_search = 1; |
0ef3e66b | 729 | factor = 10; |
31f3c99b CM |
730 | goto again; |
731 | } | |
be744175 | 732 | found: |
d2fb3437 | 733 | return group_start; |
925baedd | 734 | } |
0f9dd46c | 735 | |
e02119d5 | 736 | /* simple helper to search for an existing extent at a given offset */ |
31840ae1 | 737 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
e02119d5 CM |
738 | { |
739 | int ret; | |
740 | struct btrfs_key key; | |
31840ae1 | 741 | struct btrfs_path *path; |
e02119d5 | 742 | |
31840ae1 | 743 | path = btrfs_alloc_path(); |
d8926bb3 MF |
744 | if (!path) |
745 | return -ENOMEM; | |
746 | ||
e02119d5 CM |
747 | key.objectid = start; |
748 | key.offset = len; | |
3173a18f | 749 | key.type = BTRFS_EXTENT_ITEM_KEY; |
e02119d5 CM |
750 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, |
751 | 0, 0); | |
3173a18f JB |
752 | if (ret > 0) { |
753 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); | |
754 | if (key.objectid == start && | |
755 | key.type == BTRFS_METADATA_ITEM_KEY) | |
756 | ret = 0; | |
757 | } | |
31840ae1 | 758 | btrfs_free_path(path); |
7bb86316 CM |
759 | return ret; |
760 | } | |
761 | ||
a22285a6 | 762 | /* |
3173a18f | 763 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
764 | * |
765 | * the head node for delayed ref is used to store the sum of all the | |
766 | * reference count modifications queued up in the rbtree. the head | |
767 | * node may also store the extent flags to set. This way you can check | |
768 | * to see what the reference count and extent flags would be if all of | |
769 | * the delayed refs are not processed. | |
770 | */ | |
771 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
772 | struct btrfs_root *root, u64 bytenr, | |
3173a18f | 773 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
774 | { |
775 | struct btrfs_delayed_ref_head *head; | |
776 | struct btrfs_delayed_ref_root *delayed_refs; | |
777 | struct btrfs_path *path; | |
778 | struct btrfs_extent_item *ei; | |
779 | struct extent_buffer *leaf; | |
780 | struct btrfs_key key; | |
781 | u32 item_size; | |
782 | u64 num_refs; | |
783 | u64 extent_flags; | |
784 | int ret; | |
785 | ||
3173a18f JB |
786 | /* |
787 | * If we don't have skinny metadata, don't bother doing anything | |
788 | * different | |
789 | */ | |
790 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) { | |
791 | offset = root->leafsize; | |
792 | metadata = 0; | |
793 | } | |
794 | ||
a22285a6 YZ |
795 | path = btrfs_alloc_path(); |
796 | if (!path) | |
797 | return -ENOMEM; | |
798 | ||
3173a18f JB |
799 | if (metadata) { |
800 | key.objectid = bytenr; | |
801 | key.type = BTRFS_METADATA_ITEM_KEY; | |
802 | key.offset = offset; | |
803 | } else { | |
804 | key.objectid = bytenr; | |
805 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
806 | key.offset = offset; | |
807 | } | |
808 | ||
a22285a6 YZ |
809 | if (!trans) { |
810 | path->skip_locking = 1; | |
811 | path->search_commit_root = 1; | |
812 | } | |
813 | again: | |
814 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, | |
815 | &key, path, 0, 0); | |
816 | if (ret < 0) | |
817 | goto out_free; | |
818 | ||
3173a18f JB |
819 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
820 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
821 | key.offset = root->leafsize; | |
822 | btrfs_release_path(path); | |
823 | goto again; | |
824 | } | |
825 | ||
a22285a6 YZ |
826 | if (ret == 0) { |
827 | leaf = path->nodes[0]; | |
828 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
829 | if (item_size >= sizeof(*ei)) { | |
830 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
831 | struct btrfs_extent_item); | |
832 | num_refs = btrfs_extent_refs(leaf, ei); | |
833 | extent_flags = btrfs_extent_flags(leaf, ei); | |
834 | } else { | |
835 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
836 | struct btrfs_extent_item_v0 *ei0; | |
837 | BUG_ON(item_size != sizeof(*ei0)); | |
838 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
839 | struct btrfs_extent_item_v0); | |
840 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
841 | /* FIXME: this isn't correct for data */ | |
842 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
843 | #else | |
844 | BUG(); | |
845 | #endif | |
846 | } | |
847 | BUG_ON(num_refs == 0); | |
848 | } else { | |
849 | num_refs = 0; | |
850 | extent_flags = 0; | |
851 | ret = 0; | |
852 | } | |
853 | ||
854 | if (!trans) | |
855 | goto out; | |
856 | ||
857 | delayed_refs = &trans->transaction->delayed_refs; | |
858 | spin_lock(&delayed_refs->lock); | |
859 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
860 | if (head) { | |
861 | if (!mutex_trylock(&head->mutex)) { | |
862 | atomic_inc(&head->node.refs); | |
863 | spin_unlock(&delayed_refs->lock); | |
864 | ||
b3b4aa74 | 865 | btrfs_release_path(path); |
a22285a6 | 866 | |
8cc33e5c DS |
867 | /* |
868 | * Mutex was contended, block until it's released and try | |
869 | * again | |
870 | */ | |
a22285a6 YZ |
871 | mutex_lock(&head->mutex); |
872 | mutex_unlock(&head->mutex); | |
873 | btrfs_put_delayed_ref(&head->node); | |
874 | goto again; | |
875 | } | |
876 | if (head->extent_op && head->extent_op->update_flags) | |
877 | extent_flags |= head->extent_op->flags_to_set; | |
878 | else | |
879 | BUG_ON(num_refs == 0); | |
880 | ||
881 | num_refs += head->node.ref_mod; | |
882 | mutex_unlock(&head->mutex); | |
883 | } | |
884 | spin_unlock(&delayed_refs->lock); | |
885 | out: | |
886 | WARN_ON(num_refs == 0); | |
887 | if (refs) | |
888 | *refs = num_refs; | |
889 | if (flags) | |
890 | *flags = extent_flags; | |
891 | out_free: | |
892 | btrfs_free_path(path); | |
893 | return ret; | |
894 | } | |
895 | ||
d8d5f3e1 CM |
896 | /* |
897 | * Back reference rules. Back refs have three main goals: | |
898 | * | |
899 | * 1) differentiate between all holders of references to an extent so that | |
900 | * when a reference is dropped we can make sure it was a valid reference | |
901 | * before freeing the extent. | |
902 | * | |
903 | * 2) Provide enough information to quickly find the holders of an extent | |
904 | * if we notice a given block is corrupted or bad. | |
905 | * | |
906 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
907 | * maintenance. This is actually the same as #2, but with a slightly | |
908 | * different use case. | |
909 | * | |
5d4f98a2 YZ |
910 | * There are two kinds of back refs. The implicit back refs is optimized |
911 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
912 | * back refs of this kind provide information about the block's owner tree | |
913 | * and the pointer's key. These information allow us to find the block by | |
914 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
915 | * referenced by their owner trees. The location of tree block is recorded | |
916 | * in the back refs. Actually the full back refs is generic, and can be | |
917 | * used in all cases the implicit back refs is used. The major shortcoming | |
918 | * of the full back refs is its overhead. Every time a tree block gets | |
919 | * COWed, we have to update back refs entry for all pointers in it. | |
920 | * | |
921 | * For a newly allocated tree block, we use implicit back refs for | |
922 | * pointers in it. This means most tree related operations only involve | |
923 | * implicit back refs. For a tree block created in old transaction, the | |
924 | * only way to drop a reference to it is COW it. So we can detect the | |
925 | * event that tree block loses its owner tree's reference and do the | |
926 | * back refs conversion. | |
927 | * | |
928 | * When a tree block is COW'd through a tree, there are four cases: | |
929 | * | |
930 | * The reference count of the block is one and the tree is the block's | |
931 | * owner tree. Nothing to do in this case. | |
932 | * | |
933 | * The reference count of the block is one and the tree is not the | |
934 | * block's owner tree. In this case, full back refs is used for pointers | |
935 | * in the block. Remove these full back refs, add implicit back refs for | |
936 | * every pointers in the new block. | |
937 | * | |
938 | * The reference count of the block is greater than one and the tree is | |
939 | * the block's owner tree. In this case, implicit back refs is used for | |
940 | * pointers in the block. Add full back refs for every pointers in the | |
941 | * block, increase lower level extents' reference counts. The original | |
942 | * implicit back refs are entailed to the new block. | |
943 | * | |
944 | * The reference count of the block is greater than one and the tree is | |
945 | * not the block's owner tree. Add implicit back refs for every pointer in | |
946 | * the new block, increase lower level extents' reference count. | |
947 | * | |
948 | * Back Reference Key composing: | |
949 | * | |
950 | * The key objectid corresponds to the first byte in the extent, | |
951 | * The key type is used to differentiate between types of back refs. | |
952 | * There are different meanings of the key offset for different types | |
953 | * of back refs. | |
954 | * | |
d8d5f3e1 CM |
955 | * File extents can be referenced by: |
956 | * | |
957 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 958 | * - different files inside a single subvolume |
d8d5f3e1 CM |
959 | * - different offsets inside a file (bookend extents in file.c) |
960 | * | |
5d4f98a2 | 961 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
962 | * |
963 | * - Objectid of the subvolume root | |
d8d5f3e1 | 964 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
965 | * - original offset in the file |
966 | * - how many bookend extents | |
d8d5f3e1 | 967 | * |
5d4f98a2 YZ |
968 | * The key offset for the implicit back refs is hash of the first |
969 | * three fields. | |
d8d5f3e1 | 970 | * |
5d4f98a2 | 971 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 972 | * |
5d4f98a2 | 973 | * - number of pointers in the tree leaf |
d8d5f3e1 | 974 | * |
5d4f98a2 YZ |
975 | * The key offset for the implicit back refs is the first byte of |
976 | * the tree leaf | |
d8d5f3e1 | 977 | * |
5d4f98a2 YZ |
978 | * When a file extent is allocated, The implicit back refs is used. |
979 | * the fields are filled in: | |
d8d5f3e1 | 980 | * |
5d4f98a2 | 981 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 982 | * |
5d4f98a2 YZ |
983 | * When a file extent is removed file truncation, we find the |
984 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 985 | * |
5d4f98a2 | 986 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 987 | * |
5d4f98a2 | 988 | * Btree extents can be referenced by: |
d8d5f3e1 | 989 | * |
5d4f98a2 | 990 | * - Different subvolumes |
d8d5f3e1 | 991 | * |
5d4f98a2 YZ |
992 | * Both the implicit back refs and the full back refs for tree blocks |
993 | * only consist of key. The key offset for the implicit back refs is | |
994 | * objectid of block's owner tree. The key offset for the full back refs | |
995 | * is the first byte of parent block. | |
d8d5f3e1 | 996 | * |
5d4f98a2 YZ |
997 | * When implicit back refs is used, information about the lowest key and |
998 | * level of the tree block are required. These information are stored in | |
999 | * tree block info structure. | |
d8d5f3e1 | 1000 | */ |
31840ae1 | 1001 | |
5d4f98a2 YZ |
1002 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
1003 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
1004 | struct btrfs_root *root, | |
1005 | struct btrfs_path *path, | |
1006 | u64 owner, u32 extra_size) | |
7bb86316 | 1007 | { |
5d4f98a2 YZ |
1008 | struct btrfs_extent_item *item; |
1009 | struct btrfs_extent_item_v0 *ei0; | |
1010 | struct btrfs_extent_ref_v0 *ref0; | |
1011 | struct btrfs_tree_block_info *bi; | |
1012 | struct extent_buffer *leaf; | |
7bb86316 | 1013 | struct btrfs_key key; |
5d4f98a2 YZ |
1014 | struct btrfs_key found_key; |
1015 | u32 new_size = sizeof(*item); | |
1016 | u64 refs; | |
1017 | int ret; | |
1018 | ||
1019 | leaf = path->nodes[0]; | |
1020 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
1021 | ||
1022 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1023 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
1024 | struct btrfs_extent_item_v0); | |
1025 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
1026 | ||
1027 | if (owner == (u64)-1) { | |
1028 | while (1) { | |
1029 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1030 | ret = btrfs_next_leaf(root, path); | |
1031 | if (ret < 0) | |
1032 | return ret; | |
79787eaa | 1033 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
1034 | leaf = path->nodes[0]; |
1035 | } | |
1036 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1037 | path->slots[0]); | |
1038 | BUG_ON(key.objectid != found_key.objectid); | |
1039 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
1040 | path->slots[0]++; | |
1041 | continue; | |
1042 | } | |
1043 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1044 | struct btrfs_extent_ref_v0); | |
1045 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
1046 | break; | |
1047 | } | |
1048 | } | |
b3b4aa74 | 1049 | btrfs_release_path(path); |
5d4f98a2 YZ |
1050 | |
1051 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1052 | new_size += sizeof(*bi); | |
1053 | ||
1054 | new_size -= sizeof(*ei0); | |
1055 | ret = btrfs_search_slot(trans, root, &key, path, | |
1056 | new_size + extra_size, 1); | |
1057 | if (ret < 0) | |
1058 | return ret; | |
79787eaa | 1059 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 1060 | |
4b90c680 | 1061 | btrfs_extend_item(root, path, new_size); |
5d4f98a2 YZ |
1062 | |
1063 | leaf = path->nodes[0]; | |
1064 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1065 | btrfs_set_extent_refs(leaf, item, refs); | |
1066 | /* FIXME: get real generation */ | |
1067 | btrfs_set_extent_generation(leaf, item, 0); | |
1068 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1069 | btrfs_set_extent_flags(leaf, item, | |
1070 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1071 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1072 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1073 | /* FIXME: get first key of the block */ | |
1074 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); | |
1075 | btrfs_set_tree_block_level(leaf, bi, (int)owner); | |
1076 | } else { | |
1077 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1078 | } | |
1079 | btrfs_mark_buffer_dirty(leaf); | |
1080 | return 0; | |
1081 | } | |
1082 | #endif | |
1083 | ||
1084 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) | |
1085 | { | |
1086 | u32 high_crc = ~(u32)0; | |
1087 | u32 low_crc = ~(u32)0; | |
1088 | __le64 lenum; | |
1089 | ||
1090 | lenum = cpu_to_le64(root_objectid); | |
163e783e | 1091 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1092 | lenum = cpu_to_le64(owner); |
163e783e | 1093 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1094 | lenum = cpu_to_le64(offset); |
163e783e | 1095 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1096 | |
1097 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1098 | } | |
1099 | ||
1100 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1101 | struct btrfs_extent_data_ref *ref) | |
1102 | { | |
1103 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1104 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1105 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1106 | } | |
1107 | ||
1108 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1109 | struct btrfs_extent_data_ref *ref, | |
1110 | u64 root_objectid, u64 owner, u64 offset) | |
1111 | { | |
1112 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1113 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1114 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1115 | return 0; | |
1116 | return 1; | |
1117 | } | |
1118 | ||
1119 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
1120 | struct btrfs_root *root, | |
1121 | struct btrfs_path *path, | |
1122 | u64 bytenr, u64 parent, | |
1123 | u64 root_objectid, | |
1124 | u64 owner, u64 offset) | |
1125 | { | |
1126 | struct btrfs_key key; | |
1127 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1128 | struct extent_buffer *leaf; |
5d4f98a2 | 1129 | u32 nritems; |
74493f7a | 1130 | int ret; |
5d4f98a2 YZ |
1131 | int recow; |
1132 | int err = -ENOENT; | |
74493f7a | 1133 | |
31840ae1 | 1134 | key.objectid = bytenr; |
5d4f98a2 YZ |
1135 | if (parent) { |
1136 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1137 | key.offset = parent; | |
1138 | } else { | |
1139 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1140 | key.offset = hash_extent_data_ref(root_objectid, | |
1141 | owner, offset); | |
1142 | } | |
1143 | again: | |
1144 | recow = 0; | |
1145 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1146 | if (ret < 0) { | |
1147 | err = ret; | |
1148 | goto fail; | |
1149 | } | |
31840ae1 | 1150 | |
5d4f98a2 YZ |
1151 | if (parent) { |
1152 | if (!ret) | |
1153 | return 0; | |
1154 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1155 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1156 | btrfs_release_path(path); |
5d4f98a2 YZ |
1157 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1158 | if (ret < 0) { | |
1159 | err = ret; | |
1160 | goto fail; | |
1161 | } | |
1162 | if (!ret) | |
1163 | return 0; | |
1164 | #endif | |
1165 | goto fail; | |
31840ae1 ZY |
1166 | } |
1167 | ||
1168 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1169 | nritems = btrfs_header_nritems(leaf); |
1170 | while (1) { | |
1171 | if (path->slots[0] >= nritems) { | |
1172 | ret = btrfs_next_leaf(root, path); | |
1173 | if (ret < 0) | |
1174 | err = ret; | |
1175 | if (ret) | |
1176 | goto fail; | |
1177 | ||
1178 | leaf = path->nodes[0]; | |
1179 | nritems = btrfs_header_nritems(leaf); | |
1180 | recow = 1; | |
1181 | } | |
1182 | ||
1183 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1184 | if (key.objectid != bytenr || | |
1185 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1186 | goto fail; | |
1187 | ||
1188 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1189 | struct btrfs_extent_data_ref); | |
1190 | ||
1191 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1192 | owner, offset)) { | |
1193 | if (recow) { | |
b3b4aa74 | 1194 | btrfs_release_path(path); |
5d4f98a2 YZ |
1195 | goto again; |
1196 | } | |
1197 | err = 0; | |
1198 | break; | |
1199 | } | |
1200 | path->slots[0]++; | |
31840ae1 | 1201 | } |
5d4f98a2 YZ |
1202 | fail: |
1203 | return err; | |
31840ae1 ZY |
1204 | } |
1205 | ||
5d4f98a2 YZ |
1206 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
1207 | struct btrfs_root *root, | |
1208 | struct btrfs_path *path, | |
1209 | u64 bytenr, u64 parent, | |
1210 | u64 root_objectid, u64 owner, | |
1211 | u64 offset, int refs_to_add) | |
31840ae1 ZY |
1212 | { |
1213 | struct btrfs_key key; | |
1214 | struct extent_buffer *leaf; | |
5d4f98a2 | 1215 | u32 size; |
31840ae1 ZY |
1216 | u32 num_refs; |
1217 | int ret; | |
74493f7a | 1218 | |
74493f7a | 1219 | key.objectid = bytenr; |
5d4f98a2 YZ |
1220 | if (parent) { |
1221 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1222 | key.offset = parent; | |
1223 | size = sizeof(struct btrfs_shared_data_ref); | |
1224 | } else { | |
1225 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1226 | key.offset = hash_extent_data_ref(root_objectid, | |
1227 | owner, offset); | |
1228 | size = sizeof(struct btrfs_extent_data_ref); | |
1229 | } | |
74493f7a | 1230 | |
5d4f98a2 YZ |
1231 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1232 | if (ret && ret != -EEXIST) | |
1233 | goto fail; | |
1234 | ||
1235 | leaf = path->nodes[0]; | |
1236 | if (parent) { | |
1237 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1238 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1239 | struct btrfs_shared_data_ref); |
1240 | if (ret == 0) { | |
1241 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1242 | } else { | |
1243 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1244 | num_refs += refs_to_add; | |
1245 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1246 | } |
5d4f98a2 YZ |
1247 | } else { |
1248 | struct btrfs_extent_data_ref *ref; | |
1249 | while (ret == -EEXIST) { | |
1250 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1251 | struct btrfs_extent_data_ref); | |
1252 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1253 | owner, offset)) | |
1254 | break; | |
b3b4aa74 | 1255 | btrfs_release_path(path); |
5d4f98a2 YZ |
1256 | key.offset++; |
1257 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1258 | size); | |
1259 | if (ret && ret != -EEXIST) | |
1260 | goto fail; | |
31840ae1 | 1261 | |
5d4f98a2 YZ |
1262 | leaf = path->nodes[0]; |
1263 | } | |
1264 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1265 | struct btrfs_extent_data_ref); | |
1266 | if (ret == 0) { | |
1267 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1268 | root_objectid); | |
1269 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1270 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1271 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1272 | } else { | |
1273 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1274 | num_refs += refs_to_add; | |
1275 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1276 | } |
31840ae1 | 1277 | } |
5d4f98a2 YZ |
1278 | btrfs_mark_buffer_dirty(leaf); |
1279 | ret = 0; | |
1280 | fail: | |
b3b4aa74 | 1281 | btrfs_release_path(path); |
7bb86316 | 1282 | return ret; |
74493f7a CM |
1283 | } |
1284 | ||
5d4f98a2 YZ |
1285 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
1286 | struct btrfs_root *root, | |
1287 | struct btrfs_path *path, | |
1288 | int refs_to_drop) | |
31840ae1 | 1289 | { |
5d4f98a2 YZ |
1290 | struct btrfs_key key; |
1291 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1292 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1293 | struct extent_buffer *leaf; |
5d4f98a2 | 1294 | u32 num_refs = 0; |
31840ae1 ZY |
1295 | int ret = 0; |
1296 | ||
1297 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1298 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1299 | ||
1300 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1301 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1302 | struct btrfs_extent_data_ref); | |
1303 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1304 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1305 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1306 | struct btrfs_shared_data_ref); | |
1307 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1308 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1309 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1310 | struct btrfs_extent_ref_v0 *ref0; | |
1311 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1312 | struct btrfs_extent_ref_v0); | |
1313 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1314 | #endif | |
1315 | } else { | |
1316 | BUG(); | |
1317 | } | |
1318 | ||
56bec294 CM |
1319 | BUG_ON(num_refs < refs_to_drop); |
1320 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1321 | |
31840ae1 ZY |
1322 | if (num_refs == 0) { |
1323 | ret = btrfs_del_item(trans, root, path); | |
1324 | } else { | |
5d4f98a2 YZ |
1325 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1326 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1327 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1328 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1329 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1330 | else { | |
1331 | struct btrfs_extent_ref_v0 *ref0; | |
1332 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1333 | struct btrfs_extent_ref_v0); | |
1334 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1335 | } | |
1336 | #endif | |
31840ae1 ZY |
1337 | btrfs_mark_buffer_dirty(leaf); |
1338 | } | |
31840ae1 ZY |
1339 | return ret; |
1340 | } | |
1341 | ||
5d4f98a2 YZ |
1342 | static noinline u32 extent_data_ref_count(struct btrfs_root *root, |
1343 | struct btrfs_path *path, | |
1344 | struct btrfs_extent_inline_ref *iref) | |
15916de8 | 1345 | { |
5d4f98a2 YZ |
1346 | struct btrfs_key key; |
1347 | struct extent_buffer *leaf; | |
1348 | struct btrfs_extent_data_ref *ref1; | |
1349 | struct btrfs_shared_data_ref *ref2; | |
1350 | u32 num_refs = 0; | |
1351 | ||
1352 | leaf = path->nodes[0]; | |
1353 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1354 | if (iref) { | |
1355 | if (btrfs_extent_inline_ref_type(leaf, iref) == | |
1356 | BTRFS_EXTENT_DATA_REF_KEY) { | |
1357 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1358 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1359 | } else { | |
1360 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1361 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1362 | } | |
1363 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1364 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1365 | struct btrfs_extent_data_ref); | |
1366 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1367 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1368 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1369 | struct btrfs_shared_data_ref); | |
1370 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1371 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1372 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1373 | struct btrfs_extent_ref_v0 *ref0; | |
1374 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1375 | struct btrfs_extent_ref_v0); | |
1376 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1377 | #endif |
5d4f98a2 YZ |
1378 | } else { |
1379 | WARN_ON(1); | |
1380 | } | |
1381 | return num_refs; | |
1382 | } | |
15916de8 | 1383 | |
5d4f98a2 YZ |
1384 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
1385 | struct btrfs_root *root, | |
1386 | struct btrfs_path *path, | |
1387 | u64 bytenr, u64 parent, | |
1388 | u64 root_objectid) | |
1f3c79a2 | 1389 | { |
5d4f98a2 | 1390 | struct btrfs_key key; |
1f3c79a2 | 1391 | int ret; |
1f3c79a2 | 1392 | |
5d4f98a2 YZ |
1393 | key.objectid = bytenr; |
1394 | if (parent) { | |
1395 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1396 | key.offset = parent; | |
1397 | } else { | |
1398 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1399 | key.offset = root_objectid; | |
1f3c79a2 LH |
1400 | } |
1401 | ||
5d4f98a2 YZ |
1402 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1403 | if (ret > 0) | |
1404 | ret = -ENOENT; | |
1405 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1406 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1407 | btrfs_release_path(path); |
5d4f98a2 YZ |
1408 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1409 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1410 | if (ret > 0) | |
1411 | ret = -ENOENT; | |
1412 | } | |
1f3c79a2 | 1413 | #endif |
5d4f98a2 | 1414 | return ret; |
1f3c79a2 LH |
1415 | } |
1416 | ||
5d4f98a2 YZ |
1417 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
1418 | struct btrfs_root *root, | |
1419 | struct btrfs_path *path, | |
1420 | u64 bytenr, u64 parent, | |
1421 | u64 root_objectid) | |
31840ae1 | 1422 | { |
5d4f98a2 | 1423 | struct btrfs_key key; |
31840ae1 | 1424 | int ret; |
31840ae1 | 1425 | |
5d4f98a2 YZ |
1426 | key.objectid = bytenr; |
1427 | if (parent) { | |
1428 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1429 | key.offset = parent; | |
1430 | } else { | |
1431 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1432 | key.offset = root_objectid; | |
1433 | } | |
1434 | ||
1435 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); | |
b3b4aa74 | 1436 | btrfs_release_path(path); |
31840ae1 ZY |
1437 | return ret; |
1438 | } | |
1439 | ||
5d4f98a2 | 1440 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1441 | { |
5d4f98a2 YZ |
1442 | int type; |
1443 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1444 | if (parent > 0) | |
1445 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1446 | else | |
1447 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1448 | } else { | |
1449 | if (parent > 0) | |
1450 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1451 | else | |
1452 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1453 | } | |
1454 | return type; | |
31840ae1 | 1455 | } |
56bec294 | 1456 | |
2c47e605 YZ |
1457 | static int find_next_key(struct btrfs_path *path, int level, |
1458 | struct btrfs_key *key) | |
56bec294 | 1459 | |
02217ed2 | 1460 | { |
2c47e605 | 1461 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1462 | if (!path->nodes[level]) |
1463 | break; | |
5d4f98a2 YZ |
1464 | if (path->slots[level] + 1 >= |
1465 | btrfs_header_nritems(path->nodes[level])) | |
1466 | continue; | |
1467 | if (level == 0) | |
1468 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1469 | path->slots[level] + 1); | |
1470 | else | |
1471 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1472 | path->slots[level] + 1); | |
1473 | return 0; | |
1474 | } | |
1475 | return 1; | |
1476 | } | |
037e6390 | 1477 | |
5d4f98a2 YZ |
1478 | /* |
1479 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1480 | * to the address of inline back ref, and 0 is returned. | |
1481 | * | |
1482 | * if back ref isn't found, *ref_ret is set to the address where it | |
1483 | * should be inserted, and -ENOENT is returned. | |
1484 | * | |
1485 | * if insert is true and there are too many inline back refs, the path | |
1486 | * points to the extent item, and -EAGAIN is returned. | |
1487 | * | |
1488 | * NOTE: inline back refs are ordered in the same way that back ref | |
1489 | * items in the tree are ordered. | |
1490 | */ | |
1491 | static noinline_for_stack | |
1492 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1493 | struct btrfs_root *root, | |
1494 | struct btrfs_path *path, | |
1495 | struct btrfs_extent_inline_ref **ref_ret, | |
1496 | u64 bytenr, u64 num_bytes, | |
1497 | u64 parent, u64 root_objectid, | |
1498 | u64 owner, u64 offset, int insert) | |
1499 | { | |
1500 | struct btrfs_key key; | |
1501 | struct extent_buffer *leaf; | |
1502 | struct btrfs_extent_item *ei; | |
1503 | struct btrfs_extent_inline_ref *iref; | |
1504 | u64 flags; | |
1505 | u64 item_size; | |
1506 | unsigned long ptr; | |
1507 | unsigned long end; | |
1508 | int extra_size; | |
1509 | int type; | |
1510 | int want; | |
1511 | int ret; | |
1512 | int err = 0; | |
3173a18f JB |
1513 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
1514 | SKINNY_METADATA); | |
26b8003f | 1515 | |
db94535d | 1516 | key.objectid = bytenr; |
31840ae1 | 1517 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1518 | key.offset = num_bytes; |
31840ae1 | 1519 | |
5d4f98a2 YZ |
1520 | want = extent_ref_type(parent, owner); |
1521 | if (insert) { | |
1522 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1523 | path->keep_locks = 1; |
5d4f98a2 YZ |
1524 | } else |
1525 | extra_size = -1; | |
3173a18f JB |
1526 | |
1527 | /* | |
1528 | * Owner is our parent level, so we can just add one to get the level | |
1529 | * for the block we are interested in. | |
1530 | */ | |
1531 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1532 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1533 | key.offset = owner; | |
1534 | } | |
1535 | ||
1536 | again: | |
5d4f98a2 | 1537 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1538 | if (ret < 0) { |
5d4f98a2 YZ |
1539 | err = ret; |
1540 | goto out; | |
1541 | } | |
3173a18f JB |
1542 | |
1543 | /* | |
1544 | * We may be a newly converted file system which still has the old fat | |
1545 | * extent entries for metadata, so try and see if we have one of those. | |
1546 | */ | |
1547 | if (ret > 0 && skinny_metadata) { | |
1548 | skinny_metadata = false; | |
1549 | if (path->slots[0]) { | |
1550 | path->slots[0]--; | |
1551 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1552 | path->slots[0]); | |
1553 | if (key.objectid == bytenr && | |
1554 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1555 | key.offset == num_bytes) | |
1556 | ret = 0; | |
1557 | } | |
1558 | if (ret) { | |
1559 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
1560 | key.offset = num_bytes; | |
1561 | btrfs_release_path(path); | |
1562 | goto again; | |
1563 | } | |
1564 | } | |
1565 | ||
79787eaa JM |
1566 | if (ret && !insert) { |
1567 | err = -ENOENT; | |
1568 | goto out; | |
492104c8 JB |
1569 | } else if (ret) { |
1570 | err = -EIO; | |
1571 | WARN_ON(1); | |
1572 | goto out; | |
79787eaa | 1573 | } |
5d4f98a2 YZ |
1574 | |
1575 | leaf = path->nodes[0]; | |
1576 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1577 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1578 | if (item_size < sizeof(*ei)) { | |
1579 | if (!insert) { | |
1580 | err = -ENOENT; | |
1581 | goto out; | |
1582 | } | |
1583 | ret = convert_extent_item_v0(trans, root, path, owner, | |
1584 | extra_size); | |
1585 | if (ret < 0) { | |
1586 | err = ret; | |
1587 | goto out; | |
1588 | } | |
1589 | leaf = path->nodes[0]; | |
1590 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1591 | } | |
1592 | #endif | |
1593 | BUG_ON(item_size < sizeof(*ei)); | |
1594 | ||
5d4f98a2 YZ |
1595 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1596 | flags = btrfs_extent_flags(leaf, ei); | |
1597 | ||
1598 | ptr = (unsigned long)(ei + 1); | |
1599 | end = (unsigned long)ei + item_size; | |
1600 | ||
3173a18f | 1601 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1602 | ptr += sizeof(struct btrfs_tree_block_info); |
1603 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1604 | } |
1605 | ||
1606 | err = -ENOENT; | |
1607 | while (1) { | |
1608 | if (ptr >= end) { | |
1609 | WARN_ON(ptr > end); | |
1610 | break; | |
1611 | } | |
1612 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1613 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1614 | if (want < type) | |
1615 | break; | |
1616 | if (want > type) { | |
1617 | ptr += btrfs_extent_inline_ref_size(type); | |
1618 | continue; | |
1619 | } | |
1620 | ||
1621 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1622 | struct btrfs_extent_data_ref *dref; | |
1623 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1624 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1625 | owner, offset)) { | |
1626 | err = 0; | |
1627 | break; | |
1628 | } | |
1629 | if (hash_extent_data_ref_item(leaf, dref) < | |
1630 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1631 | break; | |
1632 | } else { | |
1633 | u64 ref_offset; | |
1634 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1635 | if (parent > 0) { | |
1636 | if (parent == ref_offset) { | |
1637 | err = 0; | |
1638 | break; | |
1639 | } | |
1640 | if (ref_offset < parent) | |
1641 | break; | |
1642 | } else { | |
1643 | if (root_objectid == ref_offset) { | |
1644 | err = 0; | |
1645 | break; | |
1646 | } | |
1647 | if (ref_offset < root_objectid) | |
1648 | break; | |
1649 | } | |
1650 | } | |
1651 | ptr += btrfs_extent_inline_ref_size(type); | |
1652 | } | |
1653 | if (err == -ENOENT && insert) { | |
1654 | if (item_size + extra_size >= | |
1655 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1656 | err = -EAGAIN; | |
1657 | goto out; | |
1658 | } | |
1659 | /* | |
1660 | * To add new inline back ref, we have to make sure | |
1661 | * there is no corresponding back ref item. | |
1662 | * For simplicity, we just do not add new inline back | |
1663 | * ref if there is any kind of item for this block | |
1664 | */ | |
2c47e605 YZ |
1665 | if (find_next_key(path, 0, &key) == 0 && |
1666 | key.objectid == bytenr && | |
85d4198e | 1667 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1668 | err = -EAGAIN; |
1669 | goto out; | |
1670 | } | |
1671 | } | |
1672 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1673 | out: | |
85d4198e | 1674 | if (insert) { |
5d4f98a2 YZ |
1675 | path->keep_locks = 0; |
1676 | btrfs_unlock_up_safe(path, 1); | |
1677 | } | |
1678 | return err; | |
1679 | } | |
1680 | ||
1681 | /* | |
1682 | * helper to add new inline back ref | |
1683 | */ | |
1684 | static noinline_for_stack | |
fd279fae | 1685 | void setup_inline_extent_backref(struct btrfs_root *root, |
143bede5 JM |
1686 | struct btrfs_path *path, |
1687 | struct btrfs_extent_inline_ref *iref, | |
1688 | u64 parent, u64 root_objectid, | |
1689 | u64 owner, u64 offset, int refs_to_add, | |
1690 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1691 | { |
1692 | struct extent_buffer *leaf; | |
1693 | struct btrfs_extent_item *ei; | |
1694 | unsigned long ptr; | |
1695 | unsigned long end; | |
1696 | unsigned long item_offset; | |
1697 | u64 refs; | |
1698 | int size; | |
1699 | int type; | |
5d4f98a2 YZ |
1700 | |
1701 | leaf = path->nodes[0]; | |
1702 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1703 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1704 | ||
1705 | type = extent_ref_type(parent, owner); | |
1706 | size = btrfs_extent_inline_ref_size(type); | |
1707 | ||
4b90c680 | 1708 | btrfs_extend_item(root, path, size); |
5d4f98a2 YZ |
1709 | |
1710 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1711 | refs = btrfs_extent_refs(leaf, ei); | |
1712 | refs += refs_to_add; | |
1713 | btrfs_set_extent_refs(leaf, ei, refs); | |
1714 | if (extent_op) | |
1715 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1716 | ||
1717 | ptr = (unsigned long)ei + item_offset; | |
1718 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1719 | if (ptr < end - size) | |
1720 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1721 | end - size - ptr); | |
1722 | ||
1723 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1724 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1725 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1726 | struct btrfs_extent_data_ref *dref; | |
1727 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1728 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1729 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1730 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1731 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1732 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1733 | struct btrfs_shared_data_ref *sref; | |
1734 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1735 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1736 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1737 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1738 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1739 | } else { | |
1740 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1741 | } | |
1742 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1743 | } |
1744 | ||
1745 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
1746 | struct btrfs_root *root, | |
1747 | struct btrfs_path *path, | |
1748 | struct btrfs_extent_inline_ref **ref_ret, | |
1749 | u64 bytenr, u64 num_bytes, u64 parent, | |
1750 | u64 root_objectid, u64 owner, u64 offset) | |
1751 | { | |
1752 | int ret; | |
1753 | ||
1754 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, | |
1755 | bytenr, num_bytes, parent, | |
1756 | root_objectid, owner, offset, 0); | |
1757 | if (ret != -ENOENT) | |
54aa1f4d | 1758 | return ret; |
5d4f98a2 | 1759 | |
b3b4aa74 | 1760 | btrfs_release_path(path); |
5d4f98a2 YZ |
1761 | *ref_ret = NULL; |
1762 | ||
1763 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1764 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, | |
1765 | root_objectid); | |
1766 | } else { | |
1767 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, | |
1768 | root_objectid, owner, offset); | |
b9473439 | 1769 | } |
5d4f98a2 YZ |
1770 | return ret; |
1771 | } | |
31840ae1 | 1772 | |
5d4f98a2 YZ |
1773 | /* |
1774 | * helper to update/remove inline back ref | |
1775 | */ | |
1776 | static noinline_for_stack | |
afe5fea7 | 1777 | void update_inline_extent_backref(struct btrfs_root *root, |
143bede5 JM |
1778 | struct btrfs_path *path, |
1779 | struct btrfs_extent_inline_ref *iref, | |
1780 | int refs_to_mod, | |
1781 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1782 | { |
1783 | struct extent_buffer *leaf; | |
1784 | struct btrfs_extent_item *ei; | |
1785 | struct btrfs_extent_data_ref *dref = NULL; | |
1786 | struct btrfs_shared_data_ref *sref = NULL; | |
1787 | unsigned long ptr; | |
1788 | unsigned long end; | |
1789 | u32 item_size; | |
1790 | int size; | |
1791 | int type; | |
5d4f98a2 YZ |
1792 | u64 refs; |
1793 | ||
1794 | leaf = path->nodes[0]; | |
1795 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1796 | refs = btrfs_extent_refs(leaf, ei); | |
1797 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1798 | refs += refs_to_mod; | |
1799 | btrfs_set_extent_refs(leaf, ei, refs); | |
1800 | if (extent_op) | |
1801 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1802 | ||
1803 | type = btrfs_extent_inline_ref_type(leaf, iref); | |
1804 | ||
1805 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1806 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1807 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1808 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1809 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1810 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1811 | } else { | |
1812 | refs = 1; | |
1813 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1814 | } |
31840ae1 | 1815 | |
5d4f98a2 YZ |
1816 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1817 | refs += refs_to_mod; | |
1818 | ||
1819 | if (refs > 0) { | |
1820 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1821 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1822 | else | |
1823 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1824 | } else { | |
1825 | size = btrfs_extent_inline_ref_size(type); | |
1826 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1827 | ptr = (unsigned long)iref; | |
1828 | end = (unsigned long)ei + item_size; | |
1829 | if (ptr + size < end) | |
1830 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1831 | end - ptr - size); | |
1832 | item_size -= size; | |
afe5fea7 | 1833 | btrfs_truncate_item(root, path, item_size, 1); |
5d4f98a2 YZ |
1834 | } |
1835 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1836 | } |
1837 | ||
1838 | static noinline_for_stack | |
1839 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
1840 | struct btrfs_root *root, | |
1841 | struct btrfs_path *path, | |
1842 | u64 bytenr, u64 num_bytes, u64 parent, | |
1843 | u64 root_objectid, u64 owner, | |
1844 | u64 offset, int refs_to_add, | |
1845 | struct btrfs_delayed_extent_op *extent_op) | |
1846 | { | |
1847 | struct btrfs_extent_inline_ref *iref; | |
1848 | int ret; | |
1849 | ||
1850 | ret = lookup_inline_extent_backref(trans, root, path, &iref, | |
1851 | bytenr, num_bytes, parent, | |
1852 | root_objectid, owner, offset, 1); | |
1853 | if (ret == 0) { | |
1854 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
afe5fea7 | 1855 | update_inline_extent_backref(root, path, iref, |
143bede5 | 1856 | refs_to_add, extent_op); |
5d4f98a2 | 1857 | } else if (ret == -ENOENT) { |
fd279fae | 1858 | setup_inline_extent_backref(root, path, iref, parent, |
143bede5 JM |
1859 | root_objectid, owner, offset, |
1860 | refs_to_add, extent_op); | |
1861 | ret = 0; | |
771ed689 | 1862 | } |
5d4f98a2 YZ |
1863 | return ret; |
1864 | } | |
31840ae1 | 1865 | |
5d4f98a2 YZ |
1866 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
1867 | struct btrfs_root *root, | |
1868 | struct btrfs_path *path, | |
1869 | u64 bytenr, u64 parent, u64 root_objectid, | |
1870 | u64 owner, u64 offset, int refs_to_add) | |
1871 | { | |
1872 | int ret; | |
1873 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1874 | BUG_ON(refs_to_add != 1); | |
1875 | ret = insert_tree_block_ref(trans, root, path, bytenr, | |
1876 | parent, root_objectid); | |
1877 | } else { | |
1878 | ret = insert_extent_data_ref(trans, root, path, bytenr, | |
1879 | parent, root_objectid, | |
1880 | owner, offset, refs_to_add); | |
1881 | } | |
1882 | return ret; | |
1883 | } | |
56bec294 | 1884 | |
5d4f98a2 YZ |
1885 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
1886 | struct btrfs_root *root, | |
1887 | struct btrfs_path *path, | |
1888 | struct btrfs_extent_inline_ref *iref, | |
1889 | int refs_to_drop, int is_data) | |
1890 | { | |
143bede5 | 1891 | int ret = 0; |
b9473439 | 1892 | |
5d4f98a2 YZ |
1893 | BUG_ON(!is_data && refs_to_drop != 1); |
1894 | if (iref) { | |
afe5fea7 | 1895 | update_inline_extent_backref(root, path, iref, |
143bede5 | 1896 | -refs_to_drop, NULL); |
5d4f98a2 YZ |
1897 | } else if (is_data) { |
1898 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop); | |
1899 | } else { | |
1900 | ret = btrfs_del_item(trans, root, path); | |
1901 | } | |
1902 | return ret; | |
1903 | } | |
1904 | ||
5378e607 | 1905 | static int btrfs_issue_discard(struct block_device *bdev, |
5d4f98a2 YZ |
1906 | u64 start, u64 len) |
1907 | { | |
5378e607 | 1908 | return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0); |
5d4f98a2 | 1909 | } |
5d4f98a2 YZ |
1910 | |
1911 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 1912 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1913 | { |
5d4f98a2 | 1914 | int ret; |
5378e607 | 1915 | u64 discarded_bytes = 0; |
a1d3c478 | 1916 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1917 | |
e244a0ae | 1918 | |
5d4f98a2 | 1919 | /* Tell the block device(s) that the sectors can be discarded */ |
3ec706c8 | 1920 | ret = btrfs_map_block(root->fs_info, REQ_DISCARD, |
a1d3c478 | 1921 | bytenr, &num_bytes, &bbio, 0); |
79787eaa | 1922 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 1923 | if (!ret) { |
a1d3c478 | 1924 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
1925 | int i; |
1926 | ||
5d4f98a2 | 1927 | |
a1d3c478 | 1928 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d5e2003c JB |
1929 | if (!stripe->dev->can_discard) |
1930 | continue; | |
1931 | ||
5378e607 LD |
1932 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1933 | stripe->physical, | |
1934 | stripe->length); | |
1935 | if (!ret) | |
1936 | discarded_bytes += stripe->length; | |
1937 | else if (ret != -EOPNOTSUPP) | |
79787eaa | 1938 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
1939 | |
1940 | /* | |
1941 | * Just in case we get back EOPNOTSUPP for some reason, | |
1942 | * just ignore the return value so we don't screw up | |
1943 | * people calling discard_extent. | |
1944 | */ | |
1945 | ret = 0; | |
5d4f98a2 | 1946 | } |
a1d3c478 | 1947 | kfree(bbio); |
5d4f98a2 | 1948 | } |
5378e607 LD |
1949 | |
1950 | if (actual_bytes) | |
1951 | *actual_bytes = discarded_bytes; | |
1952 | ||
5d4f98a2 | 1953 | |
53b381b3 DW |
1954 | if (ret == -EOPNOTSUPP) |
1955 | ret = 0; | |
5d4f98a2 | 1956 | return ret; |
5d4f98a2 YZ |
1957 | } |
1958 | ||
79787eaa | 1959 | /* Can return -ENOMEM */ |
5d4f98a2 YZ |
1960 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
1961 | struct btrfs_root *root, | |
1962 | u64 bytenr, u64 num_bytes, u64 parent, | |
66d7e7f0 | 1963 | u64 root_objectid, u64 owner, u64 offset, int for_cow) |
5d4f98a2 YZ |
1964 | { |
1965 | int ret; | |
66d7e7f0 AJ |
1966 | struct btrfs_fs_info *fs_info = root->fs_info; |
1967 | ||
5d4f98a2 YZ |
1968 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
1969 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
1970 | ||
1971 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
66d7e7f0 AJ |
1972 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
1973 | num_bytes, | |
5d4f98a2 | 1974 | parent, root_objectid, (int)owner, |
66d7e7f0 | 1975 | BTRFS_ADD_DELAYED_REF, NULL, for_cow); |
5d4f98a2 | 1976 | } else { |
66d7e7f0 AJ |
1977 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
1978 | num_bytes, | |
5d4f98a2 | 1979 | parent, root_objectid, owner, offset, |
66d7e7f0 | 1980 | BTRFS_ADD_DELAYED_REF, NULL, for_cow); |
5d4f98a2 YZ |
1981 | } |
1982 | return ret; | |
1983 | } | |
1984 | ||
1985 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
1986 | struct btrfs_root *root, | |
1987 | u64 bytenr, u64 num_bytes, | |
1988 | u64 parent, u64 root_objectid, | |
1989 | u64 owner, u64 offset, int refs_to_add, | |
1990 | struct btrfs_delayed_extent_op *extent_op) | |
1991 | { | |
1992 | struct btrfs_path *path; | |
1993 | struct extent_buffer *leaf; | |
1994 | struct btrfs_extent_item *item; | |
1995 | u64 refs; | |
1996 | int ret; | |
1997 | int err = 0; | |
1998 | ||
1999 | path = btrfs_alloc_path(); | |
2000 | if (!path) | |
2001 | return -ENOMEM; | |
2002 | ||
2003 | path->reada = 1; | |
2004 | path->leave_spinning = 1; | |
2005 | /* this will setup the path even if it fails to insert the back ref */ | |
2006 | ret = insert_inline_extent_backref(trans, root->fs_info->extent_root, | |
2007 | path, bytenr, num_bytes, parent, | |
2008 | root_objectid, owner, offset, | |
2009 | refs_to_add, extent_op); | |
2010 | if (ret == 0) | |
2011 | goto out; | |
2012 | ||
2013 | if (ret != -EAGAIN) { | |
2014 | err = ret; | |
2015 | goto out; | |
2016 | } | |
2017 | ||
2018 | leaf = path->nodes[0]; | |
2019 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2020 | refs = btrfs_extent_refs(leaf, item); | |
2021 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2022 | if (extent_op) | |
2023 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2024 | |
5d4f98a2 | 2025 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2026 | btrfs_release_path(path); |
56bec294 CM |
2027 | |
2028 | path->reada = 1; | |
b9473439 CM |
2029 | path->leave_spinning = 1; |
2030 | ||
56bec294 CM |
2031 | /* now insert the actual backref */ |
2032 | ret = insert_extent_backref(trans, root->fs_info->extent_root, | |
5d4f98a2 YZ |
2033 | path, bytenr, parent, root_objectid, |
2034 | owner, offset, refs_to_add); | |
79787eaa JM |
2035 | if (ret) |
2036 | btrfs_abort_transaction(trans, root, ret); | |
5d4f98a2 | 2037 | out: |
56bec294 | 2038 | btrfs_free_path(path); |
5d4f98a2 | 2039 | return err; |
56bec294 CM |
2040 | } |
2041 | ||
5d4f98a2 YZ |
2042 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
2043 | struct btrfs_root *root, | |
2044 | struct btrfs_delayed_ref_node *node, | |
2045 | struct btrfs_delayed_extent_op *extent_op, | |
2046 | int insert_reserved) | |
56bec294 | 2047 | { |
5d4f98a2 YZ |
2048 | int ret = 0; |
2049 | struct btrfs_delayed_data_ref *ref; | |
2050 | struct btrfs_key ins; | |
2051 | u64 parent = 0; | |
2052 | u64 ref_root = 0; | |
2053 | u64 flags = 0; | |
2054 | ||
2055 | ins.objectid = node->bytenr; | |
2056 | ins.offset = node->num_bytes; | |
2057 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2058 | ||
2059 | ref = btrfs_delayed_node_to_data_ref(node); | |
2060 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2061 | parent = ref->parent; | |
2062 | else | |
2063 | ref_root = ref->root; | |
2064 | ||
2065 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2066 | if (extent_op) |
5d4f98a2 | 2067 | flags |= extent_op->flags_to_set; |
5d4f98a2 YZ |
2068 | ret = alloc_reserved_file_extent(trans, root, |
2069 | parent, ref_root, flags, | |
2070 | ref->objectid, ref->offset, | |
2071 | &ins, node->ref_mod); | |
5d4f98a2 YZ |
2072 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2073 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
2074 | node->num_bytes, parent, | |
2075 | ref_root, ref->objectid, | |
2076 | ref->offset, node->ref_mod, | |
2077 | extent_op); | |
2078 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
2079 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
2080 | node->num_bytes, parent, | |
2081 | ref_root, ref->objectid, | |
2082 | ref->offset, node->ref_mod, | |
2083 | extent_op); | |
2084 | } else { | |
2085 | BUG(); | |
2086 | } | |
2087 | return ret; | |
2088 | } | |
2089 | ||
2090 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2091 | struct extent_buffer *leaf, | |
2092 | struct btrfs_extent_item *ei) | |
2093 | { | |
2094 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2095 | if (extent_op->update_flags) { | |
2096 | flags |= extent_op->flags_to_set; | |
2097 | btrfs_set_extent_flags(leaf, ei, flags); | |
2098 | } | |
2099 | ||
2100 | if (extent_op->update_key) { | |
2101 | struct btrfs_tree_block_info *bi; | |
2102 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2103 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2104 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2105 | } | |
2106 | } | |
2107 | ||
2108 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2109 | struct btrfs_root *root, | |
2110 | struct btrfs_delayed_ref_node *node, | |
2111 | struct btrfs_delayed_extent_op *extent_op) | |
2112 | { | |
2113 | struct btrfs_key key; | |
2114 | struct btrfs_path *path; | |
2115 | struct btrfs_extent_item *ei; | |
2116 | struct extent_buffer *leaf; | |
2117 | u32 item_size; | |
56bec294 | 2118 | int ret; |
5d4f98a2 | 2119 | int err = 0; |
3173a18f JB |
2120 | int metadata = (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2121 | node->type == BTRFS_SHARED_BLOCK_REF_KEY); | |
5d4f98a2 | 2122 | |
79787eaa JM |
2123 | if (trans->aborted) |
2124 | return 0; | |
2125 | ||
3173a18f JB |
2126 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) |
2127 | metadata = 0; | |
2128 | ||
5d4f98a2 YZ |
2129 | path = btrfs_alloc_path(); |
2130 | if (!path) | |
2131 | return -ENOMEM; | |
2132 | ||
2133 | key.objectid = node->bytenr; | |
5d4f98a2 | 2134 | |
3173a18f JB |
2135 | if (metadata) { |
2136 | struct btrfs_delayed_tree_ref *tree_ref; | |
2137 | ||
2138 | tree_ref = btrfs_delayed_node_to_tree_ref(node); | |
2139 | key.type = BTRFS_METADATA_ITEM_KEY; | |
2140 | key.offset = tree_ref->level; | |
2141 | } else { | |
2142 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2143 | key.offset = node->num_bytes; | |
2144 | } | |
2145 | ||
2146 | again: | |
5d4f98a2 YZ |
2147 | path->reada = 1; |
2148 | path->leave_spinning = 1; | |
2149 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, | |
2150 | path, 0, 1); | |
2151 | if (ret < 0) { | |
2152 | err = ret; | |
2153 | goto out; | |
2154 | } | |
2155 | if (ret > 0) { | |
3173a18f JB |
2156 | if (metadata) { |
2157 | btrfs_release_path(path); | |
2158 | metadata = 0; | |
2159 | ||
2160 | key.offset = node->num_bytes; | |
2161 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
2162 | goto again; | |
2163 | } | |
5d4f98a2 YZ |
2164 | err = -EIO; |
2165 | goto out; | |
2166 | } | |
2167 | ||
2168 | leaf = path->nodes[0]; | |
2169 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2170 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2171 | if (item_size < sizeof(*ei)) { | |
2172 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, | |
2173 | path, (u64)-1, 0); | |
2174 | if (ret < 0) { | |
2175 | err = ret; | |
2176 | goto out; | |
2177 | } | |
2178 | leaf = path->nodes[0]; | |
2179 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2180 | } | |
2181 | #endif | |
2182 | BUG_ON(item_size < sizeof(*ei)); | |
2183 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2184 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2185 | |
5d4f98a2 YZ |
2186 | btrfs_mark_buffer_dirty(leaf); |
2187 | out: | |
2188 | btrfs_free_path(path); | |
2189 | return err; | |
56bec294 CM |
2190 | } |
2191 | ||
5d4f98a2 YZ |
2192 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2193 | struct btrfs_root *root, | |
2194 | struct btrfs_delayed_ref_node *node, | |
2195 | struct btrfs_delayed_extent_op *extent_op, | |
2196 | int insert_reserved) | |
56bec294 CM |
2197 | { |
2198 | int ret = 0; | |
5d4f98a2 YZ |
2199 | struct btrfs_delayed_tree_ref *ref; |
2200 | struct btrfs_key ins; | |
2201 | u64 parent = 0; | |
2202 | u64 ref_root = 0; | |
3173a18f JB |
2203 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
2204 | SKINNY_METADATA); | |
56bec294 | 2205 | |
5d4f98a2 YZ |
2206 | ref = btrfs_delayed_node_to_tree_ref(node); |
2207 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2208 | parent = ref->parent; | |
2209 | else | |
2210 | ref_root = ref->root; | |
2211 | ||
3173a18f JB |
2212 | ins.objectid = node->bytenr; |
2213 | if (skinny_metadata) { | |
2214 | ins.offset = ref->level; | |
2215 | ins.type = BTRFS_METADATA_ITEM_KEY; | |
2216 | } else { | |
2217 | ins.offset = node->num_bytes; | |
2218 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2219 | } | |
2220 | ||
5d4f98a2 YZ |
2221 | BUG_ON(node->ref_mod != 1); |
2222 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2223 | BUG_ON(!extent_op || !extent_op->update_flags); |
5d4f98a2 YZ |
2224 | ret = alloc_reserved_tree_block(trans, root, |
2225 | parent, ref_root, | |
2226 | extent_op->flags_to_set, | |
2227 | &extent_op->key, | |
2228 | ref->level, &ins); | |
5d4f98a2 YZ |
2229 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2230 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, | |
2231 | node->num_bytes, parent, ref_root, | |
2232 | ref->level, 0, 1, extent_op); | |
2233 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { | |
2234 | ret = __btrfs_free_extent(trans, root, node->bytenr, | |
2235 | node->num_bytes, parent, ref_root, | |
2236 | ref->level, 0, 1, extent_op); | |
2237 | } else { | |
2238 | BUG(); | |
2239 | } | |
56bec294 CM |
2240 | return ret; |
2241 | } | |
2242 | ||
2243 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 YZ |
2244 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2245 | struct btrfs_root *root, | |
2246 | struct btrfs_delayed_ref_node *node, | |
2247 | struct btrfs_delayed_extent_op *extent_op, | |
2248 | int insert_reserved) | |
56bec294 | 2249 | { |
79787eaa JM |
2250 | int ret = 0; |
2251 | ||
2252 | if (trans->aborted) | |
2253 | return 0; | |
2254 | ||
5d4f98a2 | 2255 | if (btrfs_delayed_ref_is_head(node)) { |
56bec294 CM |
2256 | struct btrfs_delayed_ref_head *head; |
2257 | /* | |
2258 | * we've hit the end of the chain and we were supposed | |
2259 | * to insert this extent into the tree. But, it got | |
2260 | * deleted before we ever needed to insert it, so all | |
2261 | * we have to do is clean up the accounting | |
2262 | */ | |
5d4f98a2 YZ |
2263 | BUG_ON(extent_op); |
2264 | head = btrfs_delayed_node_to_head(node); | |
56bec294 | 2265 | if (insert_reserved) { |
f0486c68 YZ |
2266 | btrfs_pin_extent(root, node->bytenr, |
2267 | node->num_bytes, 1); | |
5d4f98a2 YZ |
2268 | if (head->is_data) { |
2269 | ret = btrfs_del_csums(trans, root, | |
2270 | node->bytenr, | |
2271 | node->num_bytes); | |
5d4f98a2 | 2272 | } |
56bec294 | 2273 | } |
79787eaa | 2274 | return ret; |
56bec294 CM |
2275 | } |
2276 | ||
5d4f98a2 YZ |
2277 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2278 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2279 | ret = run_delayed_tree_ref(trans, root, node, extent_op, | |
2280 | insert_reserved); | |
2281 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2282 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2283 | ret = run_delayed_data_ref(trans, root, node, extent_op, | |
2284 | insert_reserved); | |
2285 | else | |
2286 | BUG(); | |
2287 | return ret; | |
56bec294 CM |
2288 | } |
2289 | ||
2290 | static noinline struct btrfs_delayed_ref_node * | |
2291 | select_delayed_ref(struct btrfs_delayed_ref_head *head) | |
2292 | { | |
2293 | struct rb_node *node; | |
2294 | struct btrfs_delayed_ref_node *ref; | |
2295 | int action = BTRFS_ADD_DELAYED_REF; | |
2296 | again: | |
2297 | /* | |
2298 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2299 | * this prevents ref count from going down to zero when | |
2300 | * there still are pending delayed ref. | |
2301 | */ | |
2302 | node = rb_prev(&head->node.rb_node); | |
2303 | while (1) { | |
2304 | if (!node) | |
2305 | break; | |
2306 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2307 | rb_node); | |
2308 | if (ref->bytenr != head->node.bytenr) | |
2309 | break; | |
5d4f98a2 | 2310 | if (ref->action == action) |
56bec294 CM |
2311 | return ref; |
2312 | node = rb_prev(node); | |
2313 | } | |
2314 | if (action == BTRFS_ADD_DELAYED_REF) { | |
2315 | action = BTRFS_DROP_DELAYED_REF; | |
2316 | goto again; | |
2317 | } | |
2318 | return NULL; | |
2319 | } | |
2320 | ||
79787eaa JM |
2321 | /* |
2322 | * Returns 0 on success or if called with an already aborted transaction. | |
2323 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2324 | */ | |
c3e69d58 CM |
2325 | static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, |
2326 | struct btrfs_root *root, | |
2327 | struct list_head *cluster) | |
56bec294 | 2328 | { |
56bec294 CM |
2329 | struct btrfs_delayed_ref_root *delayed_refs; |
2330 | struct btrfs_delayed_ref_node *ref; | |
2331 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2332 | struct btrfs_delayed_extent_op *extent_op; |
097b8a7c | 2333 | struct btrfs_fs_info *fs_info = root->fs_info; |
56bec294 | 2334 | int ret; |
c3e69d58 | 2335 | int count = 0; |
56bec294 | 2336 | int must_insert_reserved = 0; |
56bec294 CM |
2337 | |
2338 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2339 | while (1) { |
2340 | if (!locked_ref) { | |
c3e69d58 CM |
2341 | /* pick a new head ref from the cluster list */ |
2342 | if (list_empty(cluster)) | |
56bec294 | 2343 | break; |
56bec294 | 2344 | |
c3e69d58 CM |
2345 | locked_ref = list_entry(cluster->next, |
2346 | struct btrfs_delayed_ref_head, cluster); | |
2347 | ||
2348 | /* grab the lock that says we are going to process | |
2349 | * all the refs for this head */ | |
2350 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
2351 | ||
2352 | /* | |
2353 | * we may have dropped the spin lock to get the head | |
2354 | * mutex lock, and that might have given someone else | |
2355 | * time to free the head. If that's true, it has been | |
2356 | * removed from our list and we can move on. | |
2357 | */ | |
2358 | if (ret == -EAGAIN) { | |
2359 | locked_ref = NULL; | |
2360 | count++; | |
2361 | continue; | |
56bec294 CM |
2362 | } |
2363 | } | |
a28ec197 | 2364 | |
ae1e206b JB |
2365 | /* |
2366 | * We need to try and merge add/drops of the same ref since we | |
2367 | * can run into issues with relocate dropping the implicit ref | |
2368 | * and then it being added back again before the drop can | |
2369 | * finish. If we merged anything we need to re-loop so we can | |
2370 | * get a good ref. | |
2371 | */ | |
2372 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, | |
2373 | locked_ref); | |
2374 | ||
d1270cd9 AJ |
2375 | /* |
2376 | * locked_ref is the head node, so we have to go one | |
2377 | * node back for any delayed ref updates | |
2378 | */ | |
2379 | ref = select_delayed_ref(locked_ref); | |
2380 | ||
2381 | if (ref && ref->seq && | |
097b8a7c | 2382 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d1270cd9 AJ |
2383 | /* |
2384 | * there are still refs with lower seq numbers in the | |
2385 | * process of being added. Don't run this ref yet. | |
2386 | */ | |
2387 | list_del_init(&locked_ref->cluster); | |
093486c4 | 2388 | btrfs_delayed_ref_unlock(locked_ref); |
d1270cd9 AJ |
2389 | locked_ref = NULL; |
2390 | delayed_refs->num_heads_ready++; | |
2391 | spin_unlock(&delayed_refs->lock); | |
2392 | cond_resched(); | |
2393 | spin_lock(&delayed_refs->lock); | |
2394 | continue; | |
2395 | } | |
2396 | ||
56bec294 CM |
2397 | /* |
2398 | * record the must insert reserved flag before we | |
2399 | * drop the spin lock. | |
2400 | */ | |
2401 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2402 | locked_ref->must_insert_reserved = 0; | |
7bb86316 | 2403 | |
5d4f98a2 YZ |
2404 | extent_op = locked_ref->extent_op; |
2405 | locked_ref->extent_op = NULL; | |
2406 | ||
56bec294 CM |
2407 | if (!ref) { |
2408 | /* All delayed refs have been processed, Go ahead | |
2409 | * and send the head node to run_one_delayed_ref, | |
2410 | * so that any accounting fixes can happen | |
2411 | */ | |
2412 | ref = &locked_ref->node; | |
5d4f98a2 YZ |
2413 | |
2414 | if (extent_op && must_insert_reserved) { | |
78a6184a | 2415 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2416 | extent_op = NULL; |
2417 | } | |
2418 | ||
2419 | if (extent_op) { | |
2420 | spin_unlock(&delayed_refs->lock); | |
2421 | ||
2422 | ret = run_delayed_extent_op(trans, root, | |
2423 | ref, extent_op); | |
78a6184a | 2424 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 | 2425 | |
79787eaa | 2426 | if (ret) { |
c2cf52eb | 2427 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); |
253beebd | 2428 | spin_lock(&delayed_refs->lock); |
093486c4 | 2429 | btrfs_delayed_ref_unlock(locked_ref); |
79787eaa JM |
2430 | return ret; |
2431 | } | |
2432 | ||
203bf287 | 2433 | goto next; |
5d4f98a2 | 2434 | } |
56bec294 | 2435 | } |
02217ed2 | 2436 | |
56bec294 CM |
2437 | ref->in_tree = 0; |
2438 | rb_erase(&ref->rb_node, &delayed_refs->root); | |
2439 | delayed_refs->num_entries--; | |
093486c4 | 2440 | if (!btrfs_delayed_ref_is_head(ref)) { |
22cd2e7d AJ |
2441 | /* |
2442 | * when we play the delayed ref, also correct the | |
2443 | * ref_mod on head | |
2444 | */ | |
2445 | switch (ref->action) { | |
2446 | case BTRFS_ADD_DELAYED_REF: | |
2447 | case BTRFS_ADD_DELAYED_EXTENT: | |
2448 | locked_ref->node.ref_mod -= ref->ref_mod; | |
2449 | break; | |
2450 | case BTRFS_DROP_DELAYED_REF: | |
2451 | locked_ref->node.ref_mod += ref->ref_mod; | |
2452 | break; | |
2453 | default: | |
2454 | WARN_ON(1); | |
2455 | } | |
2456 | } | |
56bec294 | 2457 | spin_unlock(&delayed_refs->lock); |
925baedd | 2458 | |
5d4f98a2 | 2459 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
56bec294 | 2460 | must_insert_reserved); |
eb099670 | 2461 | |
78a6184a | 2462 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2463 | if (ret) { |
093486c4 MX |
2464 | btrfs_delayed_ref_unlock(locked_ref); |
2465 | btrfs_put_delayed_ref(ref); | |
c2cf52eb | 2466 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret); |
253beebd | 2467 | spin_lock(&delayed_refs->lock); |
79787eaa JM |
2468 | return ret; |
2469 | } | |
2470 | ||
093486c4 MX |
2471 | /* |
2472 | * If this node is a head, that means all the refs in this head | |
2473 | * have been dealt with, and we will pick the next head to deal | |
2474 | * with, so we must unlock the head and drop it from the cluster | |
2475 | * list before we release it. | |
2476 | */ | |
2477 | if (btrfs_delayed_ref_is_head(ref)) { | |
2478 | list_del_init(&locked_ref->cluster); | |
2479 | btrfs_delayed_ref_unlock(locked_ref); | |
2480 | locked_ref = NULL; | |
2481 | } | |
2482 | btrfs_put_delayed_ref(ref); | |
2483 | count++; | |
203bf287 | 2484 | next: |
c3e69d58 CM |
2485 | cond_resched(); |
2486 | spin_lock(&delayed_refs->lock); | |
2487 | } | |
2488 | return count; | |
2489 | } | |
2490 | ||
709c0486 AJ |
2491 | #ifdef SCRAMBLE_DELAYED_REFS |
2492 | /* | |
2493 | * Normally delayed refs get processed in ascending bytenr order. This | |
2494 | * correlates in most cases to the order added. To expose dependencies on this | |
2495 | * order, we start to process the tree in the middle instead of the beginning | |
2496 | */ | |
2497 | static u64 find_middle(struct rb_root *root) | |
2498 | { | |
2499 | struct rb_node *n = root->rb_node; | |
2500 | struct btrfs_delayed_ref_node *entry; | |
2501 | int alt = 1; | |
2502 | u64 middle; | |
2503 | u64 first = 0, last = 0; | |
2504 | ||
2505 | n = rb_first(root); | |
2506 | if (n) { | |
2507 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2508 | first = entry->bytenr; | |
2509 | } | |
2510 | n = rb_last(root); | |
2511 | if (n) { | |
2512 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2513 | last = entry->bytenr; | |
2514 | } | |
2515 | n = root->rb_node; | |
2516 | ||
2517 | while (n) { | |
2518 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2519 | WARN_ON(!entry->in_tree); | |
2520 | ||
2521 | middle = entry->bytenr; | |
2522 | ||
2523 | if (alt) | |
2524 | n = n->rb_left; | |
2525 | else | |
2526 | n = n->rb_right; | |
2527 | ||
2528 | alt = 1 - alt; | |
2529 | } | |
2530 | return middle; | |
2531 | } | |
2532 | #endif | |
2533 | ||
bed92eae AJ |
2534 | int btrfs_delayed_refs_qgroup_accounting(struct btrfs_trans_handle *trans, |
2535 | struct btrfs_fs_info *fs_info) | |
2536 | { | |
2537 | struct qgroup_update *qgroup_update; | |
2538 | int ret = 0; | |
2539 | ||
2540 | if (list_empty(&trans->qgroup_ref_list) != | |
2541 | !trans->delayed_ref_elem.seq) { | |
2542 | /* list without seq or seq without list */ | |
c2cf52eb SK |
2543 | btrfs_err(fs_info, |
2544 | "qgroup accounting update error, list is%s empty, seq is %llu", | |
bed92eae AJ |
2545 | list_empty(&trans->qgroup_ref_list) ? "" : " not", |
2546 | trans->delayed_ref_elem.seq); | |
2547 | BUG(); | |
2548 | } | |
2549 | ||
2550 | if (!trans->delayed_ref_elem.seq) | |
2551 | return 0; | |
2552 | ||
2553 | while (!list_empty(&trans->qgroup_ref_list)) { | |
2554 | qgroup_update = list_first_entry(&trans->qgroup_ref_list, | |
2555 | struct qgroup_update, list); | |
2556 | list_del(&qgroup_update->list); | |
2557 | if (!ret) | |
2558 | ret = btrfs_qgroup_account_ref( | |
2559 | trans, fs_info, qgroup_update->node, | |
2560 | qgroup_update->extent_op); | |
2561 | kfree(qgroup_update); | |
2562 | } | |
2563 | ||
2564 | btrfs_put_tree_mod_seq(fs_info, &trans->delayed_ref_elem); | |
2565 | ||
2566 | return ret; | |
2567 | } | |
2568 | ||
bb721703 CM |
2569 | static int refs_newer(struct btrfs_delayed_ref_root *delayed_refs, int seq, |
2570 | int count) | |
2571 | { | |
2572 | int val = atomic_read(&delayed_refs->ref_seq); | |
2573 | ||
2574 | if (val < seq || val >= seq + count) | |
2575 | return 1; | |
2576 | return 0; | |
2577 | } | |
2578 | ||
c3e69d58 CM |
2579 | /* |
2580 | * this starts processing the delayed reference count updates and | |
2581 | * extent insertions we have queued up so far. count can be | |
2582 | * 0, which means to process everything in the tree at the start | |
2583 | * of the run (but not newly added entries), or it can be some target | |
2584 | * number you'd like to process. | |
79787eaa JM |
2585 | * |
2586 | * Returns 0 on success or if called with an aborted transaction | |
2587 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2588 | */ |
2589 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2590 | struct btrfs_root *root, unsigned long count) | |
2591 | { | |
2592 | struct rb_node *node; | |
2593 | struct btrfs_delayed_ref_root *delayed_refs; | |
2594 | struct btrfs_delayed_ref_node *ref; | |
2595 | struct list_head cluster; | |
2596 | int ret; | |
a168650c | 2597 | u64 delayed_start; |
c3e69d58 CM |
2598 | int run_all = count == (unsigned long)-1; |
2599 | int run_most = 0; | |
1fa11e26 | 2600 | int loops; |
c3e69d58 | 2601 | |
79787eaa JM |
2602 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2603 | if (trans->aborted) | |
2604 | return 0; | |
2605 | ||
c3e69d58 CM |
2606 | if (root == root->fs_info->extent_root) |
2607 | root = root->fs_info->tree_root; | |
2608 | ||
edf39272 JS |
2609 | btrfs_delayed_refs_qgroup_accounting(trans, root->fs_info); |
2610 | ||
c3e69d58 CM |
2611 | delayed_refs = &trans->transaction->delayed_refs; |
2612 | INIT_LIST_HEAD(&cluster); | |
bb721703 CM |
2613 | if (count == 0) { |
2614 | count = delayed_refs->num_entries * 2; | |
2615 | run_most = 1; | |
2616 | } | |
2617 | ||
2618 | if (!run_all && !run_most) { | |
2619 | int old; | |
2620 | int seq = atomic_read(&delayed_refs->ref_seq); | |
2621 | ||
2622 | progress: | |
2623 | old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1); | |
2624 | if (old) { | |
2625 | DEFINE_WAIT(__wait); | |
2626 | if (delayed_refs->num_entries < 16348) | |
2627 | return 0; | |
2628 | ||
2629 | prepare_to_wait(&delayed_refs->wait, &__wait, | |
2630 | TASK_UNINTERRUPTIBLE); | |
2631 | ||
2632 | old = atomic_cmpxchg(&delayed_refs->procs_running_refs, 0, 1); | |
2633 | if (old) { | |
2634 | schedule(); | |
2635 | finish_wait(&delayed_refs->wait, &__wait); | |
2636 | ||
2637 | if (!refs_newer(delayed_refs, seq, 256)) | |
2638 | goto progress; | |
2639 | else | |
2640 | return 0; | |
2641 | } else { | |
2642 | finish_wait(&delayed_refs->wait, &__wait); | |
2643 | goto again; | |
2644 | } | |
2645 | } | |
2646 | ||
2647 | } else { | |
2648 | atomic_inc(&delayed_refs->procs_running_refs); | |
2649 | } | |
2650 | ||
c3e69d58 | 2651 | again: |
1fa11e26 | 2652 | loops = 0; |
c3e69d58 | 2653 | spin_lock(&delayed_refs->lock); |
097b8a7c | 2654 | |
709c0486 AJ |
2655 | #ifdef SCRAMBLE_DELAYED_REFS |
2656 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2657 | #endif | |
2658 | ||
c3e69d58 CM |
2659 | while (1) { |
2660 | if (!(run_all || run_most) && | |
2661 | delayed_refs->num_heads_ready < 64) | |
2662 | break; | |
eb099670 | 2663 | |
56bec294 | 2664 | /* |
c3e69d58 CM |
2665 | * go find something we can process in the rbtree. We start at |
2666 | * the beginning of the tree, and then build a cluster | |
2667 | * of refs to process starting at the first one we are able to | |
2668 | * lock | |
56bec294 | 2669 | */ |
a168650c | 2670 | delayed_start = delayed_refs->run_delayed_start; |
c3e69d58 CM |
2671 | ret = btrfs_find_ref_cluster(trans, &cluster, |
2672 | delayed_refs->run_delayed_start); | |
2673 | if (ret) | |
56bec294 CM |
2674 | break; |
2675 | ||
c3e69d58 | 2676 | ret = run_clustered_refs(trans, root, &cluster); |
79787eaa | 2677 | if (ret < 0) { |
093486c4 | 2678 | btrfs_release_ref_cluster(&cluster); |
79787eaa JM |
2679 | spin_unlock(&delayed_refs->lock); |
2680 | btrfs_abort_transaction(trans, root, ret); | |
bb721703 | 2681 | atomic_dec(&delayed_refs->procs_running_refs); |
79787eaa JM |
2682 | return ret; |
2683 | } | |
c3e69d58 | 2684 | |
bb721703 CM |
2685 | atomic_add(ret, &delayed_refs->ref_seq); |
2686 | ||
c3e69d58 CM |
2687 | count -= min_t(unsigned long, ret, count); |
2688 | ||
2689 | if (count == 0) | |
2690 | break; | |
a168650c | 2691 | |
1fa11e26 AJ |
2692 | if (delayed_start >= delayed_refs->run_delayed_start) { |
2693 | if (loops == 0) { | |
2694 | /* | |
2695 | * btrfs_find_ref_cluster looped. let's do one | |
2696 | * more cycle. if we don't run any delayed ref | |
2697 | * during that cycle (because we can't because | |
2698 | * all of them are blocked), bail out. | |
2699 | */ | |
2700 | loops = 1; | |
2701 | } else { | |
2702 | /* | |
2703 | * no runnable refs left, stop trying | |
2704 | */ | |
2705 | BUG_ON(run_all); | |
2706 | break; | |
2707 | } | |
2708 | } | |
2709 | if (ret) { | |
a168650c | 2710 | /* refs were run, let's reset staleness detection */ |
1fa11e26 | 2711 | loops = 0; |
a168650c | 2712 | } |
eb099670 | 2713 | } |
c3e69d58 | 2714 | |
56bec294 | 2715 | if (run_all) { |
ea658bad JB |
2716 | if (!list_empty(&trans->new_bgs)) { |
2717 | spin_unlock(&delayed_refs->lock); | |
2718 | btrfs_create_pending_block_groups(trans, root); | |
2719 | spin_lock(&delayed_refs->lock); | |
2720 | } | |
2721 | ||
56bec294 | 2722 | node = rb_first(&delayed_refs->root); |
c3e69d58 | 2723 | if (!node) |
56bec294 | 2724 | goto out; |
c3e69d58 | 2725 | count = (unsigned long)-1; |
e9d0b13b | 2726 | |
56bec294 CM |
2727 | while (node) { |
2728 | ref = rb_entry(node, struct btrfs_delayed_ref_node, | |
2729 | rb_node); | |
2730 | if (btrfs_delayed_ref_is_head(ref)) { | |
2731 | struct btrfs_delayed_ref_head *head; | |
5caf2a00 | 2732 | |
56bec294 CM |
2733 | head = btrfs_delayed_node_to_head(ref); |
2734 | atomic_inc(&ref->refs); | |
2735 | ||
2736 | spin_unlock(&delayed_refs->lock); | |
8cc33e5c DS |
2737 | /* |
2738 | * Mutex was contended, block until it's | |
2739 | * released and try again | |
2740 | */ | |
56bec294 CM |
2741 | mutex_lock(&head->mutex); |
2742 | mutex_unlock(&head->mutex); | |
2743 | ||
2744 | btrfs_put_delayed_ref(ref); | |
1887be66 | 2745 | cond_resched(); |
56bec294 CM |
2746 | goto again; |
2747 | } | |
2748 | node = rb_next(node); | |
2749 | } | |
2750 | spin_unlock(&delayed_refs->lock); | |
56bec294 CM |
2751 | schedule_timeout(1); |
2752 | goto again; | |
5f39d397 | 2753 | } |
54aa1f4d | 2754 | out: |
bb721703 CM |
2755 | atomic_dec(&delayed_refs->procs_running_refs); |
2756 | smp_mb(); | |
2757 | if (waitqueue_active(&delayed_refs->wait)) | |
2758 | wake_up(&delayed_refs->wait); | |
2759 | ||
c3e69d58 | 2760 | spin_unlock(&delayed_refs->lock); |
edf39272 | 2761 | assert_qgroups_uptodate(trans); |
a28ec197 CM |
2762 | return 0; |
2763 | } | |
2764 | ||
5d4f98a2 YZ |
2765 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2766 | struct btrfs_root *root, | |
2767 | u64 bytenr, u64 num_bytes, u64 flags, | |
2768 | int is_data) | |
2769 | { | |
2770 | struct btrfs_delayed_extent_op *extent_op; | |
2771 | int ret; | |
2772 | ||
78a6184a | 2773 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2774 | if (!extent_op) |
2775 | return -ENOMEM; | |
2776 | ||
2777 | extent_op->flags_to_set = flags; | |
2778 | extent_op->update_flags = 1; | |
2779 | extent_op->update_key = 0; | |
2780 | extent_op->is_data = is_data ? 1 : 0; | |
2781 | ||
66d7e7f0 AJ |
2782 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, |
2783 | num_bytes, extent_op); | |
5d4f98a2 | 2784 | if (ret) |
78a6184a | 2785 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2786 | return ret; |
2787 | } | |
2788 | ||
2789 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, | |
2790 | struct btrfs_root *root, | |
2791 | struct btrfs_path *path, | |
2792 | u64 objectid, u64 offset, u64 bytenr) | |
2793 | { | |
2794 | struct btrfs_delayed_ref_head *head; | |
2795 | struct btrfs_delayed_ref_node *ref; | |
2796 | struct btrfs_delayed_data_ref *data_ref; | |
2797 | struct btrfs_delayed_ref_root *delayed_refs; | |
2798 | struct rb_node *node; | |
2799 | int ret = 0; | |
2800 | ||
2801 | ret = -ENOENT; | |
2802 | delayed_refs = &trans->transaction->delayed_refs; | |
2803 | spin_lock(&delayed_refs->lock); | |
2804 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
2805 | if (!head) | |
2806 | goto out; | |
2807 | ||
2808 | if (!mutex_trylock(&head->mutex)) { | |
2809 | atomic_inc(&head->node.refs); | |
2810 | spin_unlock(&delayed_refs->lock); | |
2811 | ||
b3b4aa74 | 2812 | btrfs_release_path(path); |
5d4f98a2 | 2813 | |
8cc33e5c DS |
2814 | /* |
2815 | * Mutex was contended, block until it's released and let | |
2816 | * caller try again | |
2817 | */ | |
5d4f98a2 YZ |
2818 | mutex_lock(&head->mutex); |
2819 | mutex_unlock(&head->mutex); | |
2820 | btrfs_put_delayed_ref(&head->node); | |
2821 | return -EAGAIN; | |
2822 | } | |
2823 | ||
2824 | node = rb_prev(&head->node.rb_node); | |
2825 | if (!node) | |
2826 | goto out_unlock; | |
2827 | ||
2828 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
2829 | ||
2830 | if (ref->bytenr != bytenr) | |
2831 | goto out_unlock; | |
2832 | ||
2833 | ret = 1; | |
2834 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) | |
2835 | goto out_unlock; | |
2836 | ||
2837 | data_ref = btrfs_delayed_node_to_data_ref(ref); | |
2838 | ||
2839 | node = rb_prev(node); | |
2840 | if (node) { | |
df57dbe6 LB |
2841 | int seq = ref->seq; |
2842 | ||
5d4f98a2 | 2843 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); |
df57dbe6 | 2844 | if (ref->bytenr == bytenr && ref->seq == seq) |
5d4f98a2 YZ |
2845 | goto out_unlock; |
2846 | } | |
2847 | ||
2848 | if (data_ref->root != root->root_key.objectid || | |
2849 | data_ref->objectid != objectid || data_ref->offset != offset) | |
2850 | goto out_unlock; | |
2851 | ||
2852 | ret = 0; | |
2853 | out_unlock: | |
2854 | mutex_unlock(&head->mutex); | |
2855 | out: | |
2856 | spin_unlock(&delayed_refs->lock); | |
2857 | return ret; | |
2858 | } | |
2859 | ||
2860 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, | |
2861 | struct btrfs_root *root, | |
2862 | struct btrfs_path *path, | |
2863 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d CM |
2864 | { |
2865 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
f321e491 | 2866 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2867 | struct btrfs_extent_data_ref *ref; |
2868 | struct btrfs_extent_inline_ref *iref; | |
2869 | struct btrfs_extent_item *ei; | |
f321e491 | 2870 | struct btrfs_key key; |
5d4f98a2 | 2871 | u32 item_size; |
be20aa9d | 2872 | int ret; |
925baedd | 2873 | |
be20aa9d | 2874 | key.objectid = bytenr; |
31840ae1 | 2875 | key.offset = (u64)-1; |
f321e491 | 2876 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2877 | |
be20aa9d CM |
2878 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2879 | if (ret < 0) | |
2880 | goto out; | |
79787eaa | 2881 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2882 | |
2883 | ret = -ENOENT; | |
2884 | if (path->slots[0] == 0) | |
31840ae1 | 2885 | goto out; |
be20aa9d | 2886 | |
31840ae1 | 2887 | path->slots[0]--; |
f321e491 | 2888 | leaf = path->nodes[0]; |
5d4f98a2 | 2889 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 2890 | |
5d4f98a2 | 2891 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 2892 | goto out; |
f321e491 | 2893 | |
5d4f98a2 YZ |
2894 | ret = 1; |
2895 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2896 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2897 | if (item_size < sizeof(*ei)) { | |
2898 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
2899 | goto out; | |
2900 | } | |
2901 | #endif | |
2902 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 2903 | |
5d4f98a2 YZ |
2904 | if (item_size != sizeof(*ei) + |
2905 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
2906 | goto out; | |
be20aa9d | 2907 | |
5d4f98a2 YZ |
2908 | if (btrfs_extent_generation(leaf, ei) <= |
2909 | btrfs_root_last_snapshot(&root->root_item)) | |
2910 | goto out; | |
2911 | ||
2912 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
2913 | if (btrfs_extent_inline_ref_type(leaf, iref) != | |
2914 | BTRFS_EXTENT_DATA_REF_KEY) | |
2915 | goto out; | |
2916 | ||
2917 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
2918 | if (btrfs_extent_refs(leaf, ei) != | |
2919 | btrfs_extent_data_ref_count(leaf, ref) || | |
2920 | btrfs_extent_data_ref_root(leaf, ref) != | |
2921 | root->root_key.objectid || | |
2922 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
2923 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
2924 | goto out; | |
2925 | ||
2926 | ret = 0; | |
2927 | out: | |
2928 | return ret; | |
2929 | } | |
2930 | ||
2931 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, | |
2932 | struct btrfs_root *root, | |
2933 | u64 objectid, u64 offset, u64 bytenr) | |
2934 | { | |
2935 | struct btrfs_path *path; | |
2936 | int ret; | |
2937 | int ret2; | |
2938 | ||
2939 | path = btrfs_alloc_path(); | |
2940 | if (!path) | |
2941 | return -ENOENT; | |
2942 | ||
2943 | do { | |
2944 | ret = check_committed_ref(trans, root, path, objectid, | |
2945 | offset, bytenr); | |
2946 | if (ret && ret != -ENOENT) | |
f321e491 | 2947 | goto out; |
80ff3856 | 2948 | |
5d4f98a2 YZ |
2949 | ret2 = check_delayed_ref(trans, root, path, objectid, |
2950 | offset, bytenr); | |
2951 | } while (ret2 == -EAGAIN); | |
2952 | ||
2953 | if (ret2 && ret2 != -ENOENT) { | |
2954 | ret = ret2; | |
2955 | goto out; | |
f321e491 | 2956 | } |
5d4f98a2 YZ |
2957 | |
2958 | if (ret != -ENOENT || ret2 != -ENOENT) | |
2959 | ret = 0; | |
be20aa9d | 2960 | out: |
80ff3856 | 2961 | btrfs_free_path(path); |
f0486c68 YZ |
2962 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
2963 | WARN_ON(ret > 0); | |
f321e491 | 2964 | return ret; |
be20aa9d | 2965 | } |
c5739bba | 2966 | |
5d4f98a2 | 2967 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 2968 | struct btrfs_root *root, |
5d4f98a2 | 2969 | struct extent_buffer *buf, |
66d7e7f0 | 2970 | int full_backref, int inc, int for_cow) |
31840ae1 ZY |
2971 | { |
2972 | u64 bytenr; | |
5d4f98a2 YZ |
2973 | u64 num_bytes; |
2974 | u64 parent; | |
31840ae1 | 2975 | u64 ref_root; |
31840ae1 | 2976 | u32 nritems; |
31840ae1 ZY |
2977 | struct btrfs_key key; |
2978 | struct btrfs_file_extent_item *fi; | |
2979 | int i; | |
2980 | int level; | |
2981 | int ret = 0; | |
31840ae1 | 2982 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
66d7e7f0 | 2983 | u64, u64, u64, u64, u64, u64, int); |
31840ae1 ZY |
2984 | |
2985 | ref_root = btrfs_header_owner(buf); | |
31840ae1 ZY |
2986 | nritems = btrfs_header_nritems(buf); |
2987 | level = btrfs_header_level(buf); | |
2988 | ||
5d4f98a2 YZ |
2989 | if (!root->ref_cows && level == 0) |
2990 | return 0; | |
31840ae1 | 2991 | |
5d4f98a2 YZ |
2992 | if (inc) |
2993 | process_func = btrfs_inc_extent_ref; | |
2994 | else | |
2995 | process_func = btrfs_free_extent; | |
31840ae1 | 2996 | |
5d4f98a2 YZ |
2997 | if (full_backref) |
2998 | parent = buf->start; | |
2999 | else | |
3000 | parent = 0; | |
3001 | ||
3002 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3003 | if (level == 0) { |
5d4f98a2 | 3004 | btrfs_item_key_to_cpu(buf, &key, i); |
31840ae1 ZY |
3005 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
3006 | continue; | |
5d4f98a2 | 3007 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3008 | struct btrfs_file_extent_item); |
3009 | if (btrfs_file_extent_type(buf, fi) == | |
3010 | BTRFS_FILE_EXTENT_INLINE) | |
3011 | continue; | |
3012 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3013 | if (bytenr == 0) | |
3014 | continue; | |
5d4f98a2 YZ |
3015 | |
3016 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3017 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
3018 | ret = process_func(trans, root, bytenr, num_bytes, | |
3019 | parent, ref_root, key.objectid, | |
66d7e7f0 | 3020 | key.offset, for_cow); |
31840ae1 ZY |
3021 | if (ret) |
3022 | goto fail; | |
3023 | } else { | |
5d4f98a2 YZ |
3024 | bytenr = btrfs_node_blockptr(buf, i); |
3025 | num_bytes = btrfs_level_size(root, level - 1); | |
3026 | ret = process_func(trans, root, bytenr, num_bytes, | |
66d7e7f0 AJ |
3027 | parent, ref_root, level - 1, 0, |
3028 | for_cow); | |
31840ae1 ZY |
3029 | if (ret) |
3030 | goto fail; | |
3031 | } | |
3032 | } | |
3033 | return 0; | |
3034 | fail: | |
5d4f98a2 YZ |
3035 | return ret; |
3036 | } | |
3037 | ||
3038 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 3039 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 3040 | { |
66d7e7f0 | 3041 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow); |
5d4f98a2 YZ |
3042 | } |
3043 | ||
3044 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
66d7e7f0 | 3045 | struct extent_buffer *buf, int full_backref, int for_cow) |
5d4f98a2 | 3046 | { |
66d7e7f0 | 3047 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow); |
31840ae1 ZY |
3048 | } |
3049 | ||
9078a3e1 CM |
3050 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
3051 | struct btrfs_root *root, | |
3052 | struct btrfs_path *path, | |
3053 | struct btrfs_block_group_cache *cache) | |
3054 | { | |
3055 | int ret; | |
9078a3e1 | 3056 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
5f39d397 CM |
3057 | unsigned long bi; |
3058 | struct extent_buffer *leaf; | |
9078a3e1 | 3059 | |
9078a3e1 | 3060 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
54aa1f4d CM |
3061 | if (ret < 0) |
3062 | goto fail; | |
79787eaa | 3063 | BUG_ON(ret); /* Corruption */ |
5f39d397 CM |
3064 | |
3065 | leaf = path->nodes[0]; | |
3066 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3067 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3068 | btrfs_mark_buffer_dirty(leaf); | |
b3b4aa74 | 3069 | btrfs_release_path(path); |
54aa1f4d | 3070 | fail: |
79787eaa JM |
3071 | if (ret) { |
3072 | btrfs_abort_transaction(trans, root, ret); | |
9078a3e1 | 3073 | return ret; |
79787eaa | 3074 | } |
9078a3e1 CM |
3075 | return 0; |
3076 | ||
3077 | } | |
3078 | ||
4a8c9a62 YZ |
3079 | static struct btrfs_block_group_cache * |
3080 | next_block_group(struct btrfs_root *root, | |
3081 | struct btrfs_block_group_cache *cache) | |
3082 | { | |
3083 | struct rb_node *node; | |
3084 | spin_lock(&root->fs_info->block_group_cache_lock); | |
3085 | node = rb_next(&cache->cache_node); | |
3086 | btrfs_put_block_group(cache); | |
3087 | if (node) { | |
3088 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3089 | cache_node); | |
11dfe35a | 3090 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3091 | } else |
3092 | cache = NULL; | |
3093 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
3094 | return cache; | |
3095 | } | |
3096 | ||
0af3d00b JB |
3097 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3098 | struct btrfs_trans_handle *trans, | |
3099 | struct btrfs_path *path) | |
3100 | { | |
3101 | struct btrfs_root *root = block_group->fs_info->tree_root; | |
3102 | struct inode *inode = NULL; | |
3103 | u64 alloc_hint = 0; | |
2b20982e | 3104 | int dcs = BTRFS_DC_ERROR; |
0af3d00b JB |
3105 | int num_pages = 0; |
3106 | int retries = 0; | |
3107 | int ret = 0; | |
3108 | ||
3109 | /* | |
3110 | * If this block group is smaller than 100 megs don't bother caching the | |
3111 | * block group. | |
3112 | */ | |
3113 | if (block_group->key.offset < (100 * 1024 * 1024)) { | |
3114 | spin_lock(&block_group->lock); | |
3115 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3116 | spin_unlock(&block_group->lock); | |
3117 | return 0; | |
3118 | } | |
3119 | ||
3120 | again: | |
3121 | inode = lookup_free_space_inode(root, block_group, path); | |
3122 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { | |
3123 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3124 | btrfs_release_path(path); |
0af3d00b JB |
3125 | goto out; |
3126 | } | |
3127 | ||
3128 | if (IS_ERR(inode)) { | |
3129 | BUG_ON(retries); | |
3130 | retries++; | |
3131 | ||
3132 | if (block_group->ro) | |
3133 | goto out_free; | |
3134 | ||
3135 | ret = create_free_space_inode(root, trans, block_group, path); | |
3136 | if (ret) | |
3137 | goto out_free; | |
3138 | goto again; | |
3139 | } | |
3140 | ||
5b0e95bf JB |
3141 | /* We've already setup this transaction, go ahead and exit */ |
3142 | if (block_group->cache_generation == trans->transid && | |
3143 | i_size_read(inode)) { | |
3144 | dcs = BTRFS_DC_SETUP; | |
3145 | goto out_put; | |
3146 | } | |
3147 | ||
0af3d00b JB |
3148 | /* |
3149 | * We want to set the generation to 0, that way if anything goes wrong | |
3150 | * from here on out we know not to trust this cache when we load up next | |
3151 | * time. | |
3152 | */ | |
3153 | BTRFS_I(inode)->generation = 0; | |
3154 | ret = btrfs_update_inode(trans, root, inode); | |
3155 | WARN_ON(ret); | |
3156 | ||
3157 | if (i_size_read(inode) > 0) { | |
3158 | ret = btrfs_truncate_free_space_cache(root, trans, path, | |
3159 | inode); | |
3160 | if (ret) | |
3161 | goto out_put; | |
3162 | } | |
3163 | ||
3164 | spin_lock(&block_group->lock); | |
cf7c1ef6 LB |
3165 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
3166 | !btrfs_test_opt(root, SPACE_CACHE)) { | |
3167 | /* | |
3168 | * don't bother trying to write stuff out _if_ | |
3169 | * a) we're not cached, | |
3170 | * b) we're with nospace_cache mount option. | |
3171 | */ | |
2b20982e | 3172 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3173 | spin_unlock(&block_group->lock); |
3174 | goto out_put; | |
3175 | } | |
3176 | spin_unlock(&block_group->lock); | |
3177 | ||
6fc823b1 JB |
3178 | /* |
3179 | * Try to preallocate enough space based on how big the block group is. | |
3180 | * Keep in mind this has to include any pinned space which could end up | |
3181 | * taking up quite a bit since it's not folded into the other space | |
3182 | * cache. | |
3183 | */ | |
3184 | num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024); | |
0af3d00b JB |
3185 | if (!num_pages) |
3186 | num_pages = 1; | |
3187 | ||
0af3d00b JB |
3188 | num_pages *= 16; |
3189 | num_pages *= PAGE_CACHE_SIZE; | |
3190 | ||
3191 | ret = btrfs_check_data_free_space(inode, num_pages); | |
3192 | if (ret) | |
3193 | goto out_put; | |
3194 | ||
3195 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3196 | num_pages, num_pages, | |
3197 | &alloc_hint); | |
2b20982e JB |
3198 | if (!ret) |
3199 | dcs = BTRFS_DC_SETUP; | |
0af3d00b | 3200 | btrfs_free_reserved_data_space(inode, num_pages); |
c09544e0 | 3201 | |
0af3d00b JB |
3202 | out_put: |
3203 | iput(inode); | |
3204 | out_free: | |
b3b4aa74 | 3205 | btrfs_release_path(path); |
0af3d00b JB |
3206 | out: |
3207 | spin_lock(&block_group->lock); | |
e65cbb94 | 3208 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3209 | block_group->cache_generation = trans->transid; |
2b20982e | 3210 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3211 | spin_unlock(&block_group->lock); |
3212 | ||
3213 | return ret; | |
3214 | } | |
3215 | ||
96b5179d CM |
3216 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
3217 | struct btrfs_root *root) | |
9078a3e1 | 3218 | { |
4a8c9a62 | 3219 | struct btrfs_block_group_cache *cache; |
9078a3e1 | 3220 | int err = 0; |
9078a3e1 | 3221 | struct btrfs_path *path; |
96b5179d | 3222 | u64 last = 0; |
9078a3e1 CM |
3223 | |
3224 | path = btrfs_alloc_path(); | |
3225 | if (!path) | |
3226 | return -ENOMEM; | |
3227 | ||
0af3d00b JB |
3228 | again: |
3229 | while (1) { | |
3230 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
3231 | while (cache) { | |
3232 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3233 | break; | |
3234 | cache = next_block_group(root, cache); | |
3235 | } | |
3236 | if (!cache) { | |
3237 | if (last == 0) | |
3238 | break; | |
3239 | last = 0; | |
3240 | continue; | |
3241 | } | |
3242 | err = cache_save_setup(cache, trans, path); | |
3243 | last = cache->key.objectid + cache->key.offset; | |
3244 | btrfs_put_block_group(cache); | |
3245 | } | |
3246 | ||
d397712b | 3247 | while (1) { |
4a8c9a62 YZ |
3248 | if (last == 0) { |
3249 | err = btrfs_run_delayed_refs(trans, root, | |
3250 | (unsigned long)-1); | |
79787eaa JM |
3251 | if (err) /* File system offline */ |
3252 | goto out; | |
0f9dd46c | 3253 | } |
54aa1f4d | 3254 | |
4a8c9a62 YZ |
3255 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
3256 | while (cache) { | |
0af3d00b JB |
3257 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) { |
3258 | btrfs_put_block_group(cache); | |
3259 | goto again; | |
3260 | } | |
3261 | ||
4a8c9a62 YZ |
3262 | if (cache->dirty) |
3263 | break; | |
3264 | cache = next_block_group(root, cache); | |
3265 | } | |
3266 | if (!cache) { | |
3267 | if (last == 0) | |
3268 | break; | |
3269 | last = 0; | |
3270 | continue; | |
3271 | } | |
0f9dd46c | 3272 | |
0cb59c99 JB |
3273 | if (cache->disk_cache_state == BTRFS_DC_SETUP) |
3274 | cache->disk_cache_state = BTRFS_DC_NEED_WRITE; | |
e8569813 | 3275 | cache->dirty = 0; |
4a8c9a62 | 3276 | last = cache->key.objectid + cache->key.offset; |
0f9dd46c | 3277 | |
4a8c9a62 | 3278 | err = write_one_cache_group(trans, root, path, cache); |
79787eaa JM |
3279 | if (err) /* File system offline */ |
3280 | goto out; | |
3281 | ||
4a8c9a62 | 3282 | btrfs_put_block_group(cache); |
9078a3e1 | 3283 | } |
4a8c9a62 | 3284 | |
0cb59c99 JB |
3285 | while (1) { |
3286 | /* | |
3287 | * I don't think this is needed since we're just marking our | |
3288 | * preallocated extent as written, but just in case it can't | |
3289 | * hurt. | |
3290 | */ | |
3291 | if (last == 0) { | |
3292 | err = btrfs_run_delayed_refs(trans, root, | |
3293 | (unsigned long)-1); | |
79787eaa JM |
3294 | if (err) /* File system offline */ |
3295 | goto out; | |
0cb59c99 JB |
3296 | } |
3297 | ||
3298 | cache = btrfs_lookup_first_block_group(root->fs_info, last); | |
3299 | while (cache) { | |
3300 | /* | |
3301 | * Really this shouldn't happen, but it could if we | |
3302 | * couldn't write the entire preallocated extent and | |
3303 | * splitting the extent resulted in a new block. | |
3304 | */ | |
3305 | if (cache->dirty) { | |
3306 | btrfs_put_block_group(cache); | |
3307 | goto again; | |
3308 | } | |
3309 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) | |
3310 | break; | |
3311 | cache = next_block_group(root, cache); | |
3312 | } | |
3313 | if (!cache) { | |
3314 | if (last == 0) | |
3315 | break; | |
3316 | last = 0; | |
3317 | continue; | |
3318 | } | |
3319 | ||
79787eaa | 3320 | err = btrfs_write_out_cache(root, trans, cache, path); |
0cb59c99 JB |
3321 | |
3322 | /* | |
3323 | * If we didn't have an error then the cache state is still | |
3324 | * NEED_WRITE, so we can set it to WRITTEN. | |
3325 | */ | |
79787eaa | 3326 | if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE) |
0cb59c99 JB |
3327 | cache->disk_cache_state = BTRFS_DC_WRITTEN; |
3328 | last = cache->key.objectid + cache->key.offset; | |
3329 | btrfs_put_block_group(cache); | |
3330 | } | |
79787eaa | 3331 | out: |
0cb59c99 | 3332 | |
9078a3e1 | 3333 | btrfs_free_path(path); |
79787eaa | 3334 | return err; |
9078a3e1 CM |
3335 | } |
3336 | ||
d2fb3437 YZ |
3337 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
3338 | { | |
3339 | struct btrfs_block_group_cache *block_group; | |
3340 | int readonly = 0; | |
3341 | ||
3342 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); | |
3343 | if (!block_group || block_group->ro) | |
3344 | readonly = 1; | |
3345 | if (block_group) | |
fa9c0d79 | 3346 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3347 | return readonly; |
3348 | } | |
3349 | ||
593060d7 CM |
3350 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
3351 | u64 total_bytes, u64 bytes_used, | |
3352 | struct btrfs_space_info **space_info) | |
3353 | { | |
3354 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3355 | int i; |
3356 | int factor; | |
3357 | ||
3358 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3359 | BTRFS_BLOCK_GROUP_RAID10)) | |
3360 | factor = 2; | |
3361 | else | |
3362 | factor = 1; | |
593060d7 CM |
3363 | |
3364 | found = __find_space_info(info, flags); | |
3365 | if (found) { | |
25179201 | 3366 | spin_lock(&found->lock); |
593060d7 | 3367 | found->total_bytes += total_bytes; |
89a55897 | 3368 | found->disk_total += total_bytes * factor; |
593060d7 | 3369 | found->bytes_used += bytes_used; |
b742bb82 | 3370 | found->disk_used += bytes_used * factor; |
8f18cf13 | 3371 | found->full = 0; |
25179201 | 3372 | spin_unlock(&found->lock); |
593060d7 CM |
3373 | *space_info = found; |
3374 | return 0; | |
3375 | } | |
c146afad | 3376 | found = kzalloc(sizeof(*found), GFP_NOFS); |
593060d7 CM |
3377 | if (!found) |
3378 | return -ENOMEM; | |
3379 | ||
b742bb82 YZ |
3380 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
3381 | INIT_LIST_HEAD(&found->block_groups[i]); | |
80eb234a | 3382 | init_rwsem(&found->groups_sem); |
0f9dd46c | 3383 | spin_lock_init(&found->lock); |
52ba6929 | 3384 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
593060d7 | 3385 | found->total_bytes = total_bytes; |
89a55897 | 3386 | found->disk_total = total_bytes * factor; |
593060d7 | 3387 | found->bytes_used = bytes_used; |
b742bb82 | 3388 | found->disk_used = bytes_used * factor; |
593060d7 | 3389 | found->bytes_pinned = 0; |
e8569813 | 3390 | found->bytes_reserved = 0; |
c146afad | 3391 | found->bytes_readonly = 0; |
f0486c68 | 3392 | found->bytes_may_use = 0; |
593060d7 | 3393 | found->full = 0; |
0e4f8f88 | 3394 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
6d74119f | 3395 | found->chunk_alloc = 0; |
fdb5effd JB |
3396 | found->flush = 0; |
3397 | init_waitqueue_head(&found->wait); | |
593060d7 | 3398 | *space_info = found; |
4184ea7f | 3399 | list_add_rcu(&found->list, &info->space_info); |
b4d7c3c9 LZ |
3400 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3401 | info->data_sinfo = found; | |
593060d7 CM |
3402 | return 0; |
3403 | } | |
3404 | ||
8790d502 CM |
3405 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3406 | { | |
899c81ea ID |
3407 | u64 extra_flags = chunk_to_extended(flags) & |
3408 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 3409 | |
de98ced9 | 3410 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
3411 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3412 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3413 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3414 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3415 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3416 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 3417 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 3418 | } |
593060d7 | 3419 | |
fc67c450 ID |
3420 | /* |
3421 | * returns target flags in extended format or 0 if restripe for this | |
3422 | * chunk_type is not in progress | |
c6664b42 ID |
3423 | * |
3424 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
3425 | */ |
3426 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
3427 | { | |
3428 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3429 | u64 target = 0; | |
3430 | ||
fc67c450 ID |
3431 | if (!bctl) |
3432 | return 0; | |
3433 | ||
3434 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
3435 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3436 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3437 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
3438 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3439 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3440 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
3441 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3442 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3443 | } | |
3444 | ||
3445 | return target; | |
3446 | } | |
3447 | ||
a46d11a8 ID |
3448 | /* |
3449 | * @flags: available profiles in extended format (see ctree.h) | |
3450 | * | |
e4d8ec0f ID |
3451 | * Returns reduced profile in chunk format. If profile changing is in |
3452 | * progress (either running or paused) picks the target profile (if it's | |
3453 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 3454 | */ |
2b82032c | 3455 | u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
ec44a35c | 3456 | { |
cd02dca5 CM |
3457 | /* |
3458 | * we add in the count of missing devices because we want | |
3459 | * to make sure that any RAID levels on a degraded FS | |
3460 | * continue to be honored. | |
3461 | */ | |
3462 | u64 num_devices = root->fs_info->fs_devices->rw_devices + | |
3463 | root->fs_info->fs_devices->missing_devices; | |
fc67c450 | 3464 | u64 target; |
53b381b3 | 3465 | u64 tmp; |
a061fc8d | 3466 | |
fc67c450 ID |
3467 | /* |
3468 | * see if restripe for this chunk_type is in progress, if so | |
3469 | * try to reduce to the target profile | |
3470 | */ | |
e4d8ec0f | 3471 | spin_lock(&root->fs_info->balance_lock); |
fc67c450 ID |
3472 | target = get_restripe_target(root->fs_info, flags); |
3473 | if (target) { | |
3474 | /* pick target profile only if it's already available */ | |
3475 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
e4d8ec0f | 3476 | spin_unlock(&root->fs_info->balance_lock); |
fc67c450 | 3477 | return extended_to_chunk(target); |
e4d8ec0f ID |
3478 | } |
3479 | } | |
3480 | spin_unlock(&root->fs_info->balance_lock); | |
3481 | ||
53b381b3 | 3482 | /* First, mask out the RAID levels which aren't possible */ |
a061fc8d | 3483 | if (num_devices == 1) |
53b381b3 DW |
3484 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 | |
3485 | BTRFS_BLOCK_GROUP_RAID5); | |
3486 | if (num_devices < 3) | |
3487 | flags &= ~BTRFS_BLOCK_GROUP_RAID6; | |
a061fc8d CM |
3488 | if (num_devices < 4) |
3489 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; | |
3490 | ||
53b381b3 DW |
3491 | tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 | |
3492 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 | | |
3493 | BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10); | |
3494 | flags &= ~tmp; | |
ec44a35c | 3495 | |
53b381b3 DW |
3496 | if (tmp & BTRFS_BLOCK_GROUP_RAID6) |
3497 | tmp = BTRFS_BLOCK_GROUP_RAID6; | |
3498 | else if (tmp & BTRFS_BLOCK_GROUP_RAID5) | |
3499 | tmp = BTRFS_BLOCK_GROUP_RAID5; | |
3500 | else if (tmp & BTRFS_BLOCK_GROUP_RAID10) | |
3501 | tmp = BTRFS_BLOCK_GROUP_RAID10; | |
3502 | else if (tmp & BTRFS_BLOCK_GROUP_RAID1) | |
3503 | tmp = BTRFS_BLOCK_GROUP_RAID1; | |
3504 | else if (tmp & BTRFS_BLOCK_GROUP_RAID0) | |
3505 | tmp = BTRFS_BLOCK_GROUP_RAID0; | |
a46d11a8 | 3506 | |
53b381b3 | 3507 | return extended_to_chunk(flags | tmp); |
ec44a35c CM |
3508 | } |
3509 | ||
b742bb82 | 3510 | static u64 get_alloc_profile(struct btrfs_root *root, u64 flags) |
6a63209f | 3511 | { |
de98ced9 MX |
3512 | unsigned seq; |
3513 | ||
3514 | do { | |
3515 | seq = read_seqbegin(&root->fs_info->profiles_lock); | |
3516 | ||
3517 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3518 | flags |= root->fs_info->avail_data_alloc_bits; | |
3519 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3520 | flags |= root->fs_info->avail_system_alloc_bits; | |
3521 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3522 | flags |= root->fs_info->avail_metadata_alloc_bits; | |
3523 | } while (read_seqretry(&root->fs_info->profiles_lock, seq)); | |
6fef8df1 | 3524 | |
b742bb82 | 3525 | return btrfs_reduce_alloc_profile(root, flags); |
6a63209f JB |
3526 | } |
3527 | ||
6d07bcec | 3528 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
9ed74f2d | 3529 | { |
b742bb82 | 3530 | u64 flags; |
53b381b3 | 3531 | u64 ret; |
9ed74f2d | 3532 | |
b742bb82 YZ |
3533 | if (data) |
3534 | flags = BTRFS_BLOCK_GROUP_DATA; | |
3535 | else if (root == root->fs_info->chunk_root) | |
3536 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
9ed74f2d | 3537 | else |
b742bb82 | 3538 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3539 | |
53b381b3 DW |
3540 | ret = get_alloc_profile(root, flags); |
3541 | return ret; | |
6a63209f | 3542 | } |
9ed74f2d | 3543 | |
6a63209f | 3544 | /* |
6a63209f JB |
3545 | * This will check the space that the inode allocates from to make sure we have |
3546 | * enough space for bytes. | |
6a63209f | 3547 | */ |
0ca1f7ce | 3548 | int btrfs_check_data_free_space(struct inode *inode, u64 bytes) |
6a63209f | 3549 | { |
6a63209f | 3550 | struct btrfs_space_info *data_sinfo; |
0ca1f7ce | 3551 | struct btrfs_root *root = BTRFS_I(inode)->root; |
b4d7c3c9 | 3552 | struct btrfs_fs_info *fs_info = root->fs_info; |
ab6e2410 | 3553 | u64 used; |
0af3d00b | 3554 | int ret = 0, committed = 0, alloc_chunk = 1; |
6a63209f | 3555 | |
6a63209f | 3556 | /* make sure bytes are sectorsize aligned */ |
fda2832f | 3557 | bytes = ALIGN(bytes, root->sectorsize); |
6a63209f | 3558 | |
82d5902d LZ |
3559 | if (root == root->fs_info->tree_root || |
3560 | BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID) { | |
0af3d00b JB |
3561 | alloc_chunk = 0; |
3562 | committed = 1; | |
3563 | } | |
3564 | ||
b4d7c3c9 | 3565 | data_sinfo = fs_info->data_sinfo; |
33b4d47f CM |
3566 | if (!data_sinfo) |
3567 | goto alloc; | |
9ed74f2d | 3568 | |
6a63209f JB |
3569 | again: |
3570 | /* make sure we have enough space to handle the data first */ | |
3571 | spin_lock(&data_sinfo->lock); | |
8929ecfa YZ |
3572 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
3573 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + | |
3574 | data_sinfo->bytes_may_use; | |
ab6e2410 JB |
3575 | |
3576 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 3577 | struct btrfs_trans_handle *trans; |
9ed74f2d | 3578 | |
6a63209f JB |
3579 | /* |
3580 | * if we don't have enough free bytes in this space then we need | |
3581 | * to alloc a new chunk. | |
3582 | */ | |
0af3d00b | 3583 | if (!data_sinfo->full && alloc_chunk) { |
6a63209f | 3584 | u64 alloc_target; |
9ed74f2d | 3585 | |
0e4f8f88 | 3586 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 3587 | spin_unlock(&data_sinfo->lock); |
33b4d47f | 3588 | alloc: |
6a63209f | 3589 | alloc_target = btrfs_get_alloc_profile(root, 1); |
7a7eaa40 | 3590 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3591 | if (IS_ERR(trans)) |
3592 | return PTR_ERR(trans); | |
9ed74f2d | 3593 | |
6a63209f | 3594 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
0e4f8f88 CM |
3595 | alloc_target, |
3596 | CHUNK_ALLOC_NO_FORCE); | |
6a63209f | 3597 | btrfs_end_transaction(trans, root); |
d52a5b5f MX |
3598 | if (ret < 0) { |
3599 | if (ret != -ENOSPC) | |
3600 | return ret; | |
3601 | else | |
3602 | goto commit_trans; | |
3603 | } | |
9ed74f2d | 3604 | |
b4d7c3c9 LZ |
3605 | if (!data_sinfo) |
3606 | data_sinfo = fs_info->data_sinfo; | |
3607 | ||
6a63209f JB |
3608 | goto again; |
3609 | } | |
f2bb8f5c JB |
3610 | |
3611 | /* | |
3612 | * If we have less pinned bytes than we want to allocate then | |
3613 | * don't bother committing the transaction, it won't help us. | |
3614 | */ | |
3615 | if (data_sinfo->bytes_pinned < bytes) | |
3616 | committed = 1; | |
6a63209f | 3617 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3618 | |
4e06bdd6 | 3619 | /* commit the current transaction and try again */ |
d52a5b5f | 3620 | commit_trans: |
a4abeea4 JB |
3621 | if (!committed && |
3622 | !atomic_read(&root->fs_info->open_ioctl_trans)) { | |
4e06bdd6 | 3623 | committed = 1; |
7a7eaa40 | 3624 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3625 | if (IS_ERR(trans)) |
3626 | return PTR_ERR(trans); | |
4e06bdd6 JB |
3627 | ret = btrfs_commit_transaction(trans, root); |
3628 | if (ret) | |
3629 | return ret; | |
3630 | goto again; | |
3631 | } | |
9ed74f2d | 3632 | |
6a63209f JB |
3633 | return -ENOSPC; |
3634 | } | |
3635 | data_sinfo->bytes_may_use += bytes; | |
8c2a3ca2 | 3636 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3637 | data_sinfo->flags, bytes, 1); |
6a63209f | 3638 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3639 | |
9ed74f2d | 3640 | return 0; |
9ed74f2d | 3641 | } |
6a63209f | 3642 | |
6a63209f | 3643 | /* |
fb25e914 | 3644 | * Called if we need to clear a data reservation for this inode. |
6a63209f | 3645 | */ |
0ca1f7ce | 3646 | void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes) |
e3ccfa98 | 3647 | { |
0ca1f7ce | 3648 | struct btrfs_root *root = BTRFS_I(inode)->root; |
6a63209f | 3649 | struct btrfs_space_info *data_sinfo; |
e3ccfa98 | 3650 | |
6a63209f | 3651 | /* make sure bytes are sectorsize aligned */ |
fda2832f | 3652 | bytes = ALIGN(bytes, root->sectorsize); |
e3ccfa98 | 3653 | |
b4d7c3c9 | 3654 | data_sinfo = root->fs_info->data_sinfo; |
6a63209f JB |
3655 | spin_lock(&data_sinfo->lock); |
3656 | data_sinfo->bytes_may_use -= bytes; | |
8c2a3ca2 | 3657 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
2bcc0328 | 3658 | data_sinfo->flags, bytes, 0); |
6a63209f | 3659 | spin_unlock(&data_sinfo->lock); |
e3ccfa98 JB |
3660 | } |
3661 | ||
97e728d4 | 3662 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 3663 | { |
97e728d4 JB |
3664 | struct list_head *head = &info->space_info; |
3665 | struct btrfs_space_info *found; | |
e3ccfa98 | 3666 | |
97e728d4 JB |
3667 | rcu_read_lock(); |
3668 | list_for_each_entry_rcu(found, head, list) { | |
3669 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 3670 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 3671 | } |
97e728d4 | 3672 | rcu_read_unlock(); |
e3ccfa98 JB |
3673 | } |
3674 | ||
e5bc2458 | 3675 | static int should_alloc_chunk(struct btrfs_root *root, |
698d0082 | 3676 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 3677 | { |
fb25e914 | 3678 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
424499db | 3679 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
0e4f8f88 | 3680 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
e5bc2458 | 3681 | u64 thresh; |
e3ccfa98 | 3682 | |
0e4f8f88 CM |
3683 | if (force == CHUNK_ALLOC_FORCE) |
3684 | return 1; | |
3685 | ||
fb25e914 JB |
3686 | /* |
3687 | * We need to take into account the global rsv because for all intents | |
3688 | * and purposes it's used space. Don't worry about locking the | |
3689 | * global_rsv, it doesn't change except when the transaction commits. | |
3690 | */ | |
54338b5c JB |
3691 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
3692 | num_allocated += global_rsv->size; | |
fb25e914 | 3693 | |
0e4f8f88 CM |
3694 | /* |
3695 | * in limited mode, we want to have some free space up to | |
3696 | * about 1% of the FS size. | |
3697 | */ | |
3698 | if (force == CHUNK_ALLOC_LIMITED) { | |
6c41761f | 3699 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
0e4f8f88 CM |
3700 | thresh = max_t(u64, 64 * 1024 * 1024, |
3701 | div_factor_fine(thresh, 1)); | |
3702 | ||
3703 | if (num_bytes - num_allocated < thresh) | |
3704 | return 1; | |
3705 | } | |
0e4f8f88 | 3706 | |
698d0082 | 3707 | if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8)) |
14ed0ca6 | 3708 | return 0; |
424499db | 3709 | return 1; |
32c00aff JB |
3710 | } |
3711 | ||
15d1ff81 LB |
3712 | static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type) |
3713 | { | |
3714 | u64 num_dev; | |
3715 | ||
53b381b3 DW |
3716 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
3717 | BTRFS_BLOCK_GROUP_RAID0 | | |
3718 | BTRFS_BLOCK_GROUP_RAID5 | | |
3719 | BTRFS_BLOCK_GROUP_RAID6)) | |
15d1ff81 LB |
3720 | num_dev = root->fs_info->fs_devices->rw_devices; |
3721 | else if (type & BTRFS_BLOCK_GROUP_RAID1) | |
3722 | num_dev = 2; | |
3723 | else | |
3724 | num_dev = 1; /* DUP or single */ | |
3725 | ||
3726 | /* metadata for updaing devices and chunk tree */ | |
3727 | return btrfs_calc_trans_metadata_size(root, num_dev + 1); | |
3728 | } | |
3729 | ||
3730 | static void check_system_chunk(struct btrfs_trans_handle *trans, | |
3731 | struct btrfs_root *root, u64 type) | |
3732 | { | |
3733 | struct btrfs_space_info *info; | |
3734 | u64 left; | |
3735 | u64 thresh; | |
3736 | ||
3737 | info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
3738 | spin_lock(&info->lock); | |
3739 | left = info->total_bytes - info->bytes_used - info->bytes_pinned - | |
3740 | info->bytes_reserved - info->bytes_readonly; | |
3741 | spin_unlock(&info->lock); | |
3742 | ||
3743 | thresh = get_system_chunk_thresh(root, type); | |
3744 | if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
c2cf52eb SK |
3745 | btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu", |
3746 | left, thresh, type); | |
15d1ff81 LB |
3747 | dump_space_info(info, 0, 0); |
3748 | } | |
3749 | ||
3750 | if (left < thresh) { | |
3751 | u64 flags; | |
3752 | ||
3753 | flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0); | |
3754 | btrfs_alloc_chunk(trans, root, flags); | |
3755 | } | |
3756 | } | |
3757 | ||
6324fbf3 | 3758 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
698d0082 | 3759 | struct btrfs_root *extent_root, u64 flags, int force) |
9ed74f2d | 3760 | { |
6324fbf3 | 3761 | struct btrfs_space_info *space_info; |
97e728d4 | 3762 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
6d74119f | 3763 | int wait_for_alloc = 0; |
9ed74f2d | 3764 | int ret = 0; |
9ed74f2d | 3765 | |
c6b305a8 JB |
3766 | /* Don't re-enter if we're already allocating a chunk */ |
3767 | if (trans->allocating_chunk) | |
3768 | return -ENOSPC; | |
3769 | ||
6324fbf3 | 3770 | space_info = __find_space_info(extent_root->fs_info, flags); |
593060d7 CM |
3771 | if (!space_info) { |
3772 | ret = update_space_info(extent_root->fs_info, flags, | |
3773 | 0, 0, &space_info); | |
79787eaa | 3774 | BUG_ON(ret); /* -ENOMEM */ |
9ed74f2d | 3775 | } |
79787eaa | 3776 | BUG_ON(!space_info); /* Logic error */ |
9ed74f2d | 3777 | |
6d74119f | 3778 | again: |
25179201 | 3779 | spin_lock(&space_info->lock); |
9e622d6b | 3780 | if (force < space_info->force_alloc) |
0e4f8f88 | 3781 | force = space_info->force_alloc; |
25179201 JB |
3782 | if (space_info->full) { |
3783 | spin_unlock(&space_info->lock); | |
6d74119f | 3784 | return 0; |
9ed74f2d JB |
3785 | } |
3786 | ||
698d0082 | 3787 | if (!should_alloc_chunk(extent_root, space_info, force)) { |
25179201 | 3788 | spin_unlock(&space_info->lock); |
6d74119f JB |
3789 | return 0; |
3790 | } else if (space_info->chunk_alloc) { | |
3791 | wait_for_alloc = 1; | |
3792 | } else { | |
3793 | space_info->chunk_alloc = 1; | |
9ed74f2d | 3794 | } |
0e4f8f88 | 3795 | |
25179201 | 3796 | spin_unlock(&space_info->lock); |
9ed74f2d | 3797 | |
6d74119f JB |
3798 | mutex_lock(&fs_info->chunk_mutex); |
3799 | ||
3800 | /* | |
3801 | * The chunk_mutex is held throughout the entirety of a chunk | |
3802 | * allocation, so once we've acquired the chunk_mutex we know that the | |
3803 | * other guy is done and we need to recheck and see if we should | |
3804 | * allocate. | |
3805 | */ | |
3806 | if (wait_for_alloc) { | |
3807 | mutex_unlock(&fs_info->chunk_mutex); | |
3808 | wait_for_alloc = 0; | |
3809 | goto again; | |
3810 | } | |
3811 | ||
c6b305a8 JB |
3812 | trans->allocating_chunk = true; |
3813 | ||
67377734 JB |
3814 | /* |
3815 | * If we have mixed data/metadata chunks we want to make sure we keep | |
3816 | * allocating mixed chunks instead of individual chunks. | |
3817 | */ | |
3818 | if (btrfs_mixed_space_info(space_info)) | |
3819 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
3820 | ||
97e728d4 JB |
3821 | /* |
3822 | * if we're doing a data chunk, go ahead and make sure that | |
3823 | * we keep a reasonable number of metadata chunks allocated in the | |
3824 | * FS as well. | |
3825 | */ | |
9ed74f2d | 3826 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
3827 | fs_info->data_chunk_allocations++; |
3828 | if (!(fs_info->data_chunk_allocations % | |
3829 | fs_info->metadata_ratio)) | |
3830 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
3831 | } |
3832 | ||
15d1ff81 LB |
3833 | /* |
3834 | * Check if we have enough space in SYSTEM chunk because we may need | |
3835 | * to update devices. | |
3836 | */ | |
3837 | check_system_chunk(trans, extent_root, flags); | |
3838 | ||
2b82032c | 3839 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
c6b305a8 | 3840 | trans->allocating_chunk = false; |
92b8e897 | 3841 | |
9ed74f2d | 3842 | spin_lock(&space_info->lock); |
a81cb9a2 AO |
3843 | if (ret < 0 && ret != -ENOSPC) |
3844 | goto out; | |
9ed74f2d | 3845 | if (ret) |
6324fbf3 | 3846 | space_info->full = 1; |
424499db YZ |
3847 | else |
3848 | ret = 1; | |
6d74119f | 3849 | |
0e4f8f88 | 3850 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 3851 | out: |
6d74119f | 3852 | space_info->chunk_alloc = 0; |
9ed74f2d | 3853 | spin_unlock(&space_info->lock); |
a25c75d5 | 3854 | mutex_unlock(&fs_info->chunk_mutex); |
0f9dd46c | 3855 | return ret; |
6324fbf3 | 3856 | } |
9ed74f2d | 3857 | |
a80c8dcf JB |
3858 | static int can_overcommit(struct btrfs_root *root, |
3859 | struct btrfs_space_info *space_info, u64 bytes, | |
08e007d2 | 3860 | enum btrfs_reserve_flush_enum flush) |
a80c8dcf | 3861 | { |
96f1bb57 | 3862 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
a80c8dcf | 3863 | u64 profile = btrfs_get_alloc_profile(root, 0); |
96f1bb57 | 3864 | u64 rsv_size = 0; |
a80c8dcf JB |
3865 | u64 avail; |
3866 | u64 used; | |
70afa399 | 3867 | u64 to_add; |
a80c8dcf JB |
3868 | |
3869 | used = space_info->bytes_used + space_info->bytes_reserved + | |
96f1bb57 JB |
3870 | space_info->bytes_pinned + space_info->bytes_readonly; |
3871 | ||
3872 | spin_lock(&global_rsv->lock); | |
3873 | rsv_size = global_rsv->size; | |
3874 | spin_unlock(&global_rsv->lock); | |
3875 | ||
3876 | /* | |
3877 | * We only want to allow over committing if we have lots of actual space | |
3878 | * free, but if we don't have enough space to handle the global reserve | |
3879 | * space then we could end up having a real enospc problem when trying | |
3880 | * to allocate a chunk or some other such important allocation. | |
3881 | */ | |
3882 | rsv_size <<= 1; | |
3883 | if (used + rsv_size >= space_info->total_bytes) | |
3884 | return 0; | |
3885 | ||
3886 | used += space_info->bytes_may_use; | |
a80c8dcf JB |
3887 | |
3888 | spin_lock(&root->fs_info->free_chunk_lock); | |
3889 | avail = root->fs_info->free_chunk_space; | |
3890 | spin_unlock(&root->fs_info->free_chunk_lock); | |
3891 | ||
3892 | /* | |
3893 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
3894 | * space is actually useable. For raid56, the space info used |
3895 | * doesn't include the parity drive, so we don't have to | |
3896 | * change the math | |
a80c8dcf JB |
3897 | */ |
3898 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
3899 | BTRFS_BLOCK_GROUP_RAID1 | | |
3900 | BTRFS_BLOCK_GROUP_RAID10)) | |
3901 | avail >>= 1; | |
3902 | ||
70afa399 JB |
3903 | to_add = space_info->total_bytes; |
3904 | ||
a80c8dcf | 3905 | /* |
561c294d MX |
3906 | * If we aren't flushing all things, let us overcommit up to |
3907 | * 1/2th of the space. If we can flush, don't let us overcommit | |
3908 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 3909 | */ |
08e007d2 | 3910 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
70afa399 | 3911 | to_add >>= 3; |
a80c8dcf | 3912 | else |
70afa399 | 3913 | to_add >>= 1; |
a80c8dcf | 3914 | |
70afa399 JB |
3915 | /* |
3916 | * Limit the overcommit to the amount of free space we could possibly | |
3917 | * allocate for chunks. | |
3918 | */ | |
3919 | to_add = min(avail, to_add); | |
a80c8dcf | 3920 | |
70afa399 | 3921 | if (used + bytes < space_info->total_bytes + to_add) |
a80c8dcf JB |
3922 | return 1; |
3923 | return 0; | |
3924 | } | |
3925 | ||
da633a42 MX |
3926 | void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root, |
3927 | unsigned long nr_pages) | |
3928 | { | |
3929 | struct super_block *sb = root->fs_info->sb; | |
3930 | int started; | |
3931 | ||
3932 | /* If we can not start writeback, just sync all the delalloc file. */ | |
b695188d | 3933 | started = try_to_writeback_inodes_sb_nr(sb, nr_pages, |
da633a42 MX |
3934 | WB_REASON_FS_FREE_SPACE); |
3935 | if (!started) { | |
3936 | /* | |
3937 | * We needn't worry the filesystem going from r/w to r/o though | |
3938 | * we don't acquire ->s_umount mutex, because the filesystem | |
3939 | * should guarantee the delalloc inodes list be empty after | |
3940 | * the filesystem is readonly(all dirty pages are written to | |
3941 | * the disk). | |
3942 | */ | |
3943 | btrfs_start_delalloc_inodes(root, 0); | |
98ad69cf JB |
3944 | if (!current->journal_info) |
3945 | btrfs_wait_ordered_extents(root, 0); | |
da633a42 MX |
3946 | } |
3947 | } | |
3948 | ||
9ed74f2d | 3949 | /* |
5da9d01b | 3950 | * shrink metadata reservation for delalloc |
9ed74f2d | 3951 | */ |
f4c738c2 JB |
3952 | static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig, |
3953 | bool wait_ordered) | |
5da9d01b | 3954 | { |
0ca1f7ce | 3955 | struct btrfs_block_rsv *block_rsv; |
0019f10d | 3956 | struct btrfs_space_info *space_info; |
663350ac | 3957 | struct btrfs_trans_handle *trans; |
f4c738c2 | 3958 | u64 delalloc_bytes; |
5da9d01b | 3959 | u64 max_reclaim; |
b1953bce | 3960 | long time_left; |
877da174 | 3961 | unsigned long nr_pages = (2 * 1024 * 1024) >> PAGE_CACHE_SHIFT; |
b1953bce | 3962 | int loops = 0; |
08e007d2 | 3963 | enum btrfs_reserve_flush_enum flush; |
5da9d01b | 3964 | |
663350ac | 3965 | trans = (struct btrfs_trans_handle *)current->journal_info; |
0ca1f7ce | 3966 | block_rsv = &root->fs_info->delalloc_block_rsv; |
0019f10d | 3967 | space_info = block_rsv->space_info; |
bf9022e0 CM |
3968 | |
3969 | smp_mb(); | |
963d678b MX |
3970 | delalloc_bytes = percpu_counter_sum_positive( |
3971 | &root->fs_info->delalloc_bytes); | |
f4c738c2 | 3972 | if (delalloc_bytes == 0) { |
fdb5effd | 3973 | if (trans) |
f4c738c2 | 3974 | return; |
6bbe3a9c | 3975 | btrfs_wait_ordered_extents(root, 0); |
f4c738c2 | 3976 | return; |
fdb5effd JB |
3977 | } |
3978 | ||
f4c738c2 JB |
3979 | while (delalloc_bytes && loops < 3) { |
3980 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
3981 | nr_pages = max_reclaim >> PAGE_CACHE_SHIFT; | |
da633a42 | 3982 | btrfs_writeback_inodes_sb_nr(root, nr_pages); |
dea31f52 JB |
3983 | /* |
3984 | * We need to wait for the async pages to actually start before | |
3985 | * we do anything. | |
3986 | */ | |
3987 | wait_event(root->fs_info->async_submit_wait, | |
3988 | !atomic_read(&root->fs_info->async_delalloc_pages)); | |
3989 | ||
08e007d2 MX |
3990 | if (!trans) |
3991 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
3992 | else | |
3993 | flush = BTRFS_RESERVE_NO_FLUSH; | |
0019f10d | 3994 | spin_lock(&space_info->lock); |
08e007d2 | 3995 | if (can_overcommit(root, space_info, orig, flush)) { |
f4c738c2 JB |
3996 | spin_unlock(&space_info->lock); |
3997 | break; | |
3998 | } | |
0019f10d | 3999 | spin_unlock(&space_info->lock); |
5da9d01b | 4000 | |
36e39c40 | 4001 | loops++; |
f104d044 | 4002 | if (wait_ordered && !trans) { |
6bbe3a9c | 4003 | btrfs_wait_ordered_extents(root, 0); |
f104d044 | 4004 | } else { |
f4c738c2 | 4005 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4006 | if (time_left) |
4007 | break; | |
4008 | } | |
f4c738c2 | 4009 | smp_mb(); |
963d678b MX |
4010 | delalloc_bytes = percpu_counter_sum_positive( |
4011 | &root->fs_info->delalloc_bytes); | |
5da9d01b | 4012 | } |
5da9d01b YZ |
4013 | } |
4014 | ||
663350ac JB |
4015 | /** |
4016 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4017 | * @root - the root we're allocating for | |
4018 | * @bytes - the number of bytes we want to reserve | |
4019 | * @force - force the commit | |
8bb8ab2e | 4020 | * |
663350ac JB |
4021 | * This will check to make sure that committing the transaction will actually |
4022 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4023 | * will return -ENOSPC. | |
8bb8ab2e | 4024 | */ |
663350ac JB |
4025 | static int may_commit_transaction(struct btrfs_root *root, |
4026 | struct btrfs_space_info *space_info, | |
4027 | u64 bytes, int force) | |
4028 | { | |
4029 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; | |
4030 | struct btrfs_trans_handle *trans; | |
4031 | ||
4032 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4033 | if (trans) | |
4034 | return -EAGAIN; | |
4035 | ||
4036 | if (force) | |
4037 | goto commit; | |
4038 | ||
4039 | /* See if there is enough pinned space to make this reservation */ | |
4040 | spin_lock(&space_info->lock); | |
4041 | if (space_info->bytes_pinned >= bytes) { | |
4042 | spin_unlock(&space_info->lock); | |
4043 | goto commit; | |
4044 | } | |
4045 | spin_unlock(&space_info->lock); | |
4046 | ||
4047 | /* | |
4048 | * See if there is some space in the delayed insertion reservation for | |
4049 | * this reservation. | |
4050 | */ | |
4051 | if (space_info != delayed_rsv->space_info) | |
4052 | return -ENOSPC; | |
4053 | ||
d9b0218f | 4054 | spin_lock(&space_info->lock); |
663350ac | 4055 | spin_lock(&delayed_rsv->lock); |
d9b0218f | 4056 | if (space_info->bytes_pinned + delayed_rsv->size < bytes) { |
663350ac | 4057 | spin_unlock(&delayed_rsv->lock); |
d9b0218f | 4058 | spin_unlock(&space_info->lock); |
663350ac JB |
4059 | return -ENOSPC; |
4060 | } | |
4061 | spin_unlock(&delayed_rsv->lock); | |
d9b0218f | 4062 | spin_unlock(&space_info->lock); |
663350ac JB |
4063 | |
4064 | commit: | |
4065 | trans = btrfs_join_transaction(root); | |
4066 | if (IS_ERR(trans)) | |
4067 | return -ENOSPC; | |
4068 | ||
4069 | return btrfs_commit_transaction(trans, root); | |
4070 | } | |
4071 | ||
96c3f433 | 4072 | enum flush_state { |
67b0fd63 JB |
4073 | FLUSH_DELAYED_ITEMS_NR = 1, |
4074 | FLUSH_DELAYED_ITEMS = 2, | |
4075 | FLUSH_DELALLOC = 3, | |
4076 | FLUSH_DELALLOC_WAIT = 4, | |
ea658bad JB |
4077 | ALLOC_CHUNK = 5, |
4078 | COMMIT_TRANS = 6, | |
96c3f433 JB |
4079 | }; |
4080 | ||
4081 | static int flush_space(struct btrfs_root *root, | |
4082 | struct btrfs_space_info *space_info, u64 num_bytes, | |
4083 | u64 orig_bytes, int state) | |
4084 | { | |
4085 | struct btrfs_trans_handle *trans; | |
4086 | int nr; | |
f4c738c2 | 4087 | int ret = 0; |
96c3f433 JB |
4088 | |
4089 | switch (state) { | |
96c3f433 JB |
4090 | case FLUSH_DELAYED_ITEMS_NR: |
4091 | case FLUSH_DELAYED_ITEMS: | |
4092 | if (state == FLUSH_DELAYED_ITEMS_NR) { | |
4093 | u64 bytes = btrfs_calc_trans_metadata_size(root, 1); | |
4094 | ||
4095 | nr = (int)div64_u64(num_bytes, bytes); | |
4096 | if (!nr) | |
4097 | nr = 1; | |
4098 | nr *= 2; | |
4099 | } else { | |
4100 | nr = -1; | |
4101 | } | |
4102 | trans = btrfs_join_transaction(root); | |
4103 | if (IS_ERR(trans)) { | |
4104 | ret = PTR_ERR(trans); | |
4105 | break; | |
4106 | } | |
4107 | ret = btrfs_run_delayed_items_nr(trans, root, nr); | |
4108 | btrfs_end_transaction(trans, root); | |
4109 | break; | |
67b0fd63 JB |
4110 | case FLUSH_DELALLOC: |
4111 | case FLUSH_DELALLOC_WAIT: | |
4112 | shrink_delalloc(root, num_bytes, orig_bytes, | |
4113 | state == FLUSH_DELALLOC_WAIT); | |
4114 | break; | |
ea658bad JB |
4115 | case ALLOC_CHUNK: |
4116 | trans = btrfs_join_transaction(root); | |
4117 | if (IS_ERR(trans)) { | |
4118 | ret = PTR_ERR(trans); | |
4119 | break; | |
4120 | } | |
4121 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, | |
ea658bad JB |
4122 | btrfs_get_alloc_profile(root, 0), |
4123 | CHUNK_ALLOC_NO_FORCE); | |
4124 | btrfs_end_transaction(trans, root); | |
4125 | if (ret == -ENOSPC) | |
4126 | ret = 0; | |
4127 | break; | |
96c3f433 JB |
4128 | case COMMIT_TRANS: |
4129 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); | |
4130 | break; | |
4131 | default: | |
4132 | ret = -ENOSPC; | |
4133 | break; | |
4134 | } | |
4135 | ||
4136 | return ret; | |
4137 | } | |
4a92b1b8 JB |
4138 | /** |
4139 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
4140 | * @root - the root we're allocating for | |
4141 | * @block_rsv - the block_rsv we're allocating for | |
4142 | * @orig_bytes - the number of bytes we want | |
48fc7f7e | 4143 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 4144 | * |
4a92b1b8 JB |
4145 | * This will reserve orgi_bytes number of bytes from the space info associated |
4146 | * with the block_rsv. If there is not enough space it will make an attempt to | |
4147 | * flush out space to make room. It will do this by flushing delalloc if | |
4148 | * possible or committing the transaction. If flush is 0 then no attempts to | |
4149 | * regain reservations will be made and this will fail if there is not enough | |
4150 | * space already. | |
8bb8ab2e | 4151 | */ |
4a92b1b8 | 4152 | static int reserve_metadata_bytes(struct btrfs_root *root, |
8bb8ab2e | 4153 | struct btrfs_block_rsv *block_rsv, |
08e007d2 MX |
4154 | u64 orig_bytes, |
4155 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 4156 | { |
f0486c68 | 4157 | struct btrfs_space_info *space_info = block_rsv->space_info; |
2bf64758 | 4158 | u64 used; |
8bb8ab2e | 4159 | u64 num_bytes = orig_bytes; |
67b0fd63 | 4160 | int flush_state = FLUSH_DELAYED_ITEMS_NR; |
8bb8ab2e | 4161 | int ret = 0; |
fdb5effd | 4162 | bool flushing = false; |
9ed74f2d | 4163 | |
8bb8ab2e | 4164 | again: |
fdb5effd | 4165 | ret = 0; |
8bb8ab2e | 4166 | spin_lock(&space_info->lock); |
fdb5effd | 4167 | /* |
08e007d2 MX |
4168 | * We only want to wait if somebody other than us is flushing and we |
4169 | * are actually allowed to flush all things. | |
fdb5effd | 4170 | */ |
08e007d2 MX |
4171 | while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing && |
4172 | space_info->flush) { | |
fdb5effd JB |
4173 | spin_unlock(&space_info->lock); |
4174 | /* | |
4175 | * If we have a trans handle we can't wait because the flusher | |
4176 | * may have to commit the transaction, which would mean we would | |
4177 | * deadlock since we are waiting for the flusher to finish, but | |
4178 | * hold the current transaction open. | |
4179 | */ | |
663350ac | 4180 | if (current->journal_info) |
fdb5effd | 4181 | return -EAGAIN; |
b9688bb8 AJ |
4182 | ret = wait_event_killable(space_info->wait, !space_info->flush); |
4183 | /* Must have been killed, return */ | |
4184 | if (ret) | |
fdb5effd JB |
4185 | return -EINTR; |
4186 | ||
4187 | spin_lock(&space_info->lock); | |
4188 | } | |
4189 | ||
4190 | ret = -ENOSPC; | |
2bf64758 JB |
4191 | used = space_info->bytes_used + space_info->bytes_reserved + |
4192 | space_info->bytes_pinned + space_info->bytes_readonly + | |
4193 | space_info->bytes_may_use; | |
9ed74f2d | 4194 | |
8bb8ab2e JB |
4195 | /* |
4196 | * The idea here is that we've not already over-reserved the block group | |
4197 | * then we can go ahead and save our reservation first and then start | |
4198 | * flushing if we need to. Otherwise if we've already overcommitted | |
4199 | * lets start flushing stuff first and then come back and try to make | |
4200 | * our reservation. | |
4201 | */ | |
2bf64758 JB |
4202 | if (used <= space_info->total_bytes) { |
4203 | if (used + orig_bytes <= space_info->total_bytes) { | |
fb25e914 | 4204 | space_info->bytes_may_use += orig_bytes; |
8c2a3ca2 | 4205 | trace_btrfs_space_reservation(root->fs_info, |
2bcc0328 | 4206 | "space_info", space_info->flags, orig_bytes, 1); |
8bb8ab2e JB |
4207 | ret = 0; |
4208 | } else { | |
4209 | /* | |
4210 | * Ok set num_bytes to orig_bytes since we aren't | |
4211 | * overocmmitted, this way we only try and reclaim what | |
4212 | * we need. | |
4213 | */ | |
4214 | num_bytes = orig_bytes; | |
4215 | } | |
4216 | } else { | |
4217 | /* | |
4218 | * Ok we're over committed, set num_bytes to the overcommitted | |
4219 | * amount plus the amount of bytes that we need for this | |
4220 | * reservation. | |
4221 | */ | |
2bf64758 | 4222 | num_bytes = used - space_info->total_bytes + |
96c3f433 | 4223 | (orig_bytes * 2); |
8bb8ab2e | 4224 | } |
9ed74f2d | 4225 | |
44734ed1 JB |
4226 | if (ret && can_overcommit(root, space_info, orig_bytes, flush)) { |
4227 | space_info->bytes_may_use += orig_bytes; | |
4228 | trace_btrfs_space_reservation(root->fs_info, "space_info", | |
4229 | space_info->flags, orig_bytes, | |
4230 | 1); | |
4231 | ret = 0; | |
2bf64758 JB |
4232 | } |
4233 | ||
8bb8ab2e JB |
4234 | /* |
4235 | * Couldn't make our reservation, save our place so while we're trying | |
4236 | * to reclaim space we can actually use it instead of somebody else | |
4237 | * stealing it from us. | |
08e007d2 MX |
4238 | * |
4239 | * We make the other tasks wait for the flush only when we can flush | |
4240 | * all things. | |
8bb8ab2e | 4241 | */ |
72bcd99d | 4242 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
fdb5effd JB |
4243 | flushing = true; |
4244 | space_info->flush = 1; | |
8bb8ab2e | 4245 | } |
9ed74f2d | 4246 | |
f0486c68 | 4247 | spin_unlock(&space_info->lock); |
9ed74f2d | 4248 | |
08e007d2 | 4249 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
8bb8ab2e | 4250 | goto out; |
f0486c68 | 4251 | |
96c3f433 JB |
4252 | ret = flush_space(root, space_info, num_bytes, orig_bytes, |
4253 | flush_state); | |
4254 | flush_state++; | |
08e007d2 MX |
4255 | |
4256 | /* | |
4257 | * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock | |
4258 | * would happen. So skip delalloc flush. | |
4259 | */ | |
4260 | if (flush == BTRFS_RESERVE_FLUSH_LIMIT && | |
4261 | (flush_state == FLUSH_DELALLOC || | |
4262 | flush_state == FLUSH_DELALLOC_WAIT)) | |
4263 | flush_state = ALLOC_CHUNK; | |
4264 | ||
96c3f433 | 4265 | if (!ret) |
8bb8ab2e | 4266 | goto again; |
08e007d2 MX |
4267 | else if (flush == BTRFS_RESERVE_FLUSH_LIMIT && |
4268 | flush_state < COMMIT_TRANS) | |
4269 | goto again; | |
4270 | else if (flush == BTRFS_RESERVE_FLUSH_ALL && | |
4271 | flush_state <= COMMIT_TRANS) | |
8bb8ab2e JB |
4272 | goto again; |
4273 | ||
4274 | out: | |
5d80366e JB |
4275 | if (ret == -ENOSPC && |
4276 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
4277 | struct btrfs_block_rsv *global_rsv = | |
4278 | &root->fs_info->global_block_rsv; | |
4279 | ||
4280 | if (block_rsv != global_rsv && | |
4281 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
4282 | ret = 0; | |
4283 | } | |
fdb5effd | 4284 | if (flushing) { |
8bb8ab2e | 4285 | spin_lock(&space_info->lock); |
fdb5effd JB |
4286 | space_info->flush = 0; |
4287 | wake_up_all(&space_info->wait); | |
8bb8ab2e | 4288 | spin_unlock(&space_info->lock); |
f0486c68 | 4289 | } |
f0486c68 YZ |
4290 | return ret; |
4291 | } | |
4292 | ||
79787eaa JM |
4293 | static struct btrfs_block_rsv *get_block_rsv( |
4294 | const struct btrfs_trans_handle *trans, | |
4295 | const struct btrfs_root *root) | |
f0486c68 | 4296 | { |
4c13d758 JB |
4297 | struct btrfs_block_rsv *block_rsv = NULL; |
4298 | ||
0e721106 JB |
4299 | if (root->ref_cows) |
4300 | block_rsv = trans->block_rsv; | |
4301 | ||
4302 | if (root == root->fs_info->csum_root && trans->adding_csums) | |
f0486c68 | 4303 | block_rsv = trans->block_rsv; |
4c13d758 JB |
4304 | |
4305 | if (!block_rsv) | |
f0486c68 YZ |
4306 | block_rsv = root->block_rsv; |
4307 | ||
4308 | if (!block_rsv) | |
4309 | block_rsv = &root->fs_info->empty_block_rsv; | |
4310 | ||
4311 | return block_rsv; | |
4312 | } | |
4313 | ||
4314 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
4315 | u64 num_bytes) | |
4316 | { | |
4317 | int ret = -ENOSPC; | |
4318 | spin_lock(&block_rsv->lock); | |
4319 | if (block_rsv->reserved >= num_bytes) { | |
4320 | block_rsv->reserved -= num_bytes; | |
4321 | if (block_rsv->reserved < block_rsv->size) | |
4322 | block_rsv->full = 0; | |
4323 | ret = 0; | |
4324 | } | |
4325 | spin_unlock(&block_rsv->lock); | |
4326 | return ret; | |
4327 | } | |
4328 | ||
4329 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
4330 | u64 num_bytes, int update_size) | |
4331 | { | |
4332 | spin_lock(&block_rsv->lock); | |
4333 | block_rsv->reserved += num_bytes; | |
4334 | if (update_size) | |
4335 | block_rsv->size += num_bytes; | |
4336 | else if (block_rsv->reserved >= block_rsv->size) | |
4337 | block_rsv->full = 1; | |
4338 | spin_unlock(&block_rsv->lock); | |
4339 | } | |
4340 | ||
8c2a3ca2 JB |
4341 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
4342 | struct btrfs_block_rsv *block_rsv, | |
62a45b60 | 4343 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
4344 | { |
4345 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
4346 | ||
4347 | spin_lock(&block_rsv->lock); | |
4348 | if (num_bytes == (u64)-1) | |
4349 | num_bytes = block_rsv->size; | |
4350 | block_rsv->size -= num_bytes; | |
4351 | if (block_rsv->reserved >= block_rsv->size) { | |
4352 | num_bytes = block_rsv->reserved - block_rsv->size; | |
4353 | block_rsv->reserved = block_rsv->size; | |
4354 | block_rsv->full = 1; | |
4355 | } else { | |
4356 | num_bytes = 0; | |
4357 | } | |
4358 | spin_unlock(&block_rsv->lock); | |
4359 | ||
4360 | if (num_bytes > 0) { | |
4361 | if (dest) { | |
e9e22899 JB |
4362 | spin_lock(&dest->lock); |
4363 | if (!dest->full) { | |
4364 | u64 bytes_to_add; | |
4365 | ||
4366 | bytes_to_add = dest->size - dest->reserved; | |
4367 | bytes_to_add = min(num_bytes, bytes_to_add); | |
4368 | dest->reserved += bytes_to_add; | |
4369 | if (dest->reserved >= dest->size) | |
4370 | dest->full = 1; | |
4371 | num_bytes -= bytes_to_add; | |
4372 | } | |
4373 | spin_unlock(&dest->lock); | |
4374 | } | |
4375 | if (num_bytes) { | |
f0486c68 | 4376 | spin_lock(&space_info->lock); |
fb25e914 | 4377 | space_info->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4378 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4379 | space_info->flags, num_bytes, 0); |
36e39c40 | 4380 | space_info->reservation_progress++; |
f0486c68 | 4381 | spin_unlock(&space_info->lock); |
4e06bdd6 | 4382 | } |
9ed74f2d | 4383 | } |
f0486c68 | 4384 | } |
4e06bdd6 | 4385 | |
f0486c68 YZ |
4386 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
4387 | struct btrfs_block_rsv *dst, u64 num_bytes) | |
4388 | { | |
4389 | int ret; | |
9ed74f2d | 4390 | |
f0486c68 YZ |
4391 | ret = block_rsv_use_bytes(src, num_bytes); |
4392 | if (ret) | |
4393 | return ret; | |
9ed74f2d | 4394 | |
f0486c68 | 4395 | block_rsv_add_bytes(dst, num_bytes, 1); |
9ed74f2d JB |
4396 | return 0; |
4397 | } | |
4398 | ||
66d8f3dd | 4399 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 4400 | { |
f0486c68 YZ |
4401 | memset(rsv, 0, sizeof(*rsv)); |
4402 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 4403 | rsv->type = type; |
f0486c68 YZ |
4404 | } |
4405 | ||
66d8f3dd MX |
4406 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root, |
4407 | unsigned short type) | |
f0486c68 YZ |
4408 | { |
4409 | struct btrfs_block_rsv *block_rsv; | |
4410 | struct btrfs_fs_info *fs_info = root->fs_info; | |
9ed74f2d | 4411 | |
f0486c68 YZ |
4412 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
4413 | if (!block_rsv) | |
4414 | return NULL; | |
9ed74f2d | 4415 | |
66d8f3dd | 4416 | btrfs_init_block_rsv(block_rsv, type); |
f0486c68 YZ |
4417 | block_rsv->space_info = __find_space_info(fs_info, |
4418 | BTRFS_BLOCK_GROUP_METADATA); | |
f0486c68 YZ |
4419 | return block_rsv; |
4420 | } | |
9ed74f2d | 4421 | |
f0486c68 YZ |
4422 | void btrfs_free_block_rsv(struct btrfs_root *root, |
4423 | struct btrfs_block_rsv *rsv) | |
4424 | { | |
2aaa6655 JB |
4425 | if (!rsv) |
4426 | return; | |
dabdb640 JB |
4427 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
4428 | kfree(rsv); | |
9ed74f2d JB |
4429 | } |
4430 | ||
08e007d2 MX |
4431 | int btrfs_block_rsv_add(struct btrfs_root *root, |
4432 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
4433 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 4434 | { |
f0486c68 | 4435 | int ret; |
9ed74f2d | 4436 | |
f0486c68 YZ |
4437 | if (num_bytes == 0) |
4438 | return 0; | |
8bb8ab2e | 4439 | |
61b520a9 | 4440 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
4441 | if (!ret) { |
4442 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
4443 | return 0; | |
4444 | } | |
9ed74f2d | 4445 | |
f0486c68 | 4446 | return ret; |
f0486c68 | 4447 | } |
9ed74f2d | 4448 | |
4a92b1b8 | 4449 | int btrfs_block_rsv_check(struct btrfs_root *root, |
36ba022a | 4450 | struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
4451 | { |
4452 | u64 num_bytes = 0; | |
f0486c68 | 4453 | int ret = -ENOSPC; |
9ed74f2d | 4454 | |
f0486c68 YZ |
4455 | if (!block_rsv) |
4456 | return 0; | |
9ed74f2d | 4457 | |
f0486c68 | 4458 | spin_lock(&block_rsv->lock); |
36ba022a JB |
4459 | num_bytes = div_factor(block_rsv->size, min_factor); |
4460 | if (block_rsv->reserved >= num_bytes) | |
4461 | ret = 0; | |
4462 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 4463 | |
36ba022a JB |
4464 | return ret; |
4465 | } | |
4466 | ||
08e007d2 MX |
4467 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
4468 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
4469 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
4470 | { |
4471 | u64 num_bytes = 0; | |
4472 | int ret = -ENOSPC; | |
4473 | ||
4474 | if (!block_rsv) | |
4475 | return 0; | |
4476 | ||
4477 | spin_lock(&block_rsv->lock); | |
4478 | num_bytes = min_reserved; | |
13553e52 | 4479 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 4480 | ret = 0; |
13553e52 | 4481 | else |
f0486c68 | 4482 | num_bytes -= block_rsv->reserved; |
f0486c68 | 4483 | spin_unlock(&block_rsv->lock); |
13553e52 | 4484 | |
f0486c68 YZ |
4485 | if (!ret) |
4486 | return 0; | |
4487 | ||
aa38a711 | 4488 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
4489 | if (!ret) { |
4490 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 4491 | return 0; |
6a63209f | 4492 | } |
9ed74f2d | 4493 | |
13553e52 | 4494 | return ret; |
f0486c68 YZ |
4495 | } |
4496 | ||
4497 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, | |
4498 | struct btrfs_block_rsv *dst_rsv, | |
4499 | u64 num_bytes) | |
4500 | { | |
4501 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); | |
4502 | } | |
4503 | ||
4504 | void btrfs_block_rsv_release(struct btrfs_root *root, | |
4505 | struct btrfs_block_rsv *block_rsv, | |
4506 | u64 num_bytes) | |
4507 | { | |
4508 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; | |
4509 | if (global_rsv->full || global_rsv == block_rsv || | |
4510 | block_rsv->space_info != global_rsv->space_info) | |
4511 | global_rsv = NULL; | |
8c2a3ca2 JB |
4512 | block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv, |
4513 | num_bytes); | |
6a63209f JB |
4514 | } |
4515 | ||
4516 | /* | |
8929ecfa YZ |
4517 | * helper to calculate size of global block reservation. |
4518 | * the desired value is sum of space used by extent tree, | |
4519 | * checksum tree and root tree | |
6a63209f | 4520 | */ |
8929ecfa | 4521 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
6a63209f | 4522 | { |
8929ecfa YZ |
4523 | struct btrfs_space_info *sinfo; |
4524 | u64 num_bytes; | |
4525 | u64 meta_used; | |
4526 | u64 data_used; | |
6c41761f | 4527 | int csum_size = btrfs_super_csum_size(fs_info->super_copy); |
6a63209f | 4528 | |
8929ecfa YZ |
4529 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
4530 | spin_lock(&sinfo->lock); | |
4531 | data_used = sinfo->bytes_used; | |
4532 | spin_unlock(&sinfo->lock); | |
33b4d47f | 4533 | |
8929ecfa YZ |
4534 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
4535 | spin_lock(&sinfo->lock); | |
6d48755d JB |
4536 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
4537 | data_used = 0; | |
8929ecfa YZ |
4538 | meta_used = sinfo->bytes_used; |
4539 | spin_unlock(&sinfo->lock); | |
ab6e2410 | 4540 | |
8929ecfa YZ |
4541 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
4542 | csum_size * 2; | |
4543 | num_bytes += div64_u64(data_used + meta_used, 50); | |
4e06bdd6 | 4544 | |
8929ecfa | 4545 | if (num_bytes * 3 > meta_used) |
8e62c2de | 4546 | num_bytes = div64_u64(meta_used, 3); |
ab6e2410 | 4547 | |
8929ecfa YZ |
4548 | return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10); |
4549 | } | |
6a63209f | 4550 | |
8929ecfa YZ |
4551 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
4552 | { | |
4553 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
4554 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
4555 | u64 num_bytes; | |
6a63209f | 4556 | |
8929ecfa | 4557 | num_bytes = calc_global_metadata_size(fs_info); |
33b4d47f | 4558 | |
8929ecfa | 4559 | spin_lock(&sinfo->lock); |
1f699d38 | 4560 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 4561 | |
fdf30d1c | 4562 | block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024); |
4e06bdd6 | 4563 | |
8929ecfa | 4564 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
6d48755d JB |
4565 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
4566 | sinfo->bytes_may_use; | |
8929ecfa YZ |
4567 | |
4568 | if (sinfo->total_bytes > num_bytes) { | |
4569 | num_bytes = sinfo->total_bytes - num_bytes; | |
4570 | block_rsv->reserved += num_bytes; | |
fb25e914 | 4571 | sinfo->bytes_may_use += num_bytes; |
8c2a3ca2 | 4572 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4573 | sinfo->flags, num_bytes, 1); |
6a63209f | 4574 | } |
6a63209f | 4575 | |
8929ecfa YZ |
4576 | if (block_rsv->reserved >= block_rsv->size) { |
4577 | num_bytes = block_rsv->reserved - block_rsv->size; | |
fb25e914 | 4578 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 4579 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4580 | sinfo->flags, num_bytes, 0); |
36e39c40 | 4581 | sinfo->reservation_progress++; |
8929ecfa YZ |
4582 | block_rsv->reserved = block_rsv->size; |
4583 | block_rsv->full = 1; | |
4584 | } | |
182608c8 | 4585 | |
8929ecfa | 4586 | spin_unlock(&block_rsv->lock); |
1f699d38 | 4587 | spin_unlock(&sinfo->lock); |
6a63209f JB |
4588 | } |
4589 | ||
f0486c68 | 4590 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4591 | { |
f0486c68 | 4592 | struct btrfs_space_info *space_info; |
6a63209f | 4593 | |
f0486c68 YZ |
4594 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
4595 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 4596 | |
f0486c68 | 4597 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 4598 | fs_info->global_block_rsv.space_info = space_info; |
8929ecfa | 4599 | fs_info->delalloc_block_rsv.space_info = space_info; |
f0486c68 YZ |
4600 | fs_info->trans_block_rsv.space_info = space_info; |
4601 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 4602 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 4603 | |
8929ecfa YZ |
4604 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
4605 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
4606 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
4607 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 4608 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 4609 | |
8929ecfa | 4610 | update_global_block_rsv(fs_info); |
6a63209f JB |
4611 | } |
4612 | ||
8929ecfa | 4613 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 4614 | { |
8c2a3ca2 JB |
4615 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
4616 | (u64)-1); | |
8929ecfa YZ |
4617 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
4618 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); | |
4619 | WARN_ON(fs_info->trans_block_rsv.size > 0); | |
4620 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
4621 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
4622 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
4623 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
4624 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
4625 | } |
4626 | ||
a22285a6 YZ |
4627 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
4628 | struct btrfs_root *root) | |
6a63209f | 4629 | { |
0e721106 JB |
4630 | if (!trans->block_rsv) |
4631 | return; | |
4632 | ||
a22285a6 YZ |
4633 | if (!trans->bytes_reserved) |
4634 | return; | |
6a63209f | 4635 | |
e77266e4 | 4636 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
2bcc0328 | 4637 | trans->transid, trans->bytes_reserved, 0); |
b24e03db | 4638 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); |
a22285a6 YZ |
4639 | trans->bytes_reserved = 0; |
4640 | } | |
6a63209f | 4641 | |
79787eaa | 4642 | /* Can only return 0 or -ENOSPC */ |
d68fc57b YZ |
4643 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
4644 | struct inode *inode) | |
4645 | { | |
4646 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4647 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); | |
4648 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; | |
4649 | ||
4650 | /* | |
fcb80c2a JB |
4651 | * We need to hold space in order to delete our orphan item once we've |
4652 | * added it, so this takes the reservation so we can release it later | |
4653 | * when we are truly done with the orphan item. | |
d68fc57b | 4654 | */ |
ff5714cc | 4655 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4656 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4657 | btrfs_ino(inode), num_bytes, 1); | |
d68fc57b | 4658 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
6a63209f JB |
4659 | } |
4660 | ||
d68fc57b | 4661 | void btrfs_orphan_release_metadata(struct inode *inode) |
97e728d4 | 4662 | { |
d68fc57b | 4663 | struct btrfs_root *root = BTRFS_I(inode)->root; |
ff5714cc | 4664 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
8c2a3ca2 JB |
4665 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
4666 | btrfs_ino(inode), num_bytes, 0); | |
d68fc57b YZ |
4667 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
4668 | } | |
97e728d4 | 4669 | |
d5c12070 MX |
4670 | /* |
4671 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
4672 | * root: the root of the parent directory | |
4673 | * rsv: block reservation | |
4674 | * items: the number of items that we need do reservation | |
4675 | * qgroup_reserved: used to return the reserved size in qgroup | |
4676 | * | |
4677 | * This function is used to reserve the space for snapshot/subvolume | |
4678 | * creation and deletion. Those operations are different with the | |
4679 | * common file/directory operations, they change two fs/file trees | |
4680 | * and root tree, the number of items that the qgroup reserves is | |
4681 | * different with the free space reservation. So we can not use | |
4682 | * the space reseravtion mechanism in start_transaction(). | |
4683 | */ | |
4684 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
4685 | struct btrfs_block_rsv *rsv, | |
4686 | int items, | |
4687 | u64 *qgroup_reserved) | |
a22285a6 | 4688 | { |
d5c12070 MX |
4689 | u64 num_bytes; |
4690 | int ret; | |
4691 | ||
4692 | if (root->fs_info->quota_enabled) { | |
4693 | /* One for parent inode, two for dir entries */ | |
4694 | num_bytes = 3 * root->leafsize; | |
4695 | ret = btrfs_qgroup_reserve(root, num_bytes); | |
4696 | if (ret) | |
4697 | return ret; | |
4698 | } else { | |
4699 | num_bytes = 0; | |
4700 | } | |
4701 | ||
4702 | *qgroup_reserved = num_bytes; | |
4703 | ||
4704 | num_bytes = btrfs_calc_trans_metadata_size(root, items); | |
4705 | rsv->space_info = __find_space_info(root->fs_info, | |
4706 | BTRFS_BLOCK_GROUP_METADATA); | |
4707 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
4708 | BTRFS_RESERVE_FLUSH_ALL); | |
4709 | if (ret) { | |
4710 | if (*qgroup_reserved) | |
4711 | btrfs_qgroup_free(root, *qgroup_reserved); | |
4712 | } | |
4713 | ||
4714 | return ret; | |
4715 | } | |
4716 | ||
4717 | void btrfs_subvolume_release_metadata(struct btrfs_root *root, | |
4718 | struct btrfs_block_rsv *rsv, | |
4719 | u64 qgroup_reserved) | |
4720 | { | |
4721 | btrfs_block_rsv_release(root, rsv, (u64)-1); | |
4722 | if (qgroup_reserved) | |
4723 | btrfs_qgroup_free(root, qgroup_reserved); | |
97e728d4 JB |
4724 | } |
4725 | ||
7709cde3 JB |
4726 | /** |
4727 | * drop_outstanding_extent - drop an outstanding extent | |
4728 | * @inode: the inode we're dropping the extent for | |
4729 | * | |
4730 | * This is called when we are freeing up an outstanding extent, either called | |
4731 | * after an error or after an extent is written. This will return the number of | |
4732 | * reserved extents that need to be freed. This must be called with | |
4733 | * BTRFS_I(inode)->lock held. | |
4734 | */ | |
9e0baf60 JB |
4735 | static unsigned drop_outstanding_extent(struct inode *inode) |
4736 | { | |
7fd2ae21 | 4737 | unsigned drop_inode_space = 0; |
9e0baf60 JB |
4738 | unsigned dropped_extents = 0; |
4739 | ||
9e0baf60 JB |
4740 | BUG_ON(!BTRFS_I(inode)->outstanding_extents); |
4741 | BTRFS_I(inode)->outstanding_extents--; | |
4742 | ||
7fd2ae21 | 4743 | if (BTRFS_I(inode)->outstanding_extents == 0 && |
72ac3c0d JB |
4744 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4745 | &BTRFS_I(inode)->runtime_flags)) | |
7fd2ae21 | 4746 | drop_inode_space = 1; |
7fd2ae21 | 4747 | |
9e0baf60 JB |
4748 | /* |
4749 | * If we have more or the same amount of outsanding extents than we have | |
4750 | * reserved then we need to leave the reserved extents count alone. | |
4751 | */ | |
4752 | if (BTRFS_I(inode)->outstanding_extents >= | |
4753 | BTRFS_I(inode)->reserved_extents) | |
7fd2ae21 | 4754 | return drop_inode_space; |
9e0baf60 JB |
4755 | |
4756 | dropped_extents = BTRFS_I(inode)->reserved_extents - | |
4757 | BTRFS_I(inode)->outstanding_extents; | |
4758 | BTRFS_I(inode)->reserved_extents -= dropped_extents; | |
7fd2ae21 | 4759 | return dropped_extents + drop_inode_space; |
9e0baf60 JB |
4760 | } |
4761 | ||
7709cde3 JB |
4762 | /** |
4763 | * calc_csum_metadata_size - return the amount of metada space that must be | |
4764 | * reserved/free'd for the given bytes. | |
4765 | * @inode: the inode we're manipulating | |
4766 | * @num_bytes: the number of bytes in question | |
4767 | * @reserve: 1 if we are reserving space, 0 if we are freeing space | |
4768 | * | |
4769 | * This adjusts the number of csum_bytes in the inode and then returns the | |
4770 | * correct amount of metadata that must either be reserved or freed. We | |
4771 | * calculate how many checksums we can fit into one leaf and then divide the | |
4772 | * number of bytes that will need to be checksumed by this value to figure out | |
4773 | * how many checksums will be required. If we are adding bytes then the number | |
4774 | * may go up and we will return the number of additional bytes that must be | |
4775 | * reserved. If it is going down we will return the number of bytes that must | |
4776 | * be freed. | |
4777 | * | |
4778 | * This must be called with BTRFS_I(inode)->lock held. | |
4779 | */ | |
4780 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, | |
4781 | int reserve) | |
6324fbf3 | 4782 | { |
7709cde3 JB |
4783 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4784 | u64 csum_size; | |
4785 | int num_csums_per_leaf; | |
4786 | int num_csums; | |
4787 | int old_csums; | |
4788 | ||
4789 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && | |
4790 | BTRFS_I(inode)->csum_bytes == 0) | |
4791 | return 0; | |
4792 | ||
4793 | old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4794 | if (reserve) | |
4795 | BTRFS_I(inode)->csum_bytes += num_bytes; | |
4796 | else | |
4797 | BTRFS_I(inode)->csum_bytes -= num_bytes; | |
4798 | csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item); | |
4799 | num_csums_per_leaf = (int)div64_u64(csum_size, | |
4800 | sizeof(struct btrfs_csum_item) + | |
4801 | sizeof(struct btrfs_disk_key)); | |
4802 | num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); | |
4803 | num_csums = num_csums + num_csums_per_leaf - 1; | |
4804 | num_csums = num_csums / num_csums_per_leaf; | |
4805 | ||
4806 | old_csums = old_csums + num_csums_per_leaf - 1; | |
4807 | old_csums = old_csums / num_csums_per_leaf; | |
4808 | ||
4809 | /* No change, no need to reserve more */ | |
4810 | if (old_csums == num_csums) | |
4811 | return 0; | |
4812 | ||
4813 | if (reserve) | |
4814 | return btrfs_calc_trans_metadata_size(root, | |
4815 | num_csums - old_csums); | |
4816 | ||
4817 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); | |
0ca1f7ce | 4818 | } |
c146afad | 4819 | |
0ca1f7ce YZ |
4820 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
4821 | { | |
4822 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
4823 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; | |
9e0baf60 | 4824 | u64 to_reserve = 0; |
660d3f6c | 4825 | u64 csum_bytes; |
9e0baf60 | 4826 | unsigned nr_extents = 0; |
660d3f6c | 4827 | int extra_reserve = 0; |
08e007d2 | 4828 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 4829 | int ret = 0; |
c64c2bd8 | 4830 | bool delalloc_lock = true; |
88e081bf WS |
4831 | u64 to_free = 0; |
4832 | unsigned dropped; | |
6324fbf3 | 4833 | |
c64c2bd8 JB |
4834 | /* If we are a free space inode we need to not flush since we will be in |
4835 | * the middle of a transaction commit. We also don't need the delalloc | |
4836 | * mutex since we won't race with anybody. We need this mostly to make | |
4837 | * lockdep shut its filthy mouth. | |
4838 | */ | |
4839 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 4840 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 JB |
4841 | delalloc_lock = false; |
4842 | } | |
c09544e0 | 4843 | |
08e007d2 MX |
4844 | if (flush != BTRFS_RESERVE_NO_FLUSH && |
4845 | btrfs_transaction_in_commit(root->fs_info)) | |
0ca1f7ce | 4846 | schedule_timeout(1); |
ec44a35c | 4847 | |
c64c2bd8 JB |
4848 | if (delalloc_lock) |
4849 | mutex_lock(&BTRFS_I(inode)->delalloc_mutex); | |
4850 | ||
0ca1f7ce | 4851 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
8bb8ab2e | 4852 | |
9e0baf60 JB |
4853 | spin_lock(&BTRFS_I(inode)->lock); |
4854 | BTRFS_I(inode)->outstanding_extents++; | |
4855 | ||
4856 | if (BTRFS_I(inode)->outstanding_extents > | |
660d3f6c | 4857 | BTRFS_I(inode)->reserved_extents) |
9e0baf60 JB |
4858 | nr_extents = BTRFS_I(inode)->outstanding_extents - |
4859 | BTRFS_I(inode)->reserved_extents; | |
57a45ced | 4860 | |
7fd2ae21 JB |
4861 | /* |
4862 | * Add an item to reserve for updating the inode when we complete the | |
4863 | * delalloc io. | |
4864 | */ | |
72ac3c0d JB |
4865 | if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4866 | &BTRFS_I(inode)->runtime_flags)) { | |
7fd2ae21 | 4867 | nr_extents++; |
660d3f6c | 4868 | extra_reserve = 1; |
593060d7 | 4869 | } |
7fd2ae21 JB |
4870 | |
4871 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); | |
7709cde3 | 4872 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
660d3f6c | 4873 | csum_bytes = BTRFS_I(inode)->csum_bytes; |
9e0baf60 | 4874 | spin_unlock(&BTRFS_I(inode)->lock); |
57a45ced | 4875 | |
88e081bf | 4876 | if (root->fs_info->quota_enabled) { |
c5567237 AJ |
4877 | ret = btrfs_qgroup_reserve(root, num_bytes + |
4878 | nr_extents * root->leafsize); | |
88e081bf WS |
4879 | if (ret) |
4880 | goto out_fail; | |
4881 | } | |
c5567237 | 4882 | |
88e081bf WS |
4883 | ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush); |
4884 | if (unlikely(ret)) { | |
4885 | if (root->fs_info->quota_enabled) | |
4b5829a8 MX |
4886 | btrfs_qgroup_free(root, num_bytes + |
4887 | nr_extents * root->leafsize); | |
88e081bf | 4888 | goto out_fail; |
9e0baf60 | 4889 | } |
25179201 | 4890 | |
660d3f6c JB |
4891 | spin_lock(&BTRFS_I(inode)->lock); |
4892 | if (extra_reserve) { | |
72ac3c0d JB |
4893 | set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
4894 | &BTRFS_I(inode)->runtime_flags); | |
660d3f6c JB |
4895 | nr_extents--; |
4896 | } | |
4897 | BTRFS_I(inode)->reserved_extents += nr_extents; | |
4898 | spin_unlock(&BTRFS_I(inode)->lock); | |
c64c2bd8 JB |
4899 | |
4900 | if (delalloc_lock) | |
4901 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
660d3f6c | 4902 | |
8c2a3ca2 JB |
4903 | if (to_reserve) |
4904 | trace_btrfs_space_reservation(root->fs_info,"delalloc", | |
4905 | btrfs_ino(inode), to_reserve, 1); | |
0ca1f7ce YZ |
4906 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
4907 | ||
0ca1f7ce | 4908 | return 0; |
88e081bf WS |
4909 | |
4910 | out_fail: | |
4911 | spin_lock(&BTRFS_I(inode)->lock); | |
4912 | dropped = drop_outstanding_extent(inode); | |
4913 | /* | |
4914 | * If the inodes csum_bytes is the same as the original | |
4915 | * csum_bytes then we know we haven't raced with any free()ers | |
4916 | * so we can just reduce our inodes csum bytes and carry on. | |
88e081bf | 4917 | */ |
f4881bc7 | 4918 | if (BTRFS_I(inode)->csum_bytes == csum_bytes) { |
88e081bf | 4919 | calc_csum_metadata_size(inode, num_bytes, 0); |
f4881bc7 JB |
4920 | } else { |
4921 | u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes; | |
4922 | u64 bytes; | |
4923 | ||
4924 | /* | |
4925 | * This is tricky, but first we need to figure out how much we | |
4926 | * free'd from any free-ers that occured during this | |
4927 | * reservation, so we reset ->csum_bytes to the csum_bytes | |
4928 | * before we dropped our lock, and then call the free for the | |
4929 | * number of bytes that were freed while we were trying our | |
4930 | * reservation. | |
4931 | */ | |
4932 | bytes = csum_bytes - BTRFS_I(inode)->csum_bytes; | |
4933 | BTRFS_I(inode)->csum_bytes = csum_bytes; | |
4934 | to_free = calc_csum_metadata_size(inode, bytes, 0); | |
4935 | ||
4936 | ||
4937 | /* | |
4938 | * Now we need to see how much we would have freed had we not | |
4939 | * been making this reservation and our ->csum_bytes were not | |
4940 | * artificially inflated. | |
4941 | */ | |
4942 | BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes; | |
4943 | bytes = csum_bytes - orig_csum_bytes; | |
4944 | bytes = calc_csum_metadata_size(inode, bytes, 0); | |
4945 | ||
4946 | /* | |
4947 | * Now reset ->csum_bytes to what it should be. If bytes is | |
4948 | * more than to_free then we would have free'd more space had we | |
4949 | * not had an artificially high ->csum_bytes, so we need to free | |
4950 | * the remainder. If bytes is the same or less then we don't | |
4951 | * need to do anything, the other free-ers did the correct | |
4952 | * thing. | |
4953 | */ | |
4954 | BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes; | |
4955 | if (bytes > to_free) | |
4956 | to_free = bytes - to_free; | |
4957 | else | |
4958 | to_free = 0; | |
4959 | } | |
88e081bf WS |
4960 | spin_unlock(&BTRFS_I(inode)->lock); |
4961 | if (dropped) | |
4962 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
4963 | ||
4964 | if (to_free) { | |
4965 | btrfs_block_rsv_release(root, block_rsv, to_free); | |
4966 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
4967 | btrfs_ino(inode), to_free, 0); | |
4968 | } | |
4969 | if (delalloc_lock) | |
4970 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); | |
4971 | return ret; | |
0ca1f7ce YZ |
4972 | } |
4973 | ||
7709cde3 JB |
4974 | /** |
4975 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
4976 | * @inode: the inode to release the reservation for | |
4977 | * @num_bytes: the number of bytes we're releasing | |
4978 | * | |
4979 | * This will release the metadata reservation for an inode. This can be called | |
4980 | * once we complete IO for a given set of bytes to release their metadata | |
4981 | * reservations. | |
4982 | */ | |
0ca1f7ce YZ |
4983 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) |
4984 | { | |
4985 | struct btrfs_root *root = BTRFS_I(inode)->root; | |
9e0baf60 JB |
4986 | u64 to_free = 0; |
4987 | unsigned dropped; | |
0ca1f7ce YZ |
4988 | |
4989 | num_bytes = ALIGN(num_bytes, root->sectorsize); | |
7709cde3 | 4990 | spin_lock(&BTRFS_I(inode)->lock); |
9e0baf60 | 4991 | dropped = drop_outstanding_extent(inode); |
97e728d4 | 4992 | |
0934856d MX |
4993 | if (num_bytes) |
4994 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); | |
7709cde3 | 4995 | spin_unlock(&BTRFS_I(inode)->lock); |
9e0baf60 JB |
4996 | if (dropped > 0) |
4997 | to_free += btrfs_calc_trans_metadata_size(root, dropped); | |
0ca1f7ce | 4998 | |
8c2a3ca2 JB |
4999 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
5000 | btrfs_ino(inode), to_free, 0); | |
c5567237 AJ |
5001 | if (root->fs_info->quota_enabled) { |
5002 | btrfs_qgroup_free(root, num_bytes + | |
5003 | dropped * root->leafsize); | |
5004 | } | |
5005 | ||
0ca1f7ce YZ |
5006 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, |
5007 | to_free); | |
5008 | } | |
5009 | ||
7709cde3 JB |
5010 | /** |
5011 | * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc | |
5012 | * @inode: inode we're writing to | |
5013 | * @num_bytes: the number of bytes we want to allocate | |
5014 | * | |
5015 | * This will do the following things | |
5016 | * | |
5017 | * o reserve space in the data space info for num_bytes | |
5018 | * o reserve space in the metadata space info based on number of outstanding | |
5019 | * extents and how much csums will be needed | |
5020 | * o add to the inodes ->delalloc_bytes | |
5021 | * o add it to the fs_info's delalloc inodes list. | |
5022 | * | |
5023 | * This will return 0 for success and -ENOSPC if there is no space left. | |
5024 | */ | |
0ca1f7ce YZ |
5025 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes) |
5026 | { | |
5027 | int ret; | |
5028 | ||
5029 | ret = btrfs_check_data_free_space(inode, num_bytes); | |
d397712b | 5030 | if (ret) |
0ca1f7ce YZ |
5031 | return ret; |
5032 | ||
5033 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes); | |
5034 | if (ret) { | |
5035 | btrfs_free_reserved_data_space(inode, num_bytes); | |
5036 | return ret; | |
5037 | } | |
5038 | ||
5039 | return 0; | |
5040 | } | |
5041 | ||
7709cde3 JB |
5042 | /** |
5043 | * btrfs_delalloc_release_space - release data and metadata space for delalloc | |
5044 | * @inode: inode we're releasing space for | |
5045 | * @num_bytes: the number of bytes we want to free up | |
5046 | * | |
5047 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is | |
5048 | * called in the case that we don't need the metadata AND data reservations | |
5049 | * anymore. So if there is an error or we insert an inline extent. | |
5050 | * | |
5051 | * This function will release the metadata space that was not used and will | |
5052 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
5053 | * list if there are no delalloc bytes left. | |
5054 | */ | |
0ca1f7ce YZ |
5055 | void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes) |
5056 | { | |
5057 | btrfs_delalloc_release_metadata(inode, num_bytes); | |
5058 | btrfs_free_reserved_data_space(inode, num_bytes); | |
6324fbf3 CM |
5059 | } |
5060 | ||
c53d613e | 5061 | static int update_block_group(struct btrfs_root *root, |
f0486c68 | 5062 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 5063 | { |
0af3d00b | 5064 | struct btrfs_block_group_cache *cache = NULL; |
9078a3e1 | 5065 | struct btrfs_fs_info *info = root->fs_info; |
db94535d | 5066 | u64 total = num_bytes; |
9078a3e1 | 5067 | u64 old_val; |
db94535d | 5068 | u64 byte_in_group; |
0af3d00b | 5069 | int factor; |
3e1ad54f | 5070 | |
5d4f98a2 YZ |
5071 | /* block accounting for super block */ |
5072 | spin_lock(&info->delalloc_lock); | |
6c41761f | 5073 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
5074 | if (alloc) |
5075 | old_val += num_bytes; | |
5076 | else | |
5077 | old_val -= num_bytes; | |
6c41761f | 5078 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
5d4f98a2 YZ |
5079 | spin_unlock(&info->delalloc_lock); |
5080 | ||
d397712b | 5081 | while (total) { |
db94535d | 5082 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 5083 | if (!cache) |
79787eaa | 5084 | return -ENOENT; |
b742bb82 YZ |
5085 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
5086 | BTRFS_BLOCK_GROUP_RAID1 | | |
5087 | BTRFS_BLOCK_GROUP_RAID10)) | |
5088 | factor = 2; | |
5089 | else | |
5090 | factor = 1; | |
9d66e233 JB |
5091 | /* |
5092 | * If this block group has free space cache written out, we | |
5093 | * need to make sure to load it if we are removing space. This | |
5094 | * is because we need the unpinning stage to actually add the | |
5095 | * space back to the block group, otherwise we will leak space. | |
5096 | */ | |
5097 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 5098 | cache_block_group(cache, 1); |
0af3d00b | 5099 | |
db94535d CM |
5100 | byte_in_group = bytenr - cache->key.objectid; |
5101 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 5102 | |
25179201 | 5103 | spin_lock(&cache->space_info->lock); |
c286ac48 | 5104 | spin_lock(&cache->lock); |
0af3d00b | 5105 | |
73bc1876 | 5106 | if (btrfs_test_opt(root, SPACE_CACHE) && |
0af3d00b JB |
5107 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
5108 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
5109 | ||
0f9dd46c | 5110 | cache->dirty = 1; |
9078a3e1 | 5111 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 5112 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 5113 | if (alloc) { |
db94535d | 5114 | old_val += num_bytes; |
11833d66 YZ |
5115 | btrfs_set_block_group_used(&cache->item, old_val); |
5116 | cache->reserved -= num_bytes; | |
11833d66 | 5117 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
5118 | cache->space_info->bytes_used += num_bytes; |
5119 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 5120 | spin_unlock(&cache->lock); |
25179201 | 5121 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 5122 | } else { |
db94535d | 5123 | old_val -= num_bytes; |
c286ac48 | 5124 | btrfs_set_block_group_used(&cache->item, old_val); |
f0486c68 YZ |
5125 | cache->pinned += num_bytes; |
5126 | cache->space_info->bytes_pinned += num_bytes; | |
6324fbf3 | 5127 | cache->space_info->bytes_used -= num_bytes; |
b742bb82 | 5128 | cache->space_info->disk_used -= num_bytes * factor; |
c286ac48 | 5129 | spin_unlock(&cache->lock); |
25179201 | 5130 | spin_unlock(&cache->space_info->lock); |
1f3c79a2 | 5131 | |
f0486c68 YZ |
5132 | set_extent_dirty(info->pinned_extents, |
5133 | bytenr, bytenr + num_bytes - 1, | |
5134 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 5135 | } |
fa9c0d79 | 5136 | btrfs_put_block_group(cache); |
db94535d CM |
5137 | total -= num_bytes; |
5138 | bytenr += num_bytes; | |
9078a3e1 CM |
5139 | } |
5140 | return 0; | |
5141 | } | |
6324fbf3 | 5142 | |
a061fc8d CM |
5143 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
5144 | { | |
0f9dd46c | 5145 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 5146 | u64 bytenr; |
0f9dd46c | 5147 | |
a1897fdd LB |
5148 | spin_lock(&root->fs_info->block_group_cache_lock); |
5149 | bytenr = root->fs_info->first_logical_byte; | |
5150 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
5151 | ||
5152 | if (bytenr < (u64)-1) | |
5153 | return bytenr; | |
5154 | ||
0f9dd46c JB |
5155 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); |
5156 | if (!cache) | |
a061fc8d | 5157 | return 0; |
0f9dd46c | 5158 | |
d2fb3437 | 5159 | bytenr = cache->key.objectid; |
fa9c0d79 | 5160 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
5161 | |
5162 | return bytenr; | |
a061fc8d CM |
5163 | } |
5164 | ||
f0486c68 YZ |
5165 | static int pin_down_extent(struct btrfs_root *root, |
5166 | struct btrfs_block_group_cache *cache, | |
5167 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 5168 | { |
11833d66 YZ |
5169 | spin_lock(&cache->space_info->lock); |
5170 | spin_lock(&cache->lock); | |
5171 | cache->pinned += num_bytes; | |
5172 | cache->space_info->bytes_pinned += num_bytes; | |
5173 | if (reserved) { | |
5174 | cache->reserved -= num_bytes; | |
5175 | cache->space_info->bytes_reserved -= num_bytes; | |
5176 | } | |
5177 | spin_unlock(&cache->lock); | |
5178 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 5179 | |
f0486c68 YZ |
5180 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
5181 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); | |
5182 | return 0; | |
5183 | } | |
68b38550 | 5184 | |
f0486c68 YZ |
5185 | /* |
5186 | * this function must be called within transaction | |
5187 | */ | |
5188 | int btrfs_pin_extent(struct btrfs_root *root, | |
5189 | u64 bytenr, u64 num_bytes, int reserved) | |
5190 | { | |
5191 | struct btrfs_block_group_cache *cache; | |
68b38550 | 5192 | |
f0486c68 | 5193 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
79787eaa | 5194 | BUG_ON(!cache); /* Logic error */ |
f0486c68 YZ |
5195 | |
5196 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); | |
5197 | ||
5198 | btrfs_put_block_group(cache); | |
11833d66 YZ |
5199 | return 0; |
5200 | } | |
5201 | ||
f0486c68 | 5202 | /* |
e688b725 CM |
5203 | * this function must be called within transaction |
5204 | */ | |
dcfac415 | 5205 | int btrfs_pin_extent_for_log_replay(struct btrfs_root *root, |
e688b725 CM |
5206 | u64 bytenr, u64 num_bytes) |
5207 | { | |
5208 | struct btrfs_block_group_cache *cache; | |
5209 | ||
5210 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); | |
79787eaa | 5211 | BUG_ON(!cache); /* Logic error */ |
e688b725 CM |
5212 | |
5213 | /* | |
5214 | * pull in the free space cache (if any) so that our pin | |
5215 | * removes the free space from the cache. We have load_only set | |
5216 | * to one because the slow code to read in the free extents does check | |
5217 | * the pinned extents. | |
5218 | */ | |
f6373bf3 | 5219 | cache_block_group(cache, 1); |
e688b725 CM |
5220 | |
5221 | pin_down_extent(root, cache, bytenr, num_bytes, 0); | |
5222 | ||
5223 | /* remove us from the free space cache (if we're there at all) */ | |
5224 | btrfs_remove_free_space(cache, bytenr, num_bytes); | |
5225 | btrfs_put_block_group(cache); | |
5226 | return 0; | |
5227 | } | |
5228 | ||
fb25e914 JB |
5229 | /** |
5230 | * btrfs_update_reserved_bytes - update the block_group and space info counters | |
5231 | * @cache: The cache we are manipulating | |
5232 | * @num_bytes: The number of bytes in question | |
5233 | * @reserve: One of the reservation enums | |
5234 | * | |
5235 | * This is called by the allocator when it reserves space, or by somebody who is | |
5236 | * freeing space that was never actually used on disk. For example if you | |
5237 | * reserve some space for a new leaf in transaction A and before transaction A | |
5238 | * commits you free that leaf, you call this with reserve set to 0 in order to | |
5239 | * clear the reservation. | |
5240 | * | |
5241 | * Metadata reservations should be called with RESERVE_ALLOC so we do the proper | |
5242 | * ENOSPC accounting. For data we handle the reservation through clearing the | |
5243 | * delalloc bits in the io_tree. We have to do this since we could end up | |
5244 | * allocating less disk space for the amount of data we have reserved in the | |
5245 | * case of compression. | |
5246 | * | |
5247 | * If this is a reservation and the block group has become read only we cannot | |
5248 | * make the reservation and return -EAGAIN, otherwise this function always | |
5249 | * succeeds. | |
f0486c68 | 5250 | */ |
fb25e914 JB |
5251 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
5252 | u64 num_bytes, int reserve) | |
11833d66 | 5253 | { |
fb25e914 | 5254 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 5255 | int ret = 0; |
79787eaa | 5256 | |
fb25e914 JB |
5257 | spin_lock(&space_info->lock); |
5258 | spin_lock(&cache->lock); | |
5259 | if (reserve != RESERVE_FREE) { | |
f0486c68 YZ |
5260 | if (cache->ro) { |
5261 | ret = -EAGAIN; | |
5262 | } else { | |
fb25e914 JB |
5263 | cache->reserved += num_bytes; |
5264 | space_info->bytes_reserved += num_bytes; | |
5265 | if (reserve == RESERVE_ALLOC) { | |
8c2a3ca2 | 5266 | trace_btrfs_space_reservation(cache->fs_info, |
2bcc0328 LB |
5267 | "space_info", space_info->flags, |
5268 | num_bytes, 0); | |
fb25e914 JB |
5269 | space_info->bytes_may_use -= num_bytes; |
5270 | } | |
f0486c68 | 5271 | } |
fb25e914 JB |
5272 | } else { |
5273 | if (cache->ro) | |
5274 | space_info->bytes_readonly += num_bytes; | |
5275 | cache->reserved -= num_bytes; | |
5276 | space_info->bytes_reserved -= num_bytes; | |
5277 | space_info->reservation_progress++; | |
324ae4df | 5278 | } |
fb25e914 JB |
5279 | spin_unlock(&cache->lock); |
5280 | spin_unlock(&space_info->lock); | |
f0486c68 | 5281 | return ret; |
324ae4df | 5282 | } |
9078a3e1 | 5283 | |
143bede5 | 5284 | void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
11833d66 | 5285 | struct btrfs_root *root) |
e8569813 | 5286 | { |
e8569813 | 5287 | struct btrfs_fs_info *fs_info = root->fs_info; |
11833d66 YZ |
5288 | struct btrfs_caching_control *next; |
5289 | struct btrfs_caching_control *caching_ctl; | |
5290 | struct btrfs_block_group_cache *cache; | |
e8569813 | 5291 | |
11833d66 | 5292 | down_write(&fs_info->extent_commit_sem); |
25179201 | 5293 | |
11833d66 YZ |
5294 | list_for_each_entry_safe(caching_ctl, next, |
5295 | &fs_info->caching_block_groups, list) { | |
5296 | cache = caching_ctl->block_group; | |
5297 | if (block_group_cache_done(cache)) { | |
5298 | cache->last_byte_to_unpin = (u64)-1; | |
5299 | list_del_init(&caching_ctl->list); | |
5300 | put_caching_control(caching_ctl); | |
e8569813 | 5301 | } else { |
11833d66 | 5302 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 5303 | } |
e8569813 | 5304 | } |
11833d66 YZ |
5305 | |
5306 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
5307 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
5308 | else | |
5309 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
5310 | ||
5311 | up_write(&fs_info->extent_commit_sem); | |
8929ecfa YZ |
5312 | |
5313 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
5314 | } |
5315 | ||
11833d66 | 5316 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
ccd467d6 | 5317 | { |
11833d66 YZ |
5318 | struct btrfs_fs_info *fs_info = root->fs_info; |
5319 | struct btrfs_block_group_cache *cache = NULL; | |
7b398f8e JB |
5320 | struct btrfs_space_info *space_info; |
5321 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
11833d66 | 5322 | u64 len; |
7b398f8e | 5323 | bool readonly; |
ccd467d6 | 5324 | |
11833d66 | 5325 | while (start <= end) { |
7b398f8e | 5326 | readonly = false; |
11833d66 YZ |
5327 | if (!cache || |
5328 | start >= cache->key.objectid + cache->key.offset) { | |
5329 | if (cache) | |
5330 | btrfs_put_block_group(cache); | |
5331 | cache = btrfs_lookup_block_group(fs_info, start); | |
79787eaa | 5332 | BUG_ON(!cache); /* Logic error */ |
11833d66 YZ |
5333 | } |
5334 | ||
5335 | len = cache->key.objectid + cache->key.offset - start; | |
5336 | len = min(len, end + 1 - start); | |
5337 | ||
5338 | if (start < cache->last_byte_to_unpin) { | |
5339 | len = min(len, cache->last_byte_to_unpin - start); | |
5340 | btrfs_add_free_space(cache, start, len); | |
5341 | } | |
5342 | ||
f0486c68 | 5343 | start += len; |
7b398f8e | 5344 | space_info = cache->space_info; |
f0486c68 | 5345 | |
7b398f8e | 5346 | spin_lock(&space_info->lock); |
11833d66 YZ |
5347 | spin_lock(&cache->lock); |
5348 | cache->pinned -= len; | |
7b398f8e JB |
5349 | space_info->bytes_pinned -= len; |
5350 | if (cache->ro) { | |
5351 | space_info->bytes_readonly += len; | |
5352 | readonly = true; | |
5353 | } | |
11833d66 | 5354 | spin_unlock(&cache->lock); |
7b398f8e JB |
5355 | if (!readonly && global_rsv->space_info == space_info) { |
5356 | spin_lock(&global_rsv->lock); | |
5357 | if (!global_rsv->full) { | |
5358 | len = min(len, global_rsv->size - | |
5359 | global_rsv->reserved); | |
5360 | global_rsv->reserved += len; | |
5361 | space_info->bytes_may_use += len; | |
5362 | if (global_rsv->reserved >= global_rsv->size) | |
5363 | global_rsv->full = 1; | |
5364 | } | |
5365 | spin_unlock(&global_rsv->lock); | |
5366 | } | |
5367 | spin_unlock(&space_info->lock); | |
ccd467d6 | 5368 | } |
11833d66 YZ |
5369 | |
5370 | if (cache) | |
5371 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
5372 | return 0; |
5373 | } | |
5374 | ||
5375 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
11833d66 | 5376 | struct btrfs_root *root) |
a28ec197 | 5377 | { |
11833d66 YZ |
5378 | struct btrfs_fs_info *fs_info = root->fs_info; |
5379 | struct extent_io_tree *unpin; | |
1a5bc167 CM |
5380 | u64 start; |
5381 | u64 end; | |
a28ec197 | 5382 | int ret; |
a28ec197 | 5383 | |
79787eaa JM |
5384 | if (trans->aborted) |
5385 | return 0; | |
5386 | ||
11833d66 YZ |
5387 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
5388 | unpin = &fs_info->freed_extents[1]; | |
5389 | else | |
5390 | unpin = &fs_info->freed_extents[0]; | |
5391 | ||
d397712b | 5392 | while (1) { |
1a5bc167 | 5393 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 5394 | EXTENT_DIRTY, NULL); |
1a5bc167 | 5395 | if (ret) |
a28ec197 | 5396 | break; |
1f3c79a2 | 5397 | |
5378e607 LD |
5398 | if (btrfs_test_opt(root, DISCARD)) |
5399 | ret = btrfs_discard_extent(root, start, | |
5400 | end + 1 - start, NULL); | |
1f3c79a2 | 5401 | |
1a5bc167 | 5402 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
11833d66 | 5403 | unpin_extent_range(root, start, end); |
b9473439 | 5404 | cond_resched(); |
a28ec197 | 5405 | } |
817d52f8 | 5406 | |
e20d96d6 CM |
5407 | return 0; |
5408 | } | |
5409 | ||
5d4f98a2 YZ |
5410 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
5411 | struct btrfs_root *root, | |
5412 | u64 bytenr, u64 num_bytes, u64 parent, | |
5413 | u64 root_objectid, u64 owner_objectid, | |
5414 | u64 owner_offset, int refs_to_drop, | |
5415 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 5416 | { |
e2fa7227 | 5417 | struct btrfs_key key; |
5d4f98a2 | 5418 | struct btrfs_path *path; |
1261ec42 CM |
5419 | struct btrfs_fs_info *info = root->fs_info; |
5420 | struct btrfs_root *extent_root = info->extent_root; | |
5f39d397 | 5421 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
5422 | struct btrfs_extent_item *ei; |
5423 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 5424 | int ret; |
5d4f98a2 | 5425 | int is_data; |
952fccac CM |
5426 | int extent_slot = 0; |
5427 | int found_extent = 0; | |
5428 | int num_to_del = 1; | |
5d4f98a2 YZ |
5429 | u32 item_size; |
5430 | u64 refs; | |
3173a18f JB |
5431 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
5432 | SKINNY_METADATA); | |
037e6390 | 5433 | |
5caf2a00 | 5434 | path = btrfs_alloc_path(); |
54aa1f4d CM |
5435 | if (!path) |
5436 | return -ENOMEM; | |
5f26f772 | 5437 | |
3c12ac72 | 5438 | path->reada = 1; |
b9473439 | 5439 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5440 | |
5441 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
5442 | BUG_ON(!is_data && refs_to_drop != 1); | |
5443 | ||
3173a18f JB |
5444 | if (is_data) |
5445 | skinny_metadata = 0; | |
5446 | ||
5d4f98a2 YZ |
5447 | ret = lookup_extent_backref(trans, extent_root, path, &iref, |
5448 | bytenr, num_bytes, parent, | |
5449 | root_objectid, owner_objectid, | |
5450 | owner_offset); | |
7bb86316 | 5451 | if (ret == 0) { |
952fccac | 5452 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
5453 | while (extent_slot >= 0) { |
5454 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 5455 | extent_slot); |
5d4f98a2 | 5456 | if (key.objectid != bytenr) |
952fccac | 5457 | break; |
5d4f98a2 YZ |
5458 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
5459 | key.offset == num_bytes) { | |
952fccac CM |
5460 | found_extent = 1; |
5461 | break; | |
5462 | } | |
3173a18f JB |
5463 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
5464 | key.offset == owner_objectid) { | |
5465 | found_extent = 1; | |
5466 | break; | |
5467 | } | |
952fccac CM |
5468 | if (path->slots[0] - extent_slot > 5) |
5469 | break; | |
5d4f98a2 | 5470 | extent_slot--; |
952fccac | 5471 | } |
5d4f98a2 YZ |
5472 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
5473 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
5474 | if (found_extent && item_size < sizeof(*ei)) | |
5475 | found_extent = 0; | |
5476 | #endif | |
31840ae1 | 5477 | if (!found_extent) { |
5d4f98a2 | 5478 | BUG_ON(iref); |
56bec294 | 5479 | ret = remove_extent_backref(trans, extent_root, path, |
5d4f98a2 YZ |
5480 | NULL, refs_to_drop, |
5481 | is_data); | |
005d6427 DS |
5482 | if (ret) { |
5483 | btrfs_abort_transaction(trans, extent_root, ret); | |
5484 | goto out; | |
5485 | } | |
b3b4aa74 | 5486 | btrfs_release_path(path); |
b9473439 | 5487 | path->leave_spinning = 1; |
5d4f98a2 YZ |
5488 | |
5489 | key.objectid = bytenr; | |
5490 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5491 | key.offset = num_bytes; | |
5492 | ||
3173a18f JB |
5493 | if (!is_data && skinny_metadata) { |
5494 | key.type = BTRFS_METADATA_ITEM_KEY; | |
5495 | key.offset = owner_objectid; | |
5496 | } | |
5497 | ||
31840ae1 ZY |
5498 | ret = btrfs_search_slot(trans, extent_root, |
5499 | &key, path, -1, 1); | |
3173a18f JB |
5500 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
5501 | /* | |
5502 | * Couldn't find our skinny metadata item, | |
5503 | * see if we have ye olde extent item. | |
5504 | */ | |
5505 | path->slots[0]--; | |
5506 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
5507 | path->slots[0]); | |
5508 | if (key.objectid == bytenr && | |
5509 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
5510 | key.offset == num_bytes) | |
5511 | ret = 0; | |
5512 | } | |
5513 | ||
5514 | if (ret > 0 && skinny_metadata) { | |
5515 | skinny_metadata = false; | |
5516 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5517 | key.offset = num_bytes; | |
5518 | btrfs_release_path(path); | |
5519 | ret = btrfs_search_slot(trans, extent_root, | |
5520 | &key, path, -1, 1); | |
5521 | } | |
5522 | ||
f3465ca4 | 5523 | if (ret) { |
c2cf52eb SK |
5524 | btrfs_err(info, "umm, got %d back from search, was looking for %llu", |
5525 | ret, (unsigned long long)bytenr); | |
b783e62d JB |
5526 | if (ret > 0) |
5527 | btrfs_print_leaf(extent_root, | |
5528 | path->nodes[0]); | |
f3465ca4 | 5529 | } |
005d6427 DS |
5530 | if (ret < 0) { |
5531 | btrfs_abort_transaction(trans, extent_root, ret); | |
5532 | goto out; | |
5533 | } | |
31840ae1 ZY |
5534 | extent_slot = path->slots[0]; |
5535 | } | |
79787eaa | 5536 | } else if (ret == -ENOENT) { |
7bb86316 CM |
5537 | btrfs_print_leaf(extent_root, path->nodes[0]); |
5538 | WARN_ON(1); | |
c2cf52eb SK |
5539 | btrfs_err(info, |
5540 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
5541 | (unsigned long long)bytenr, | |
5542 | (unsigned long long)parent, | |
5543 | (unsigned long long)root_objectid, | |
5544 | (unsigned long long)owner_objectid, | |
5545 | (unsigned long long)owner_offset); | |
79787eaa | 5546 | } else { |
005d6427 DS |
5547 | btrfs_abort_transaction(trans, extent_root, ret); |
5548 | goto out; | |
7bb86316 | 5549 | } |
5f39d397 CM |
5550 | |
5551 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
5552 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
5553 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
5554 | if (item_size < sizeof(*ei)) { | |
5555 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
5556 | ret = convert_extent_item_v0(trans, extent_root, path, | |
5557 | owner_objectid, 0); | |
005d6427 DS |
5558 | if (ret < 0) { |
5559 | btrfs_abort_transaction(trans, extent_root, ret); | |
5560 | goto out; | |
5561 | } | |
5d4f98a2 | 5562 | |
b3b4aa74 | 5563 | btrfs_release_path(path); |
5d4f98a2 YZ |
5564 | path->leave_spinning = 1; |
5565 | ||
5566 | key.objectid = bytenr; | |
5567 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
5568 | key.offset = num_bytes; | |
5569 | ||
5570 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
5571 | -1, 1); | |
5572 | if (ret) { | |
c2cf52eb SK |
5573 | btrfs_err(info, "umm, got %d back from search, was looking for %llu", |
5574 | ret, (unsigned long long)bytenr); | |
5d4f98a2 YZ |
5575 | btrfs_print_leaf(extent_root, path->nodes[0]); |
5576 | } | |
005d6427 DS |
5577 | if (ret < 0) { |
5578 | btrfs_abort_transaction(trans, extent_root, ret); | |
5579 | goto out; | |
5580 | } | |
5581 | ||
5d4f98a2 YZ |
5582 | extent_slot = path->slots[0]; |
5583 | leaf = path->nodes[0]; | |
5584 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
5585 | } | |
5586 | #endif | |
5587 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 5588 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 5589 | struct btrfs_extent_item); |
3173a18f JB |
5590 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
5591 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
5592 | struct btrfs_tree_block_info *bi; |
5593 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
5594 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
5595 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
5596 | } | |
56bec294 | 5597 | |
5d4f98a2 | 5598 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 JB |
5599 | if (refs < refs_to_drop) { |
5600 | btrfs_err(info, "trying to drop %d refs but we only have %Lu " | |
5601 | "for bytenr %Lu\n", refs_to_drop, refs, bytenr); | |
5602 | ret = -EINVAL; | |
5603 | btrfs_abort_transaction(trans, extent_root, ret); | |
5604 | goto out; | |
5605 | } | |
56bec294 | 5606 | refs -= refs_to_drop; |
5f39d397 | 5607 | |
5d4f98a2 YZ |
5608 | if (refs > 0) { |
5609 | if (extent_op) | |
5610 | __run_delayed_extent_op(extent_op, leaf, ei); | |
5611 | /* | |
5612 | * In the case of inline back ref, reference count will | |
5613 | * be updated by remove_extent_backref | |
952fccac | 5614 | */ |
5d4f98a2 YZ |
5615 | if (iref) { |
5616 | BUG_ON(!found_extent); | |
5617 | } else { | |
5618 | btrfs_set_extent_refs(leaf, ei, refs); | |
5619 | btrfs_mark_buffer_dirty(leaf); | |
5620 | } | |
5621 | if (found_extent) { | |
5622 | ret = remove_extent_backref(trans, extent_root, path, | |
5623 | iref, refs_to_drop, | |
5624 | is_data); | |
005d6427 DS |
5625 | if (ret) { |
5626 | btrfs_abort_transaction(trans, extent_root, ret); | |
5627 | goto out; | |
5628 | } | |
952fccac | 5629 | } |
5d4f98a2 | 5630 | } else { |
5d4f98a2 YZ |
5631 | if (found_extent) { |
5632 | BUG_ON(is_data && refs_to_drop != | |
5633 | extent_data_ref_count(root, path, iref)); | |
5634 | if (iref) { | |
5635 | BUG_ON(path->slots[0] != extent_slot); | |
5636 | } else { | |
5637 | BUG_ON(path->slots[0] != extent_slot + 1); | |
5638 | path->slots[0] = extent_slot; | |
5639 | num_to_del = 2; | |
5640 | } | |
78fae27e | 5641 | } |
b9473439 | 5642 | |
952fccac CM |
5643 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
5644 | num_to_del); | |
005d6427 DS |
5645 | if (ret) { |
5646 | btrfs_abort_transaction(trans, extent_root, ret); | |
5647 | goto out; | |
5648 | } | |
b3b4aa74 | 5649 | btrfs_release_path(path); |
21af804c | 5650 | |
5d4f98a2 | 5651 | if (is_data) { |
459931ec | 5652 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
005d6427 DS |
5653 | if (ret) { |
5654 | btrfs_abort_transaction(trans, extent_root, ret); | |
5655 | goto out; | |
5656 | } | |
459931ec CM |
5657 | } |
5658 | ||
c53d613e | 5659 | ret = update_block_group(root, bytenr, num_bytes, 0); |
005d6427 DS |
5660 | if (ret) { |
5661 | btrfs_abort_transaction(trans, extent_root, ret); | |
5662 | goto out; | |
5663 | } | |
a28ec197 | 5664 | } |
79787eaa | 5665 | out: |
5caf2a00 | 5666 | btrfs_free_path(path); |
a28ec197 CM |
5667 | return ret; |
5668 | } | |
5669 | ||
1887be66 | 5670 | /* |
f0486c68 | 5671 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
5672 | * delayed ref for that extent as well. This searches the delayed ref tree for |
5673 | * a given extent, and if there are no other delayed refs to be processed, it | |
5674 | * removes it from the tree. | |
5675 | */ | |
5676 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
5677 | struct btrfs_root *root, u64 bytenr) | |
5678 | { | |
5679 | struct btrfs_delayed_ref_head *head; | |
5680 | struct btrfs_delayed_ref_root *delayed_refs; | |
5681 | struct btrfs_delayed_ref_node *ref; | |
5682 | struct rb_node *node; | |
f0486c68 | 5683 | int ret = 0; |
1887be66 CM |
5684 | |
5685 | delayed_refs = &trans->transaction->delayed_refs; | |
5686 | spin_lock(&delayed_refs->lock); | |
5687 | head = btrfs_find_delayed_ref_head(trans, bytenr); | |
5688 | if (!head) | |
5689 | goto out; | |
5690 | ||
5691 | node = rb_prev(&head->node.rb_node); | |
5692 | if (!node) | |
5693 | goto out; | |
5694 | ||
5695 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); | |
5696 | ||
5697 | /* there are still entries for this ref, we can't drop it */ | |
5698 | if (ref->bytenr == bytenr) | |
5699 | goto out; | |
5700 | ||
5d4f98a2 YZ |
5701 | if (head->extent_op) { |
5702 | if (!head->must_insert_reserved) | |
5703 | goto out; | |
78a6184a | 5704 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
5705 | head->extent_op = NULL; |
5706 | } | |
5707 | ||
1887be66 CM |
5708 | /* |
5709 | * waiting for the lock here would deadlock. If someone else has it | |
5710 | * locked they are already in the process of dropping it anyway | |
5711 | */ | |
5712 | if (!mutex_trylock(&head->mutex)) | |
5713 | goto out; | |
5714 | ||
5715 | /* | |
5716 | * at this point we have a head with no other entries. Go | |
5717 | * ahead and process it. | |
5718 | */ | |
5719 | head->node.in_tree = 0; | |
5720 | rb_erase(&head->node.rb_node, &delayed_refs->root); | |
c3e69d58 | 5721 | |
1887be66 CM |
5722 | delayed_refs->num_entries--; |
5723 | ||
5724 | /* | |
5725 | * we don't take a ref on the node because we're removing it from the | |
5726 | * tree, so we just steal the ref the tree was holding. | |
5727 | */ | |
c3e69d58 CM |
5728 | delayed_refs->num_heads--; |
5729 | if (list_empty(&head->cluster)) | |
5730 | delayed_refs->num_heads_ready--; | |
5731 | ||
5732 | list_del_init(&head->cluster); | |
1887be66 CM |
5733 | spin_unlock(&delayed_refs->lock); |
5734 | ||
f0486c68 YZ |
5735 | BUG_ON(head->extent_op); |
5736 | if (head->must_insert_reserved) | |
5737 | ret = 1; | |
5738 | ||
5739 | mutex_unlock(&head->mutex); | |
1887be66 | 5740 | btrfs_put_delayed_ref(&head->node); |
f0486c68 | 5741 | return ret; |
1887be66 CM |
5742 | out: |
5743 | spin_unlock(&delayed_refs->lock); | |
5744 | return 0; | |
5745 | } | |
5746 | ||
f0486c68 YZ |
5747 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
5748 | struct btrfs_root *root, | |
5749 | struct extent_buffer *buf, | |
5581a51a | 5750 | u64 parent, int last_ref) |
f0486c68 | 5751 | { |
f0486c68 YZ |
5752 | struct btrfs_block_group_cache *cache = NULL; |
5753 | int ret; | |
5754 | ||
5755 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
66d7e7f0 AJ |
5756 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
5757 | buf->start, buf->len, | |
5758 | parent, root->root_key.objectid, | |
5759 | btrfs_header_level(buf), | |
5581a51a | 5760 | BTRFS_DROP_DELAYED_REF, NULL, 0); |
79787eaa | 5761 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 YZ |
5762 | } |
5763 | ||
5764 | if (!last_ref) | |
5765 | return; | |
5766 | ||
f0486c68 | 5767 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); |
f0486c68 YZ |
5768 | |
5769 | if (btrfs_header_generation(buf) == trans->transid) { | |
5770 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
5771 | ret = check_ref_cleanup(trans, root, buf->start); | |
5772 | if (!ret) | |
37be25bc | 5773 | goto out; |
f0486c68 YZ |
5774 | } |
5775 | ||
5776 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { | |
5777 | pin_down_extent(root, cache, buf->start, buf->len, 1); | |
37be25bc | 5778 | goto out; |
f0486c68 YZ |
5779 | } |
5780 | ||
5781 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
5782 | ||
5783 | btrfs_add_free_space(cache, buf->start, buf->len); | |
fb25e914 | 5784 | btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE); |
f0486c68 YZ |
5785 | } |
5786 | out: | |
a826d6dc JB |
5787 | /* |
5788 | * Deleting the buffer, clear the corrupt flag since it doesn't matter | |
5789 | * anymore. | |
5790 | */ | |
5791 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
f0486c68 YZ |
5792 | btrfs_put_block_group(cache); |
5793 | } | |
5794 | ||
79787eaa | 5795 | /* Can return -ENOMEM */ |
66d7e7f0 AJ |
5796 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
5797 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, | |
5798 | u64 owner, u64 offset, int for_cow) | |
925baedd CM |
5799 | { |
5800 | int ret; | |
66d7e7f0 | 5801 | struct btrfs_fs_info *fs_info = root->fs_info; |
925baedd | 5802 | |
56bec294 CM |
5803 | /* |
5804 | * tree log blocks never actually go into the extent allocation | |
5805 | * tree, just update pinning info and exit early. | |
56bec294 | 5806 | */ |
5d4f98a2 YZ |
5807 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
5808 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 5809 | /* unlocks the pinned mutex */ |
11833d66 | 5810 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
56bec294 | 5811 | ret = 0; |
5d4f98a2 | 5812 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 AJ |
5813 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
5814 | num_bytes, | |
5d4f98a2 | 5815 | parent, root_objectid, (int)owner, |
66d7e7f0 | 5816 | BTRFS_DROP_DELAYED_REF, NULL, for_cow); |
5d4f98a2 | 5817 | } else { |
66d7e7f0 AJ |
5818 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
5819 | num_bytes, | |
5820 | parent, root_objectid, owner, | |
5821 | offset, BTRFS_DROP_DELAYED_REF, | |
5822 | NULL, for_cow); | |
56bec294 | 5823 | } |
925baedd CM |
5824 | return ret; |
5825 | } | |
5826 | ||
53b381b3 DW |
5827 | static u64 stripe_align(struct btrfs_root *root, |
5828 | struct btrfs_block_group_cache *cache, | |
5829 | u64 val, u64 num_bytes) | |
87ee04eb | 5830 | { |
fda2832f | 5831 | u64 ret = ALIGN(val, root->stripesize); |
87ee04eb CM |
5832 | return ret; |
5833 | } | |
5834 | ||
817d52f8 JB |
5835 | /* |
5836 | * when we wait for progress in the block group caching, its because | |
5837 | * our allocation attempt failed at least once. So, we must sleep | |
5838 | * and let some progress happen before we try again. | |
5839 | * | |
5840 | * This function will sleep at least once waiting for new free space to | |
5841 | * show up, and then it will check the block group free space numbers | |
5842 | * for our min num_bytes. Another option is to have it go ahead | |
5843 | * and look in the rbtree for a free extent of a given size, but this | |
5844 | * is a good start. | |
5845 | */ | |
5846 | static noinline int | |
5847 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, | |
5848 | u64 num_bytes) | |
5849 | { | |
11833d66 | 5850 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 5851 | |
11833d66 YZ |
5852 | caching_ctl = get_caching_control(cache); |
5853 | if (!caching_ctl) | |
817d52f8 | 5854 | return 0; |
817d52f8 | 5855 | |
11833d66 | 5856 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 5857 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
5858 | |
5859 | put_caching_control(caching_ctl); | |
5860 | return 0; | |
5861 | } | |
5862 | ||
5863 | static noinline int | |
5864 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
5865 | { | |
5866 | struct btrfs_caching_control *caching_ctl; | |
11833d66 YZ |
5867 | |
5868 | caching_ctl = get_caching_control(cache); | |
5869 | if (!caching_ctl) | |
5870 | return 0; | |
5871 | ||
5872 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
5873 | ||
5874 | put_caching_control(caching_ctl); | |
817d52f8 JB |
5875 | return 0; |
5876 | } | |
5877 | ||
31e50229 | 5878 | int __get_raid_index(u64 flags) |
b742bb82 | 5879 | { |
7738a53a | 5880 | if (flags & BTRFS_BLOCK_GROUP_RAID10) |
e6ec716f | 5881 | return BTRFS_RAID_RAID10; |
7738a53a | 5882 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
e6ec716f | 5883 | return BTRFS_RAID_RAID1; |
7738a53a | 5884 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
e6ec716f | 5885 | return BTRFS_RAID_DUP; |
7738a53a | 5886 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
e6ec716f | 5887 | return BTRFS_RAID_RAID0; |
53b381b3 | 5888 | else if (flags & BTRFS_BLOCK_GROUP_RAID5) |
e942f883 | 5889 | return BTRFS_RAID_RAID5; |
53b381b3 | 5890 | else if (flags & BTRFS_BLOCK_GROUP_RAID6) |
e942f883 | 5891 | return BTRFS_RAID_RAID6; |
7738a53a | 5892 | |
e942f883 | 5893 | return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ |
b742bb82 YZ |
5894 | } |
5895 | ||
7738a53a ID |
5896 | static int get_block_group_index(struct btrfs_block_group_cache *cache) |
5897 | { | |
31e50229 | 5898 | return __get_raid_index(cache->flags); |
7738a53a ID |
5899 | } |
5900 | ||
817d52f8 | 5901 | enum btrfs_loop_type { |
285ff5af JB |
5902 | LOOP_CACHING_NOWAIT = 0, |
5903 | LOOP_CACHING_WAIT = 1, | |
5904 | LOOP_ALLOC_CHUNK = 2, | |
5905 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
5906 | }; |
5907 | ||
fec577fb CM |
5908 | /* |
5909 | * walks the btree of allocated extents and find a hole of a given size. | |
5910 | * The key ins is changed to record the hole: | |
5911 | * ins->objectid == block start | |
62e2749e | 5912 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
fec577fb CM |
5913 | * ins->offset == number of blocks |
5914 | * Any available blocks before search_start are skipped. | |
5915 | */ | |
d397712b | 5916 | static noinline int find_free_extent(struct btrfs_trans_handle *trans, |
98ed5174 CM |
5917 | struct btrfs_root *orig_root, |
5918 | u64 num_bytes, u64 empty_size, | |
98ed5174 | 5919 | u64 hint_byte, struct btrfs_key *ins, |
e0f54067 | 5920 | u64 data) |
fec577fb | 5921 | { |
80eb234a | 5922 | int ret = 0; |
d397712b | 5923 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
fa9c0d79 | 5924 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 5925 | struct btrfs_block_group_cache *block_group = NULL; |
274bd4fb | 5926 | struct btrfs_block_group_cache *used_block_group; |
81c9ad23 | 5927 | u64 search_start = 0; |
239b14b3 | 5928 | int empty_cluster = 2 * 1024 * 1024; |
80eb234a | 5929 | struct btrfs_space_info *space_info; |
fa9c0d79 | 5930 | int loop = 0; |
ac5c9300 | 5931 | int index = __get_raid_index(data); |
fb25e914 JB |
5932 | int alloc_type = (data & BTRFS_BLOCK_GROUP_DATA) ? |
5933 | RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC; | |
817d52f8 | 5934 | bool found_uncached_bg = false; |
0a24325e | 5935 | bool failed_cluster_refill = false; |
1cdda9b8 | 5936 | bool failed_alloc = false; |
67377734 | 5937 | bool use_cluster = true; |
60d2adbb | 5938 | bool have_caching_bg = false; |
fec577fb | 5939 | |
db94535d | 5940 | WARN_ON(num_bytes < root->sectorsize); |
b1a4d965 | 5941 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
80eb234a JB |
5942 | ins->objectid = 0; |
5943 | ins->offset = 0; | |
b1a4d965 | 5944 | |
3f7de037 JB |
5945 | trace_find_free_extent(orig_root, num_bytes, empty_size, data); |
5946 | ||
2552d17e | 5947 | space_info = __find_space_info(root->fs_info, data); |
1b1d1f66 | 5948 | if (!space_info) { |
c2cf52eb | 5949 | btrfs_err(root->fs_info, "No space info for %llu", data); |
1b1d1f66 JB |
5950 | return -ENOSPC; |
5951 | } | |
2552d17e | 5952 | |
67377734 JB |
5953 | /* |
5954 | * If the space info is for both data and metadata it means we have a | |
5955 | * small filesystem and we can't use the clustering stuff. | |
5956 | */ | |
5957 | if (btrfs_mixed_space_info(space_info)) | |
5958 | use_cluster = false; | |
5959 | ||
67377734 | 5960 | if (data & BTRFS_BLOCK_GROUP_METADATA && use_cluster) { |
fa9c0d79 | 5961 | last_ptr = &root->fs_info->meta_alloc_cluster; |
536ac8ae CM |
5962 | if (!btrfs_test_opt(root, SSD)) |
5963 | empty_cluster = 64 * 1024; | |
239b14b3 CM |
5964 | } |
5965 | ||
67377734 JB |
5966 | if ((data & BTRFS_BLOCK_GROUP_DATA) && use_cluster && |
5967 | btrfs_test_opt(root, SSD)) { | |
fa9c0d79 CM |
5968 | last_ptr = &root->fs_info->data_alloc_cluster; |
5969 | } | |
0f9dd46c | 5970 | |
239b14b3 | 5971 | if (last_ptr) { |
fa9c0d79 CM |
5972 | spin_lock(&last_ptr->lock); |
5973 | if (last_ptr->block_group) | |
5974 | hint_byte = last_ptr->window_start; | |
5975 | spin_unlock(&last_ptr->lock); | |
239b14b3 | 5976 | } |
fa9c0d79 | 5977 | |
a061fc8d | 5978 | search_start = max(search_start, first_logical_byte(root, 0)); |
239b14b3 | 5979 | search_start = max(search_start, hint_byte); |
0b86a832 | 5980 | |
817d52f8 | 5981 | if (!last_ptr) |
fa9c0d79 | 5982 | empty_cluster = 0; |
fa9c0d79 | 5983 | |
2552d17e | 5984 | if (search_start == hint_byte) { |
2552d17e JB |
5985 | block_group = btrfs_lookup_block_group(root->fs_info, |
5986 | search_start); | |
274bd4fb | 5987 | used_block_group = block_group; |
817d52f8 JB |
5988 | /* |
5989 | * we don't want to use the block group if it doesn't match our | |
5990 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
5991 | * |
5992 | * However if we are re-searching with an ideal block group | |
5993 | * picked out then we don't care that the block group is cached. | |
817d52f8 JB |
5994 | */ |
5995 | if (block_group && block_group_bits(block_group, data) && | |
285ff5af | 5996 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 5997 | down_read(&space_info->groups_sem); |
44fb5511 CM |
5998 | if (list_empty(&block_group->list) || |
5999 | block_group->ro) { | |
6000 | /* | |
6001 | * someone is removing this block group, | |
6002 | * we can't jump into the have_block_group | |
6003 | * target because our list pointers are not | |
6004 | * valid | |
6005 | */ | |
6006 | btrfs_put_block_group(block_group); | |
6007 | up_read(&space_info->groups_sem); | |
ccf0e725 | 6008 | } else { |
b742bb82 | 6009 | index = get_block_group_index(block_group); |
44fb5511 | 6010 | goto have_block_group; |
ccf0e725 | 6011 | } |
2552d17e | 6012 | } else if (block_group) { |
fa9c0d79 | 6013 | btrfs_put_block_group(block_group); |
2552d17e | 6014 | } |
42e70e7a | 6015 | } |
2552d17e | 6016 | search: |
60d2adbb | 6017 | have_caching_bg = false; |
80eb234a | 6018 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
6019 | list_for_each_entry(block_group, &space_info->block_groups[index], |
6020 | list) { | |
6226cb0a | 6021 | u64 offset; |
817d52f8 | 6022 | int cached; |
8a1413a2 | 6023 | |
274bd4fb | 6024 | used_block_group = block_group; |
11dfe35a | 6025 | btrfs_get_block_group(block_group); |
2552d17e | 6026 | search_start = block_group->key.objectid; |
42e70e7a | 6027 | |
83a50de9 CM |
6028 | /* |
6029 | * this can happen if we end up cycling through all the | |
6030 | * raid types, but we want to make sure we only allocate | |
6031 | * for the proper type. | |
6032 | */ | |
6033 | if (!block_group_bits(block_group, data)) { | |
6034 | u64 extra = BTRFS_BLOCK_GROUP_DUP | | |
6035 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
6036 | BTRFS_BLOCK_GROUP_RAID5 | |
6037 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
6038 | BTRFS_BLOCK_GROUP_RAID10; |
6039 | ||
6040 | /* | |
6041 | * if they asked for extra copies and this block group | |
6042 | * doesn't provide them, bail. This does allow us to | |
6043 | * fill raid0 from raid1. | |
6044 | */ | |
6045 | if ((data & extra) && !(block_group->flags & extra)) | |
6046 | goto loop; | |
6047 | } | |
6048 | ||
2552d17e | 6049 | have_block_group: |
291c7d2f JB |
6050 | cached = block_group_cache_done(block_group); |
6051 | if (unlikely(!cached)) { | |
291c7d2f | 6052 | found_uncached_bg = true; |
f6373bf3 | 6053 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
6054 | BUG_ON(ret < 0); |
6055 | ret = 0; | |
817d52f8 JB |
6056 | } |
6057 | ||
ea6a478e | 6058 | if (unlikely(block_group->ro)) |
2552d17e | 6059 | goto loop; |
0f9dd46c | 6060 | |
0a24325e | 6061 | /* |
062c05c4 AO |
6062 | * Ok we want to try and use the cluster allocator, so |
6063 | * lets look there | |
0a24325e | 6064 | */ |
062c05c4 | 6065 | if (last_ptr) { |
8de972b4 | 6066 | unsigned long aligned_cluster; |
fa9c0d79 CM |
6067 | /* |
6068 | * the refill lock keeps out other | |
6069 | * people trying to start a new cluster | |
6070 | */ | |
6071 | spin_lock(&last_ptr->refill_lock); | |
274bd4fb AO |
6072 | used_block_group = last_ptr->block_group; |
6073 | if (used_block_group != block_group && | |
6074 | (!used_block_group || | |
6075 | used_block_group->ro || | |
6076 | !block_group_bits(used_block_group, data))) { | |
6077 | used_block_group = block_group; | |
44fb5511 | 6078 | goto refill_cluster; |
274bd4fb AO |
6079 | } |
6080 | ||
6081 | if (used_block_group != block_group) | |
6082 | btrfs_get_block_group(used_block_group); | |
44fb5511 | 6083 | |
274bd4fb AO |
6084 | offset = btrfs_alloc_from_cluster(used_block_group, |
6085 | last_ptr, num_bytes, used_block_group->key.objectid); | |
fa9c0d79 CM |
6086 | if (offset) { |
6087 | /* we have a block, we're done */ | |
6088 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
6089 | trace_btrfs_reserve_extent_cluster(root, |
6090 | block_group, search_start, num_bytes); | |
fa9c0d79 CM |
6091 | goto checks; |
6092 | } | |
6093 | ||
274bd4fb AO |
6094 | WARN_ON(last_ptr->block_group != used_block_group); |
6095 | if (used_block_group != block_group) { | |
6096 | btrfs_put_block_group(used_block_group); | |
6097 | used_block_group = block_group; | |
fa9c0d79 | 6098 | } |
44fb5511 | 6099 | refill_cluster: |
274bd4fb | 6100 | BUG_ON(used_block_group != block_group); |
062c05c4 AO |
6101 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
6102 | * set up a new clusters, so lets just skip it | |
6103 | * and let the allocator find whatever block | |
6104 | * it can find. If we reach this point, we | |
6105 | * will have tried the cluster allocator | |
6106 | * plenty of times and not have found | |
6107 | * anything, so we are likely way too | |
6108 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
6109 | * anything. |
6110 | * | |
6111 | * However, if the cluster is taken from the | |
6112 | * current block group, release the cluster | |
6113 | * first, so that we stand a better chance of | |
6114 | * succeeding in the unclustered | |
6115 | * allocation. */ | |
6116 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
6117 | last_ptr->block_group != block_group) { | |
062c05c4 AO |
6118 | spin_unlock(&last_ptr->refill_lock); |
6119 | goto unclustered_alloc; | |
6120 | } | |
6121 | ||
fa9c0d79 CM |
6122 | /* |
6123 | * this cluster didn't work out, free it and | |
6124 | * start over | |
6125 | */ | |
6126 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
6127 | ||
a5f6f719 AO |
6128 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
6129 | spin_unlock(&last_ptr->refill_lock); | |
6130 | goto unclustered_alloc; | |
6131 | } | |
6132 | ||
8de972b4 CM |
6133 | aligned_cluster = max_t(unsigned long, |
6134 | empty_cluster + empty_size, | |
6135 | block_group->full_stripe_len); | |
6136 | ||
fa9c0d79 | 6137 | /* allocate a cluster in this block group */ |
451d7585 | 6138 | ret = btrfs_find_space_cluster(trans, root, |
fa9c0d79 | 6139 | block_group, last_ptr, |
1b22bad7 | 6140 | search_start, num_bytes, |
8de972b4 | 6141 | aligned_cluster); |
fa9c0d79 CM |
6142 | if (ret == 0) { |
6143 | /* | |
6144 | * now pull our allocation out of this | |
6145 | * cluster | |
6146 | */ | |
6147 | offset = btrfs_alloc_from_cluster(block_group, | |
6148 | last_ptr, num_bytes, | |
6149 | search_start); | |
6150 | if (offset) { | |
6151 | /* we found one, proceed */ | |
6152 | spin_unlock(&last_ptr->refill_lock); | |
3f7de037 JB |
6153 | trace_btrfs_reserve_extent_cluster(root, |
6154 | block_group, search_start, | |
6155 | num_bytes); | |
fa9c0d79 CM |
6156 | goto checks; |
6157 | } | |
0a24325e JB |
6158 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
6159 | && !failed_cluster_refill) { | |
817d52f8 JB |
6160 | spin_unlock(&last_ptr->refill_lock); |
6161 | ||
0a24325e | 6162 | failed_cluster_refill = true; |
817d52f8 JB |
6163 | wait_block_group_cache_progress(block_group, |
6164 | num_bytes + empty_cluster + empty_size); | |
6165 | goto have_block_group; | |
fa9c0d79 | 6166 | } |
817d52f8 | 6167 | |
fa9c0d79 CM |
6168 | /* |
6169 | * at this point we either didn't find a cluster | |
6170 | * or we weren't able to allocate a block from our | |
6171 | * cluster. Free the cluster we've been trying | |
6172 | * to use, and go to the next block group | |
6173 | */ | |
0a24325e | 6174 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 6175 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 6176 | goto loop; |
fa9c0d79 CM |
6177 | } |
6178 | ||
062c05c4 | 6179 | unclustered_alloc: |
a5f6f719 AO |
6180 | spin_lock(&block_group->free_space_ctl->tree_lock); |
6181 | if (cached && | |
6182 | block_group->free_space_ctl->free_space < | |
6183 | num_bytes + empty_cluster + empty_size) { | |
6184 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
6185 | goto loop; | |
6186 | } | |
6187 | spin_unlock(&block_group->free_space_ctl->tree_lock); | |
6188 | ||
6226cb0a JB |
6189 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
6190 | num_bytes, empty_size); | |
1cdda9b8 JB |
6191 | /* |
6192 | * If we didn't find a chunk, and we haven't failed on this | |
6193 | * block group before, and this block group is in the middle of | |
6194 | * caching and we are ok with waiting, then go ahead and wait | |
6195 | * for progress to be made, and set failed_alloc to true. | |
6196 | * | |
6197 | * If failed_alloc is true then we've already waited on this | |
6198 | * block group once and should move on to the next block group. | |
6199 | */ | |
6200 | if (!offset && !failed_alloc && !cached && | |
6201 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 6202 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
6203 | num_bytes + empty_size); |
6204 | failed_alloc = true; | |
817d52f8 | 6205 | goto have_block_group; |
1cdda9b8 | 6206 | } else if (!offset) { |
60d2adbb MX |
6207 | if (!cached) |
6208 | have_caching_bg = true; | |
1cdda9b8 | 6209 | goto loop; |
817d52f8 | 6210 | } |
fa9c0d79 | 6211 | checks: |
53b381b3 DW |
6212 | search_start = stripe_align(root, used_block_group, |
6213 | offset, num_bytes); | |
25179201 | 6214 | |
2552d17e JB |
6215 | /* move on to the next group */ |
6216 | if (search_start + num_bytes > | |
274bd4fb AO |
6217 | used_block_group->key.objectid + used_block_group->key.offset) { |
6218 | btrfs_add_free_space(used_block_group, offset, num_bytes); | |
2552d17e | 6219 | goto loop; |
6226cb0a | 6220 | } |
f5a31e16 | 6221 | |
f0486c68 | 6222 | if (offset < search_start) |
274bd4fb | 6223 | btrfs_add_free_space(used_block_group, offset, |
f0486c68 YZ |
6224 | search_start - offset); |
6225 | BUG_ON(offset > search_start); | |
2552d17e | 6226 | |
274bd4fb | 6227 | ret = btrfs_update_reserved_bytes(used_block_group, num_bytes, |
fb25e914 | 6228 | alloc_type); |
f0486c68 | 6229 | if (ret == -EAGAIN) { |
274bd4fb | 6230 | btrfs_add_free_space(used_block_group, offset, num_bytes); |
2552d17e | 6231 | goto loop; |
0f9dd46c | 6232 | } |
0b86a832 | 6233 | |
f0486c68 | 6234 | /* we are all good, lets return */ |
2552d17e JB |
6235 | ins->objectid = search_start; |
6236 | ins->offset = num_bytes; | |
d2fb3437 | 6237 | |
3f7de037 JB |
6238 | trace_btrfs_reserve_extent(orig_root, block_group, |
6239 | search_start, num_bytes); | |
274bd4fb AO |
6240 | if (used_block_group != block_group) |
6241 | btrfs_put_block_group(used_block_group); | |
d82a6f1d | 6242 | btrfs_put_block_group(block_group); |
2552d17e JB |
6243 | break; |
6244 | loop: | |
0a24325e | 6245 | failed_cluster_refill = false; |
1cdda9b8 | 6246 | failed_alloc = false; |
b742bb82 | 6247 | BUG_ON(index != get_block_group_index(block_group)); |
274bd4fb AO |
6248 | if (used_block_group != block_group) |
6249 | btrfs_put_block_group(used_block_group); | |
fa9c0d79 | 6250 | btrfs_put_block_group(block_group); |
2552d17e JB |
6251 | } |
6252 | up_read(&space_info->groups_sem); | |
6253 | ||
60d2adbb MX |
6254 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
6255 | goto search; | |
6256 | ||
b742bb82 YZ |
6257 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
6258 | goto search; | |
6259 | ||
285ff5af | 6260 | /* |
ccf0e725 JB |
6261 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
6262 | * caching kthreads as we move along | |
817d52f8 JB |
6263 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
6264 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
6265 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
6266 | * again | |
fa9c0d79 | 6267 | */ |
723bda20 | 6268 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 6269 | index = 0; |
723bda20 | 6270 | loop++; |
817d52f8 | 6271 | if (loop == LOOP_ALLOC_CHUNK) { |
698d0082 | 6272 | ret = do_chunk_alloc(trans, root, data, |
ea658bad JB |
6273 | CHUNK_ALLOC_FORCE); |
6274 | /* | |
6275 | * Do not bail out on ENOSPC since we | |
6276 | * can do more things. | |
6277 | */ | |
6278 | if (ret < 0 && ret != -ENOSPC) { | |
6279 | btrfs_abort_transaction(trans, | |
6280 | root, ret); | |
6281 | goto out; | |
723bda20 | 6282 | } |
2552d17e JB |
6283 | } |
6284 | ||
723bda20 JB |
6285 | if (loop == LOOP_NO_EMPTY_SIZE) { |
6286 | empty_size = 0; | |
6287 | empty_cluster = 0; | |
fa9c0d79 | 6288 | } |
723bda20 JB |
6289 | |
6290 | goto search; | |
2552d17e JB |
6291 | } else if (!ins->objectid) { |
6292 | ret = -ENOSPC; | |
d82a6f1d | 6293 | } else if (ins->objectid) { |
80eb234a | 6294 | ret = 0; |
be744175 | 6295 | } |
79787eaa | 6296 | out: |
be744175 | 6297 | |
0f70abe2 | 6298 | return ret; |
fec577fb | 6299 | } |
ec44a35c | 6300 | |
9ed74f2d JB |
6301 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
6302 | int dump_block_groups) | |
0f9dd46c JB |
6303 | { |
6304 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 6305 | int index = 0; |
0f9dd46c | 6306 | |
9ed74f2d | 6307 | spin_lock(&info->lock); |
fb25e914 JB |
6308 | printk(KERN_INFO "space_info %llu has %llu free, is %sfull\n", |
6309 | (unsigned long long)info->flags, | |
d397712b | 6310 | (unsigned long long)(info->total_bytes - info->bytes_used - |
9ed74f2d | 6311 | info->bytes_pinned - info->bytes_reserved - |
8929ecfa | 6312 | info->bytes_readonly), |
d397712b | 6313 | (info->full) ? "" : "not "); |
8929ecfa YZ |
6314 | printk(KERN_INFO "space_info total=%llu, used=%llu, pinned=%llu, " |
6315 | "reserved=%llu, may_use=%llu, readonly=%llu\n", | |
21380931 | 6316 | (unsigned long long)info->total_bytes, |
8929ecfa | 6317 | (unsigned long long)info->bytes_used, |
21380931 | 6318 | (unsigned long long)info->bytes_pinned, |
8929ecfa | 6319 | (unsigned long long)info->bytes_reserved, |
21380931 | 6320 | (unsigned long long)info->bytes_may_use, |
8929ecfa | 6321 | (unsigned long long)info->bytes_readonly); |
9ed74f2d JB |
6322 | spin_unlock(&info->lock); |
6323 | ||
6324 | if (!dump_block_groups) | |
6325 | return; | |
0f9dd46c | 6326 | |
80eb234a | 6327 | down_read(&info->groups_sem); |
b742bb82 YZ |
6328 | again: |
6329 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 6330 | spin_lock(&cache->lock); |
799ffc3c | 6331 | printk(KERN_INFO "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s\n", |
d397712b CM |
6332 | (unsigned long long)cache->key.objectid, |
6333 | (unsigned long long)cache->key.offset, | |
6334 | (unsigned long long)btrfs_block_group_used(&cache->item), | |
6335 | (unsigned long long)cache->pinned, | |
799ffc3c LB |
6336 | (unsigned long long)cache->reserved, |
6337 | cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
6338 | btrfs_dump_free_space(cache, bytes); |
6339 | spin_unlock(&cache->lock); | |
6340 | } | |
b742bb82 YZ |
6341 | if (++index < BTRFS_NR_RAID_TYPES) |
6342 | goto again; | |
80eb234a | 6343 | up_read(&info->groups_sem); |
0f9dd46c | 6344 | } |
e8569813 | 6345 | |
11833d66 YZ |
6346 | int btrfs_reserve_extent(struct btrfs_trans_handle *trans, |
6347 | struct btrfs_root *root, | |
6348 | u64 num_bytes, u64 min_alloc_size, | |
6349 | u64 empty_size, u64 hint_byte, | |
81c9ad23 | 6350 | struct btrfs_key *ins, u64 data) |
fec577fb | 6351 | { |
9e622d6b | 6352 | bool final_tried = false; |
fec577fb | 6353 | int ret; |
925baedd | 6354 | |
6a63209f | 6355 | data = btrfs_get_alloc_profile(root, data); |
98d20f67 | 6356 | again: |
db94535d CM |
6357 | WARN_ON(num_bytes < root->sectorsize); |
6358 | ret = find_free_extent(trans, root, num_bytes, empty_size, | |
81c9ad23 | 6359 | hint_byte, ins, data); |
3b951516 | 6360 | |
9e622d6b MX |
6361 | if (ret == -ENOSPC) { |
6362 | if (!final_tried) { | |
6363 | num_bytes = num_bytes >> 1; | |
24542bf7 | 6364 | num_bytes = round_down(num_bytes, root->sectorsize); |
9e622d6b | 6365 | num_bytes = max(num_bytes, min_alloc_size); |
9e622d6b MX |
6366 | if (num_bytes == min_alloc_size) |
6367 | final_tried = true; | |
6368 | goto again; | |
6369 | } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) { | |
6370 | struct btrfs_space_info *sinfo; | |
6371 | ||
6372 | sinfo = __find_space_info(root->fs_info, data); | |
c2cf52eb SK |
6373 | btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu", |
6374 | (unsigned long long)data, | |
6375 | (unsigned long long)num_bytes); | |
53804280 JM |
6376 | if (sinfo) |
6377 | dump_space_info(sinfo, num_bytes, 1); | |
9e622d6b | 6378 | } |
925baedd | 6379 | } |
0f9dd46c | 6380 | |
1abe9b8a | 6381 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
6382 | ||
0f9dd46c | 6383 | return ret; |
e6dcd2dc CM |
6384 | } |
6385 | ||
e688b725 CM |
6386 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, |
6387 | u64 start, u64 len, int pin) | |
65b51a00 | 6388 | { |
0f9dd46c | 6389 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 6390 | int ret = 0; |
0f9dd46c | 6391 | |
0f9dd46c JB |
6392 | cache = btrfs_lookup_block_group(root->fs_info, start); |
6393 | if (!cache) { | |
c2cf52eb SK |
6394 | btrfs_err(root->fs_info, "Unable to find block group for %llu", |
6395 | (unsigned long long)start); | |
0f9dd46c JB |
6396 | return -ENOSPC; |
6397 | } | |
1f3c79a2 | 6398 | |
5378e607 LD |
6399 | if (btrfs_test_opt(root, DISCARD)) |
6400 | ret = btrfs_discard_extent(root, start, len, NULL); | |
1f3c79a2 | 6401 | |
e688b725 CM |
6402 | if (pin) |
6403 | pin_down_extent(root, cache, start, len, 1); | |
6404 | else { | |
6405 | btrfs_add_free_space(cache, start, len); | |
6406 | btrfs_update_reserved_bytes(cache, len, RESERVE_FREE); | |
6407 | } | |
fa9c0d79 | 6408 | btrfs_put_block_group(cache); |
817d52f8 | 6409 | |
1abe9b8a | 6410 | trace_btrfs_reserved_extent_free(root, start, len); |
6411 | ||
e6dcd2dc CM |
6412 | return ret; |
6413 | } | |
6414 | ||
e688b725 CM |
6415 | int btrfs_free_reserved_extent(struct btrfs_root *root, |
6416 | u64 start, u64 len) | |
6417 | { | |
6418 | return __btrfs_free_reserved_extent(root, start, len, 0); | |
6419 | } | |
6420 | ||
6421 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, | |
6422 | u64 start, u64 len) | |
6423 | { | |
6424 | return __btrfs_free_reserved_extent(root, start, len, 1); | |
6425 | } | |
6426 | ||
5d4f98a2 YZ |
6427 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
6428 | struct btrfs_root *root, | |
6429 | u64 parent, u64 root_objectid, | |
6430 | u64 flags, u64 owner, u64 offset, | |
6431 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
6432 | { |
6433 | int ret; | |
5d4f98a2 | 6434 | struct btrfs_fs_info *fs_info = root->fs_info; |
e6dcd2dc | 6435 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 6436 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 6437 | struct btrfs_path *path; |
5d4f98a2 YZ |
6438 | struct extent_buffer *leaf; |
6439 | int type; | |
6440 | u32 size; | |
26b8003f | 6441 | |
5d4f98a2 YZ |
6442 | if (parent > 0) |
6443 | type = BTRFS_SHARED_DATA_REF_KEY; | |
6444 | else | |
6445 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 6446 | |
5d4f98a2 | 6447 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
6448 | |
6449 | path = btrfs_alloc_path(); | |
db5b493a TI |
6450 | if (!path) |
6451 | return -ENOMEM; | |
47e4bb98 | 6452 | |
b9473439 | 6453 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6454 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
6455 | ins, size); | |
79787eaa JM |
6456 | if (ret) { |
6457 | btrfs_free_path(path); | |
6458 | return ret; | |
6459 | } | |
0f9dd46c | 6460 | |
5d4f98a2 YZ |
6461 | leaf = path->nodes[0]; |
6462 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 6463 | struct btrfs_extent_item); |
5d4f98a2 YZ |
6464 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
6465 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
6466 | btrfs_set_extent_flags(leaf, extent_item, | |
6467 | flags | BTRFS_EXTENT_FLAG_DATA); | |
6468 | ||
6469 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
6470 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
6471 | if (parent > 0) { | |
6472 | struct btrfs_shared_data_ref *ref; | |
6473 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
6474 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
6475 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
6476 | } else { | |
6477 | struct btrfs_extent_data_ref *ref; | |
6478 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
6479 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
6480 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
6481 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
6482 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
6483 | } | |
47e4bb98 CM |
6484 | |
6485 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 6486 | btrfs_free_path(path); |
f510cfec | 6487 | |
c53d613e | 6488 | ret = update_block_group(root, ins->objectid, ins->offset, 1); |
79787eaa | 6489 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb SK |
6490 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
6491 | (unsigned long long)ins->objectid, | |
6492 | (unsigned long long)ins->offset); | |
f5947066 CM |
6493 | BUG(); |
6494 | } | |
e6dcd2dc CM |
6495 | return ret; |
6496 | } | |
6497 | ||
5d4f98a2 YZ |
6498 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
6499 | struct btrfs_root *root, | |
6500 | u64 parent, u64 root_objectid, | |
6501 | u64 flags, struct btrfs_disk_key *key, | |
6502 | int level, struct btrfs_key *ins) | |
e6dcd2dc CM |
6503 | { |
6504 | int ret; | |
5d4f98a2 YZ |
6505 | struct btrfs_fs_info *fs_info = root->fs_info; |
6506 | struct btrfs_extent_item *extent_item; | |
6507 | struct btrfs_tree_block_info *block_info; | |
6508 | struct btrfs_extent_inline_ref *iref; | |
6509 | struct btrfs_path *path; | |
6510 | struct extent_buffer *leaf; | |
3173a18f JB |
6511 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
6512 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, | |
6513 | SKINNY_METADATA); | |
6514 | ||
6515 | if (!skinny_metadata) | |
6516 | size += sizeof(*block_info); | |
1c2308f8 | 6517 | |
5d4f98a2 | 6518 | path = btrfs_alloc_path(); |
d8926bb3 MF |
6519 | if (!path) |
6520 | return -ENOMEM; | |
56bec294 | 6521 | |
5d4f98a2 YZ |
6522 | path->leave_spinning = 1; |
6523 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
6524 | ins, size); | |
79787eaa JM |
6525 | if (ret) { |
6526 | btrfs_free_path(path); | |
6527 | return ret; | |
6528 | } | |
5d4f98a2 YZ |
6529 | |
6530 | leaf = path->nodes[0]; | |
6531 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
6532 | struct btrfs_extent_item); | |
6533 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
6534 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
6535 | btrfs_set_extent_flags(leaf, extent_item, | |
6536 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 6537 | |
3173a18f JB |
6538 | if (skinny_metadata) { |
6539 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
6540 | } else { | |
6541 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
6542 | btrfs_set_tree_block_key(leaf, block_info, key); | |
6543 | btrfs_set_tree_block_level(leaf, block_info, level); | |
6544 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
6545 | } | |
5d4f98a2 | 6546 | |
5d4f98a2 YZ |
6547 | if (parent > 0) { |
6548 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
6549 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6550 | BTRFS_SHARED_BLOCK_REF_KEY); | |
6551 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
6552 | } else { | |
6553 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
6554 | BTRFS_TREE_BLOCK_REF_KEY); | |
6555 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
6556 | } | |
6557 | ||
6558 | btrfs_mark_buffer_dirty(leaf); | |
6559 | btrfs_free_path(path); | |
6560 | ||
3173a18f | 6561 | ret = update_block_group(root, ins->objectid, root->leafsize, 1); |
79787eaa | 6562 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb SK |
6563 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
6564 | (unsigned long long)ins->objectid, | |
6565 | (unsigned long long)ins->offset); | |
5d4f98a2 YZ |
6566 | BUG(); |
6567 | } | |
6568 | return ret; | |
6569 | } | |
6570 | ||
6571 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
6572 | struct btrfs_root *root, | |
6573 | u64 root_objectid, u64 owner, | |
6574 | u64 offset, struct btrfs_key *ins) | |
6575 | { | |
6576 | int ret; | |
6577 | ||
6578 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
6579 | ||
66d7e7f0 AJ |
6580 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, |
6581 | ins->offset, 0, | |
6582 | root_objectid, owner, offset, | |
6583 | BTRFS_ADD_DELAYED_EXTENT, NULL, 0); | |
e6dcd2dc CM |
6584 | return ret; |
6585 | } | |
e02119d5 CM |
6586 | |
6587 | /* | |
6588 | * this is used by the tree logging recovery code. It records that | |
6589 | * an extent has been allocated and makes sure to clear the free | |
6590 | * space cache bits as well | |
6591 | */ | |
5d4f98a2 YZ |
6592 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
6593 | struct btrfs_root *root, | |
6594 | u64 root_objectid, u64 owner, u64 offset, | |
6595 | struct btrfs_key *ins) | |
e02119d5 CM |
6596 | { |
6597 | int ret; | |
6598 | struct btrfs_block_group_cache *block_group; | |
11833d66 YZ |
6599 | struct btrfs_caching_control *caching_ctl; |
6600 | u64 start = ins->objectid; | |
6601 | u64 num_bytes = ins->offset; | |
e02119d5 | 6602 | |
e02119d5 | 6603 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
f6373bf3 | 6604 | cache_block_group(block_group, 0); |
11833d66 | 6605 | caching_ctl = get_caching_control(block_group); |
e02119d5 | 6606 | |
11833d66 YZ |
6607 | if (!caching_ctl) { |
6608 | BUG_ON(!block_group_cache_done(block_group)); | |
6609 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
79787eaa | 6610 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6611 | } else { |
6612 | mutex_lock(&caching_ctl->mutex); | |
6613 | ||
6614 | if (start >= caching_ctl->progress) { | |
6615 | ret = add_excluded_extent(root, start, num_bytes); | |
79787eaa | 6616 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6617 | } else if (start + num_bytes <= caching_ctl->progress) { |
6618 | ret = btrfs_remove_free_space(block_group, | |
6619 | start, num_bytes); | |
79787eaa | 6620 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6621 | } else { |
6622 | num_bytes = caching_ctl->progress - start; | |
6623 | ret = btrfs_remove_free_space(block_group, | |
6624 | start, num_bytes); | |
79787eaa | 6625 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6626 | |
6627 | start = caching_ctl->progress; | |
6628 | num_bytes = ins->objectid + ins->offset - | |
6629 | caching_ctl->progress; | |
6630 | ret = add_excluded_extent(root, start, num_bytes); | |
79787eaa | 6631 | BUG_ON(ret); /* -ENOMEM */ |
11833d66 YZ |
6632 | } |
6633 | ||
6634 | mutex_unlock(&caching_ctl->mutex); | |
6635 | put_caching_control(caching_ctl); | |
6636 | } | |
6637 | ||
fb25e914 JB |
6638 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, |
6639 | RESERVE_ALLOC_NO_ACCOUNT); | |
79787eaa | 6640 | BUG_ON(ret); /* logic error */ |
fa9c0d79 | 6641 | btrfs_put_block_group(block_group); |
5d4f98a2 YZ |
6642 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
6643 | 0, owner, offset, ins, 1); | |
e02119d5 CM |
6644 | return ret; |
6645 | } | |
6646 | ||
65b51a00 CM |
6647 | struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans, |
6648 | struct btrfs_root *root, | |
4008c04a CM |
6649 | u64 bytenr, u32 blocksize, |
6650 | int level) | |
65b51a00 CM |
6651 | { |
6652 | struct extent_buffer *buf; | |
6653 | ||
6654 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
6655 | if (!buf) | |
6656 | return ERR_PTR(-ENOMEM); | |
6657 | btrfs_set_header_generation(buf, trans->transid); | |
85d4e461 | 6658 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 CM |
6659 | btrfs_tree_lock(buf); |
6660 | clean_tree_block(trans, root, buf); | |
3083ee2e | 6661 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
6662 | |
6663 | btrfs_set_lock_blocking(buf); | |
65b51a00 | 6664 | btrfs_set_buffer_uptodate(buf); |
b4ce94de | 6665 | |
d0c803c4 | 6666 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
8cef4e16 YZ |
6667 | /* |
6668 | * we allow two log transactions at a time, use different | |
6669 | * EXENT bit to differentiate dirty pages. | |
6670 | */ | |
6671 | if (root->log_transid % 2 == 0) | |
6672 | set_extent_dirty(&root->dirty_log_pages, buf->start, | |
6673 | buf->start + buf->len - 1, GFP_NOFS); | |
6674 | else | |
6675 | set_extent_new(&root->dirty_log_pages, buf->start, | |
6676 | buf->start + buf->len - 1, GFP_NOFS); | |
d0c803c4 CM |
6677 | } else { |
6678 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, | |
65b51a00 | 6679 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 6680 | } |
65b51a00 | 6681 | trans->blocks_used++; |
b4ce94de | 6682 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
6683 | return buf; |
6684 | } | |
6685 | ||
f0486c68 YZ |
6686 | static struct btrfs_block_rsv * |
6687 | use_block_rsv(struct btrfs_trans_handle *trans, | |
6688 | struct btrfs_root *root, u32 blocksize) | |
6689 | { | |
6690 | struct btrfs_block_rsv *block_rsv; | |
68a82277 | 6691 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
f0486c68 YZ |
6692 | int ret; |
6693 | ||
6694 | block_rsv = get_block_rsv(trans, root); | |
6695 | ||
6696 | if (block_rsv->size == 0) { | |
08e007d2 MX |
6697 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, |
6698 | BTRFS_RESERVE_NO_FLUSH); | |
68a82277 JB |
6699 | /* |
6700 | * If we couldn't reserve metadata bytes try and use some from | |
6701 | * the global reserve. | |
6702 | */ | |
6703 | if (ret && block_rsv != global_rsv) { | |
6704 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
6705 | if (!ret) | |
6706 | return global_rsv; | |
f0486c68 | 6707 | return ERR_PTR(ret); |
68a82277 | 6708 | } else if (ret) { |
f0486c68 | 6709 | return ERR_PTR(ret); |
68a82277 | 6710 | } |
f0486c68 YZ |
6711 | return block_rsv; |
6712 | } | |
6713 | ||
6714 | ret = block_rsv_use_bytes(block_rsv, blocksize); | |
6715 | if (!ret) | |
6716 | return block_rsv; | |
ca7e70f5 | 6717 | if (ret && !block_rsv->failfast) { |
b069e0c3 DS |
6718 | if (btrfs_test_opt(root, ENOSPC_DEBUG)) { |
6719 | static DEFINE_RATELIMIT_STATE(_rs, | |
6720 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
6721 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
6722 | if (__ratelimit(&_rs)) | |
6723 | WARN(1, KERN_DEBUG | |
6724 | "btrfs: block rsv returned %d\n", ret); | |
6725 | } | |
08e007d2 MX |
6726 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, |
6727 | BTRFS_RESERVE_NO_FLUSH); | |
68a82277 | 6728 | if (!ret) { |
68a82277 JB |
6729 | return block_rsv; |
6730 | } else if (ret && block_rsv != global_rsv) { | |
6731 | ret = block_rsv_use_bytes(global_rsv, blocksize); | |
6732 | if (!ret) | |
6733 | return global_rsv; | |
6734 | } | |
6735 | } | |
f0486c68 | 6736 | |
f0486c68 YZ |
6737 | return ERR_PTR(-ENOSPC); |
6738 | } | |
6739 | ||
8c2a3ca2 JB |
6740 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
6741 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
6742 | { |
6743 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 6744 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
6745 | } |
6746 | ||
fec577fb | 6747 | /* |
f0486c68 YZ |
6748 | * finds a free extent and does all the dirty work required for allocation |
6749 | * returns the key for the extent through ins, and a tree buffer for | |
6750 | * the first block of the extent through buf. | |
6751 | * | |
fec577fb CM |
6752 | * returns the tree buffer or NULL. |
6753 | */ | |
5f39d397 | 6754 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
6755 | struct btrfs_root *root, u32 blocksize, |
6756 | u64 parent, u64 root_objectid, | |
6757 | struct btrfs_disk_key *key, int level, | |
5581a51a | 6758 | u64 hint, u64 empty_size) |
fec577fb | 6759 | { |
e2fa7227 | 6760 | struct btrfs_key ins; |
f0486c68 | 6761 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 6762 | struct extent_buffer *buf; |
f0486c68 YZ |
6763 | u64 flags = 0; |
6764 | int ret; | |
3173a18f JB |
6765 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
6766 | SKINNY_METADATA); | |
fec577fb | 6767 | |
f0486c68 YZ |
6768 | block_rsv = use_block_rsv(trans, root, blocksize); |
6769 | if (IS_ERR(block_rsv)) | |
6770 | return ERR_CAST(block_rsv); | |
6771 | ||
6772 | ret = btrfs_reserve_extent(trans, root, blocksize, blocksize, | |
81c9ad23 | 6773 | empty_size, hint, &ins, 0); |
fec577fb | 6774 | if (ret) { |
8c2a3ca2 | 6775 | unuse_block_rsv(root->fs_info, block_rsv, blocksize); |
54aa1f4d | 6776 | return ERR_PTR(ret); |
fec577fb | 6777 | } |
55c69072 | 6778 | |
4008c04a CM |
6779 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
6780 | blocksize, level); | |
79787eaa | 6781 | BUG_ON(IS_ERR(buf)); /* -ENOMEM */ |
f0486c68 YZ |
6782 | |
6783 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
6784 | if (parent == 0) | |
6785 | parent = ins.objectid; | |
6786 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
6787 | } else | |
6788 | BUG_ON(parent > 0); | |
6789 | ||
6790 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
6791 | struct btrfs_delayed_extent_op *extent_op; | |
78a6184a | 6792 | extent_op = btrfs_alloc_delayed_extent_op(); |
79787eaa | 6793 | BUG_ON(!extent_op); /* -ENOMEM */ |
f0486c68 YZ |
6794 | if (key) |
6795 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
6796 | else | |
6797 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
6798 | extent_op->flags_to_set = flags; | |
3173a18f JB |
6799 | if (skinny_metadata) |
6800 | extent_op->update_key = 0; | |
6801 | else | |
6802 | extent_op->update_key = 1; | |
f0486c68 YZ |
6803 | extent_op->update_flags = 1; |
6804 | extent_op->is_data = 0; | |
6805 | ||
66d7e7f0 AJ |
6806 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
6807 | ins.objectid, | |
f0486c68 YZ |
6808 | ins.offset, parent, root_objectid, |
6809 | level, BTRFS_ADD_DELAYED_EXTENT, | |
5581a51a | 6810 | extent_op, 0); |
79787eaa | 6811 | BUG_ON(ret); /* -ENOMEM */ |
f0486c68 | 6812 | } |
fec577fb CM |
6813 | return buf; |
6814 | } | |
a28ec197 | 6815 | |
2c47e605 YZ |
6816 | struct walk_control { |
6817 | u64 refs[BTRFS_MAX_LEVEL]; | |
6818 | u64 flags[BTRFS_MAX_LEVEL]; | |
6819 | struct btrfs_key update_progress; | |
6820 | int stage; | |
6821 | int level; | |
6822 | int shared_level; | |
6823 | int update_ref; | |
6824 | int keep_locks; | |
1c4850e2 YZ |
6825 | int reada_slot; |
6826 | int reada_count; | |
66d7e7f0 | 6827 | int for_reloc; |
2c47e605 YZ |
6828 | }; |
6829 | ||
6830 | #define DROP_REFERENCE 1 | |
6831 | #define UPDATE_BACKREF 2 | |
6832 | ||
1c4850e2 YZ |
6833 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
6834 | struct btrfs_root *root, | |
6835 | struct walk_control *wc, | |
6836 | struct btrfs_path *path) | |
6407bf6d | 6837 | { |
1c4850e2 YZ |
6838 | u64 bytenr; |
6839 | u64 generation; | |
6840 | u64 refs; | |
94fcca9f | 6841 | u64 flags; |
5d4f98a2 | 6842 | u32 nritems; |
1c4850e2 YZ |
6843 | u32 blocksize; |
6844 | struct btrfs_key key; | |
6845 | struct extent_buffer *eb; | |
6407bf6d | 6846 | int ret; |
1c4850e2 YZ |
6847 | int slot; |
6848 | int nread = 0; | |
6407bf6d | 6849 | |
1c4850e2 YZ |
6850 | if (path->slots[wc->level] < wc->reada_slot) { |
6851 | wc->reada_count = wc->reada_count * 2 / 3; | |
6852 | wc->reada_count = max(wc->reada_count, 2); | |
6853 | } else { | |
6854 | wc->reada_count = wc->reada_count * 3 / 2; | |
6855 | wc->reada_count = min_t(int, wc->reada_count, | |
6856 | BTRFS_NODEPTRS_PER_BLOCK(root)); | |
6857 | } | |
7bb86316 | 6858 | |
1c4850e2 YZ |
6859 | eb = path->nodes[wc->level]; |
6860 | nritems = btrfs_header_nritems(eb); | |
6861 | blocksize = btrfs_level_size(root, wc->level - 1); | |
bd56b302 | 6862 | |
1c4850e2 YZ |
6863 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
6864 | if (nread >= wc->reada_count) | |
6865 | break; | |
bd56b302 | 6866 | |
2dd3e67b | 6867 | cond_resched(); |
1c4850e2 YZ |
6868 | bytenr = btrfs_node_blockptr(eb, slot); |
6869 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 6870 | |
1c4850e2 YZ |
6871 | if (slot == path->slots[wc->level]) |
6872 | goto reada; | |
5d4f98a2 | 6873 | |
1c4850e2 YZ |
6874 | if (wc->stage == UPDATE_BACKREF && |
6875 | generation <= root->root_key.offset) | |
bd56b302 CM |
6876 | continue; |
6877 | ||
94fcca9f | 6878 | /* We don't lock the tree block, it's OK to be racy here */ |
3173a18f JB |
6879 | ret = btrfs_lookup_extent_info(trans, root, bytenr, |
6880 | wc->level - 1, 1, &refs, | |
6881 | &flags); | |
79787eaa JM |
6882 | /* We don't care about errors in readahead. */ |
6883 | if (ret < 0) | |
6884 | continue; | |
94fcca9f YZ |
6885 | BUG_ON(refs == 0); |
6886 | ||
1c4850e2 | 6887 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
6888 | if (refs == 1) |
6889 | goto reada; | |
bd56b302 | 6890 | |
94fcca9f YZ |
6891 | if (wc->level == 1 && |
6892 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6893 | continue; | |
1c4850e2 YZ |
6894 | if (!wc->update_ref || |
6895 | generation <= root->root_key.offset) | |
6896 | continue; | |
6897 | btrfs_node_key_to_cpu(eb, &key, slot); | |
6898 | ret = btrfs_comp_cpu_keys(&key, | |
6899 | &wc->update_progress); | |
6900 | if (ret < 0) | |
6901 | continue; | |
94fcca9f YZ |
6902 | } else { |
6903 | if (wc->level == 1 && | |
6904 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
6905 | continue; | |
6407bf6d | 6906 | } |
1c4850e2 YZ |
6907 | reada: |
6908 | ret = readahead_tree_block(root, bytenr, blocksize, | |
6909 | generation); | |
6910 | if (ret) | |
bd56b302 | 6911 | break; |
1c4850e2 | 6912 | nread++; |
20524f02 | 6913 | } |
1c4850e2 | 6914 | wc->reada_slot = slot; |
20524f02 | 6915 | } |
2c47e605 | 6916 | |
f82d02d9 | 6917 | /* |
2c016dc2 | 6918 | * helper to process tree block while walking down the tree. |
2c47e605 | 6919 | * |
2c47e605 YZ |
6920 | * when wc->stage == UPDATE_BACKREF, this function updates |
6921 | * back refs for pointers in the block. | |
6922 | * | |
6923 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 6924 | */ |
2c47e605 | 6925 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 6926 | struct btrfs_root *root, |
2c47e605 | 6927 | struct btrfs_path *path, |
94fcca9f | 6928 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 6929 | { |
2c47e605 YZ |
6930 | int level = wc->level; |
6931 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 6932 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
6933 | int ret; |
6934 | ||
2c47e605 YZ |
6935 | if (wc->stage == UPDATE_BACKREF && |
6936 | btrfs_header_owner(eb) != root->root_key.objectid) | |
6937 | return 1; | |
f82d02d9 | 6938 | |
2c47e605 YZ |
6939 | /* |
6940 | * when reference count of tree block is 1, it won't increase | |
6941 | * again. once full backref flag is set, we never clear it. | |
6942 | */ | |
94fcca9f YZ |
6943 | if (lookup_info && |
6944 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
6945 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 YZ |
6946 | BUG_ON(!path->locks[level]); |
6947 | ret = btrfs_lookup_extent_info(trans, root, | |
3173a18f | 6948 | eb->start, level, 1, |
2c47e605 YZ |
6949 | &wc->refs[level], |
6950 | &wc->flags[level]); | |
79787eaa JM |
6951 | BUG_ON(ret == -ENOMEM); |
6952 | if (ret) | |
6953 | return ret; | |
2c47e605 YZ |
6954 | BUG_ON(wc->refs[level] == 0); |
6955 | } | |
5d4f98a2 | 6956 | |
2c47e605 YZ |
6957 | if (wc->stage == DROP_REFERENCE) { |
6958 | if (wc->refs[level] > 1) | |
6959 | return 1; | |
f82d02d9 | 6960 | |
2c47e605 | 6961 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 6962 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6963 | path->locks[level] = 0; |
6964 | } | |
6965 | return 0; | |
6966 | } | |
f82d02d9 | 6967 | |
2c47e605 YZ |
6968 | /* wc->stage == UPDATE_BACKREF */ |
6969 | if (!(wc->flags[level] & flag)) { | |
6970 | BUG_ON(!path->locks[level]); | |
66d7e7f0 | 6971 | ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc); |
79787eaa | 6972 | BUG_ON(ret); /* -ENOMEM */ |
66d7e7f0 | 6973 | ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc); |
79787eaa | 6974 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6975 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, |
6976 | eb->len, flag, 0); | |
79787eaa | 6977 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
6978 | wc->flags[level] |= flag; |
6979 | } | |
6980 | ||
6981 | /* | |
6982 | * the block is shared by multiple trees, so it's not good to | |
6983 | * keep the tree lock | |
6984 | */ | |
6985 | if (path->locks[level] && level > 0) { | |
bd681513 | 6986 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
6987 | path->locks[level] = 0; |
6988 | } | |
6989 | return 0; | |
6990 | } | |
6991 | ||
1c4850e2 | 6992 | /* |
2c016dc2 | 6993 | * helper to process tree block pointer. |
1c4850e2 YZ |
6994 | * |
6995 | * when wc->stage == DROP_REFERENCE, this function checks | |
6996 | * reference count of the block pointed to. if the block | |
6997 | * is shared and we need update back refs for the subtree | |
6998 | * rooted at the block, this function changes wc->stage to | |
6999 | * UPDATE_BACKREF. if the block is shared and there is no | |
7000 | * need to update back, this function drops the reference | |
7001 | * to the block. | |
7002 | * | |
7003 | * NOTE: return value 1 means we should stop walking down. | |
7004 | */ | |
7005 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
7006 | struct btrfs_root *root, | |
7007 | struct btrfs_path *path, | |
94fcca9f | 7008 | struct walk_control *wc, int *lookup_info) |
1c4850e2 YZ |
7009 | { |
7010 | u64 bytenr; | |
7011 | u64 generation; | |
7012 | u64 parent; | |
7013 | u32 blocksize; | |
7014 | struct btrfs_key key; | |
7015 | struct extent_buffer *next; | |
7016 | int level = wc->level; | |
7017 | int reada = 0; | |
7018 | int ret = 0; | |
7019 | ||
7020 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
7021 | path->slots[level]); | |
7022 | /* | |
7023 | * if the lower level block was created before the snapshot | |
7024 | * was created, we know there is no need to update back refs | |
7025 | * for the subtree | |
7026 | */ | |
7027 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
7028 | generation <= root->root_key.offset) { |
7029 | *lookup_info = 1; | |
1c4850e2 | 7030 | return 1; |
94fcca9f | 7031 | } |
1c4850e2 YZ |
7032 | |
7033 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
7034 | blocksize = btrfs_level_size(root, level - 1); | |
7035 | ||
7036 | next = btrfs_find_tree_block(root, bytenr, blocksize); | |
7037 | if (!next) { | |
7038 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); | |
90d2c51d MX |
7039 | if (!next) |
7040 | return -ENOMEM; | |
1c4850e2 YZ |
7041 | reada = 1; |
7042 | } | |
7043 | btrfs_tree_lock(next); | |
7044 | btrfs_set_lock_blocking(next); | |
7045 | ||
3173a18f | 7046 | ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1, |
94fcca9f YZ |
7047 | &wc->refs[level - 1], |
7048 | &wc->flags[level - 1]); | |
79787eaa JM |
7049 | if (ret < 0) { |
7050 | btrfs_tree_unlock(next); | |
7051 | return ret; | |
7052 | } | |
7053 | ||
c2cf52eb SK |
7054 | if (unlikely(wc->refs[level - 1] == 0)) { |
7055 | btrfs_err(root->fs_info, "Missing references."); | |
7056 | BUG(); | |
7057 | } | |
94fcca9f | 7058 | *lookup_info = 0; |
1c4850e2 | 7059 | |
94fcca9f | 7060 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 7061 | if (wc->refs[level - 1] > 1) { |
94fcca9f YZ |
7062 | if (level == 1 && |
7063 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
7064 | goto skip; | |
7065 | ||
1c4850e2 YZ |
7066 | if (!wc->update_ref || |
7067 | generation <= root->root_key.offset) | |
7068 | goto skip; | |
7069 | ||
7070 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
7071 | path->slots[level]); | |
7072 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
7073 | if (ret < 0) | |
7074 | goto skip; | |
7075 | ||
7076 | wc->stage = UPDATE_BACKREF; | |
7077 | wc->shared_level = level - 1; | |
7078 | } | |
94fcca9f YZ |
7079 | } else { |
7080 | if (level == 1 && | |
7081 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
7082 | goto skip; | |
1c4850e2 YZ |
7083 | } |
7084 | ||
b9fab919 | 7085 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
7086 | btrfs_tree_unlock(next); |
7087 | free_extent_buffer(next); | |
7088 | next = NULL; | |
94fcca9f | 7089 | *lookup_info = 1; |
1c4850e2 YZ |
7090 | } |
7091 | ||
7092 | if (!next) { | |
7093 | if (reada && level == 1) | |
7094 | reada_walk_down(trans, root, wc, path); | |
7095 | next = read_tree_block(root, bytenr, blocksize, generation); | |
416bc658 JB |
7096 | if (!next || !extent_buffer_uptodate(next)) { |
7097 | free_extent_buffer(next); | |
97d9a8a4 | 7098 | return -EIO; |
416bc658 | 7099 | } |
1c4850e2 YZ |
7100 | btrfs_tree_lock(next); |
7101 | btrfs_set_lock_blocking(next); | |
7102 | } | |
7103 | ||
7104 | level--; | |
7105 | BUG_ON(level != btrfs_header_level(next)); | |
7106 | path->nodes[level] = next; | |
7107 | path->slots[level] = 0; | |
bd681513 | 7108 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
7109 | wc->level = level; |
7110 | if (wc->level == 1) | |
7111 | wc->reada_slot = 0; | |
7112 | return 0; | |
7113 | skip: | |
7114 | wc->refs[level - 1] = 0; | |
7115 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
7116 | if (wc->stage == DROP_REFERENCE) { |
7117 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
7118 | parent = path->nodes[level]->start; | |
7119 | } else { | |
7120 | BUG_ON(root->root_key.objectid != | |
7121 | btrfs_header_owner(path->nodes[level])); | |
7122 | parent = 0; | |
7123 | } | |
1c4850e2 | 7124 | |
94fcca9f | 7125 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
66d7e7f0 | 7126 | root->root_key.objectid, level - 1, 0, 0); |
79787eaa | 7127 | BUG_ON(ret); /* -ENOMEM */ |
1c4850e2 | 7128 | } |
1c4850e2 YZ |
7129 | btrfs_tree_unlock(next); |
7130 | free_extent_buffer(next); | |
94fcca9f | 7131 | *lookup_info = 1; |
1c4850e2 YZ |
7132 | return 1; |
7133 | } | |
7134 | ||
2c47e605 | 7135 | /* |
2c016dc2 | 7136 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
7137 | * |
7138 | * when wc->stage == DROP_REFERENCE, this function drops | |
7139 | * reference count on the block. | |
7140 | * | |
7141 | * when wc->stage == UPDATE_BACKREF, this function changes | |
7142 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
7143 | * to UPDATE_BACKREF previously while processing the block. | |
7144 | * | |
7145 | * NOTE: return value 1 means we should stop walking up. | |
7146 | */ | |
7147 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
7148 | struct btrfs_root *root, | |
7149 | struct btrfs_path *path, | |
7150 | struct walk_control *wc) | |
7151 | { | |
f0486c68 | 7152 | int ret; |
2c47e605 YZ |
7153 | int level = wc->level; |
7154 | struct extent_buffer *eb = path->nodes[level]; | |
7155 | u64 parent = 0; | |
7156 | ||
7157 | if (wc->stage == UPDATE_BACKREF) { | |
7158 | BUG_ON(wc->shared_level < level); | |
7159 | if (level < wc->shared_level) | |
7160 | goto out; | |
7161 | ||
2c47e605 YZ |
7162 | ret = find_next_key(path, level + 1, &wc->update_progress); |
7163 | if (ret > 0) | |
7164 | wc->update_ref = 0; | |
7165 | ||
7166 | wc->stage = DROP_REFERENCE; | |
7167 | wc->shared_level = -1; | |
7168 | path->slots[level] = 0; | |
7169 | ||
7170 | /* | |
7171 | * check reference count again if the block isn't locked. | |
7172 | * we should start walking down the tree again if reference | |
7173 | * count is one. | |
7174 | */ | |
7175 | if (!path->locks[level]) { | |
7176 | BUG_ON(level == 0); | |
7177 | btrfs_tree_lock(eb); | |
7178 | btrfs_set_lock_blocking(eb); | |
bd681513 | 7179 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7180 | |
7181 | ret = btrfs_lookup_extent_info(trans, root, | |
3173a18f | 7182 | eb->start, level, 1, |
2c47e605 YZ |
7183 | &wc->refs[level], |
7184 | &wc->flags[level]); | |
79787eaa JM |
7185 | if (ret < 0) { |
7186 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 7187 | path->locks[level] = 0; |
79787eaa JM |
7188 | return ret; |
7189 | } | |
2c47e605 YZ |
7190 | BUG_ON(wc->refs[level] == 0); |
7191 | if (wc->refs[level] == 1) { | |
bd681513 | 7192 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 7193 | path->locks[level] = 0; |
2c47e605 YZ |
7194 | return 1; |
7195 | } | |
f82d02d9 | 7196 | } |
2c47e605 | 7197 | } |
f82d02d9 | 7198 | |
2c47e605 YZ |
7199 | /* wc->stage == DROP_REFERENCE */ |
7200 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 7201 | |
2c47e605 YZ |
7202 | if (wc->refs[level] == 1) { |
7203 | if (level == 0) { | |
7204 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
66d7e7f0 AJ |
7205 | ret = btrfs_dec_ref(trans, root, eb, 1, |
7206 | wc->for_reloc); | |
2c47e605 | 7207 | else |
66d7e7f0 AJ |
7208 | ret = btrfs_dec_ref(trans, root, eb, 0, |
7209 | wc->for_reloc); | |
79787eaa | 7210 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
7211 | } |
7212 | /* make block locked assertion in clean_tree_block happy */ | |
7213 | if (!path->locks[level] && | |
7214 | btrfs_header_generation(eb) == trans->transid) { | |
7215 | btrfs_tree_lock(eb); | |
7216 | btrfs_set_lock_blocking(eb); | |
bd681513 | 7217 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7218 | } |
7219 | clean_tree_block(trans, root, eb); | |
7220 | } | |
7221 | ||
7222 | if (eb == root->node) { | |
7223 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
7224 | parent = eb->start; | |
7225 | else | |
7226 | BUG_ON(root->root_key.objectid != | |
7227 | btrfs_header_owner(eb)); | |
7228 | } else { | |
7229 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
7230 | parent = path->nodes[level + 1]->start; | |
7231 | else | |
7232 | BUG_ON(root->root_key.objectid != | |
7233 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 7234 | } |
f82d02d9 | 7235 | |
5581a51a | 7236 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
7237 | out: |
7238 | wc->refs[level] = 0; | |
7239 | wc->flags[level] = 0; | |
f0486c68 | 7240 | return 0; |
2c47e605 YZ |
7241 | } |
7242 | ||
7243 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
7244 | struct btrfs_root *root, | |
7245 | struct btrfs_path *path, | |
7246 | struct walk_control *wc) | |
7247 | { | |
2c47e605 | 7248 | int level = wc->level; |
94fcca9f | 7249 | int lookup_info = 1; |
2c47e605 YZ |
7250 | int ret; |
7251 | ||
7252 | while (level >= 0) { | |
94fcca9f | 7253 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
7254 | if (ret > 0) |
7255 | break; | |
7256 | ||
7257 | if (level == 0) | |
7258 | break; | |
7259 | ||
7a7965f8 YZ |
7260 | if (path->slots[level] >= |
7261 | btrfs_header_nritems(path->nodes[level])) | |
7262 | break; | |
7263 | ||
94fcca9f | 7264 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
7265 | if (ret > 0) { |
7266 | path->slots[level]++; | |
7267 | continue; | |
90d2c51d MX |
7268 | } else if (ret < 0) |
7269 | return ret; | |
1c4850e2 | 7270 | level = wc->level; |
f82d02d9 | 7271 | } |
f82d02d9 YZ |
7272 | return 0; |
7273 | } | |
7274 | ||
d397712b | 7275 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 7276 | struct btrfs_root *root, |
f82d02d9 | 7277 | struct btrfs_path *path, |
2c47e605 | 7278 | struct walk_control *wc, int max_level) |
20524f02 | 7279 | { |
2c47e605 | 7280 | int level = wc->level; |
20524f02 | 7281 | int ret; |
9f3a7427 | 7282 | |
2c47e605 YZ |
7283 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
7284 | while (level < max_level && path->nodes[level]) { | |
7285 | wc->level = level; | |
7286 | if (path->slots[level] + 1 < | |
7287 | btrfs_header_nritems(path->nodes[level])) { | |
7288 | path->slots[level]++; | |
20524f02 CM |
7289 | return 0; |
7290 | } else { | |
2c47e605 YZ |
7291 | ret = walk_up_proc(trans, root, path, wc); |
7292 | if (ret > 0) | |
7293 | return 0; | |
bd56b302 | 7294 | |
2c47e605 | 7295 | if (path->locks[level]) { |
bd681513 CM |
7296 | btrfs_tree_unlock_rw(path->nodes[level], |
7297 | path->locks[level]); | |
2c47e605 | 7298 | path->locks[level] = 0; |
f82d02d9 | 7299 | } |
2c47e605 YZ |
7300 | free_extent_buffer(path->nodes[level]); |
7301 | path->nodes[level] = NULL; | |
7302 | level++; | |
20524f02 CM |
7303 | } |
7304 | } | |
7305 | return 1; | |
7306 | } | |
7307 | ||
9aca1d51 | 7308 | /* |
2c47e605 YZ |
7309 | * drop a subvolume tree. |
7310 | * | |
7311 | * this function traverses the tree freeing any blocks that only | |
7312 | * referenced by the tree. | |
7313 | * | |
7314 | * when a shared tree block is found. this function decreases its | |
7315 | * reference count by one. if update_ref is true, this function | |
7316 | * also make sure backrefs for the shared block and all lower level | |
7317 | * blocks are properly updated. | |
9d1a2a3a DS |
7318 | * |
7319 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 7320 | */ |
2c536799 | 7321 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
7322 | struct btrfs_block_rsv *block_rsv, int update_ref, |
7323 | int for_reloc) | |
20524f02 | 7324 | { |
5caf2a00 | 7325 | struct btrfs_path *path; |
2c47e605 YZ |
7326 | struct btrfs_trans_handle *trans; |
7327 | struct btrfs_root *tree_root = root->fs_info->tree_root; | |
9f3a7427 | 7328 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
7329 | struct walk_control *wc; |
7330 | struct btrfs_key key; | |
7331 | int err = 0; | |
7332 | int ret; | |
7333 | int level; | |
20524f02 | 7334 | |
5caf2a00 | 7335 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
7336 | if (!path) { |
7337 | err = -ENOMEM; | |
7338 | goto out; | |
7339 | } | |
20524f02 | 7340 | |
2c47e605 | 7341 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
7342 | if (!wc) { |
7343 | btrfs_free_path(path); | |
cb1b69f4 TI |
7344 | err = -ENOMEM; |
7345 | goto out; | |
38a1a919 | 7346 | } |
2c47e605 | 7347 | |
a22285a6 | 7348 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
7349 | if (IS_ERR(trans)) { |
7350 | err = PTR_ERR(trans); | |
7351 | goto out_free; | |
7352 | } | |
98d5dc13 | 7353 | |
3fd0a558 YZ |
7354 | if (block_rsv) |
7355 | trans->block_rsv = block_rsv; | |
2c47e605 | 7356 | |
9f3a7427 | 7357 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 7358 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
7359 | path->nodes[level] = btrfs_lock_root_node(root); |
7360 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 7361 | path->slots[level] = 0; |
bd681513 | 7362 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7363 | memset(&wc->update_progress, 0, |
7364 | sizeof(wc->update_progress)); | |
9f3a7427 | 7365 | } else { |
9f3a7427 | 7366 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
7367 | memcpy(&wc->update_progress, &key, |
7368 | sizeof(wc->update_progress)); | |
7369 | ||
6702ed49 | 7370 | level = root_item->drop_level; |
2c47e605 | 7371 | BUG_ON(level == 0); |
6702ed49 | 7372 | path->lowest_level = level; |
2c47e605 YZ |
7373 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
7374 | path->lowest_level = 0; | |
7375 | if (ret < 0) { | |
7376 | err = ret; | |
79787eaa | 7377 | goto out_end_trans; |
9f3a7427 | 7378 | } |
1c4850e2 | 7379 | WARN_ON(ret > 0); |
2c47e605 | 7380 | |
7d9eb12c CM |
7381 | /* |
7382 | * unlock our path, this is safe because only this | |
7383 | * function is allowed to delete this snapshot | |
7384 | */ | |
5d4f98a2 | 7385 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
7386 | |
7387 | level = btrfs_header_level(root->node); | |
7388 | while (1) { | |
7389 | btrfs_tree_lock(path->nodes[level]); | |
7390 | btrfs_set_lock_blocking(path->nodes[level]); | |
7391 | ||
7392 | ret = btrfs_lookup_extent_info(trans, root, | |
7393 | path->nodes[level]->start, | |
3173a18f | 7394 | level, 1, &wc->refs[level], |
2c47e605 | 7395 | &wc->flags[level]); |
79787eaa JM |
7396 | if (ret < 0) { |
7397 | err = ret; | |
7398 | goto out_end_trans; | |
7399 | } | |
2c47e605 YZ |
7400 | BUG_ON(wc->refs[level] == 0); |
7401 | ||
7402 | if (level == root_item->drop_level) | |
7403 | break; | |
7404 | ||
7405 | btrfs_tree_unlock(path->nodes[level]); | |
7406 | WARN_ON(wc->refs[level] != 1); | |
7407 | level--; | |
7408 | } | |
9f3a7427 | 7409 | } |
2c47e605 YZ |
7410 | |
7411 | wc->level = level; | |
7412 | wc->shared_level = -1; | |
7413 | wc->stage = DROP_REFERENCE; | |
7414 | wc->update_ref = update_ref; | |
7415 | wc->keep_locks = 0; | |
66d7e7f0 | 7416 | wc->for_reloc = for_reloc; |
1c4850e2 | 7417 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
2c47e605 | 7418 | |
d397712b | 7419 | while (1) { |
9d1a2a3a DS |
7420 | if (!for_reloc && btrfs_fs_closing(root->fs_info)) { |
7421 | pr_debug("btrfs: drop snapshot early exit\n"); | |
7422 | err = -EAGAIN; | |
7423 | goto out_end_trans; | |
7424 | } | |
7425 | ||
2c47e605 YZ |
7426 | ret = walk_down_tree(trans, root, path, wc); |
7427 | if (ret < 0) { | |
7428 | err = ret; | |
20524f02 | 7429 | break; |
2c47e605 | 7430 | } |
9aca1d51 | 7431 | |
2c47e605 YZ |
7432 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
7433 | if (ret < 0) { | |
7434 | err = ret; | |
20524f02 | 7435 | break; |
2c47e605 YZ |
7436 | } |
7437 | ||
7438 | if (ret > 0) { | |
7439 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
7440 | break; |
7441 | } | |
2c47e605 YZ |
7442 | |
7443 | if (wc->stage == DROP_REFERENCE) { | |
7444 | level = wc->level; | |
7445 | btrfs_node_key(path->nodes[level], | |
7446 | &root_item->drop_progress, | |
7447 | path->slots[level]); | |
7448 | root_item->drop_level = level; | |
7449 | } | |
7450 | ||
7451 | BUG_ON(wc->level == 0); | |
3fd0a558 | 7452 | if (btrfs_should_end_transaction(trans, tree_root)) { |
2c47e605 YZ |
7453 | ret = btrfs_update_root(trans, tree_root, |
7454 | &root->root_key, | |
7455 | root_item); | |
79787eaa JM |
7456 | if (ret) { |
7457 | btrfs_abort_transaction(trans, tree_root, ret); | |
7458 | err = ret; | |
7459 | goto out_end_trans; | |
7460 | } | |
2c47e605 | 7461 | |
3fd0a558 | 7462 | btrfs_end_transaction_throttle(trans, tree_root); |
a22285a6 | 7463 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
7464 | if (IS_ERR(trans)) { |
7465 | err = PTR_ERR(trans); | |
7466 | goto out_free; | |
7467 | } | |
3fd0a558 YZ |
7468 | if (block_rsv) |
7469 | trans->block_rsv = block_rsv; | |
c3e69d58 | 7470 | } |
20524f02 | 7471 | } |
b3b4aa74 | 7472 | btrfs_release_path(path); |
79787eaa JM |
7473 | if (err) |
7474 | goto out_end_trans; | |
2c47e605 YZ |
7475 | |
7476 | ret = btrfs_del_root(trans, tree_root, &root->root_key); | |
79787eaa JM |
7477 | if (ret) { |
7478 | btrfs_abort_transaction(trans, tree_root, ret); | |
7479 | goto out_end_trans; | |
7480 | } | |
2c47e605 | 7481 | |
76dda93c YZ |
7482 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
7483 | ret = btrfs_find_last_root(tree_root, root->root_key.objectid, | |
7484 | NULL, NULL); | |
79787eaa JM |
7485 | if (ret < 0) { |
7486 | btrfs_abort_transaction(trans, tree_root, ret); | |
7487 | err = ret; | |
7488 | goto out_end_trans; | |
7489 | } else if (ret > 0) { | |
84cd948c JB |
7490 | /* if we fail to delete the orphan item this time |
7491 | * around, it'll get picked up the next time. | |
7492 | * | |
7493 | * The most common failure here is just -ENOENT. | |
7494 | */ | |
7495 | btrfs_del_orphan_item(trans, tree_root, | |
7496 | root->root_key.objectid); | |
76dda93c YZ |
7497 | } |
7498 | } | |
7499 | ||
7500 | if (root->in_radix) { | |
7501 | btrfs_free_fs_root(tree_root->fs_info, root); | |
7502 | } else { | |
7503 | free_extent_buffer(root->node); | |
7504 | free_extent_buffer(root->commit_root); | |
7505 | kfree(root); | |
7506 | } | |
79787eaa | 7507 | out_end_trans: |
3fd0a558 | 7508 | btrfs_end_transaction_throttle(trans, tree_root); |
79787eaa | 7509 | out_free: |
2c47e605 | 7510 | kfree(wc); |
5caf2a00 | 7511 | btrfs_free_path(path); |
cb1b69f4 TI |
7512 | out: |
7513 | if (err) | |
7514 | btrfs_std_error(root->fs_info, err); | |
2c536799 | 7515 | return err; |
20524f02 | 7516 | } |
9078a3e1 | 7517 | |
2c47e605 YZ |
7518 | /* |
7519 | * drop subtree rooted at tree block 'node'. | |
7520 | * | |
7521 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 7522 | * only used by relocation code |
2c47e605 | 7523 | */ |
f82d02d9 YZ |
7524 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
7525 | struct btrfs_root *root, | |
7526 | struct extent_buffer *node, | |
7527 | struct extent_buffer *parent) | |
7528 | { | |
7529 | struct btrfs_path *path; | |
2c47e605 | 7530 | struct walk_control *wc; |
f82d02d9 YZ |
7531 | int level; |
7532 | int parent_level; | |
7533 | int ret = 0; | |
7534 | int wret; | |
7535 | ||
2c47e605 YZ |
7536 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
7537 | ||
f82d02d9 | 7538 | path = btrfs_alloc_path(); |
db5b493a TI |
7539 | if (!path) |
7540 | return -ENOMEM; | |
f82d02d9 | 7541 | |
2c47e605 | 7542 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
7543 | if (!wc) { |
7544 | btrfs_free_path(path); | |
7545 | return -ENOMEM; | |
7546 | } | |
2c47e605 | 7547 | |
b9447ef8 | 7548 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
7549 | parent_level = btrfs_header_level(parent); |
7550 | extent_buffer_get(parent); | |
7551 | path->nodes[parent_level] = parent; | |
7552 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
7553 | ||
b9447ef8 | 7554 | btrfs_assert_tree_locked(node); |
f82d02d9 | 7555 | level = btrfs_header_level(node); |
f82d02d9 YZ |
7556 | path->nodes[level] = node; |
7557 | path->slots[level] = 0; | |
bd681513 | 7558 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
7559 | |
7560 | wc->refs[parent_level] = 1; | |
7561 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
7562 | wc->level = level; | |
7563 | wc->shared_level = -1; | |
7564 | wc->stage = DROP_REFERENCE; | |
7565 | wc->update_ref = 0; | |
7566 | wc->keep_locks = 1; | |
66d7e7f0 | 7567 | wc->for_reloc = 1; |
1c4850e2 | 7568 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
f82d02d9 YZ |
7569 | |
7570 | while (1) { | |
2c47e605 YZ |
7571 | wret = walk_down_tree(trans, root, path, wc); |
7572 | if (wret < 0) { | |
f82d02d9 | 7573 | ret = wret; |
f82d02d9 | 7574 | break; |
2c47e605 | 7575 | } |
f82d02d9 | 7576 | |
2c47e605 | 7577 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
7578 | if (wret < 0) |
7579 | ret = wret; | |
7580 | if (wret != 0) | |
7581 | break; | |
7582 | } | |
7583 | ||
2c47e605 | 7584 | kfree(wc); |
f82d02d9 YZ |
7585 | btrfs_free_path(path); |
7586 | return ret; | |
7587 | } | |
7588 | ||
ec44a35c CM |
7589 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
7590 | { | |
7591 | u64 num_devices; | |
fc67c450 | 7592 | u64 stripped; |
e4d8ec0f | 7593 | |
fc67c450 ID |
7594 | /* |
7595 | * if restripe for this chunk_type is on pick target profile and | |
7596 | * return, otherwise do the usual balance | |
7597 | */ | |
7598 | stripped = get_restripe_target(root->fs_info, flags); | |
7599 | if (stripped) | |
7600 | return extended_to_chunk(stripped); | |
e4d8ec0f | 7601 | |
cd02dca5 CM |
7602 | /* |
7603 | * we add in the count of missing devices because we want | |
7604 | * to make sure that any RAID levels on a degraded FS | |
7605 | * continue to be honored. | |
7606 | */ | |
7607 | num_devices = root->fs_info->fs_devices->rw_devices + | |
7608 | root->fs_info->fs_devices->missing_devices; | |
7609 | ||
fc67c450 | 7610 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 7611 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
7612 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
7613 | ||
ec44a35c CM |
7614 | if (num_devices == 1) { |
7615 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7616 | stripped = flags & ~stripped; | |
7617 | ||
7618 | /* turn raid0 into single device chunks */ | |
7619 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
7620 | return stripped; | |
7621 | ||
7622 | /* turn mirroring into duplication */ | |
7623 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7624 | BTRFS_BLOCK_GROUP_RAID10)) | |
7625 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
7626 | } else { |
7627 | /* they already had raid on here, just return */ | |
ec44a35c CM |
7628 | if (flags & stripped) |
7629 | return flags; | |
7630 | ||
7631 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
7632 | stripped = flags & ~stripped; | |
7633 | ||
7634 | /* switch duplicated blocks with raid1 */ | |
7635 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
7636 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
7637 | ||
e3176ca2 | 7638 | /* this is drive concat, leave it alone */ |
ec44a35c | 7639 | } |
e3176ca2 | 7640 | |
ec44a35c CM |
7641 | return flags; |
7642 | } | |
7643 | ||
199c36ea | 7644 | static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 7645 | { |
f0486c68 YZ |
7646 | struct btrfs_space_info *sinfo = cache->space_info; |
7647 | u64 num_bytes; | |
199c36ea | 7648 | u64 min_allocable_bytes; |
f0486c68 | 7649 | int ret = -ENOSPC; |
0ef3e66b | 7650 | |
c286ac48 | 7651 | |
199c36ea MX |
7652 | /* |
7653 | * We need some metadata space and system metadata space for | |
7654 | * allocating chunks in some corner cases until we force to set | |
7655 | * it to be readonly. | |
7656 | */ | |
7657 | if ((sinfo->flags & | |
7658 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
7659 | !force) | |
7660 | min_allocable_bytes = 1 * 1024 * 1024; | |
7661 | else | |
7662 | min_allocable_bytes = 0; | |
7663 | ||
f0486c68 YZ |
7664 | spin_lock(&sinfo->lock); |
7665 | spin_lock(&cache->lock); | |
61cfea9b W |
7666 | |
7667 | if (cache->ro) { | |
7668 | ret = 0; | |
7669 | goto out; | |
7670 | } | |
7671 | ||
f0486c68 YZ |
7672 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
7673 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7674 | ||
7675 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + | |
37be25bc JB |
7676 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + |
7677 | min_allocable_bytes <= sinfo->total_bytes) { | |
f0486c68 | 7678 | sinfo->bytes_readonly += num_bytes; |
f0486c68 YZ |
7679 | cache->ro = 1; |
7680 | ret = 0; | |
7681 | } | |
61cfea9b | 7682 | out: |
f0486c68 YZ |
7683 | spin_unlock(&cache->lock); |
7684 | spin_unlock(&sinfo->lock); | |
7685 | return ret; | |
7686 | } | |
7d9eb12c | 7687 | |
f0486c68 YZ |
7688 | int btrfs_set_block_group_ro(struct btrfs_root *root, |
7689 | struct btrfs_block_group_cache *cache) | |
c286ac48 | 7690 | |
f0486c68 YZ |
7691 | { |
7692 | struct btrfs_trans_handle *trans; | |
7693 | u64 alloc_flags; | |
7694 | int ret; | |
7d9eb12c | 7695 | |
f0486c68 | 7696 | BUG_ON(cache->ro); |
0ef3e66b | 7697 | |
ff5714cc | 7698 | trans = btrfs_join_transaction(root); |
79787eaa JM |
7699 | if (IS_ERR(trans)) |
7700 | return PTR_ERR(trans); | |
5d4f98a2 | 7701 | |
f0486c68 | 7702 | alloc_flags = update_block_group_flags(root, cache->flags); |
79787eaa | 7703 | if (alloc_flags != cache->flags) { |
698d0082 | 7704 | ret = do_chunk_alloc(trans, root, alloc_flags, |
79787eaa JM |
7705 | CHUNK_ALLOC_FORCE); |
7706 | if (ret < 0) | |
7707 | goto out; | |
7708 | } | |
5d4f98a2 | 7709 | |
199c36ea | 7710 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7711 | if (!ret) |
7712 | goto out; | |
7713 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); | |
698d0082 | 7714 | ret = do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 7715 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
7716 | if (ret < 0) |
7717 | goto out; | |
199c36ea | 7718 | ret = set_block_group_ro(cache, 0); |
f0486c68 YZ |
7719 | out: |
7720 | btrfs_end_transaction(trans, root); | |
7721 | return ret; | |
7722 | } | |
5d4f98a2 | 7723 | |
c87f08ca CM |
7724 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
7725 | struct btrfs_root *root, u64 type) | |
7726 | { | |
7727 | u64 alloc_flags = get_alloc_profile(root, type); | |
698d0082 | 7728 | return do_chunk_alloc(trans, root, alloc_flags, |
0e4f8f88 | 7729 | CHUNK_ALLOC_FORCE); |
c87f08ca CM |
7730 | } |
7731 | ||
6d07bcec MX |
7732 | /* |
7733 | * helper to account the unused space of all the readonly block group in the | |
7734 | * list. takes mirrors into account. | |
7735 | */ | |
7736 | static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list) | |
7737 | { | |
7738 | struct btrfs_block_group_cache *block_group; | |
7739 | u64 free_bytes = 0; | |
7740 | int factor; | |
7741 | ||
7742 | list_for_each_entry(block_group, groups_list, list) { | |
7743 | spin_lock(&block_group->lock); | |
7744 | ||
7745 | if (!block_group->ro) { | |
7746 | spin_unlock(&block_group->lock); | |
7747 | continue; | |
7748 | } | |
7749 | ||
7750 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
7751 | BTRFS_BLOCK_GROUP_RAID10 | | |
7752 | BTRFS_BLOCK_GROUP_DUP)) | |
7753 | factor = 2; | |
7754 | else | |
7755 | factor = 1; | |
7756 | ||
7757 | free_bytes += (block_group->key.offset - | |
7758 | btrfs_block_group_used(&block_group->item)) * | |
7759 | factor; | |
7760 | ||
7761 | spin_unlock(&block_group->lock); | |
7762 | } | |
7763 | ||
7764 | return free_bytes; | |
7765 | } | |
7766 | ||
7767 | /* | |
7768 | * helper to account the unused space of all the readonly block group in the | |
7769 | * space_info. takes mirrors into account. | |
7770 | */ | |
7771 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) | |
7772 | { | |
7773 | int i; | |
7774 | u64 free_bytes = 0; | |
7775 | ||
7776 | spin_lock(&sinfo->lock); | |
7777 | ||
7778 | for(i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
7779 | if (!list_empty(&sinfo->block_groups[i])) | |
7780 | free_bytes += __btrfs_get_ro_block_group_free_space( | |
7781 | &sinfo->block_groups[i]); | |
7782 | ||
7783 | spin_unlock(&sinfo->lock); | |
7784 | ||
7785 | return free_bytes; | |
7786 | } | |
7787 | ||
143bede5 | 7788 | void btrfs_set_block_group_rw(struct btrfs_root *root, |
f0486c68 | 7789 | struct btrfs_block_group_cache *cache) |
5d4f98a2 | 7790 | { |
f0486c68 YZ |
7791 | struct btrfs_space_info *sinfo = cache->space_info; |
7792 | u64 num_bytes; | |
7793 | ||
7794 | BUG_ON(!cache->ro); | |
7795 | ||
7796 | spin_lock(&sinfo->lock); | |
7797 | spin_lock(&cache->lock); | |
7798 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - | |
7799 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
7800 | sinfo->bytes_readonly -= num_bytes; | |
7801 | cache->ro = 0; | |
7802 | spin_unlock(&cache->lock); | |
7803 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
7804 | } |
7805 | ||
ba1bf481 JB |
7806 | /* |
7807 | * checks to see if its even possible to relocate this block group. | |
7808 | * | |
7809 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
7810 | * ok to go ahead and try. | |
7811 | */ | |
7812 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) | |
1a40e23b | 7813 | { |
ba1bf481 JB |
7814 | struct btrfs_block_group_cache *block_group; |
7815 | struct btrfs_space_info *space_info; | |
7816 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
7817 | struct btrfs_device *device; | |
cdcb725c | 7818 | u64 min_free; |
6719db6a JB |
7819 | u64 dev_min = 1; |
7820 | u64 dev_nr = 0; | |
4a5e98f5 | 7821 | u64 target; |
cdcb725c | 7822 | int index; |
ba1bf481 JB |
7823 | int full = 0; |
7824 | int ret = 0; | |
1a40e23b | 7825 | |
ba1bf481 | 7826 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
1a40e23b | 7827 | |
ba1bf481 JB |
7828 | /* odd, couldn't find the block group, leave it alone */ |
7829 | if (!block_group) | |
7830 | return -1; | |
1a40e23b | 7831 | |
cdcb725c | 7832 | min_free = btrfs_block_group_used(&block_group->item); |
7833 | ||
ba1bf481 | 7834 | /* no bytes used, we're good */ |
cdcb725c | 7835 | if (!min_free) |
1a40e23b ZY |
7836 | goto out; |
7837 | ||
ba1bf481 JB |
7838 | space_info = block_group->space_info; |
7839 | spin_lock(&space_info->lock); | |
17d217fe | 7840 | |
ba1bf481 | 7841 | full = space_info->full; |
17d217fe | 7842 | |
ba1bf481 JB |
7843 | /* |
7844 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
7845 | * relocate it unless we're able to allocate a new chunk below. |
7846 | * | |
7847 | * Otherwise, we need to make sure we have room in the space to handle | |
7848 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 7849 | */ |
7ce618db | 7850 | if ((space_info->total_bytes != block_group->key.offset) && |
cdcb725c | 7851 | (space_info->bytes_used + space_info->bytes_reserved + |
7852 | space_info->bytes_pinned + space_info->bytes_readonly + | |
7853 | min_free < space_info->total_bytes)) { | |
ba1bf481 JB |
7854 | spin_unlock(&space_info->lock); |
7855 | goto out; | |
17d217fe | 7856 | } |
ba1bf481 | 7857 | spin_unlock(&space_info->lock); |
ea8c2819 | 7858 | |
ba1bf481 JB |
7859 | /* |
7860 | * ok we don't have enough space, but maybe we have free space on our | |
7861 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
7862 | * alloc devices and guess if we have enough space. if this block |
7863 | * group is going to be restriped, run checks against the target | |
7864 | * profile instead of the current one. | |
ba1bf481 JB |
7865 | */ |
7866 | ret = -1; | |
ea8c2819 | 7867 | |
cdcb725c | 7868 | /* |
7869 | * index: | |
7870 | * 0: raid10 | |
7871 | * 1: raid1 | |
7872 | * 2: dup | |
7873 | * 3: raid0 | |
7874 | * 4: single | |
7875 | */ | |
4a5e98f5 ID |
7876 | target = get_restripe_target(root->fs_info, block_group->flags); |
7877 | if (target) { | |
31e50229 | 7878 | index = __get_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
7879 | } else { |
7880 | /* | |
7881 | * this is just a balance, so if we were marked as full | |
7882 | * we know there is no space for a new chunk | |
7883 | */ | |
7884 | if (full) | |
7885 | goto out; | |
7886 | ||
7887 | index = get_block_group_index(block_group); | |
7888 | } | |
7889 | ||
e6ec716f | 7890 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 7891 | dev_min = 4; |
6719db6a JB |
7892 | /* Divide by 2 */ |
7893 | min_free >>= 1; | |
e6ec716f | 7894 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 7895 | dev_min = 2; |
e6ec716f | 7896 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
7897 | /* Multiply by 2 */ |
7898 | min_free <<= 1; | |
e6ec716f | 7899 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 7900 | dev_min = fs_devices->rw_devices; |
6719db6a | 7901 | do_div(min_free, dev_min); |
cdcb725c | 7902 | } |
7903 | ||
ba1bf481 JB |
7904 | mutex_lock(&root->fs_info->chunk_mutex); |
7905 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { | |
7bfc837d | 7906 | u64 dev_offset; |
56bec294 | 7907 | |
ba1bf481 JB |
7908 | /* |
7909 | * check to make sure we can actually find a chunk with enough | |
7910 | * space to fit our block group in. | |
7911 | */ | |
63a212ab SB |
7912 | if (device->total_bytes > device->bytes_used + min_free && |
7913 | !device->is_tgtdev_for_dev_replace) { | |
125ccb0a | 7914 | ret = find_free_dev_extent(device, min_free, |
7bfc837d | 7915 | &dev_offset, NULL); |
ba1bf481 | 7916 | if (!ret) |
cdcb725c | 7917 | dev_nr++; |
7918 | ||
7919 | if (dev_nr >= dev_min) | |
73e48b27 | 7920 | break; |
cdcb725c | 7921 | |
ba1bf481 | 7922 | ret = -1; |
725c8463 | 7923 | } |
edbd8d4e | 7924 | } |
ba1bf481 | 7925 | mutex_unlock(&root->fs_info->chunk_mutex); |
edbd8d4e | 7926 | out: |
ba1bf481 | 7927 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
7928 | return ret; |
7929 | } | |
7930 | ||
b2950863 CH |
7931 | static int find_first_block_group(struct btrfs_root *root, |
7932 | struct btrfs_path *path, struct btrfs_key *key) | |
0b86a832 | 7933 | { |
925baedd | 7934 | int ret = 0; |
0b86a832 CM |
7935 | struct btrfs_key found_key; |
7936 | struct extent_buffer *leaf; | |
7937 | int slot; | |
edbd8d4e | 7938 | |
0b86a832 CM |
7939 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
7940 | if (ret < 0) | |
925baedd CM |
7941 | goto out; |
7942 | ||
d397712b | 7943 | while (1) { |
0b86a832 | 7944 | slot = path->slots[0]; |
edbd8d4e | 7945 | leaf = path->nodes[0]; |
0b86a832 CM |
7946 | if (slot >= btrfs_header_nritems(leaf)) { |
7947 | ret = btrfs_next_leaf(root, path); | |
7948 | if (ret == 0) | |
7949 | continue; | |
7950 | if (ret < 0) | |
925baedd | 7951 | goto out; |
0b86a832 | 7952 | break; |
edbd8d4e | 7953 | } |
0b86a832 | 7954 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 7955 | |
0b86a832 | 7956 | if (found_key.objectid >= key->objectid && |
925baedd CM |
7957 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
7958 | ret = 0; | |
7959 | goto out; | |
7960 | } | |
0b86a832 | 7961 | path->slots[0]++; |
edbd8d4e | 7962 | } |
925baedd | 7963 | out: |
0b86a832 | 7964 | return ret; |
edbd8d4e CM |
7965 | } |
7966 | ||
0af3d00b JB |
7967 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
7968 | { | |
7969 | struct btrfs_block_group_cache *block_group; | |
7970 | u64 last = 0; | |
7971 | ||
7972 | while (1) { | |
7973 | struct inode *inode; | |
7974 | ||
7975 | block_group = btrfs_lookup_first_block_group(info, last); | |
7976 | while (block_group) { | |
7977 | spin_lock(&block_group->lock); | |
7978 | if (block_group->iref) | |
7979 | break; | |
7980 | spin_unlock(&block_group->lock); | |
7981 | block_group = next_block_group(info->tree_root, | |
7982 | block_group); | |
7983 | } | |
7984 | if (!block_group) { | |
7985 | if (last == 0) | |
7986 | break; | |
7987 | last = 0; | |
7988 | continue; | |
7989 | } | |
7990 | ||
7991 | inode = block_group->inode; | |
7992 | block_group->iref = 0; | |
7993 | block_group->inode = NULL; | |
7994 | spin_unlock(&block_group->lock); | |
7995 | iput(inode); | |
7996 | last = block_group->key.objectid + block_group->key.offset; | |
7997 | btrfs_put_block_group(block_group); | |
7998 | } | |
7999 | } | |
8000 | ||
1a40e23b ZY |
8001 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
8002 | { | |
8003 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 8004 | struct btrfs_space_info *space_info; |
11833d66 | 8005 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
8006 | struct rb_node *n; |
8007 | ||
11833d66 YZ |
8008 | down_write(&info->extent_commit_sem); |
8009 | while (!list_empty(&info->caching_block_groups)) { | |
8010 | caching_ctl = list_entry(info->caching_block_groups.next, | |
8011 | struct btrfs_caching_control, list); | |
8012 | list_del(&caching_ctl->list); | |
8013 | put_caching_control(caching_ctl); | |
8014 | } | |
8015 | up_write(&info->extent_commit_sem); | |
8016 | ||
1a40e23b ZY |
8017 | spin_lock(&info->block_group_cache_lock); |
8018 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
8019 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
8020 | cache_node); | |
1a40e23b ZY |
8021 | rb_erase(&block_group->cache_node, |
8022 | &info->block_group_cache_tree); | |
d899e052 YZ |
8023 | spin_unlock(&info->block_group_cache_lock); |
8024 | ||
80eb234a | 8025 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 8026 | list_del(&block_group->list); |
80eb234a | 8027 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 8028 | |
817d52f8 | 8029 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 8030 | wait_block_group_cache_done(block_group); |
817d52f8 | 8031 | |
3c14874a JB |
8032 | /* |
8033 | * We haven't cached this block group, which means we could | |
8034 | * possibly have excluded extents on this block group. | |
8035 | */ | |
8036 | if (block_group->cached == BTRFS_CACHE_NO) | |
8037 | free_excluded_extents(info->extent_root, block_group); | |
8038 | ||
817d52f8 | 8039 | btrfs_remove_free_space_cache(block_group); |
11dfe35a | 8040 | btrfs_put_block_group(block_group); |
d899e052 YZ |
8041 | |
8042 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
8043 | } |
8044 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
8045 | |
8046 | /* now that all the block groups are freed, go through and | |
8047 | * free all the space_info structs. This is only called during | |
8048 | * the final stages of unmount, and so we know nobody is | |
8049 | * using them. We call synchronize_rcu() once before we start, | |
8050 | * just to be on the safe side. | |
8051 | */ | |
8052 | synchronize_rcu(); | |
8053 | ||
8929ecfa YZ |
8054 | release_global_block_rsv(info); |
8055 | ||
4184ea7f CM |
8056 | while(!list_empty(&info->space_info)) { |
8057 | space_info = list_entry(info->space_info.next, | |
8058 | struct btrfs_space_info, | |
8059 | list); | |
b069e0c3 DS |
8060 | if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) { |
8061 | if (space_info->bytes_pinned > 0 || | |
8062 | space_info->bytes_reserved > 0 || | |
8063 | space_info->bytes_may_use > 0) { | |
8064 | WARN_ON(1); | |
8065 | dump_space_info(space_info, 0, 0); | |
8066 | } | |
f0486c68 | 8067 | } |
4184ea7f CM |
8068 | list_del(&space_info->list); |
8069 | kfree(space_info); | |
8070 | } | |
1a40e23b ZY |
8071 | return 0; |
8072 | } | |
8073 | ||
b742bb82 YZ |
8074 | static void __link_block_group(struct btrfs_space_info *space_info, |
8075 | struct btrfs_block_group_cache *cache) | |
8076 | { | |
8077 | int index = get_block_group_index(cache); | |
8078 | ||
8079 | down_write(&space_info->groups_sem); | |
8080 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
8081 | up_write(&space_info->groups_sem); | |
8082 | } | |
8083 | ||
9078a3e1 CM |
8084 | int btrfs_read_block_groups(struct btrfs_root *root) |
8085 | { | |
8086 | struct btrfs_path *path; | |
8087 | int ret; | |
9078a3e1 | 8088 | struct btrfs_block_group_cache *cache; |
be744175 | 8089 | struct btrfs_fs_info *info = root->fs_info; |
6324fbf3 | 8090 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
8091 | struct btrfs_key key; |
8092 | struct btrfs_key found_key; | |
5f39d397 | 8093 | struct extent_buffer *leaf; |
0af3d00b JB |
8094 | int need_clear = 0; |
8095 | u64 cache_gen; | |
96b5179d | 8096 | |
be744175 | 8097 | root = info->extent_root; |
9078a3e1 | 8098 | key.objectid = 0; |
0b86a832 | 8099 | key.offset = 0; |
9078a3e1 | 8100 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
9078a3e1 CM |
8101 | path = btrfs_alloc_path(); |
8102 | if (!path) | |
8103 | return -ENOMEM; | |
026fd317 | 8104 | path->reada = 1; |
9078a3e1 | 8105 | |
6c41761f | 8106 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); |
73bc1876 | 8107 | if (btrfs_test_opt(root, SPACE_CACHE) && |
6c41761f | 8108 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) |
0af3d00b | 8109 | need_clear = 1; |
88c2ba3b JB |
8110 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
8111 | need_clear = 1; | |
0af3d00b | 8112 | |
d397712b | 8113 | while (1) { |
0b86a832 | 8114 | ret = find_first_block_group(root, path, &key); |
b742bb82 YZ |
8115 | if (ret > 0) |
8116 | break; | |
0b86a832 CM |
8117 | if (ret != 0) |
8118 | goto error; | |
5f39d397 CM |
8119 | leaf = path->nodes[0]; |
8120 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
8f18cf13 | 8121 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
9078a3e1 | 8122 | if (!cache) { |
0b86a832 | 8123 | ret = -ENOMEM; |
f0486c68 | 8124 | goto error; |
9078a3e1 | 8125 | } |
34d52cb6 LZ |
8126 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
8127 | GFP_NOFS); | |
8128 | if (!cache->free_space_ctl) { | |
8129 | kfree(cache); | |
8130 | ret = -ENOMEM; | |
8131 | goto error; | |
8132 | } | |
3e1ad54f | 8133 | |
d2fb3437 | 8134 | atomic_set(&cache->count, 1); |
c286ac48 | 8135 | spin_lock_init(&cache->lock); |
817d52f8 | 8136 | cache->fs_info = info; |
0f9dd46c | 8137 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 8138 | INIT_LIST_HEAD(&cache->cluster_list); |
96303081 | 8139 | |
cf7c1ef6 LB |
8140 | if (need_clear) { |
8141 | /* | |
8142 | * When we mount with old space cache, we need to | |
8143 | * set BTRFS_DC_CLEAR and set dirty flag. | |
8144 | * | |
8145 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
8146 | * truncate the old free space cache inode and | |
8147 | * setup a new one. | |
8148 | * b) Setting 'dirty flag' makes sure that we flush | |
8149 | * the new space cache info onto disk. | |
8150 | */ | |
0af3d00b | 8151 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 LB |
8152 | if (btrfs_test_opt(root, SPACE_CACHE)) |
8153 | cache->dirty = 1; | |
8154 | } | |
0af3d00b | 8155 | |
5f39d397 CM |
8156 | read_extent_buffer(leaf, &cache->item, |
8157 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
8158 | sizeof(cache->item)); | |
9078a3e1 | 8159 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
0b86a832 | 8160 | |
9078a3e1 | 8161 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 8162 | btrfs_release_path(path); |
0b86a832 | 8163 | cache->flags = btrfs_block_group_flags(&cache->item); |
817d52f8 | 8164 | cache->sectorsize = root->sectorsize; |
53b381b3 DW |
8165 | cache->full_stripe_len = btrfs_full_stripe_len(root, |
8166 | &root->fs_info->mapping_tree, | |
8167 | found_key.objectid); | |
34d52cb6 LZ |
8168 | btrfs_init_free_space_ctl(cache); |
8169 | ||
3c14874a JB |
8170 | /* |
8171 | * We need to exclude the super stripes now so that the space | |
8172 | * info has super bytes accounted for, otherwise we'll think | |
8173 | * we have more space than we actually do. | |
8174 | */ | |
835d974f JB |
8175 | ret = exclude_super_stripes(root, cache); |
8176 | if (ret) { | |
8177 | /* | |
8178 | * We may have excluded something, so call this just in | |
8179 | * case. | |
8180 | */ | |
8181 | free_excluded_extents(root, cache); | |
8182 | kfree(cache->free_space_ctl); | |
8183 | kfree(cache); | |
8184 | goto error; | |
8185 | } | |
3c14874a | 8186 | |
817d52f8 JB |
8187 | /* |
8188 | * check for two cases, either we are full, and therefore | |
8189 | * don't need to bother with the caching work since we won't | |
8190 | * find any space, or we are empty, and we can just add all | |
8191 | * the space in and be done with it. This saves us _alot_ of | |
8192 | * time, particularly in the full case. | |
8193 | */ | |
8194 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 8195 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 8196 | cache->cached = BTRFS_CACHE_FINISHED; |
1b2da372 | 8197 | free_excluded_extents(root, cache); |
817d52f8 | 8198 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 8199 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 JB |
8200 | cache->cached = BTRFS_CACHE_FINISHED; |
8201 | add_new_free_space(cache, root->fs_info, | |
8202 | found_key.objectid, | |
8203 | found_key.objectid + | |
8204 | found_key.offset); | |
11833d66 | 8205 | free_excluded_extents(root, cache); |
817d52f8 | 8206 | } |
96b5179d | 8207 | |
8c579fe7 JB |
8208 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
8209 | if (ret) { | |
8210 | btrfs_remove_free_space_cache(cache); | |
8211 | btrfs_put_block_group(cache); | |
8212 | goto error; | |
8213 | } | |
8214 | ||
6324fbf3 CM |
8215 | ret = update_space_info(info, cache->flags, found_key.offset, |
8216 | btrfs_block_group_used(&cache->item), | |
8217 | &space_info); | |
8c579fe7 JB |
8218 | if (ret) { |
8219 | btrfs_remove_free_space_cache(cache); | |
8220 | spin_lock(&info->block_group_cache_lock); | |
8221 | rb_erase(&cache->cache_node, | |
8222 | &info->block_group_cache_tree); | |
8223 | spin_unlock(&info->block_group_cache_lock); | |
8224 | btrfs_put_block_group(cache); | |
8225 | goto error; | |
8226 | } | |
8227 | ||
6324fbf3 | 8228 | cache->space_info = space_info; |
1b2da372 | 8229 | spin_lock(&cache->space_info->lock); |
f0486c68 | 8230 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
8231 | spin_unlock(&cache->space_info->lock); |
8232 | ||
b742bb82 | 8233 | __link_block_group(space_info, cache); |
0f9dd46c | 8234 | |
75ccf47d | 8235 | set_avail_alloc_bits(root->fs_info, cache->flags); |
2b82032c | 8236 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
199c36ea | 8237 | set_block_group_ro(cache, 1); |
9078a3e1 | 8238 | } |
b742bb82 YZ |
8239 | |
8240 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { | |
8241 | if (!(get_alloc_profile(root, space_info->flags) & | |
8242 | (BTRFS_BLOCK_GROUP_RAID10 | | |
8243 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
8244 | BTRFS_BLOCK_GROUP_RAID5 | |
8245 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
8246 | BTRFS_BLOCK_GROUP_DUP))) |
8247 | continue; | |
8248 | /* | |
8249 | * avoid allocating from un-mirrored block group if there are | |
8250 | * mirrored block groups. | |
8251 | */ | |
8252 | list_for_each_entry(cache, &space_info->block_groups[3], list) | |
199c36ea | 8253 | set_block_group_ro(cache, 1); |
b742bb82 | 8254 | list_for_each_entry(cache, &space_info->block_groups[4], list) |
199c36ea | 8255 | set_block_group_ro(cache, 1); |
9078a3e1 | 8256 | } |
f0486c68 YZ |
8257 | |
8258 | init_global_block_rsv(info); | |
0b86a832 CM |
8259 | ret = 0; |
8260 | error: | |
9078a3e1 | 8261 | btrfs_free_path(path); |
0b86a832 | 8262 | return ret; |
9078a3e1 | 8263 | } |
6324fbf3 | 8264 | |
ea658bad JB |
8265 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans, |
8266 | struct btrfs_root *root) | |
8267 | { | |
8268 | struct btrfs_block_group_cache *block_group, *tmp; | |
8269 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
8270 | struct btrfs_block_group_item item; | |
8271 | struct btrfs_key key; | |
8272 | int ret = 0; | |
8273 | ||
8274 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, | |
8275 | new_bg_list) { | |
8276 | list_del_init(&block_group->new_bg_list); | |
8277 | ||
8278 | if (ret) | |
8279 | continue; | |
8280 | ||
8281 | spin_lock(&block_group->lock); | |
8282 | memcpy(&item, &block_group->item, sizeof(item)); | |
8283 | memcpy(&key, &block_group->key, sizeof(key)); | |
8284 | spin_unlock(&block_group->lock); | |
8285 | ||
8286 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
8287 | sizeof(item)); | |
8288 | if (ret) | |
8289 | btrfs_abort_transaction(trans, extent_root, ret); | |
8290 | } | |
8291 | } | |
8292 | ||
6324fbf3 CM |
8293 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
8294 | struct btrfs_root *root, u64 bytes_used, | |
e17cade2 | 8295 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
6324fbf3 CM |
8296 | u64 size) |
8297 | { | |
8298 | int ret; | |
6324fbf3 CM |
8299 | struct btrfs_root *extent_root; |
8300 | struct btrfs_block_group_cache *cache; | |
6324fbf3 CM |
8301 | |
8302 | extent_root = root->fs_info->extent_root; | |
6324fbf3 | 8303 | |
12fcfd22 | 8304 | root->fs_info->last_trans_log_full_commit = trans->transid; |
e02119d5 | 8305 | |
8f18cf13 | 8306 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
0f9dd46c JB |
8307 | if (!cache) |
8308 | return -ENOMEM; | |
34d52cb6 LZ |
8309 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
8310 | GFP_NOFS); | |
8311 | if (!cache->free_space_ctl) { | |
8312 | kfree(cache); | |
8313 | return -ENOMEM; | |
8314 | } | |
0f9dd46c | 8315 | |
e17cade2 | 8316 | cache->key.objectid = chunk_offset; |
6324fbf3 | 8317 | cache->key.offset = size; |
d2fb3437 | 8318 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
96303081 | 8319 | cache->sectorsize = root->sectorsize; |
0af3d00b | 8320 | cache->fs_info = root->fs_info; |
53b381b3 DW |
8321 | cache->full_stripe_len = btrfs_full_stripe_len(root, |
8322 | &root->fs_info->mapping_tree, | |
8323 | chunk_offset); | |
96303081 | 8324 | |
d2fb3437 | 8325 | atomic_set(&cache->count, 1); |
c286ac48 | 8326 | spin_lock_init(&cache->lock); |
0f9dd46c | 8327 | INIT_LIST_HEAD(&cache->list); |
fa9c0d79 | 8328 | INIT_LIST_HEAD(&cache->cluster_list); |
ea658bad | 8329 | INIT_LIST_HEAD(&cache->new_bg_list); |
0ef3e66b | 8330 | |
34d52cb6 LZ |
8331 | btrfs_init_free_space_ctl(cache); |
8332 | ||
6324fbf3 | 8333 | btrfs_set_block_group_used(&cache->item, bytes_used); |
6324fbf3 CM |
8334 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
8335 | cache->flags = type; | |
8336 | btrfs_set_block_group_flags(&cache->item, type); | |
8337 | ||
11833d66 | 8338 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 8339 | cache->cached = BTRFS_CACHE_FINISHED; |
835d974f JB |
8340 | ret = exclude_super_stripes(root, cache); |
8341 | if (ret) { | |
8342 | /* | |
8343 | * We may have excluded something, so call this just in | |
8344 | * case. | |
8345 | */ | |
8346 | free_excluded_extents(root, cache); | |
8347 | kfree(cache->free_space_ctl); | |
8348 | kfree(cache); | |
8349 | return ret; | |
8350 | } | |
96303081 | 8351 | |
817d52f8 JB |
8352 | add_new_free_space(cache, root->fs_info, chunk_offset, |
8353 | chunk_offset + size); | |
8354 | ||
11833d66 YZ |
8355 | free_excluded_extents(root, cache); |
8356 | ||
8c579fe7 JB |
8357 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
8358 | if (ret) { | |
8359 | btrfs_remove_free_space_cache(cache); | |
8360 | btrfs_put_block_group(cache); | |
8361 | return ret; | |
8362 | } | |
8363 | ||
6324fbf3 CM |
8364 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
8365 | &cache->space_info); | |
8c579fe7 JB |
8366 | if (ret) { |
8367 | btrfs_remove_free_space_cache(cache); | |
8368 | spin_lock(&root->fs_info->block_group_cache_lock); | |
8369 | rb_erase(&cache->cache_node, | |
8370 | &root->fs_info->block_group_cache_tree); | |
8371 | spin_unlock(&root->fs_info->block_group_cache_lock); | |
8372 | btrfs_put_block_group(cache); | |
8373 | return ret; | |
8374 | } | |
c7c144db | 8375 | update_global_block_rsv(root->fs_info); |
1b2da372 JB |
8376 | |
8377 | spin_lock(&cache->space_info->lock); | |
f0486c68 | 8378 | cache->space_info->bytes_readonly += cache->bytes_super; |
1b2da372 JB |
8379 | spin_unlock(&cache->space_info->lock); |
8380 | ||
b742bb82 | 8381 | __link_block_group(cache->space_info, cache); |
6324fbf3 | 8382 | |
ea658bad | 8383 | list_add_tail(&cache->new_bg_list, &trans->new_bgs); |
6324fbf3 | 8384 | |
d18a2c44 | 8385 | set_avail_alloc_bits(extent_root->fs_info, type); |
925baedd | 8386 | |
6324fbf3 CM |
8387 | return 0; |
8388 | } | |
1a40e23b | 8389 | |
10ea00f5 ID |
8390 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
8391 | { | |
899c81ea ID |
8392 | u64 extra_flags = chunk_to_extended(flags) & |
8393 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 8394 | |
de98ced9 | 8395 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
8396 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
8397 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
8398 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
8399 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
8400 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
8401 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 8402 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
8403 | } |
8404 | ||
1a40e23b ZY |
8405 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
8406 | struct btrfs_root *root, u64 group_start) | |
8407 | { | |
8408 | struct btrfs_path *path; | |
8409 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 8410 | struct btrfs_free_cluster *cluster; |
0af3d00b | 8411 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
1a40e23b | 8412 | struct btrfs_key key; |
0af3d00b | 8413 | struct inode *inode; |
1a40e23b | 8414 | int ret; |
10ea00f5 | 8415 | int index; |
89a55897 | 8416 | int factor; |
1a40e23b | 8417 | |
1a40e23b ZY |
8418 | root = root->fs_info->extent_root; |
8419 | ||
8420 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); | |
8421 | BUG_ON(!block_group); | |
c146afad | 8422 | BUG_ON(!block_group->ro); |
1a40e23b | 8423 | |
9f7c43c9 | 8424 | /* |
8425 | * Free the reserved super bytes from this block group before | |
8426 | * remove it. | |
8427 | */ | |
8428 | free_excluded_extents(root, block_group); | |
8429 | ||
1a40e23b | 8430 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 8431 | index = get_block_group_index(block_group); |
89a55897 JB |
8432 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
8433 | BTRFS_BLOCK_GROUP_RAID1 | | |
8434 | BTRFS_BLOCK_GROUP_RAID10)) | |
8435 | factor = 2; | |
8436 | else | |
8437 | factor = 1; | |
1a40e23b | 8438 | |
44fb5511 CM |
8439 | /* make sure this block group isn't part of an allocation cluster */ |
8440 | cluster = &root->fs_info->data_alloc_cluster; | |
8441 | spin_lock(&cluster->refill_lock); | |
8442 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
8443 | spin_unlock(&cluster->refill_lock); | |
8444 | ||
8445 | /* | |
8446 | * make sure this block group isn't part of a metadata | |
8447 | * allocation cluster | |
8448 | */ | |
8449 | cluster = &root->fs_info->meta_alloc_cluster; | |
8450 | spin_lock(&cluster->refill_lock); | |
8451 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
8452 | spin_unlock(&cluster->refill_lock); | |
8453 | ||
1a40e23b | 8454 | path = btrfs_alloc_path(); |
d8926bb3 MF |
8455 | if (!path) { |
8456 | ret = -ENOMEM; | |
8457 | goto out; | |
8458 | } | |
1a40e23b | 8459 | |
10b2f34d | 8460 | inode = lookup_free_space_inode(tree_root, block_group, path); |
0af3d00b | 8461 | if (!IS_ERR(inode)) { |
b532402e | 8462 | ret = btrfs_orphan_add(trans, inode); |
79787eaa JM |
8463 | if (ret) { |
8464 | btrfs_add_delayed_iput(inode); | |
8465 | goto out; | |
8466 | } | |
0af3d00b JB |
8467 | clear_nlink(inode); |
8468 | /* One for the block groups ref */ | |
8469 | spin_lock(&block_group->lock); | |
8470 | if (block_group->iref) { | |
8471 | block_group->iref = 0; | |
8472 | block_group->inode = NULL; | |
8473 | spin_unlock(&block_group->lock); | |
8474 | iput(inode); | |
8475 | } else { | |
8476 | spin_unlock(&block_group->lock); | |
8477 | } | |
8478 | /* One for our lookup ref */ | |
455757c3 | 8479 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
8480 | } |
8481 | ||
8482 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
8483 | key.offset = block_group->key.objectid; | |
8484 | key.type = 0; | |
8485 | ||
8486 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
8487 | if (ret < 0) | |
8488 | goto out; | |
8489 | if (ret > 0) | |
b3b4aa74 | 8490 | btrfs_release_path(path); |
0af3d00b JB |
8491 | if (ret == 0) { |
8492 | ret = btrfs_del_item(trans, tree_root, path); | |
8493 | if (ret) | |
8494 | goto out; | |
b3b4aa74 | 8495 | btrfs_release_path(path); |
0af3d00b JB |
8496 | } |
8497 | ||
3dfdb934 | 8498 | spin_lock(&root->fs_info->block_group_cache_lock); |
1a40e23b ZY |
8499 | rb_erase(&block_group->cache_node, |
8500 | &root->fs_info->block_group_cache_tree); | |
a1897fdd LB |
8501 | |
8502 | if (root->fs_info->first_logical_byte == block_group->key.objectid) | |
8503 | root->fs_info->first_logical_byte = (u64)-1; | |
3dfdb934 | 8504 | spin_unlock(&root->fs_info->block_group_cache_lock); |
817d52f8 | 8505 | |
80eb234a | 8506 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
8507 | /* |
8508 | * we must use list_del_init so people can check to see if they | |
8509 | * are still on the list after taking the semaphore | |
8510 | */ | |
8511 | list_del_init(&block_group->list); | |
10ea00f5 ID |
8512 | if (list_empty(&block_group->space_info->block_groups[index])) |
8513 | clear_avail_alloc_bits(root->fs_info, block_group->flags); | |
80eb234a | 8514 | up_write(&block_group->space_info->groups_sem); |
1a40e23b | 8515 | |
817d52f8 | 8516 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 8517 | wait_block_group_cache_done(block_group); |
817d52f8 JB |
8518 | |
8519 | btrfs_remove_free_space_cache(block_group); | |
8520 | ||
c146afad YZ |
8521 | spin_lock(&block_group->space_info->lock); |
8522 | block_group->space_info->total_bytes -= block_group->key.offset; | |
8523 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 8524 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
c146afad | 8525 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 8526 | |
0af3d00b JB |
8527 | memcpy(&key, &block_group->key, sizeof(key)); |
8528 | ||
283bb197 | 8529 | btrfs_clear_space_info_full(root->fs_info); |
c146afad | 8530 | |
fa9c0d79 CM |
8531 | btrfs_put_block_group(block_group); |
8532 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
8533 | |
8534 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
8535 | if (ret > 0) | |
8536 | ret = -EIO; | |
8537 | if (ret < 0) | |
8538 | goto out; | |
8539 | ||
8540 | ret = btrfs_del_item(trans, root, path); | |
8541 | out: | |
8542 | btrfs_free_path(path); | |
8543 | return ret; | |
8544 | } | |
acce952b | 8545 | |
c59021f8 | 8546 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
8547 | { | |
8548 | struct btrfs_space_info *space_info; | |
1aba86d6 | 8549 | struct btrfs_super_block *disk_super; |
8550 | u64 features; | |
8551 | u64 flags; | |
8552 | int mixed = 0; | |
c59021f8 | 8553 | int ret; |
8554 | ||
6c41761f | 8555 | disk_super = fs_info->super_copy; |
1aba86d6 | 8556 | if (!btrfs_super_root(disk_super)) |
8557 | return 1; | |
c59021f8 | 8558 | |
1aba86d6 | 8559 | features = btrfs_super_incompat_flags(disk_super); |
8560 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
8561 | mixed = 1; | |
c59021f8 | 8562 | |
1aba86d6 | 8563 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
8564 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
c59021f8 | 8565 | if (ret) |
1aba86d6 | 8566 | goto out; |
c59021f8 | 8567 | |
1aba86d6 | 8568 | if (mixed) { |
8569 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
8570 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8571 | } else { | |
8572 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
8573 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8574 | if (ret) | |
8575 | goto out; | |
8576 | ||
8577 | flags = BTRFS_BLOCK_GROUP_DATA; | |
8578 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); | |
8579 | } | |
8580 | out: | |
c59021f8 | 8581 | return ret; |
8582 | } | |
8583 | ||
acce952b | 8584 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
8585 | { | |
8586 | return unpin_extent_range(root, start, end); | |
8587 | } | |
8588 | ||
8589 | int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr, | |
5378e607 | 8590 | u64 num_bytes, u64 *actual_bytes) |
acce952b | 8591 | { |
5378e607 | 8592 | return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes); |
acce952b | 8593 | } |
f7039b1d LD |
8594 | |
8595 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) | |
8596 | { | |
8597 | struct btrfs_fs_info *fs_info = root->fs_info; | |
8598 | struct btrfs_block_group_cache *cache = NULL; | |
8599 | u64 group_trimmed; | |
8600 | u64 start; | |
8601 | u64 end; | |
8602 | u64 trimmed = 0; | |
2cac13e4 | 8603 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
8604 | int ret = 0; |
8605 | ||
2cac13e4 LB |
8606 | /* |
8607 | * try to trim all FS space, our block group may start from non-zero. | |
8608 | */ | |
8609 | if (range->len == total_bytes) | |
8610 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
8611 | else | |
8612 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
8613 | |
8614 | while (cache) { | |
8615 | if (cache->key.objectid >= (range->start + range->len)) { | |
8616 | btrfs_put_block_group(cache); | |
8617 | break; | |
8618 | } | |
8619 | ||
8620 | start = max(range->start, cache->key.objectid); | |
8621 | end = min(range->start + range->len, | |
8622 | cache->key.objectid + cache->key.offset); | |
8623 | ||
8624 | if (end - start >= range->minlen) { | |
8625 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 8626 | ret = cache_block_group(cache, 0); |
f7039b1d LD |
8627 | if (!ret) |
8628 | wait_block_group_cache_done(cache); | |
8629 | } | |
8630 | ret = btrfs_trim_block_group(cache, | |
8631 | &group_trimmed, | |
8632 | start, | |
8633 | end, | |
8634 | range->minlen); | |
8635 | ||
8636 | trimmed += group_trimmed; | |
8637 | if (ret) { | |
8638 | btrfs_put_block_group(cache); | |
8639 | break; | |
8640 | } | |
8641 | } | |
8642 | ||
8643 | cache = next_block_group(fs_info->tree_root, cache); | |
8644 | } | |
8645 | ||
8646 | range->len = trimmed; | |
8647 | return ret; | |
8648 | } |