]>
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> |
f361bf4a | 19 | #include <linux/sched/signal.h> |
edbd8d4e | 20 | #include <linux/pagemap.h> |
ec44a35c | 21 | #include <linux/writeback.h> |
21af804c | 22 | #include <linux/blkdev.h> |
b7a9f29f | 23 | #include <linux/sort.h> |
4184ea7f | 24 | #include <linux/rcupdate.h> |
817d52f8 | 25 | #include <linux/kthread.h> |
5a0e3ad6 | 26 | #include <linux/slab.h> |
dff51cd1 | 27 | #include <linux/ratelimit.h> |
b150a4f1 | 28 | #include <linux/percpu_counter.h> |
69fe2d75 | 29 | #include <linux/lockdep.h> |
9678c543 | 30 | #include <linux/crc32c.h> |
995946dd | 31 | #include "tree-log.h" |
fec577fb CM |
32 | #include "disk-io.h" |
33 | #include "print-tree.h" | |
0b86a832 | 34 | #include "volumes.h" |
53b381b3 | 35 | #include "raid56.h" |
925baedd | 36 | #include "locking.h" |
fa9c0d79 | 37 | #include "free-space-cache.h" |
1e144fb8 | 38 | #include "free-space-tree.h" |
3fed40cc | 39 | #include "math.h" |
6ab0a202 | 40 | #include "sysfs.h" |
fcebe456 | 41 | #include "qgroup.h" |
fd708b81 | 42 | #include "ref-verify.h" |
fec577fb | 43 | |
709c0486 AJ |
44 | #undef SCRAMBLE_DELAYED_REFS |
45 | ||
9e622d6b MX |
46 | /* |
47 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
48 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
49 | * if we really need one. | |
50 | * | |
0e4f8f88 CM |
51 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
52 | * if we have very few chunks already allocated. This is | |
53 | * used as part of the clustering code to help make sure | |
54 | * we have a good pool of storage to cluster in, without | |
55 | * filling the FS with empty chunks | |
56 | * | |
9e622d6b MX |
57 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
58 | * | |
0e4f8f88 CM |
59 | */ |
60 | enum { | |
61 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
62 | CHUNK_ALLOC_LIMITED = 1, |
63 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
64 | }; |
65 | ||
5d4f98a2 | 66 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 67 | struct btrfs_fs_info *fs_info, |
c682f9b3 | 68 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
69 | u64 root_objectid, u64 owner_objectid, |
70 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 71 | struct btrfs_delayed_extent_op *extra_op); |
5d4f98a2 YZ |
72 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
73 | struct extent_buffer *leaf, | |
74 | struct btrfs_extent_item *ei); | |
75 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
2ff7e61e | 76 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
77 | u64 parent, u64 root_objectid, |
78 | u64 flags, u64 owner, u64 offset, | |
79 | struct btrfs_key *ins, int ref_mod); | |
80 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
2ff7e61e | 81 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
82 | u64 parent, u64 root_objectid, |
83 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 84 | int level, struct btrfs_key *ins); |
6a63209f | 85 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 86 | struct btrfs_fs_info *fs_info, u64 flags, |
698d0082 | 87 | int force); |
11833d66 YZ |
88 | static int find_next_key(struct btrfs_path *path, int level, |
89 | struct btrfs_key *key); | |
ab8d0fc4 JM |
90 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
91 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 92 | int dump_block_groups); |
5d80366e JB |
93 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
94 | u64 num_bytes); | |
957780eb JB |
95 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, |
96 | struct btrfs_space_info *space_info, | |
97 | u64 num_bytes); | |
98 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
99 | struct btrfs_space_info *space_info, | |
100 | u64 num_bytes); | |
6a63209f | 101 | |
817d52f8 JB |
102 | static noinline int |
103 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
104 | { | |
105 | smp_mb(); | |
36cce922 JB |
106 | return cache->cached == BTRFS_CACHE_FINISHED || |
107 | cache->cached == BTRFS_CACHE_ERROR; | |
817d52f8 JB |
108 | } |
109 | ||
0f9dd46c JB |
110 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
111 | { | |
112 | return (cache->flags & bits) == bits; | |
113 | } | |
114 | ||
758f2dfc | 115 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
116 | { |
117 | atomic_inc(&cache->count); | |
118 | } | |
119 | ||
120 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
121 | { | |
f0486c68 YZ |
122 | if (atomic_dec_and_test(&cache->count)) { |
123 | WARN_ON(cache->pinned > 0); | |
124 | WARN_ON(cache->reserved > 0); | |
0966a7b1 QW |
125 | |
126 | /* | |
127 | * If not empty, someone is still holding mutex of | |
128 | * full_stripe_lock, which can only be released by caller. | |
129 | * And it will definitely cause use-after-free when caller | |
130 | * tries to release full stripe lock. | |
131 | * | |
132 | * No better way to resolve, but only to warn. | |
133 | */ | |
134 | WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); | |
34d52cb6 | 135 | kfree(cache->free_space_ctl); |
11dfe35a | 136 | kfree(cache); |
f0486c68 | 137 | } |
11dfe35a JB |
138 | } |
139 | ||
0f9dd46c JB |
140 | /* |
141 | * this adds the block group to the fs_info rb tree for the block group | |
142 | * cache | |
143 | */ | |
b2950863 | 144 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
145 | struct btrfs_block_group_cache *block_group) |
146 | { | |
147 | struct rb_node **p; | |
148 | struct rb_node *parent = NULL; | |
149 | struct btrfs_block_group_cache *cache; | |
150 | ||
151 | spin_lock(&info->block_group_cache_lock); | |
152 | p = &info->block_group_cache_tree.rb_node; | |
153 | ||
154 | while (*p) { | |
155 | parent = *p; | |
156 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
157 | cache_node); | |
158 | if (block_group->key.objectid < cache->key.objectid) { | |
159 | p = &(*p)->rb_left; | |
160 | } else if (block_group->key.objectid > cache->key.objectid) { | |
161 | p = &(*p)->rb_right; | |
162 | } else { | |
163 | spin_unlock(&info->block_group_cache_lock); | |
164 | return -EEXIST; | |
165 | } | |
166 | } | |
167 | ||
168 | rb_link_node(&block_group->cache_node, parent, p); | |
169 | rb_insert_color(&block_group->cache_node, | |
170 | &info->block_group_cache_tree); | |
a1897fdd LB |
171 | |
172 | if (info->first_logical_byte > block_group->key.objectid) | |
173 | info->first_logical_byte = block_group->key.objectid; | |
174 | ||
0f9dd46c JB |
175 | spin_unlock(&info->block_group_cache_lock); |
176 | ||
177 | return 0; | |
178 | } | |
179 | ||
180 | /* | |
181 | * This will return the block group at or after bytenr if contains is 0, else | |
182 | * it will return the block group that contains the bytenr | |
183 | */ | |
184 | static struct btrfs_block_group_cache * | |
185 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
186 | int contains) | |
187 | { | |
188 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
189 | struct rb_node *n; | |
190 | u64 end, start; | |
191 | ||
192 | spin_lock(&info->block_group_cache_lock); | |
193 | n = info->block_group_cache_tree.rb_node; | |
194 | ||
195 | while (n) { | |
196 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
197 | cache_node); | |
198 | end = cache->key.objectid + cache->key.offset - 1; | |
199 | start = cache->key.objectid; | |
200 | ||
201 | if (bytenr < start) { | |
202 | if (!contains && (!ret || start < ret->key.objectid)) | |
203 | ret = cache; | |
204 | n = n->rb_left; | |
205 | } else if (bytenr > start) { | |
206 | if (contains && bytenr <= end) { | |
207 | ret = cache; | |
208 | break; | |
209 | } | |
210 | n = n->rb_right; | |
211 | } else { | |
212 | ret = cache; | |
213 | break; | |
214 | } | |
215 | } | |
a1897fdd | 216 | if (ret) { |
11dfe35a | 217 | btrfs_get_block_group(ret); |
a1897fdd LB |
218 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
219 | info->first_logical_byte = ret->key.objectid; | |
220 | } | |
0f9dd46c JB |
221 | spin_unlock(&info->block_group_cache_lock); |
222 | ||
223 | return ret; | |
224 | } | |
225 | ||
2ff7e61e | 226 | static int add_excluded_extent(struct btrfs_fs_info *fs_info, |
11833d66 | 227 | u64 start, u64 num_bytes) |
817d52f8 | 228 | { |
11833d66 | 229 | u64 end = start + num_bytes - 1; |
0b246afa | 230 | set_extent_bits(&fs_info->freed_extents[0], |
ceeb0ae7 | 231 | start, end, EXTENT_UPTODATE); |
0b246afa | 232 | set_extent_bits(&fs_info->freed_extents[1], |
ceeb0ae7 | 233 | start, end, EXTENT_UPTODATE); |
11833d66 YZ |
234 | return 0; |
235 | } | |
817d52f8 | 236 | |
2ff7e61e | 237 | static void free_excluded_extents(struct btrfs_fs_info *fs_info, |
11833d66 YZ |
238 | struct btrfs_block_group_cache *cache) |
239 | { | |
240 | u64 start, end; | |
817d52f8 | 241 | |
11833d66 YZ |
242 | start = cache->key.objectid; |
243 | end = start + cache->key.offset - 1; | |
244 | ||
0b246afa | 245 | clear_extent_bits(&fs_info->freed_extents[0], |
91166212 | 246 | start, end, EXTENT_UPTODATE); |
0b246afa | 247 | clear_extent_bits(&fs_info->freed_extents[1], |
91166212 | 248 | start, end, EXTENT_UPTODATE); |
817d52f8 JB |
249 | } |
250 | ||
2ff7e61e | 251 | static int exclude_super_stripes(struct btrfs_fs_info *fs_info, |
11833d66 | 252 | struct btrfs_block_group_cache *cache) |
817d52f8 | 253 | { |
817d52f8 JB |
254 | u64 bytenr; |
255 | u64 *logical; | |
256 | int stripe_len; | |
257 | int i, nr, ret; | |
258 | ||
06b2331f YZ |
259 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
260 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
261 | cache->bytes_super += stripe_len; | |
2ff7e61e | 262 | ret = add_excluded_extent(fs_info, cache->key.objectid, |
06b2331f | 263 | stripe_len); |
835d974f JB |
264 | if (ret) |
265 | return ret; | |
06b2331f YZ |
266 | } |
267 | ||
817d52f8 JB |
268 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
269 | bytenr = btrfs_sb_offset(i); | |
0b246afa | 270 | ret = btrfs_rmap_block(fs_info, cache->key.objectid, |
ab8d0fc4 | 271 | bytenr, 0, &logical, &nr, &stripe_len); |
835d974f JB |
272 | if (ret) |
273 | return ret; | |
11833d66 | 274 | |
817d52f8 | 275 | while (nr--) { |
51bf5f0b JB |
276 | u64 start, len; |
277 | ||
278 | if (logical[nr] > cache->key.objectid + | |
279 | cache->key.offset) | |
280 | continue; | |
281 | ||
282 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
283 | continue; | |
284 | ||
285 | start = logical[nr]; | |
286 | if (start < cache->key.objectid) { | |
287 | start = cache->key.objectid; | |
288 | len = (logical[nr] + stripe_len) - start; | |
289 | } else { | |
290 | len = min_t(u64, stripe_len, | |
291 | cache->key.objectid + | |
292 | cache->key.offset - start); | |
293 | } | |
294 | ||
295 | cache->bytes_super += len; | |
2ff7e61e | 296 | ret = add_excluded_extent(fs_info, start, len); |
835d974f JB |
297 | if (ret) { |
298 | kfree(logical); | |
299 | return ret; | |
300 | } | |
817d52f8 | 301 | } |
11833d66 | 302 | |
817d52f8 JB |
303 | kfree(logical); |
304 | } | |
817d52f8 JB |
305 | return 0; |
306 | } | |
307 | ||
11833d66 YZ |
308 | static struct btrfs_caching_control * |
309 | get_caching_control(struct btrfs_block_group_cache *cache) | |
310 | { | |
311 | struct btrfs_caching_control *ctl; | |
312 | ||
313 | spin_lock(&cache->lock); | |
dde5abee JB |
314 | if (!cache->caching_ctl) { |
315 | spin_unlock(&cache->lock); | |
11833d66 YZ |
316 | return NULL; |
317 | } | |
318 | ||
319 | ctl = cache->caching_ctl; | |
1e4f4714 | 320 | refcount_inc(&ctl->count); |
11833d66 YZ |
321 | spin_unlock(&cache->lock); |
322 | return ctl; | |
323 | } | |
324 | ||
325 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
326 | { | |
1e4f4714 | 327 | if (refcount_dec_and_test(&ctl->count)) |
11833d66 YZ |
328 | kfree(ctl); |
329 | } | |
330 | ||
d0bd4560 | 331 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 332 | static void fragment_free_space(struct btrfs_block_group_cache *block_group) |
d0bd4560 | 333 | { |
2ff7e61e | 334 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
d0bd4560 JB |
335 | u64 start = block_group->key.objectid; |
336 | u64 len = block_group->key.offset; | |
337 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? | |
0b246afa | 338 | fs_info->nodesize : fs_info->sectorsize; |
d0bd4560 JB |
339 | u64 step = chunk << 1; |
340 | ||
341 | while (len > chunk) { | |
342 | btrfs_remove_free_space(block_group, start, chunk); | |
343 | start += step; | |
344 | if (len < step) | |
345 | len = 0; | |
346 | else | |
347 | len -= step; | |
348 | } | |
349 | } | |
350 | #endif | |
351 | ||
0f9dd46c JB |
352 | /* |
353 | * this is only called by cache_block_group, since we could have freed extents | |
354 | * we need to check the pinned_extents for any extents that can't be used yet | |
355 | * since their free space will be released as soon as the transaction commits. | |
356 | */ | |
a5ed9182 OS |
357 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
358 | struct btrfs_fs_info *info, u64 start, u64 end) | |
0f9dd46c | 359 | { |
817d52f8 | 360 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
361 | int ret; |
362 | ||
363 | while (start < end) { | |
11833d66 | 364 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 365 | &extent_start, &extent_end, |
e6138876 JB |
366 | EXTENT_DIRTY | EXTENT_UPTODATE, |
367 | NULL); | |
0f9dd46c JB |
368 | if (ret) |
369 | break; | |
370 | ||
06b2331f | 371 | if (extent_start <= start) { |
0f9dd46c JB |
372 | start = extent_end + 1; |
373 | } else if (extent_start > start && extent_start < end) { | |
374 | size = extent_start - start; | |
817d52f8 | 375 | total_added += size; |
ea6a478e JB |
376 | ret = btrfs_add_free_space(block_group, start, |
377 | size); | |
79787eaa | 378 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
379 | start = extent_end + 1; |
380 | } else { | |
381 | break; | |
382 | } | |
383 | } | |
384 | ||
385 | if (start < end) { | |
386 | size = end - start; | |
817d52f8 | 387 | total_added += size; |
ea6a478e | 388 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 389 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
390 | } |
391 | ||
817d52f8 | 392 | return total_added; |
0f9dd46c JB |
393 | } |
394 | ||
73fa48b6 | 395 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) |
e37c9e69 | 396 | { |
0b246afa JM |
397 | struct btrfs_block_group_cache *block_group = caching_ctl->block_group; |
398 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
399 | struct btrfs_root *extent_root = fs_info->extent_root; | |
e37c9e69 | 400 | struct btrfs_path *path; |
5f39d397 | 401 | struct extent_buffer *leaf; |
11833d66 | 402 | struct btrfs_key key; |
817d52f8 | 403 | u64 total_found = 0; |
11833d66 YZ |
404 | u64 last = 0; |
405 | u32 nritems; | |
73fa48b6 | 406 | int ret; |
d0bd4560 | 407 | bool wakeup = true; |
f510cfec | 408 | |
e37c9e69 CM |
409 | path = btrfs_alloc_path(); |
410 | if (!path) | |
73fa48b6 | 411 | return -ENOMEM; |
7d7d6068 | 412 | |
817d52f8 | 413 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 414 | |
d0bd4560 JB |
415 | #ifdef CONFIG_BTRFS_DEBUG |
416 | /* | |
417 | * If we're fragmenting we don't want to make anybody think we can | |
418 | * allocate from this block group until we've had a chance to fragment | |
419 | * the free space. | |
420 | */ | |
2ff7e61e | 421 | if (btrfs_should_fragment_free_space(block_group)) |
d0bd4560 JB |
422 | wakeup = false; |
423 | #endif | |
5cd57b2c | 424 | /* |
817d52f8 JB |
425 | * We don't want to deadlock with somebody trying to allocate a new |
426 | * extent for the extent root while also trying to search the extent | |
427 | * root to add free space. So we skip locking and search the commit | |
428 | * root, since its read-only | |
5cd57b2c CM |
429 | */ |
430 | path->skip_locking = 1; | |
817d52f8 | 431 | path->search_commit_root = 1; |
e4058b54 | 432 | path->reada = READA_FORWARD; |
817d52f8 | 433 | |
e4404d6e | 434 | key.objectid = last; |
e37c9e69 | 435 | key.offset = 0; |
11833d66 | 436 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 437 | |
52ee28d2 | 438 | next: |
11833d66 | 439 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 440 | if (ret < 0) |
73fa48b6 | 441 | goto out; |
a512bbf8 | 442 | |
11833d66 YZ |
443 | leaf = path->nodes[0]; |
444 | nritems = btrfs_header_nritems(leaf); | |
445 | ||
d397712b | 446 | while (1) { |
7841cb28 | 447 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 448 | last = (u64)-1; |
817d52f8 | 449 | break; |
f25784b3 | 450 | } |
817d52f8 | 451 | |
11833d66 YZ |
452 | if (path->slots[0] < nritems) { |
453 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
454 | } else { | |
455 | ret = find_next_key(path, 0, &key); | |
456 | if (ret) | |
e37c9e69 | 457 | break; |
817d52f8 | 458 | |
c9ea7b24 | 459 | if (need_resched() || |
9e351cc8 | 460 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
d0bd4560 JB |
461 | if (wakeup) |
462 | caching_ctl->progress = last; | |
ff5714cc | 463 | btrfs_release_path(path); |
9e351cc8 | 464 | up_read(&fs_info->commit_root_sem); |
589d8ade | 465 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 466 | cond_resched(); |
73fa48b6 OS |
467 | mutex_lock(&caching_ctl->mutex); |
468 | down_read(&fs_info->commit_root_sem); | |
469 | goto next; | |
589d8ade | 470 | } |
0a3896d0 JB |
471 | |
472 | ret = btrfs_next_leaf(extent_root, path); | |
473 | if (ret < 0) | |
73fa48b6 | 474 | goto out; |
0a3896d0 JB |
475 | if (ret) |
476 | break; | |
589d8ade JB |
477 | leaf = path->nodes[0]; |
478 | nritems = btrfs_header_nritems(leaf); | |
479 | continue; | |
11833d66 | 480 | } |
817d52f8 | 481 | |
52ee28d2 LB |
482 | if (key.objectid < last) { |
483 | key.objectid = last; | |
484 | key.offset = 0; | |
485 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
486 | ||
d0bd4560 JB |
487 | if (wakeup) |
488 | caching_ctl->progress = last; | |
52ee28d2 LB |
489 | btrfs_release_path(path); |
490 | goto next; | |
491 | } | |
492 | ||
11833d66 YZ |
493 | if (key.objectid < block_group->key.objectid) { |
494 | path->slots[0]++; | |
817d52f8 | 495 | continue; |
e37c9e69 | 496 | } |
0f9dd46c | 497 | |
e37c9e69 | 498 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 499 | block_group->key.offset) |
e37c9e69 | 500 | break; |
7d7d6068 | 501 | |
3173a18f JB |
502 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
503 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
817d52f8 JB |
504 | total_found += add_new_free_space(block_group, |
505 | fs_info, last, | |
506 | key.objectid); | |
3173a18f JB |
507 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
508 | last = key.objectid + | |
da17066c | 509 | fs_info->nodesize; |
3173a18f JB |
510 | else |
511 | last = key.objectid + key.offset; | |
817d52f8 | 512 | |
73fa48b6 | 513 | if (total_found > CACHING_CTL_WAKE_UP) { |
11833d66 | 514 | total_found = 0; |
d0bd4560 JB |
515 | if (wakeup) |
516 | wake_up(&caching_ctl->wait); | |
11833d66 | 517 | } |
817d52f8 | 518 | } |
e37c9e69 CM |
519 | path->slots[0]++; |
520 | } | |
817d52f8 | 521 | ret = 0; |
e37c9e69 | 522 | |
817d52f8 JB |
523 | total_found += add_new_free_space(block_group, fs_info, last, |
524 | block_group->key.objectid + | |
525 | block_group->key.offset); | |
11833d66 | 526 | caching_ctl->progress = (u64)-1; |
817d52f8 | 527 | |
73fa48b6 OS |
528 | out: |
529 | btrfs_free_path(path); | |
530 | return ret; | |
531 | } | |
532 | ||
533 | static noinline void caching_thread(struct btrfs_work *work) | |
534 | { | |
535 | struct btrfs_block_group_cache *block_group; | |
536 | struct btrfs_fs_info *fs_info; | |
537 | struct btrfs_caching_control *caching_ctl; | |
b4570aa9 | 538 | struct btrfs_root *extent_root; |
73fa48b6 OS |
539 | int ret; |
540 | ||
541 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
542 | block_group = caching_ctl->block_group; | |
543 | fs_info = block_group->fs_info; | |
b4570aa9 | 544 | extent_root = fs_info->extent_root; |
73fa48b6 OS |
545 | |
546 | mutex_lock(&caching_ctl->mutex); | |
547 | down_read(&fs_info->commit_root_sem); | |
548 | ||
1e144fb8 OS |
549 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
550 | ret = load_free_space_tree(caching_ctl); | |
551 | else | |
552 | ret = load_extent_tree_free(caching_ctl); | |
73fa48b6 | 553 | |
817d52f8 | 554 | spin_lock(&block_group->lock); |
11833d66 | 555 | block_group->caching_ctl = NULL; |
73fa48b6 | 556 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; |
817d52f8 | 557 | spin_unlock(&block_group->lock); |
0f9dd46c | 558 | |
d0bd4560 | 559 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 560 | if (btrfs_should_fragment_free_space(block_group)) { |
d0bd4560 JB |
561 | u64 bytes_used; |
562 | ||
563 | spin_lock(&block_group->space_info->lock); | |
564 | spin_lock(&block_group->lock); | |
565 | bytes_used = block_group->key.offset - | |
566 | btrfs_block_group_used(&block_group->item); | |
567 | block_group->space_info->bytes_used += bytes_used >> 1; | |
568 | spin_unlock(&block_group->lock); | |
569 | spin_unlock(&block_group->space_info->lock); | |
2ff7e61e | 570 | fragment_free_space(block_group); |
d0bd4560 JB |
571 | } |
572 | #endif | |
573 | ||
574 | caching_ctl->progress = (u64)-1; | |
11833d66 | 575 | |
9e351cc8 | 576 | up_read(&fs_info->commit_root_sem); |
2ff7e61e | 577 | free_excluded_extents(fs_info, block_group); |
11833d66 | 578 | mutex_unlock(&caching_ctl->mutex); |
73fa48b6 | 579 | |
11833d66 YZ |
580 | wake_up(&caching_ctl->wait); |
581 | ||
582 | put_caching_control(caching_ctl); | |
11dfe35a | 583 | btrfs_put_block_group(block_group); |
817d52f8 JB |
584 | } |
585 | ||
9d66e233 | 586 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 587 | int load_cache_only) |
817d52f8 | 588 | { |
291c7d2f | 589 | DEFINE_WAIT(wait); |
11833d66 YZ |
590 | struct btrfs_fs_info *fs_info = cache->fs_info; |
591 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
592 | int ret = 0; |
593 | ||
291c7d2f | 594 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
595 | if (!caching_ctl) |
596 | return -ENOMEM; | |
291c7d2f JB |
597 | |
598 | INIT_LIST_HEAD(&caching_ctl->list); | |
599 | mutex_init(&caching_ctl->mutex); | |
600 | init_waitqueue_head(&caching_ctl->wait); | |
601 | caching_ctl->block_group = cache; | |
602 | caching_ctl->progress = cache->key.objectid; | |
1e4f4714 | 603 | refcount_set(&caching_ctl->count, 1); |
9e0af237 LB |
604 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, |
605 | caching_thread, NULL, NULL); | |
291c7d2f JB |
606 | |
607 | spin_lock(&cache->lock); | |
608 | /* | |
609 | * This should be a rare occasion, but this could happen I think in the | |
610 | * case where one thread starts to load the space cache info, and then | |
611 | * some other thread starts a transaction commit which tries to do an | |
612 | * allocation while the other thread is still loading the space cache | |
613 | * info. The previous loop should have kept us from choosing this block | |
614 | * group, but if we've moved to the state where we will wait on caching | |
615 | * block groups we need to first check if we're doing a fast load here, | |
616 | * so we can wait for it to finish, otherwise we could end up allocating | |
617 | * from a block group who's cache gets evicted for one reason or | |
618 | * another. | |
619 | */ | |
620 | while (cache->cached == BTRFS_CACHE_FAST) { | |
621 | struct btrfs_caching_control *ctl; | |
622 | ||
623 | ctl = cache->caching_ctl; | |
1e4f4714 | 624 | refcount_inc(&ctl->count); |
291c7d2f JB |
625 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); |
626 | spin_unlock(&cache->lock); | |
627 | ||
628 | schedule(); | |
629 | ||
630 | finish_wait(&ctl->wait, &wait); | |
631 | put_caching_control(ctl); | |
632 | spin_lock(&cache->lock); | |
633 | } | |
634 | ||
635 | if (cache->cached != BTRFS_CACHE_NO) { | |
636 | spin_unlock(&cache->lock); | |
637 | kfree(caching_ctl); | |
11833d66 | 638 | return 0; |
291c7d2f JB |
639 | } |
640 | WARN_ON(cache->caching_ctl); | |
641 | cache->caching_ctl = caching_ctl; | |
642 | cache->cached = BTRFS_CACHE_FAST; | |
643 | spin_unlock(&cache->lock); | |
11833d66 | 644 | |
d8953d69 | 645 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cb83b7b8 | 646 | mutex_lock(&caching_ctl->mutex); |
9d66e233 JB |
647 | ret = load_free_space_cache(fs_info, cache); |
648 | ||
649 | spin_lock(&cache->lock); | |
650 | if (ret == 1) { | |
291c7d2f | 651 | cache->caching_ctl = NULL; |
9d66e233 JB |
652 | cache->cached = BTRFS_CACHE_FINISHED; |
653 | cache->last_byte_to_unpin = (u64)-1; | |
cb83b7b8 | 654 | caching_ctl->progress = (u64)-1; |
9d66e233 | 655 | } else { |
291c7d2f JB |
656 | if (load_cache_only) { |
657 | cache->caching_ctl = NULL; | |
658 | cache->cached = BTRFS_CACHE_NO; | |
659 | } else { | |
660 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 661 | cache->has_caching_ctl = 1; |
291c7d2f | 662 | } |
9d66e233 JB |
663 | } |
664 | spin_unlock(&cache->lock); | |
d0bd4560 JB |
665 | #ifdef CONFIG_BTRFS_DEBUG |
666 | if (ret == 1 && | |
2ff7e61e | 667 | btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
668 | u64 bytes_used; |
669 | ||
670 | spin_lock(&cache->space_info->lock); | |
671 | spin_lock(&cache->lock); | |
672 | bytes_used = cache->key.offset - | |
673 | btrfs_block_group_used(&cache->item); | |
674 | cache->space_info->bytes_used += bytes_used >> 1; | |
675 | spin_unlock(&cache->lock); | |
676 | spin_unlock(&cache->space_info->lock); | |
2ff7e61e | 677 | fragment_free_space(cache); |
d0bd4560 JB |
678 | } |
679 | #endif | |
cb83b7b8 JB |
680 | mutex_unlock(&caching_ctl->mutex); |
681 | ||
291c7d2f | 682 | wake_up(&caching_ctl->wait); |
3c14874a | 683 | if (ret == 1) { |
291c7d2f | 684 | put_caching_control(caching_ctl); |
2ff7e61e | 685 | free_excluded_extents(fs_info, cache); |
9d66e233 | 686 | return 0; |
3c14874a | 687 | } |
291c7d2f JB |
688 | } else { |
689 | /* | |
1e144fb8 OS |
690 | * We're either using the free space tree or no caching at all. |
691 | * Set cached to the appropriate value and wakeup any waiters. | |
291c7d2f JB |
692 | */ |
693 | spin_lock(&cache->lock); | |
694 | if (load_cache_only) { | |
695 | cache->caching_ctl = NULL; | |
696 | cache->cached = BTRFS_CACHE_NO; | |
697 | } else { | |
698 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 699 | cache->has_caching_ctl = 1; |
291c7d2f JB |
700 | } |
701 | spin_unlock(&cache->lock); | |
702 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
703 | } |
704 | ||
291c7d2f JB |
705 | if (load_cache_only) { |
706 | put_caching_control(caching_ctl); | |
11833d66 | 707 | return 0; |
817d52f8 | 708 | } |
817d52f8 | 709 | |
9e351cc8 | 710 | down_write(&fs_info->commit_root_sem); |
1e4f4714 | 711 | refcount_inc(&caching_ctl->count); |
11833d66 | 712 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
9e351cc8 | 713 | up_write(&fs_info->commit_root_sem); |
11833d66 | 714 | |
11dfe35a | 715 | btrfs_get_block_group(cache); |
11833d66 | 716 | |
e66f0bb1 | 717 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 718 | |
ef8bbdfe | 719 | return ret; |
e37c9e69 CM |
720 | } |
721 | ||
0f9dd46c JB |
722 | /* |
723 | * return the block group that starts at or after bytenr | |
724 | */ | |
d397712b CM |
725 | static struct btrfs_block_group_cache * |
726 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 727 | { |
e2c89907 | 728 | return block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b CM |
729 | } |
730 | ||
0f9dd46c | 731 | /* |
9f55684c | 732 | * return the block group that contains the given bytenr |
0f9dd46c | 733 | */ |
d397712b CM |
734 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
735 | struct btrfs_fs_info *info, | |
736 | u64 bytenr) | |
be744175 | 737 | { |
e2c89907 | 738 | return block_group_cache_tree_search(info, bytenr, 1); |
be744175 | 739 | } |
0b86a832 | 740 | |
0f9dd46c JB |
741 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
742 | u64 flags) | |
6324fbf3 | 743 | { |
0f9dd46c | 744 | struct list_head *head = &info->space_info; |
0f9dd46c | 745 | struct btrfs_space_info *found; |
4184ea7f | 746 | |
52ba6929 | 747 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 748 | |
4184ea7f CM |
749 | rcu_read_lock(); |
750 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 751 | if (found->flags & flags) { |
4184ea7f | 752 | rcu_read_unlock(); |
0f9dd46c | 753 | return found; |
4184ea7f | 754 | } |
0f9dd46c | 755 | } |
4184ea7f | 756 | rcu_read_unlock(); |
0f9dd46c | 757 | return NULL; |
6324fbf3 CM |
758 | } |
759 | ||
0d9f824d OS |
760 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes, |
761 | u64 owner, u64 root_objectid) | |
762 | { | |
763 | struct btrfs_space_info *space_info; | |
764 | u64 flags; | |
765 | ||
766 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
767 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) | |
768 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
769 | else | |
770 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
771 | } else { | |
772 | flags = BTRFS_BLOCK_GROUP_DATA; | |
773 | } | |
774 | ||
775 | space_info = __find_space_info(fs_info, flags); | |
55e8196a | 776 | ASSERT(space_info); |
0d9f824d OS |
777 | percpu_counter_add(&space_info->total_bytes_pinned, num_bytes); |
778 | } | |
779 | ||
4184ea7f CM |
780 | /* |
781 | * after adding space to the filesystem, we need to clear the full flags | |
782 | * on all the space infos. | |
783 | */ | |
784 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
785 | { | |
786 | struct list_head *head = &info->space_info; | |
787 | struct btrfs_space_info *found; | |
788 | ||
789 | rcu_read_lock(); | |
790 | list_for_each_entry_rcu(found, head, list) | |
791 | found->full = 0; | |
792 | rcu_read_unlock(); | |
793 | } | |
794 | ||
1a4ed8fd | 795 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 796 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
797 | { |
798 | int ret; | |
799 | struct btrfs_key key; | |
31840ae1 | 800 | struct btrfs_path *path; |
e02119d5 | 801 | |
31840ae1 | 802 | path = btrfs_alloc_path(); |
d8926bb3 MF |
803 | if (!path) |
804 | return -ENOMEM; | |
805 | ||
e02119d5 CM |
806 | key.objectid = start; |
807 | key.offset = len; | |
3173a18f | 808 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 809 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 810 | btrfs_free_path(path); |
7bb86316 CM |
811 | return ret; |
812 | } | |
813 | ||
a22285a6 | 814 | /* |
3173a18f | 815 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
816 | * |
817 | * the head node for delayed ref is used to store the sum of all the | |
818 | * reference count modifications queued up in the rbtree. the head | |
819 | * node may also store the extent flags to set. This way you can check | |
820 | * to see what the reference count and extent flags would be if all of | |
821 | * the delayed refs are not processed. | |
822 | */ | |
823 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 824 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 825 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
826 | { |
827 | struct btrfs_delayed_ref_head *head; | |
828 | struct btrfs_delayed_ref_root *delayed_refs; | |
829 | struct btrfs_path *path; | |
830 | struct btrfs_extent_item *ei; | |
831 | struct extent_buffer *leaf; | |
832 | struct btrfs_key key; | |
833 | u32 item_size; | |
834 | u64 num_refs; | |
835 | u64 extent_flags; | |
836 | int ret; | |
837 | ||
3173a18f JB |
838 | /* |
839 | * If we don't have skinny metadata, don't bother doing anything | |
840 | * different | |
841 | */ | |
0b246afa JM |
842 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
843 | offset = fs_info->nodesize; | |
3173a18f JB |
844 | metadata = 0; |
845 | } | |
846 | ||
a22285a6 YZ |
847 | path = btrfs_alloc_path(); |
848 | if (!path) | |
849 | return -ENOMEM; | |
850 | ||
a22285a6 YZ |
851 | if (!trans) { |
852 | path->skip_locking = 1; | |
853 | path->search_commit_root = 1; | |
854 | } | |
639eefc8 FDBM |
855 | |
856 | search_again: | |
857 | key.objectid = bytenr; | |
858 | key.offset = offset; | |
859 | if (metadata) | |
860 | key.type = BTRFS_METADATA_ITEM_KEY; | |
861 | else | |
862 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
863 | ||
0b246afa | 864 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
865 | if (ret < 0) |
866 | goto out_free; | |
867 | ||
3173a18f | 868 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
869 | if (path->slots[0]) { |
870 | path->slots[0]--; | |
871 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
872 | path->slots[0]); | |
873 | if (key.objectid == bytenr && | |
874 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 875 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
876 | ret = 0; |
877 | } | |
3173a18f JB |
878 | } |
879 | ||
a22285a6 YZ |
880 | if (ret == 0) { |
881 | leaf = path->nodes[0]; | |
882 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
883 | if (item_size >= sizeof(*ei)) { | |
884 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
885 | struct btrfs_extent_item); | |
886 | num_refs = btrfs_extent_refs(leaf, ei); | |
887 | extent_flags = btrfs_extent_flags(leaf, ei); | |
888 | } else { | |
889 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
890 | struct btrfs_extent_item_v0 *ei0; | |
891 | BUG_ON(item_size != sizeof(*ei0)); | |
892 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
893 | struct btrfs_extent_item_v0); | |
894 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
895 | /* FIXME: this isn't correct for data */ | |
896 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
897 | #else | |
898 | BUG(); | |
899 | #endif | |
900 | } | |
901 | BUG_ON(num_refs == 0); | |
902 | } else { | |
903 | num_refs = 0; | |
904 | extent_flags = 0; | |
905 | ret = 0; | |
906 | } | |
907 | ||
908 | if (!trans) | |
909 | goto out; | |
910 | ||
911 | delayed_refs = &trans->transaction->delayed_refs; | |
912 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 913 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
914 | if (head) { |
915 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 916 | refcount_inc(&head->refs); |
a22285a6 YZ |
917 | spin_unlock(&delayed_refs->lock); |
918 | ||
b3b4aa74 | 919 | btrfs_release_path(path); |
a22285a6 | 920 | |
8cc33e5c DS |
921 | /* |
922 | * Mutex was contended, block until it's released and try | |
923 | * again | |
924 | */ | |
a22285a6 YZ |
925 | mutex_lock(&head->mutex); |
926 | mutex_unlock(&head->mutex); | |
d278850e | 927 | btrfs_put_delayed_ref_head(head); |
639eefc8 | 928 | goto search_again; |
a22285a6 | 929 | } |
d7df2c79 | 930 | spin_lock(&head->lock); |
a22285a6 YZ |
931 | if (head->extent_op && head->extent_op->update_flags) |
932 | extent_flags |= head->extent_op->flags_to_set; | |
933 | else | |
934 | BUG_ON(num_refs == 0); | |
935 | ||
d278850e | 936 | num_refs += head->ref_mod; |
d7df2c79 | 937 | spin_unlock(&head->lock); |
a22285a6 YZ |
938 | mutex_unlock(&head->mutex); |
939 | } | |
940 | spin_unlock(&delayed_refs->lock); | |
941 | out: | |
942 | WARN_ON(num_refs == 0); | |
943 | if (refs) | |
944 | *refs = num_refs; | |
945 | if (flags) | |
946 | *flags = extent_flags; | |
947 | out_free: | |
948 | btrfs_free_path(path); | |
949 | return ret; | |
950 | } | |
951 | ||
d8d5f3e1 CM |
952 | /* |
953 | * Back reference rules. Back refs have three main goals: | |
954 | * | |
955 | * 1) differentiate between all holders of references to an extent so that | |
956 | * when a reference is dropped we can make sure it was a valid reference | |
957 | * before freeing the extent. | |
958 | * | |
959 | * 2) Provide enough information to quickly find the holders of an extent | |
960 | * if we notice a given block is corrupted or bad. | |
961 | * | |
962 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
963 | * maintenance. This is actually the same as #2, but with a slightly | |
964 | * different use case. | |
965 | * | |
5d4f98a2 YZ |
966 | * There are two kinds of back refs. The implicit back refs is optimized |
967 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
968 | * back refs of this kind provide information about the block's owner tree | |
969 | * and the pointer's key. These information allow us to find the block by | |
970 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
971 | * referenced by their owner trees. The location of tree block is recorded | |
972 | * in the back refs. Actually the full back refs is generic, and can be | |
973 | * used in all cases the implicit back refs is used. The major shortcoming | |
974 | * of the full back refs is its overhead. Every time a tree block gets | |
975 | * COWed, we have to update back refs entry for all pointers in it. | |
976 | * | |
977 | * For a newly allocated tree block, we use implicit back refs for | |
978 | * pointers in it. This means most tree related operations only involve | |
979 | * implicit back refs. For a tree block created in old transaction, the | |
980 | * only way to drop a reference to it is COW it. So we can detect the | |
981 | * event that tree block loses its owner tree's reference and do the | |
982 | * back refs conversion. | |
983 | * | |
01327610 | 984 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
985 | * |
986 | * The reference count of the block is one and the tree is the block's | |
987 | * owner tree. Nothing to do in this case. | |
988 | * | |
989 | * The reference count of the block is one and the tree is not the | |
990 | * block's owner tree. In this case, full back refs is used for pointers | |
991 | * in the block. Remove these full back refs, add implicit back refs for | |
992 | * every pointers in the new block. | |
993 | * | |
994 | * The reference count of the block is greater than one and the tree is | |
995 | * the block's owner tree. In this case, implicit back refs is used for | |
996 | * pointers in the block. Add full back refs for every pointers in the | |
997 | * block, increase lower level extents' reference counts. The original | |
998 | * implicit back refs are entailed to the new block. | |
999 | * | |
1000 | * The reference count of the block is greater than one and the tree is | |
1001 | * not the block's owner tree. Add implicit back refs for every pointer in | |
1002 | * the new block, increase lower level extents' reference count. | |
1003 | * | |
1004 | * Back Reference Key composing: | |
1005 | * | |
1006 | * The key objectid corresponds to the first byte in the extent, | |
1007 | * The key type is used to differentiate between types of back refs. | |
1008 | * There are different meanings of the key offset for different types | |
1009 | * of back refs. | |
1010 | * | |
d8d5f3e1 CM |
1011 | * File extents can be referenced by: |
1012 | * | |
1013 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 1014 | * - different files inside a single subvolume |
d8d5f3e1 CM |
1015 | * - different offsets inside a file (bookend extents in file.c) |
1016 | * | |
5d4f98a2 | 1017 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
1018 | * |
1019 | * - Objectid of the subvolume root | |
d8d5f3e1 | 1020 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
1021 | * - original offset in the file |
1022 | * - how many bookend extents | |
d8d5f3e1 | 1023 | * |
5d4f98a2 YZ |
1024 | * The key offset for the implicit back refs is hash of the first |
1025 | * three fields. | |
d8d5f3e1 | 1026 | * |
5d4f98a2 | 1027 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 1028 | * |
5d4f98a2 | 1029 | * - number of pointers in the tree leaf |
d8d5f3e1 | 1030 | * |
5d4f98a2 YZ |
1031 | * The key offset for the implicit back refs is the first byte of |
1032 | * the tree leaf | |
d8d5f3e1 | 1033 | * |
5d4f98a2 YZ |
1034 | * When a file extent is allocated, The implicit back refs is used. |
1035 | * the fields are filled in: | |
d8d5f3e1 | 1036 | * |
5d4f98a2 | 1037 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 1038 | * |
5d4f98a2 YZ |
1039 | * When a file extent is removed file truncation, we find the |
1040 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 1041 | * |
5d4f98a2 | 1042 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 1043 | * |
5d4f98a2 | 1044 | * Btree extents can be referenced by: |
d8d5f3e1 | 1045 | * |
5d4f98a2 | 1046 | * - Different subvolumes |
d8d5f3e1 | 1047 | * |
5d4f98a2 YZ |
1048 | * Both the implicit back refs and the full back refs for tree blocks |
1049 | * only consist of key. The key offset for the implicit back refs is | |
1050 | * objectid of block's owner tree. The key offset for the full back refs | |
1051 | * is the first byte of parent block. | |
d8d5f3e1 | 1052 | * |
5d4f98a2 YZ |
1053 | * When implicit back refs is used, information about the lowest key and |
1054 | * level of the tree block are required. These information are stored in | |
1055 | * tree block info structure. | |
d8d5f3e1 | 1056 | */ |
31840ae1 | 1057 | |
5d4f98a2 YZ |
1058 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
1059 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
87bde3cd | 1060 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1061 | struct btrfs_path *path, |
1062 | u64 owner, u32 extra_size) | |
7bb86316 | 1063 | { |
87bde3cd | 1064 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1065 | struct btrfs_extent_item *item; |
1066 | struct btrfs_extent_item_v0 *ei0; | |
1067 | struct btrfs_extent_ref_v0 *ref0; | |
1068 | struct btrfs_tree_block_info *bi; | |
1069 | struct extent_buffer *leaf; | |
7bb86316 | 1070 | struct btrfs_key key; |
5d4f98a2 YZ |
1071 | struct btrfs_key found_key; |
1072 | u32 new_size = sizeof(*item); | |
1073 | u64 refs; | |
1074 | int ret; | |
1075 | ||
1076 | leaf = path->nodes[0]; | |
1077 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
1078 | ||
1079 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1080 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
1081 | struct btrfs_extent_item_v0); | |
1082 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
1083 | ||
1084 | if (owner == (u64)-1) { | |
1085 | while (1) { | |
1086 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1087 | ret = btrfs_next_leaf(root, path); | |
1088 | if (ret < 0) | |
1089 | return ret; | |
79787eaa | 1090 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
1091 | leaf = path->nodes[0]; |
1092 | } | |
1093 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1094 | path->slots[0]); | |
1095 | BUG_ON(key.objectid != found_key.objectid); | |
1096 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
1097 | path->slots[0]++; | |
1098 | continue; | |
1099 | } | |
1100 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1101 | struct btrfs_extent_ref_v0); | |
1102 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
1103 | break; | |
1104 | } | |
1105 | } | |
b3b4aa74 | 1106 | btrfs_release_path(path); |
5d4f98a2 YZ |
1107 | |
1108 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1109 | new_size += sizeof(*bi); | |
1110 | ||
1111 | new_size -= sizeof(*ei0); | |
1112 | ret = btrfs_search_slot(trans, root, &key, path, | |
1113 | new_size + extra_size, 1); | |
1114 | if (ret < 0) | |
1115 | return ret; | |
79787eaa | 1116 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 1117 | |
87bde3cd | 1118 | btrfs_extend_item(fs_info, path, new_size); |
5d4f98a2 YZ |
1119 | |
1120 | leaf = path->nodes[0]; | |
1121 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1122 | btrfs_set_extent_refs(leaf, item, refs); | |
1123 | /* FIXME: get real generation */ | |
1124 | btrfs_set_extent_generation(leaf, item, 0); | |
1125 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1126 | btrfs_set_extent_flags(leaf, item, | |
1127 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1128 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1129 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1130 | /* FIXME: get first key of the block */ | |
b159fa28 | 1131 | memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi)); |
5d4f98a2 YZ |
1132 | btrfs_set_tree_block_level(leaf, bi, (int)owner); |
1133 | } else { | |
1134 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1135 | } | |
1136 | btrfs_mark_buffer_dirty(leaf); | |
1137 | return 0; | |
1138 | } | |
1139 | #endif | |
1140 | ||
167ce953 LB |
1141 | /* |
1142 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
1143 | * is_data == BTRFS_REF_TYPE_DATA, data type is requried, | |
1144 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. | |
1145 | */ | |
1146 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
1147 | struct btrfs_extent_inline_ref *iref, | |
1148 | enum btrfs_inline_ref_type is_data) | |
1149 | { | |
1150 | int type = btrfs_extent_inline_ref_type(eb, iref); | |
64ecdb64 | 1151 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 LB |
1152 | |
1153 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
1154 | type == BTRFS_SHARED_BLOCK_REF_KEY || | |
1155 | type == BTRFS_SHARED_DATA_REF_KEY || | |
1156 | type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1157 | if (is_data == BTRFS_REF_TYPE_BLOCK) { | |
64ecdb64 | 1158 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
167ce953 | 1159 | return type; |
64ecdb64 LB |
1160 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
1161 | ASSERT(eb->fs_info); | |
1162 | /* | |
1163 | * Every shared one has parent tree | |
1164 | * block, which must be aligned to | |
1165 | * nodesize. | |
1166 | */ | |
1167 | if (offset && | |
1168 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1169 | return type; | |
1170 | } | |
167ce953 | 1171 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 1172 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 1173 | return type; |
64ecdb64 LB |
1174 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
1175 | ASSERT(eb->fs_info); | |
1176 | /* | |
1177 | * Every shared one has parent tree | |
1178 | * block, which must be aligned to | |
1179 | * nodesize. | |
1180 | */ | |
1181 | if (offset && | |
1182 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1183 | return type; | |
1184 | } | |
167ce953 LB |
1185 | } else { |
1186 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
1187 | return type; | |
1188 | } | |
1189 | } | |
1190 | ||
1191 | btrfs_print_leaf((struct extent_buffer *)eb); | |
1192 | btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d", | |
1193 | eb->start, type); | |
1194 | WARN_ON(1); | |
1195 | ||
1196 | return BTRFS_REF_TYPE_INVALID; | |
1197 | } | |
1198 | ||
5d4f98a2 YZ |
1199 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
1200 | { | |
1201 | u32 high_crc = ~(u32)0; | |
1202 | u32 low_crc = ~(u32)0; | |
1203 | __le64 lenum; | |
1204 | ||
1205 | lenum = cpu_to_le64(root_objectid); | |
9678c543 | 1206 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1207 | lenum = cpu_to_le64(owner); |
9678c543 | 1208 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1209 | lenum = cpu_to_le64(offset); |
9678c543 | 1210 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1211 | |
1212 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1213 | } | |
1214 | ||
1215 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1216 | struct btrfs_extent_data_ref *ref) | |
1217 | { | |
1218 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1219 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1220 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1221 | } | |
1222 | ||
1223 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1224 | struct btrfs_extent_data_ref *ref, | |
1225 | u64 root_objectid, u64 owner, u64 offset) | |
1226 | { | |
1227 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1228 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1229 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1230 | return 0; | |
1231 | return 1; | |
1232 | } | |
1233 | ||
1234 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1235 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1236 | struct btrfs_path *path, |
1237 | u64 bytenr, u64 parent, | |
1238 | u64 root_objectid, | |
1239 | u64 owner, u64 offset) | |
1240 | { | |
87bde3cd | 1241 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1242 | struct btrfs_key key; |
1243 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1244 | struct extent_buffer *leaf; |
5d4f98a2 | 1245 | u32 nritems; |
74493f7a | 1246 | int ret; |
5d4f98a2 YZ |
1247 | int recow; |
1248 | int err = -ENOENT; | |
74493f7a | 1249 | |
31840ae1 | 1250 | key.objectid = bytenr; |
5d4f98a2 YZ |
1251 | if (parent) { |
1252 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1253 | key.offset = parent; | |
1254 | } else { | |
1255 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1256 | key.offset = hash_extent_data_ref(root_objectid, | |
1257 | owner, offset); | |
1258 | } | |
1259 | again: | |
1260 | recow = 0; | |
1261 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1262 | if (ret < 0) { | |
1263 | err = ret; | |
1264 | goto fail; | |
1265 | } | |
31840ae1 | 1266 | |
5d4f98a2 YZ |
1267 | if (parent) { |
1268 | if (!ret) | |
1269 | return 0; | |
1270 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1271 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1272 | btrfs_release_path(path); |
5d4f98a2 YZ |
1273 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1274 | if (ret < 0) { | |
1275 | err = ret; | |
1276 | goto fail; | |
1277 | } | |
1278 | if (!ret) | |
1279 | return 0; | |
1280 | #endif | |
1281 | goto fail; | |
31840ae1 ZY |
1282 | } |
1283 | ||
1284 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1285 | nritems = btrfs_header_nritems(leaf); |
1286 | while (1) { | |
1287 | if (path->slots[0] >= nritems) { | |
1288 | ret = btrfs_next_leaf(root, path); | |
1289 | if (ret < 0) | |
1290 | err = ret; | |
1291 | if (ret) | |
1292 | goto fail; | |
1293 | ||
1294 | leaf = path->nodes[0]; | |
1295 | nritems = btrfs_header_nritems(leaf); | |
1296 | recow = 1; | |
1297 | } | |
1298 | ||
1299 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1300 | if (key.objectid != bytenr || | |
1301 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1302 | goto fail; | |
1303 | ||
1304 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1305 | struct btrfs_extent_data_ref); | |
1306 | ||
1307 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1308 | owner, offset)) { | |
1309 | if (recow) { | |
b3b4aa74 | 1310 | btrfs_release_path(path); |
5d4f98a2 YZ |
1311 | goto again; |
1312 | } | |
1313 | err = 0; | |
1314 | break; | |
1315 | } | |
1316 | path->slots[0]++; | |
31840ae1 | 1317 | } |
5d4f98a2 YZ |
1318 | fail: |
1319 | return err; | |
31840ae1 ZY |
1320 | } |
1321 | ||
5d4f98a2 | 1322 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1323 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1324 | struct btrfs_path *path, |
1325 | u64 bytenr, u64 parent, | |
1326 | u64 root_objectid, u64 owner, | |
1327 | u64 offset, int refs_to_add) | |
31840ae1 | 1328 | { |
87bde3cd | 1329 | struct btrfs_root *root = fs_info->extent_root; |
31840ae1 ZY |
1330 | struct btrfs_key key; |
1331 | struct extent_buffer *leaf; | |
5d4f98a2 | 1332 | u32 size; |
31840ae1 ZY |
1333 | u32 num_refs; |
1334 | int ret; | |
74493f7a | 1335 | |
74493f7a | 1336 | key.objectid = bytenr; |
5d4f98a2 YZ |
1337 | if (parent) { |
1338 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1339 | key.offset = parent; | |
1340 | size = sizeof(struct btrfs_shared_data_ref); | |
1341 | } else { | |
1342 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1343 | key.offset = hash_extent_data_ref(root_objectid, | |
1344 | owner, offset); | |
1345 | size = sizeof(struct btrfs_extent_data_ref); | |
1346 | } | |
74493f7a | 1347 | |
5d4f98a2 YZ |
1348 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1349 | if (ret && ret != -EEXIST) | |
1350 | goto fail; | |
1351 | ||
1352 | leaf = path->nodes[0]; | |
1353 | if (parent) { | |
1354 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1355 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1356 | struct btrfs_shared_data_ref); |
1357 | if (ret == 0) { | |
1358 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1359 | } else { | |
1360 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1361 | num_refs += refs_to_add; | |
1362 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1363 | } |
5d4f98a2 YZ |
1364 | } else { |
1365 | struct btrfs_extent_data_ref *ref; | |
1366 | while (ret == -EEXIST) { | |
1367 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1368 | struct btrfs_extent_data_ref); | |
1369 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1370 | owner, offset)) | |
1371 | break; | |
b3b4aa74 | 1372 | btrfs_release_path(path); |
5d4f98a2 YZ |
1373 | key.offset++; |
1374 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1375 | size); | |
1376 | if (ret && ret != -EEXIST) | |
1377 | goto fail; | |
31840ae1 | 1378 | |
5d4f98a2 YZ |
1379 | leaf = path->nodes[0]; |
1380 | } | |
1381 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1382 | struct btrfs_extent_data_ref); | |
1383 | if (ret == 0) { | |
1384 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1385 | root_objectid); | |
1386 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1387 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1388 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1389 | } else { | |
1390 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1391 | num_refs += refs_to_add; | |
1392 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1393 | } |
31840ae1 | 1394 | } |
5d4f98a2 YZ |
1395 | btrfs_mark_buffer_dirty(leaf); |
1396 | ret = 0; | |
1397 | fail: | |
b3b4aa74 | 1398 | btrfs_release_path(path); |
7bb86316 | 1399 | return ret; |
74493f7a CM |
1400 | } |
1401 | ||
5d4f98a2 | 1402 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1403 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 1404 | struct btrfs_path *path, |
fcebe456 | 1405 | int refs_to_drop, int *last_ref) |
31840ae1 | 1406 | { |
5d4f98a2 YZ |
1407 | struct btrfs_key key; |
1408 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1409 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1410 | struct extent_buffer *leaf; |
5d4f98a2 | 1411 | u32 num_refs = 0; |
31840ae1 ZY |
1412 | int ret = 0; |
1413 | ||
1414 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1415 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1416 | ||
1417 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1418 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1419 | struct btrfs_extent_data_ref); | |
1420 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1421 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1422 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1423 | struct btrfs_shared_data_ref); | |
1424 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1425 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1426 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1427 | struct btrfs_extent_ref_v0 *ref0; | |
1428 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1429 | struct btrfs_extent_ref_v0); | |
1430 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1431 | #endif | |
1432 | } else { | |
1433 | BUG(); | |
1434 | } | |
1435 | ||
56bec294 CM |
1436 | BUG_ON(num_refs < refs_to_drop); |
1437 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1438 | |
31840ae1 | 1439 | if (num_refs == 0) { |
87bde3cd | 1440 | ret = btrfs_del_item(trans, fs_info->extent_root, path); |
fcebe456 | 1441 | *last_ref = 1; |
31840ae1 | 1442 | } else { |
5d4f98a2 YZ |
1443 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1444 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1445 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1446 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1447 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1448 | else { | |
1449 | struct btrfs_extent_ref_v0 *ref0; | |
1450 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1451 | struct btrfs_extent_ref_v0); | |
1452 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1453 | } | |
1454 | #endif | |
31840ae1 ZY |
1455 | btrfs_mark_buffer_dirty(leaf); |
1456 | } | |
31840ae1 ZY |
1457 | return ret; |
1458 | } | |
1459 | ||
9ed0dea0 | 1460 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1461 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1462 | { |
5d4f98a2 YZ |
1463 | struct btrfs_key key; |
1464 | struct extent_buffer *leaf; | |
1465 | struct btrfs_extent_data_ref *ref1; | |
1466 | struct btrfs_shared_data_ref *ref2; | |
1467 | u32 num_refs = 0; | |
3de28d57 | 1468 | int type; |
5d4f98a2 YZ |
1469 | |
1470 | leaf = path->nodes[0]; | |
1471 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1472 | if (iref) { | |
3de28d57 LB |
1473 | /* |
1474 | * If type is invalid, we should have bailed out earlier than | |
1475 | * this call. | |
1476 | */ | |
1477 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
1478 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
1479 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
1480 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
1481 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1482 | } else { | |
1483 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1484 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1485 | } | |
1486 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1487 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1488 | struct btrfs_extent_data_ref); | |
1489 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1490 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1491 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1492 | struct btrfs_shared_data_ref); | |
1493 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1494 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1495 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1496 | struct btrfs_extent_ref_v0 *ref0; | |
1497 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1498 | struct btrfs_extent_ref_v0); | |
1499 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1500 | #endif |
5d4f98a2 YZ |
1501 | } else { |
1502 | WARN_ON(1); | |
1503 | } | |
1504 | return num_refs; | |
1505 | } | |
15916de8 | 1506 | |
5d4f98a2 | 1507 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1508 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1509 | struct btrfs_path *path, |
1510 | u64 bytenr, u64 parent, | |
1511 | u64 root_objectid) | |
1f3c79a2 | 1512 | { |
87bde3cd | 1513 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 | 1514 | struct btrfs_key key; |
1f3c79a2 | 1515 | int ret; |
1f3c79a2 | 1516 | |
5d4f98a2 YZ |
1517 | key.objectid = bytenr; |
1518 | if (parent) { | |
1519 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1520 | key.offset = parent; | |
1521 | } else { | |
1522 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1523 | key.offset = root_objectid; | |
1f3c79a2 LH |
1524 | } |
1525 | ||
5d4f98a2 YZ |
1526 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1527 | if (ret > 0) | |
1528 | ret = -ENOENT; | |
1529 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1530 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1531 | btrfs_release_path(path); |
5d4f98a2 YZ |
1532 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1533 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1534 | if (ret > 0) | |
1535 | ret = -ENOENT; | |
1536 | } | |
1f3c79a2 | 1537 | #endif |
5d4f98a2 | 1538 | return ret; |
1f3c79a2 LH |
1539 | } |
1540 | ||
5d4f98a2 | 1541 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1542 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1543 | struct btrfs_path *path, |
1544 | u64 bytenr, u64 parent, | |
1545 | u64 root_objectid) | |
31840ae1 | 1546 | { |
5d4f98a2 | 1547 | struct btrfs_key key; |
31840ae1 | 1548 | int ret; |
31840ae1 | 1549 | |
5d4f98a2 YZ |
1550 | key.objectid = bytenr; |
1551 | if (parent) { | |
1552 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1553 | key.offset = parent; | |
1554 | } else { | |
1555 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1556 | key.offset = root_objectid; | |
1557 | } | |
1558 | ||
87bde3cd JM |
1559 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, |
1560 | path, &key, 0); | |
b3b4aa74 | 1561 | btrfs_release_path(path); |
31840ae1 ZY |
1562 | return ret; |
1563 | } | |
1564 | ||
5d4f98a2 | 1565 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1566 | { |
5d4f98a2 YZ |
1567 | int type; |
1568 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1569 | if (parent > 0) | |
1570 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1571 | else | |
1572 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1573 | } else { | |
1574 | if (parent > 0) | |
1575 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1576 | else | |
1577 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1578 | } | |
1579 | return type; | |
31840ae1 | 1580 | } |
56bec294 | 1581 | |
2c47e605 YZ |
1582 | static int find_next_key(struct btrfs_path *path, int level, |
1583 | struct btrfs_key *key) | |
56bec294 | 1584 | |
02217ed2 | 1585 | { |
2c47e605 | 1586 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1587 | if (!path->nodes[level]) |
1588 | break; | |
5d4f98a2 YZ |
1589 | if (path->slots[level] + 1 >= |
1590 | btrfs_header_nritems(path->nodes[level])) | |
1591 | continue; | |
1592 | if (level == 0) | |
1593 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1594 | path->slots[level] + 1); | |
1595 | else | |
1596 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1597 | path->slots[level] + 1); | |
1598 | return 0; | |
1599 | } | |
1600 | return 1; | |
1601 | } | |
037e6390 | 1602 | |
5d4f98a2 YZ |
1603 | /* |
1604 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1605 | * to the address of inline back ref, and 0 is returned. | |
1606 | * | |
1607 | * if back ref isn't found, *ref_ret is set to the address where it | |
1608 | * should be inserted, and -ENOENT is returned. | |
1609 | * | |
1610 | * if insert is true and there are too many inline back refs, the path | |
1611 | * points to the extent item, and -EAGAIN is returned. | |
1612 | * | |
1613 | * NOTE: inline back refs are ordered in the same way that back ref | |
1614 | * items in the tree are ordered. | |
1615 | */ | |
1616 | static noinline_for_stack | |
1617 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1618 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1619 | struct btrfs_path *path, |
1620 | struct btrfs_extent_inline_ref **ref_ret, | |
1621 | u64 bytenr, u64 num_bytes, | |
1622 | u64 parent, u64 root_objectid, | |
1623 | u64 owner, u64 offset, int insert) | |
1624 | { | |
87bde3cd | 1625 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1626 | struct btrfs_key key; |
1627 | struct extent_buffer *leaf; | |
1628 | struct btrfs_extent_item *ei; | |
1629 | struct btrfs_extent_inline_ref *iref; | |
1630 | u64 flags; | |
1631 | u64 item_size; | |
1632 | unsigned long ptr; | |
1633 | unsigned long end; | |
1634 | int extra_size; | |
1635 | int type; | |
1636 | int want; | |
1637 | int ret; | |
1638 | int err = 0; | |
0b246afa | 1639 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 1640 | int needed; |
26b8003f | 1641 | |
db94535d | 1642 | key.objectid = bytenr; |
31840ae1 | 1643 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1644 | key.offset = num_bytes; |
31840ae1 | 1645 | |
5d4f98a2 YZ |
1646 | want = extent_ref_type(parent, owner); |
1647 | if (insert) { | |
1648 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1649 | path->keep_locks = 1; |
5d4f98a2 YZ |
1650 | } else |
1651 | extra_size = -1; | |
3173a18f JB |
1652 | |
1653 | /* | |
1654 | * Owner is our parent level, so we can just add one to get the level | |
1655 | * for the block we are interested in. | |
1656 | */ | |
1657 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1658 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1659 | key.offset = owner; | |
1660 | } | |
1661 | ||
1662 | again: | |
5d4f98a2 | 1663 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1664 | if (ret < 0) { |
5d4f98a2 YZ |
1665 | err = ret; |
1666 | goto out; | |
1667 | } | |
3173a18f JB |
1668 | |
1669 | /* | |
1670 | * We may be a newly converted file system which still has the old fat | |
1671 | * extent entries for metadata, so try and see if we have one of those. | |
1672 | */ | |
1673 | if (ret > 0 && skinny_metadata) { | |
1674 | skinny_metadata = false; | |
1675 | if (path->slots[0]) { | |
1676 | path->slots[0]--; | |
1677 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1678 | path->slots[0]); | |
1679 | if (key.objectid == bytenr && | |
1680 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1681 | key.offset == num_bytes) | |
1682 | ret = 0; | |
1683 | } | |
1684 | if (ret) { | |
9ce49a0b | 1685 | key.objectid = bytenr; |
3173a18f JB |
1686 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1687 | key.offset = num_bytes; | |
1688 | btrfs_release_path(path); | |
1689 | goto again; | |
1690 | } | |
1691 | } | |
1692 | ||
79787eaa JM |
1693 | if (ret && !insert) { |
1694 | err = -ENOENT; | |
1695 | goto out; | |
fae7f21c | 1696 | } else if (WARN_ON(ret)) { |
492104c8 | 1697 | err = -EIO; |
492104c8 | 1698 | goto out; |
79787eaa | 1699 | } |
5d4f98a2 YZ |
1700 | |
1701 | leaf = path->nodes[0]; | |
1702 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1703 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1704 | if (item_size < sizeof(*ei)) { | |
1705 | if (!insert) { | |
1706 | err = -ENOENT; | |
1707 | goto out; | |
1708 | } | |
87bde3cd | 1709 | ret = convert_extent_item_v0(trans, fs_info, path, owner, |
5d4f98a2 YZ |
1710 | extra_size); |
1711 | if (ret < 0) { | |
1712 | err = ret; | |
1713 | goto out; | |
1714 | } | |
1715 | leaf = path->nodes[0]; | |
1716 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1717 | } | |
1718 | #endif | |
1719 | BUG_ON(item_size < sizeof(*ei)); | |
1720 | ||
5d4f98a2 YZ |
1721 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1722 | flags = btrfs_extent_flags(leaf, ei); | |
1723 | ||
1724 | ptr = (unsigned long)(ei + 1); | |
1725 | end = (unsigned long)ei + item_size; | |
1726 | ||
3173a18f | 1727 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1728 | ptr += sizeof(struct btrfs_tree_block_info); |
1729 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1730 | } |
1731 | ||
3de28d57 LB |
1732 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
1733 | needed = BTRFS_REF_TYPE_DATA; | |
1734 | else | |
1735 | needed = BTRFS_REF_TYPE_BLOCK; | |
1736 | ||
5d4f98a2 YZ |
1737 | err = -ENOENT; |
1738 | while (1) { | |
1739 | if (ptr >= end) { | |
1740 | WARN_ON(ptr > end); | |
1741 | break; | |
1742 | } | |
1743 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
1744 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
1745 | if (type == BTRFS_REF_TYPE_INVALID) { | |
1746 | err = -EINVAL; | |
1747 | goto out; | |
1748 | } | |
1749 | ||
5d4f98a2 YZ |
1750 | if (want < type) |
1751 | break; | |
1752 | if (want > type) { | |
1753 | ptr += btrfs_extent_inline_ref_size(type); | |
1754 | continue; | |
1755 | } | |
1756 | ||
1757 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1758 | struct btrfs_extent_data_ref *dref; | |
1759 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1760 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1761 | owner, offset)) { | |
1762 | err = 0; | |
1763 | break; | |
1764 | } | |
1765 | if (hash_extent_data_ref_item(leaf, dref) < | |
1766 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1767 | break; | |
1768 | } else { | |
1769 | u64 ref_offset; | |
1770 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1771 | if (parent > 0) { | |
1772 | if (parent == ref_offset) { | |
1773 | err = 0; | |
1774 | break; | |
1775 | } | |
1776 | if (ref_offset < parent) | |
1777 | break; | |
1778 | } else { | |
1779 | if (root_objectid == ref_offset) { | |
1780 | err = 0; | |
1781 | break; | |
1782 | } | |
1783 | if (ref_offset < root_objectid) | |
1784 | break; | |
1785 | } | |
1786 | } | |
1787 | ptr += btrfs_extent_inline_ref_size(type); | |
1788 | } | |
1789 | if (err == -ENOENT && insert) { | |
1790 | if (item_size + extra_size >= | |
1791 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1792 | err = -EAGAIN; | |
1793 | goto out; | |
1794 | } | |
1795 | /* | |
1796 | * To add new inline back ref, we have to make sure | |
1797 | * there is no corresponding back ref item. | |
1798 | * For simplicity, we just do not add new inline back | |
1799 | * ref if there is any kind of item for this block | |
1800 | */ | |
2c47e605 YZ |
1801 | if (find_next_key(path, 0, &key) == 0 && |
1802 | key.objectid == bytenr && | |
85d4198e | 1803 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1804 | err = -EAGAIN; |
1805 | goto out; | |
1806 | } | |
1807 | } | |
1808 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1809 | out: | |
85d4198e | 1810 | if (insert) { |
5d4f98a2 YZ |
1811 | path->keep_locks = 0; |
1812 | btrfs_unlock_up_safe(path, 1); | |
1813 | } | |
1814 | return err; | |
1815 | } | |
1816 | ||
1817 | /* | |
1818 | * helper to add new inline back ref | |
1819 | */ | |
1820 | static noinline_for_stack | |
87bde3cd | 1821 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1822 | struct btrfs_path *path, |
1823 | struct btrfs_extent_inline_ref *iref, | |
1824 | u64 parent, u64 root_objectid, | |
1825 | u64 owner, u64 offset, int refs_to_add, | |
1826 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1827 | { |
1828 | struct extent_buffer *leaf; | |
1829 | struct btrfs_extent_item *ei; | |
1830 | unsigned long ptr; | |
1831 | unsigned long end; | |
1832 | unsigned long item_offset; | |
1833 | u64 refs; | |
1834 | int size; | |
1835 | int type; | |
5d4f98a2 YZ |
1836 | |
1837 | leaf = path->nodes[0]; | |
1838 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1839 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1840 | ||
1841 | type = extent_ref_type(parent, owner); | |
1842 | size = btrfs_extent_inline_ref_size(type); | |
1843 | ||
87bde3cd | 1844 | btrfs_extend_item(fs_info, path, size); |
5d4f98a2 YZ |
1845 | |
1846 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1847 | refs = btrfs_extent_refs(leaf, ei); | |
1848 | refs += refs_to_add; | |
1849 | btrfs_set_extent_refs(leaf, ei, refs); | |
1850 | if (extent_op) | |
1851 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1852 | ||
1853 | ptr = (unsigned long)ei + item_offset; | |
1854 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1855 | if (ptr < end - size) | |
1856 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1857 | end - size - ptr); | |
1858 | ||
1859 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1860 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1861 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1862 | struct btrfs_extent_data_ref *dref; | |
1863 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1864 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1865 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1866 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1867 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1868 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1869 | struct btrfs_shared_data_ref *sref; | |
1870 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1871 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1872 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1873 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1874 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1875 | } else { | |
1876 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1877 | } | |
1878 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1879 | } |
1880 | ||
1881 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1882 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1883 | struct btrfs_path *path, |
1884 | struct btrfs_extent_inline_ref **ref_ret, | |
1885 | u64 bytenr, u64 num_bytes, u64 parent, | |
1886 | u64 root_objectid, u64 owner, u64 offset) | |
1887 | { | |
1888 | int ret; | |
1889 | ||
87bde3cd | 1890 | ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret, |
5d4f98a2 YZ |
1891 | bytenr, num_bytes, parent, |
1892 | root_objectid, owner, offset, 0); | |
1893 | if (ret != -ENOENT) | |
54aa1f4d | 1894 | return ret; |
5d4f98a2 | 1895 | |
b3b4aa74 | 1896 | btrfs_release_path(path); |
5d4f98a2 YZ |
1897 | *ref_ret = NULL; |
1898 | ||
1899 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
87bde3cd JM |
1900 | ret = lookup_tree_block_ref(trans, fs_info, path, bytenr, |
1901 | parent, root_objectid); | |
5d4f98a2 | 1902 | } else { |
87bde3cd JM |
1903 | ret = lookup_extent_data_ref(trans, fs_info, path, bytenr, |
1904 | parent, root_objectid, owner, | |
1905 | offset); | |
b9473439 | 1906 | } |
5d4f98a2 YZ |
1907 | return ret; |
1908 | } | |
31840ae1 | 1909 | |
5d4f98a2 YZ |
1910 | /* |
1911 | * helper to update/remove inline back ref | |
1912 | */ | |
1913 | static noinline_for_stack | |
87bde3cd | 1914 | void update_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1915 | struct btrfs_path *path, |
1916 | struct btrfs_extent_inline_ref *iref, | |
1917 | int refs_to_mod, | |
fcebe456 JB |
1918 | struct btrfs_delayed_extent_op *extent_op, |
1919 | int *last_ref) | |
5d4f98a2 YZ |
1920 | { |
1921 | struct extent_buffer *leaf; | |
1922 | struct btrfs_extent_item *ei; | |
1923 | struct btrfs_extent_data_ref *dref = NULL; | |
1924 | struct btrfs_shared_data_ref *sref = NULL; | |
1925 | unsigned long ptr; | |
1926 | unsigned long end; | |
1927 | u32 item_size; | |
1928 | int size; | |
1929 | int type; | |
5d4f98a2 YZ |
1930 | u64 refs; |
1931 | ||
1932 | leaf = path->nodes[0]; | |
1933 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1934 | refs = btrfs_extent_refs(leaf, ei); | |
1935 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1936 | refs += refs_to_mod; | |
1937 | btrfs_set_extent_refs(leaf, ei, refs); | |
1938 | if (extent_op) | |
1939 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1940 | ||
3de28d57 LB |
1941 | /* |
1942 | * If type is invalid, we should have bailed out after | |
1943 | * lookup_inline_extent_backref(). | |
1944 | */ | |
1945 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
1946 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
5d4f98a2 YZ |
1947 | |
1948 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1949 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1950 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1951 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1952 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1953 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1954 | } else { | |
1955 | refs = 1; | |
1956 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1957 | } |
31840ae1 | 1958 | |
5d4f98a2 YZ |
1959 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1960 | refs += refs_to_mod; | |
1961 | ||
1962 | if (refs > 0) { | |
1963 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1964 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1965 | else | |
1966 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1967 | } else { | |
fcebe456 | 1968 | *last_ref = 1; |
5d4f98a2 YZ |
1969 | size = btrfs_extent_inline_ref_size(type); |
1970 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1971 | ptr = (unsigned long)iref; | |
1972 | end = (unsigned long)ei + item_size; | |
1973 | if (ptr + size < end) | |
1974 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1975 | end - ptr - size); | |
1976 | item_size -= size; | |
87bde3cd | 1977 | btrfs_truncate_item(fs_info, path, item_size, 1); |
5d4f98a2 YZ |
1978 | } |
1979 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1980 | } |
1981 | ||
1982 | static noinline_for_stack | |
1983 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1984 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1985 | struct btrfs_path *path, |
1986 | u64 bytenr, u64 num_bytes, u64 parent, | |
1987 | u64 root_objectid, u64 owner, | |
1988 | u64 offset, int refs_to_add, | |
1989 | struct btrfs_delayed_extent_op *extent_op) | |
1990 | { | |
1991 | struct btrfs_extent_inline_ref *iref; | |
1992 | int ret; | |
1993 | ||
87bde3cd | 1994 | ret = lookup_inline_extent_backref(trans, fs_info, path, &iref, |
5d4f98a2 YZ |
1995 | bytenr, num_bytes, parent, |
1996 | root_objectid, owner, offset, 1); | |
1997 | if (ret == 0) { | |
1998 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
87bde3cd | 1999 | update_inline_extent_backref(fs_info, path, iref, |
fcebe456 | 2000 | refs_to_add, extent_op, NULL); |
5d4f98a2 | 2001 | } else if (ret == -ENOENT) { |
87bde3cd | 2002 | setup_inline_extent_backref(fs_info, path, iref, parent, |
143bede5 JM |
2003 | root_objectid, owner, offset, |
2004 | refs_to_add, extent_op); | |
2005 | ret = 0; | |
771ed689 | 2006 | } |
5d4f98a2 YZ |
2007 | return ret; |
2008 | } | |
31840ae1 | 2009 | |
5d4f98a2 | 2010 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
87bde3cd | 2011 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2012 | struct btrfs_path *path, |
2013 | u64 bytenr, u64 parent, u64 root_objectid, | |
2014 | u64 owner, u64 offset, int refs_to_add) | |
2015 | { | |
2016 | int ret; | |
2017 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
2018 | BUG_ON(refs_to_add != 1); | |
87bde3cd | 2019 | ret = insert_tree_block_ref(trans, fs_info, path, bytenr, |
5d4f98a2 YZ |
2020 | parent, root_objectid); |
2021 | } else { | |
87bde3cd | 2022 | ret = insert_extent_data_ref(trans, fs_info, path, bytenr, |
5d4f98a2 YZ |
2023 | parent, root_objectid, |
2024 | owner, offset, refs_to_add); | |
2025 | } | |
2026 | return ret; | |
2027 | } | |
56bec294 | 2028 | |
5d4f98a2 | 2029 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
87bde3cd | 2030 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2031 | struct btrfs_path *path, |
2032 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 2033 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 2034 | { |
143bede5 | 2035 | int ret = 0; |
b9473439 | 2036 | |
5d4f98a2 YZ |
2037 | BUG_ON(!is_data && refs_to_drop != 1); |
2038 | if (iref) { | |
87bde3cd | 2039 | update_inline_extent_backref(fs_info, path, iref, |
fcebe456 | 2040 | -refs_to_drop, NULL, last_ref); |
5d4f98a2 | 2041 | } else if (is_data) { |
87bde3cd | 2042 | ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop, |
fcebe456 | 2043 | last_ref); |
5d4f98a2 | 2044 | } else { |
fcebe456 | 2045 | *last_ref = 1; |
87bde3cd | 2046 | ret = btrfs_del_item(trans, fs_info->extent_root, path); |
5d4f98a2 YZ |
2047 | } |
2048 | return ret; | |
2049 | } | |
2050 | ||
86557861 | 2051 | #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len)) |
d04c6b88 JM |
2052 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
2053 | u64 *discarded_bytes) | |
5d4f98a2 | 2054 | { |
86557861 JM |
2055 | int j, ret = 0; |
2056 | u64 bytes_left, end; | |
4d89d377 | 2057 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 2058 | |
4d89d377 JM |
2059 | if (WARN_ON(start != aligned_start)) { |
2060 | len -= aligned_start - start; | |
2061 | len = round_down(len, 1 << 9); | |
2062 | start = aligned_start; | |
2063 | } | |
d04c6b88 | 2064 | |
4d89d377 | 2065 | *discarded_bytes = 0; |
86557861 JM |
2066 | |
2067 | if (!len) | |
2068 | return 0; | |
2069 | ||
2070 | end = start + len; | |
2071 | bytes_left = len; | |
2072 | ||
2073 | /* Skip any superblocks on this device. */ | |
2074 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
2075 | u64 sb_start = btrfs_sb_offset(j); | |
2076 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
2077 | u64 size = sb_start - start; | |
2078 | ||
2079 | if (!in_range(sb_start, start, bytes_left) && | |
2080 | !in_range(sb_end, start, bytes_left) && | |
2081 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
2082 | continue; | |
2083 | ||
2084 | /* | |
2085 | * Superblock spans beginning of range. Adjust start and | |
2086 | * try again. | |
2087 | */ | |
2088 | if (sb_start <= start) { | |
2089 | start += sb_end - start; | |
2090 | if (start > end) { | |
2091 | bytes_left = 0; | |
2092 | break; | |
2093 | } | |
2094 | bytes_left = end - start; | |
2095 | continue; | |
2096 | } | |
2097 | ||
2098 | if (size) { | |
2099 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
2100 | GFP_NOFS, 0); | |
2101 | if (!ret) | |
2102 | *discarded_bytes += size; | |
2103 | else if (ret != -EOPNOTSUPP) | |
2104 | return ret; | |
2105 | } | |
2106 | ||
2107 | start = sb_end; | |
2108 | if (start > end) { | |
2109 | bytes_left = 0; | |
2110 | break; | |
2111 | } | |
2112 | bytes_left = end - start; | |
2113 | } | |
2114 | ||
2115 | if (bytes_left) { | |
2116 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
2117 | GFP_NOFS, 0); |
2118 | if (!ret) | |
86557861 | 2119 | *discarded_bytes += bytes_left; |
4d89d377 | 2120 | } |
d04c6b88 | 2121 | return ret; |
5d4f98a2 | 2122 | } |
5d4f98a2 | 2123 | |
2ff7e61e | 2124 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 2125 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 2126 | { |
5d4f98a2 | 2127 | int ret; |
5378e607 | 2128 | u64 discarded_bytes = 0; |
a1d3c478 | 2129 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 2130 | |
e244a0ae | 2131 | |
2999241d FM |
2132 | /* |
2133 | * Avoid races with device replace and make sure our bbio has devices | |
2134 | * associated to its stripes that don't go away while we are discarding. | |
2135 | */ | |
0b246afa | 2136 | btrfs_bio_counter_inc_blocked(fs_info); |
5d4f98a2 | 2137 | /* Tell the block device(s) that the sectors can be discarded */ |
0b246afa JM |
2138 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes, |
2139 | &bbio, 0); | |
79787eaa | 2140 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 2141 | if (!ret) { |
a1d3c478 | 2142 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
2143 | int i; |
2144 | ||
5d4f98a2 | 2145 | |
a1d3c478 | 2146 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 2147 | u64 bytes; |
38b5f68e AJ |
2148 | struct request_queue *req_q; |
2149 | ||
627e0873 FM |
2150 | if (!stripe->dev->bdev) { |
2151 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); | |
2152 | continue; | |
2153 | } | |
38b5f68e AJ |
2154 | req_q = bdev_get_queue(stripe->dev->bdev); |
2155 | if (!blk_queue_discard(req_q)) | |
d5e2003c JB |
2156 | continue; |
2157 | ||
5378e607 LD |
2158 | ret = btrfs_issue_discard(stripe->dev->bdev, |
2159 | stripe->physical, | |
d04c6b88 JM |
2160 | stripe->length, |
2161 | &bytes); | |
5378e607 | 2162 | if (!ret) |
d04c6b88 | 2163 | discarded_bytes += bytes; |
5378e607 | 2164 | else if (ret != -EOPNOTSUPP) |
79787eaa | 2165 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
2166 | |
2167 | /* | |
2168 | * Just in case we get back EOPNOTSUPP for some reason, | |
2169 | * just ignore the return value so we don't screw up | |
2170 | * people calling discard_extent. | |
2171 | */ | |
2172 | ret = 0; | |
5d4f98a2 | 2173 | } |
6e9606d2 | 2174 | btrfs_put_bbio(bbio); |
5d4f98a2 | 2175 | } |
0b246afa | 2176 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
2177 | |
2178 | if (actual_bytes) | |
2179 | *actual_bytes = discarded_bytes; | |
2180 | ||
5d4f98a2 | 2181 | |
53b381b3 DW |
2182 | if (ret == -EOPNOTSUPP) |
2183 | ret = 0; | |
5d4f98a2 | 2184 | return ret; |
5d4f98a2 YZ |
2185 | } |
2186 | ||
79787eaa | 2187 | /* Can return -ENOMEM */ |
5d4f98a2 | 2188 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
84f7d8e6 | 2189 | struct btrfs_root *root, |
5d4f98a2 | 2190 | u64 bytenr, u64 num_bytes, u64 parent, |
b06c4bf5 | 2191 | u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 | 2192 | { |
84f7d8e6 | 2193 | struct btrfs_fs_info *fs_info = root->fs_info; |
d7eae340 | 2194 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 2195 | int ret; |
66d7e7f0 | 2196 | |
5d4f98a2 YZ |
2197 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
2198 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
2199 | ||
fd708b81 JB |
2200 | btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid, |
2201 | owner, offset, BTRFS_ADD_DELAYED_REF); | |
2202 | ||
5d4f98a2 | 2203 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 | 2204 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
7be07912 OS |
2205 | num_bytes, parent, |
2206 | root_objectid, (int)owner, | |
2207 | BTRFS_ADD_DELAYED_REF, NULL, | |
d7eae340 | 2208 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 2209 | } else { |
66d7e7f0 | 2210 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
7be07912 OS |
2211 | num_bytes, parent, |
2212 | root_objectid, owner, offset, | |
d7eae340 OS |
2213 | 0, BTRFS_ADD_DELAYED_REF, |
2214 | &old_ref_mod, &new_ref_mod); | |
5d4f98a2 | 2215 | } |
d7eae340 OS |
2216 | |
2217 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) | |
2218 | add_pinned_bytes(fs_info, -num_bytes, owner, root_objectid); | |
2219 | ||
5d4f98a2 YZ |
2220 | return ret; |
2221 | } | |
2222 | ||
2223 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2224 | struct btrfs_fs_info *fs_info, |
c682f9b3 | 2225 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2226 | u64 parent, u64 root_objectid, |
2227 | u64 owner, u64 offset, int refs_to_add, | |
2228 | struct btrfs_delayed_extent_op *extent_op) | |
2229 | { | |
2230 | struct btrfs_path *path; | |
2231 | struct extent_buffer *leaf; | |
2232 | struct btrfs_extent_item *item; | |
fcebe456 | 2233 | struct btrfs_key key; |
c682f9b3 QW |
2234 | u64 bytenr = node->bytenr; |
2235 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2236 | u64 refs; |
2237 | int ret; | |
5d4f98a2 YZ |
2238 | |
2239 | path = btrfs_alloc_path(); | |
2240 | if (!path) | |
2241 | return -ENOMEM; | |
2242 | ||
e4058b54 | 2243 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2244 | path->leave_spinning = 1; |
2245 | /* this will setup the path even if it fails to insert the back ref */ | |
87bde3cd JM |
2246 | ret = insert_inline_extent_backref(trans, fs_info, path, bytenr, |
2247 | num_bytes, parent, root_objectid, | |
2248 | owner, offset, | |
5d4f98a2 | 2249 | refs_to_add, extent_op); |
0ed4792a | 2250 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2251 | goto out; |
fcebe456 JB |
2252 | |
2253 | /* | |
2254 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2255 | * inline extent ref, so just update the reference count and add a | |
2256 | * normal backref. | |
2257 | */ | |
5d4f98a2 | 2258 | leaf = path->nodes[0]; |
fcebe456 | 2259 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2260 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2261 | refs = btrfs_extent_refs(leaf, item); | |
2262 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2263 | if (extent_op) | |
2264 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2265 | |
5d4f98a2 | 2266 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2267 | btrfs_release_path(path); |
56bec294 | 2268 | |
e4058b54 | 2269 | path->reada = READA_FORWARD; |
b9473439 | 2270 | path->leave_spinning = 1; |
56bec294 | 2271 | /* now insert the actual backref */ |
87bde3cd JM |
2272 | ret = insert_extent_backref(trans, fs_info, path, bytenr, parent, |
2273 | root_objectid, owner, offset, refs_to_add); | |
79787eaa | 2274 | if (ret) |
66642832 | 2275 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 2276 | out: |
56bec294 | 2277 | btrfs_free_path(path); |
30d133fc | 2278 | return ret; |
56bec294 CM |
2279 | } |
2280 | ||
5d4f98a2 | 2281 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2282 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2283 | struct btrfs_delayed_ref_node *node, |
2284 | struct btrfs_delayed_extent_op *extent_op, | |
2285 | int insert_reserved) | |
56bec294 | 2286 | { |
5d4f98a2 YZ |
2287 | int ret = 0; |
2288 | struct btrfs_delayed_data_ref *ref; | |
2289 | struct btrfs_key ins; | |
2290 | u64 parent = 0; | |
2291 | u64 ref_root = 0; | |
2292 | u64 flags = 0; | |
2293 | ||
2294 | ins.objectid = node->bytenr; | |
2295 | ins.offset = node->num_bytes; | |
2296 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2297 | ||
2298 | ref = btrfs_delayed_node_to_data_ref(node); | |
0b246afa | 2299 | trace_run_delayed_data_ref(fs_info, node, ref, node->action); |
599c75ec | 2300 | |
5d4f98a2 YZ |
2301 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2302 | parent = ref->parent; | |
fcebe456 | 2303 | ref_root = ref->root; |
5d4f98a2 YZ |
2304 | |
2305 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2306 | if (extent_op) |
5d4f98a2 | 2307 | flags |= extent_op->flags_to_set; |
2ff7e61e | 2308 | ret = alloc_reserved_file_extent(trans, fs_info, |
5d4f98a2 YZ |
2309 | parent, ref_root, flags, |
2310 | ref->objectid, ref->offset, | |
2311 | &ins, node->ref_mod); | |
5d4f98a2 | 2312 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2ff7e61e | 2313 | ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent, |
5d4f98a2 YZ |
2314 | ref_root, ref->objectid, |
2315 | ref->offset, node->ref_mod, | |
c682f9b3 | 2316 | extent_op); |
5d4f98a2 | 2317 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
2ff7e61e | 2318 | ret = __btrfs_free_extent(trans, fs_info, node, parent, |
5d4f98a2 YZ |
2319 | ref_root, ref->objectid, |
2320 | ref->offset, node->ref_mod, | |
c682f9b3 | 2321 | extent_op); |
5d4f98a2 YZ |
2322 | } else { |
2323 | BUG(); | |
2324 | } | |
2325 | return ret; | |
2326 | } | |
2327 | ||
2328 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2329 | struct extent_buffer *leaf, | |
2330 | struct btrfs_extent_item *ei) | |
2331 | { | |
2332 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2333 | if (extent_op->update_flags) { | |
2334 | flags |= extent_op->flags_to_set; | |
2335 | btrfs_set_extent_flags(leaf, ei, flags); | |
2336 | } | |
2337 | ||
2338 | if (extent_op->update_key) { | |
2339 | struct btrfs_tree_block_info *bi; | |
2340 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2341 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2342 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2343 | } | |
2344 | } | |
2345 | ||
2346 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2347 | struct btrfs_fs_info *fs_info, |
d278850e | 2348 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
2349 | struct btrfs_delayed_extent_op *extent_op) |
2350 | { | |
2351 | struct btrfs_key key; | |
2352 | struct btrfs_path *path; | |
2353 | struct btrfs_extent_item *ei; | |
2354 | struct extent_buffer *leaf; | |
2355 | u32 item_size; | |
56bec294 | 2356 | int ret; |
5d4f98a2 | 2357 | int err = 0; |
b1c79e09 | 2358 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2359 | |
79787eaa JM |
2360 | if (trans->aborted) |
2361 | return 0; | |
2362 | ||
0b246afa | 2363 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
2364 | metadata = 0; |
2365 | ||
5d4f98a2 YZ |
2366 | path = btrfs_alloc_path(); |
2367 | if (!path) | |
2368 | return -ENOMEM; | |
2369 | ||
d278850e | 2370 | key.objectid = head->bytenr; |
5d4f98a2 | 2371 | |
3173a18f | 2372 | if (metadata) { |
3173a18f | 2373 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2374 | key.offset = extent_op->level; |
3173a18f JB |
2375 | } else { |
2376 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 2377 | key.offset = head->num_bytes; |
3173a18f JB |
2378 | } |
2379 | ||
2380 | again: | |
e4058b54 | 2381 | path->reada = READA_FORWARD; |
5d4f98a2 | 2382 | path->leave_spinning = 1; |
0b246afa | 2383 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
2384 | if (ret < 0) { |
2385 | err = ret; | |
2386 | goto out; | |
2387 | } | |
2388 | if (ret > 0) { | |
3173a18f | 2389 | if (metadata) { |
55994887 FDBM |
2390 | if (path->slots[0] > 0) { |
2391 | path->slots[0]--; | |
2392 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2393 | path->slots[0]); | |
d278850e | 2394 | if (key.objectid == head->bytenr && |
55994887 | 2395 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 2396 | key.offset == head->num_bytes) |
55994887 FDBM |
2397 | ret = 0; |
2398 | } | |
2399 | if (ret > 0) { | |
2400 | btrfs_release_path(path); | |
2401 | metadata = 0; | |
3173a18f | 2402 | |
d278850e JB |
2403 | key.objectid = head->bytenr; |
2404 | key.offset = head->num_bytes; | |
55994887 FDBM |
2405 | key.type = BTRFS_EXTENT_ITEM_KEY; |
2406 | goto again; | |
2407 | } | |
2408 | } else { | |
2409 | err = -EIO; | |
2410 | goto out; | |
3173a18f | 2411 | } |
5d4f98a2 YZ |
2412 | } |
2413 | ||
2414 | leaf = path->nodes[0]; | |
2415 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2416 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2417 | if (item_size < sizeof(*ei)) { | |
87bde3cd | 2418 | ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0); |
5d4f98a2 YZ |
2419 | if (ret < 0) { |
2420 | err = ret; | |
2421 | goto out; | |
2422 | } | |
2423 | leaf = path->nodes[0]; | |
2424 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2425 | } | |
2426 | #endif | |
2427 | BUG_ON(item_size < sizeof(*ei)); | |
2428 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2429 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2430 | |
5d4f98a2 YZ |
2431 | btrfs_mark_buffer_dirty(leaf); |
2432 | out: | |
2433 | btrfs_free_path(path); | |
2434 | return err; | |
56bec294 CM |
2435 | } |
2436 | ||
5d4f98a2 | 2437 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2438 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2439 | struct btrfs_delayed_ref_node *node, |
2440 | struct btrfs_delayed_extent_op *extent_op, | |
2441 | int insert_reserved) | |
56bec294 CM |
2442 | { |
2443 | int ret = 0; | |
5d4f98a2 YZ |
2444 | struct btrfs_delayed_tree_ref *ref; |
2445 | struct btrfs_key ins; | |
2446 | u64 parent = 0; | |
2447 | u64 ref_root = 0; | |
0b246afa | 2448 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
56bec294 | 2449 | |
5d4f98a2 | 2450 | ref = btrfs_delayed_node_to_tree_ref(node); |
0b246afa | 2451 | trace_run_delayed_tree_ref(fs_info, node, ref, node->action); |
599c75ec | 2452 | |
5d4f98a2 YZ |
2453 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2454 | parent = ref->parent; | |
fcebe456 | 2455 | ref_root = ref->root; |
5d4f98a2 | 2456 | |
3173a18f JB |
2457 | ins.objectid = node->bytenr; |
2458 | if (skinny_metadata) { | |
2459 | ins.offset = ref->level; | |
2460 | ins.type = BTRFS_METADATA_ITEM_KEY; | |
2461 | } else { | |
2462 | ins.offset = node->num_bytes; | |
2463 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2464 | } | |
2465 | ||
02794222 | 2466 | if (node->ref_mod != 1) { |
2ff7e61e | 2467 | btrfs_err(fs_info, |
02794222 LB |
2468 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
2469 | node->bytenr, node->ref_mod, node->action, ref_root, | |
2470 | parent); | |
2471 | return -EIO; | |
2472 | } | |
5d4f98a2 | 2473 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 2474 | BUG_ON(!extent_op || !extent_op->update_flags); |
2ff7e61e | 2475 | ret = alloc_reserved_tree_block(trans, fs_info, |
5d4f98a2 YZ |
2476 | parent, ref_root, |
2477 | extent_op->flags_to_set, | |
2478 | &extent_op->key, | |
b06c4bf5 | 2479 | ref->level, &ins); |
5d4f98a2 | 2480 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2ff7e61e | 2481 | ret = __btrfs_inc_extent_ref(trans, fs_info, node, |
c682f9b3 QW |
2482 | parent, ref_root, |
2483 | ref->level, 0, 1, | |
fcebe456 | 2484 | extent_op); |
5d4f98a2 | 2485 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
2ff7e61e | 2486 | ret = __btrfs_free_extent(trans, fs_info, node, |
c682f9b3 QW |
2487 | parent, ref_root, |
2488 | ref->level, 0, 1, extent_op); | |
5d4f98a2 YZ |
2489 | } else { |
2490 | BUG(); | |
2491 | } | |
56bec294 CM |
2492 | return ret; |
2493 | } | |
2494 | ||
2495 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 2496 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2497 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2498 | struct btrfs_delayed_ref_node *node, |
2499 | struct btrfs_delayed_extent_op *extent_op, | |
2500 | int insert_reserved) | |
56bec294 | 2501 | { |
79787eaa JM |
2502 | int ret = 0; |
2503 | ||
857cc2fc JB |
2504 | if (trans->aborted) { |
2505 | if (insert_reserved) | |
2ff7e61e | 2506 | btrfs_pin_extent(fs_info, node->bytenr, |
857cc2fc | 2507 | node->num_bytes, 1); |
79787eaa | 2508 | return 0; |
857cc2fc | 2509 | } |
79787eaa | 2510 | |
5d4f98a2 YZ |
2511 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2512 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2ff7e61e | 2513 | ret = run_delayed_tree_ref(trans, fs_info, node, extent_op, |
5d4f98a2 YZ |
2514 | insert_reserved); |
2515 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2516 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2ff7e61e | 2517 | ret = run_delayed_data_ref(trans, fs_info, node, extent_op, |
5d4f98a2 YZ |
2518 | insert_reserved); |
2519 | else | |
2520 | BUG(); | |
2521 | return ret; | |
56bec294 CM |
2522 | } |
2523 | ||
c6fc2454 | 2524 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2525 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2526 | { | |
cffc3374 FM |
2527 | struct btrfs_delayed_ref_node *ref; |
2528 | ||
0e0adbcf | 2529 | if (RB_EMPTY_ROOT(&head->ref_tree)) |
c6fc2454 | 2530 | return NULL; |
d7df2c79 | 2531 | |
cffc3374 FM |
2532 | /* |
2533 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2534 | * This is to prevent a ref count from going down to zero, which deletes | |
2535 | * the extent item from the extent tree, when there still are references | |
2536 | * to add, which would fail because they would not find the extent item. | |
2537 | */ | |
1d57ee94 WX |
2538 | if (!list_empty(&head->ref_add_list)) |
2539 | return list_first_entry(&head->ref_add_list, | |
2540 | struct btrfs_delayed_ref_node, add_list); | |
2541 | ||
0e0adbcf JB |
2542 | ref = rb_entry(rb_first(&head->ref_tree), |
2543 | struct btrfs_delayed_ref_node, ref_node); | |
1d57ee94 WX |
2544 | ASSERT(list_empty(&ref->add_list)); |
2545 | return ref; | |
56bec294 CM |
2546 | } |
2547 | ||
2eadaa22 JB |
2548 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
2549 | struct btrfs_delayed_ref_head *head) | |
2550 | { | |
2551 | spin_lock(&delayed_refs->lock); | |
2552 | head->processing = 0; | |
2553 | delayed_refs->num_heads_ready++; | |
2554 | spin_unlock(&delayed_refs->lock); | |
2555 | btrfs_delayed_ref_unlock(head); | |
2556 | } | |
2557 | ||
b00e6250 JB |
2558 | static int cleanup_extent_op(struct btrfs_trans_handle *trans, |
2559 | struct btrfs_fs_info *fs_info, | |
2560 | struct btrfs_delayed_ref_head *head) | |
2561 | { | |
2562 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
2563 | int ret; | |
2564 | ||
2565 | if (!extent_op) | |
2566 | return 0; | |
2567 | head->extent_op = NULL; | |
2568 | if (head->must_insert_reserved) { | |
2569 | btrfs_free_delayed_extent_op(extent_op); | |
2570 | return 0; | |
2571 | } | |
2572 | spin_unlock(&head->lock); | |
d278850e | 2573 | ret = run_delayed_extent_op(trans, fs_info, head, extent_op); |
b00e6250 JB |
2574 | btrfs_free_delayed_extent_op(extent_op); |
2575 | return ret ? ret : 1; | |
2576 | } | |
2577 | ||
194ab0bc JB |
2578 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
2579 | struct btrfs_fs_info *fs_info, | |
2580 | struct btrfs_delayed_ref_head *head) | |
2581 | { | |
2582 | struct btrfs_delayed_ref_root *delayed_refs; | |
2583 | int ret; | |
2584 | ||
2585 | delayed_refs = &trans->transaction->delayed_refs; | |
2586 | ||
2587 | ret = cleanup_extent_op(trans, fs_info, head); | |
2588 | if (ret < 0) { | |
2589 | unselect_delayed_ref_head(delayed_refs, head); | |
2590 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
2591 | return ret; | |
2592 | } else if (ret) { | |
2593 | return ret; | |
2594 | } | |
2595 | ||
2596 | /* | |
2597 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
2598 | * and then re-check to make sure nobody got added. | |
2599 | */ | |
2600 | spin_unlock(&head->lock); | |
2601 | spin_lock(&delayed_refs->lock); | |
2602 | spin_lock(&head->lock); | |
0e0adbcf | 2603 | if (!RB_EMPTY_ROOT(&head->ref_tree) || head->extent_op) { |
194ab0bc JB |
2604 | spin_unlock(&head->lock); |
2605 | spin_unlock(&delayed_refs->lock); | |
2606 | return 1; | |
2607 | } | |
194ab0bc JB |
2608 | delayed_refs->num_heads--; |
2609 | rb_erase(&head->href_node, &delayed_refs->href_root); | |
d278850e | 2610 | RB_CLEAR_NODE(&head->href_node); |
194ab0bc | 2611 | spin_unlock(&delayed_refs->lock); |
c1103f7a JB |
2612 | spin_unlock(&head->lock); |
2613 | atomic_dec(&delayed_refs->num_entries); | |
2614 | ||
d278850e | 2615 | trace_run_delayed_ref_head(fs_info, head, 0); |
c1103f7a JB |
2616 | |
2617 | if (head->total_ref_mod < 0) { | |
2618 | struct btrfs_block_group_cache *cache; | |
2619 | ||
d278850e | 2620 | cache = btrfs_lookup_block_group(fs_info, head->bytenr); |
c1103f7a JB |
2621 | ASSERT(cache); |
2622 | percpu_counter_add(&cache->space_info->total_bytes_pinned, | |
d278850e | 2623 | -head->num_bytes); |
c1103f7a JB |
2624 | btrfs_put_block_group(cache); |
2625 | ||
2626 | if (head->is_data) { | |
2627 | spin_lock(&delayed_refs->lock); | |
d278850e | 2628 | delayed_refs->pending_csums -= head->num_bytes; |
c1103f7a JB |
2629 | spin_unlock(&delayed_refs->lock); |
2630 | } | |
2631 | } | |
2632 | ||
2633 | if (head->must_insert_reserved) { | |
d278850e JB |
2634 | btrfs_pin_extent(fs_info, head->bytenr, |
2635 | head->num_bytes, 1); | |
c1103f7a | 2636 | if (head->is_data) { |
d278850e JB |
2637 | ret = btrfs_del_csums(trans, fs_info, head->bytenr, |
2638 | head->num_bytes); | |
c1103f7a JB |
2639 | } |
2640 | } | |
2641 | ||
2642 | /* Also free its reserved qgroup space */ | |
2643 | btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root, | |
2644 | head->qgroup_reserved); | |
2645 | btrfs_delayed_ref_unlock(head); | |
d278850e | 2646 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
2647 | return 0; |
2648 | } | |
2649 | ||
79787eaa JM |
2650 | /* |
2651 | * Returns 0 on success or if called with an already aborted transaction. | |
2652 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2653 | */ | |
d7df2c79 | 2654 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2655 | struct btrfs_fs_info *fs_info, |
d7df2c79 | 2656 | unsigned long nr) |
56bec294 | 2657 | { |
56bec294 CM |
2658 | struct btrfs_delayed_ref_root *delayed_refs; |
2659 | struct btrfs_delayed_ref_node *ref; | |
2660 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2661 | struct btrfs_delayed_extent_op *extent_op; |
0a2b2a84 | 2662 | ktime_t start = ktime_get(); |
56bec294 | 2663 | int ret; |
d7df2c79 | 2664 | unsigned long count = 0; |
0a2b2a84 | 2665 | unsigned long actual_count = 0; |
56bec294 | 2666 | int must_insert_reserved = 0; |
56bec294 CM |
2667 | |
2668 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2669 | while (1) { |
2670 | if (!locked_ref) { | |
d7df2c79 | 2671 | if (count >= nr) |
56bec294 | 2672 | break; |
56bec294 | 2673 | |
d7df2c79 JB |
2674 | spin_lock(&delayed_refs->lock); |
2675 | locked_ref = btrfs_select_ref_head(trans); | |
2676 | if (!locked_ref) { | |
2677 | spin_unlock(&delayed_refs->lock); | |
2678 | break; | |
2679 | } | |
c3e69d58 CM |
2680 | |
2681 | /* grab the lock that says we are going to process | |
2682 | * all the refs for this head */ | |
2683 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
d7df2c79 | 2684 | spin_unlock(&delayed_refs->lock); |
c3e69d58 CM |
2685 | /* |
2686 | * we may have dropped the spin lock to get the head | |
2687 | * mutex lock, and that might have given someone else | |
2688 | * time to free the head. If that's true, it has been | |
2689 | * removed from our list and we can move on. | |
2690 | */ | |
2691 | if (ret == -EAGAIN) { | |
2692 | locked_ref = NULL; | |
2693 | count++; | |
2694 | continue; | |
56bec294 CM |
2695 | } |
2696 | } | |
a28ec197 | 2697 | |
2c3cf7d5 FM |
2698 | /* |
2699 | * We need to try and merge add/drops of the same ref since we | |
2700 | * can run into issues with relocate dropping the implicit ref | |
2701 | * and then it being added back again before the drop can | |
2702 | * finish. If we merged anything we need to re-loop so we can | |
2703 | * get a good ref. | |
2704 | * Or we can get node references of the same type that weren't | |
2705 | * merged when created due to bumps in the tree mod seq, and | |
2706 | * we need to merge them to prevent adding an inline extent | |
2707 | * backref before dropping it (triggering a BUG_ON at | |
2708 | * insert_inline_extent_backref()). | |
2709 | */ | |
d7df2c79 | 2710 | spin_lock(&locked_ref->lock); |
2c3cf7d5 FM |
2711 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, |
2712 | locked_ref); | |
ae1e206b | 2713 | |
d1270cd9 AJ |
2714 | /* |
2715 | * locked_ref is the head node, so we have to go one | |
2716 | * node back for any delayed ref updates | |
2717 | */ | |
2718 | ref = select_delayed_ref(locked_ref); | |
2719 | ||
2720 | if (ref && ref->seq && | |
097b8a7c | 2721 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d7df2c79 | 2722 | spin_unlock(&locked_ref->lock); |
2eadaa22 | 2723 | unselect_delayed_ref_head(delayed_refs, locked_ref); |
d7df2c79 | 2724 | locked_ref = NULL; |
d1270cd9 | 2725 | cond_resched(); |
27a377db | 2726 | count++; |
d1270cd9 AJ |
2727 | continue; |
2728 | } | |
2729 | ||
c1103f7a JB |
2730 | /* |
2731 | * We're done processing refs in this ref_head, clean everything | |
2732 | * up and move on to the next ref_head. | |
2733 | */ | |
56bec294 | 2734 | if (!ref) { |
194ab0bc JB |
2735 | ret = cleanup_ref_head(trans, fs_info, locked_ref); |
2736 | if (ret > 0 ) { | |
b00e6250 JB |
2737 | /* We dropped our lock, we need to loop. */ |
2738 | ret = 0; | |
d7df2c79 | 2739 | continue; |
194ab0bc JB |
2740 | } else if (ret) { |
2741 | return ret; | |
5d4f98a2 | 2742 | } |
c1103f7a JB |
2743 | locked_ref = NULL; |
2744 | count++; | |
2745 | continue; | |
2746 | } | |
02217ed2 | 2747 | |
c1103f7a JB |
2748 | actual_count++; |
2749 | ref->in_tree = 0; | |
0e0adbcf JB |
2750 | rb_erase(&ref->ref_node, &locked_ref->ref_tree); |
2751 | RB_CLEAR_NODE(&ref->ref_node); | |
c1103f7a JB |
2752 | if (!list_empty(&ref->add_list)) |
2753 | list_del(&ref->add_list); | |
2754 | /* | |
2755 | * When we play the delayed ref, also correct the ref_mod on | |
2756 | * head | |
2757 | */ | |
2758 | switch (ref->action) { | |
2759 | case BTRFS_ADD_DELAYED_REF: | |
2760 | case BTRFS_ADD_DELAYED_EXTENT: | |
d278850e | 2761 | locked_ref->ref_mod -= ref->ref_mod; |
c1103f7a JB |
2762 | break; |
2763 | case BTRFS_DROP_DELAYED_REF: | |
d278850e | 2764 | locked_ref->ref_mod += ref->ref_mod; |
c1103f7a JB |
2765 | break; |
2766 | default: | |
2767 | WARN_ON(1); | |
22cd2e7d | 2768 | } |
1ce7a5ec JB |
2769 | atomic_dec(&delayed_refs->num_entries); |
2770 | ||
b00e6250 JB |
2771 | /* |
2772 | * Record the must-insert_reserved flag before we drop the spin | |
2773 | * lock. | |
2774 | */ | |
2775 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2776 | locked_ref->must_insert_reserved = 0; | |
2777 | ||
2778 | extent_op = locked_ref->extent_op; | |
2779 | locked_ref->extent_op = NULL; | |
d7df2c79 | 2780 | spin_unlock(&locked_ref->lock); |
925baedd | 2781 | |
2ff7e61e | 2782 | ret = run_one_delayed_ref(trans, fs_info, ref, extent_op, |
56bec294 | 2783 | must_insert_reserved); |
eb099670 | 2784 | |
78a6184a | 2785 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2786 | if (ret) { |
2eadaa22 | 2787 | unselect_delayed_ref_head(delayed_refs, locked_ref); |
093486c4 | 2788 | btrfs_put_delayed_ref(ref); |
5d163e0e JM |
2789 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", |
2790 | ret); | |
79787eaa JM |
2791 | return ret; |
2792 | } | |
2793 | ||
093486c4 MX |
2794 | btrfs_put_delayed_ref(ref); |
2795 | count++; | |
c3e69d58 | 2796 | cond_resched(); |
c3e69d58 | 2797 | } |
0a2b2a84 JB |
2798 | |
2799 | /* | |
2800 | * We don't want to include ref heads since we can have empty ref heads | |
2801 | * and those will drastically skew our runtime down since we just do | |
2802 | * accounting, no actual extent tree updates. | |
2803 | */ | |
2804 | if (actual_count > 0) { | |
2805 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2806 | u64 avg; | |
2807 | ||
2808 | /* | |
2809 | * We weigh the current average higher than our current runtime | |
2810 | * to avoid large swings in the average. | |
2811 | */ | |
2812 | spin_lock(&delayed_refs->lock); | |
2813 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2814 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2815 | spin_unlock(&delayed_refs->lock); |
2816 | } | |
d7df2c79 | 2817 | return 0; |
c3e69d58 CM |
2818 | } |
2819 | ||
709c0486 AJ |
2820 | #ifdef SCRAMBLE_DELAYED_REFS |
2821 | /* | |
2822 | * Normally delayed refs get processed in ascending bytenr order. This | |
2823 | * correlates in most cases to the order added. To expose dependencies on this | |
2824 | * order, we start to process the tree in the middle instead of the beginning | |
2825 | */ | |
2826 | static u64 find_middle(struct rb_root *root) | |
2827 | { | |
2828 | struct rb_node *n = root->rb_node; | |
2829 | struct btrfs_delayed_ref_node *entry; | |
2830 | int alt = 1; | |
2831 | u64 middle; | |
2832 | u64 first = 0, last = 0; | |
2833 | ||
2834 | n = rb_first(root); | |
2835 | if (n) { | |
2836 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2837 | first = entry->bytenr; | |
2838 | } | |
2839 | n = rb_last(root); | |
2840 | if (n) { | |
2841 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2842 | last = entry->bytenr; | |
2843 | } | |
2844 | n = root->rb_node; | |
2845 | ||
2846 | while (n) { | |
2847 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2848 | WARN_ON(!entry->in_tree); | |
2849 | ||
2850 | middle = entry->bytenr; | |
2851 | ||
2852 | if (alt) | |
2853 | n = n->rb_left; | |
2854 | else | |
2855 | n = n->rb_right; | |
2856 | ||
2857 | alt = 1 - alt; | |
2858 | } | |
2859 | return middle; | |
2860 | } | |
2861 | #endif | |
2862 | ||
2ff7e61e | 2863 | static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads) |
1be41b78 JB |
2864 | { |
2865 | u64 num_bytes; | |
2866 | ||
2867 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2868 | sizeof(struct btrfs_extent_inline_ref)); | |
0b246afa | 2869 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1be41b78 JB |
2870 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
2871 | ||
2872 | /* | |
2873 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2874 | * closer to what we're really going to want to use. |
1be41b78 | 2875 | */ |
0b246afa | 2876 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info)); |
1be41b78 JB |
2877 | } |
2878 | ||
1262133b JB |
2879 | /* |
2880 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2881 | * would require to store the csums for that many bytes. | |
2882 | */ | |
2ff7e61e | 2883 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2884 | { |
2885 | u64 csum_size; | |
2886 | u64 num_csums_per_leaf; | |
2887 | u64 num_csums; | |
2888 | ||
0b246afa | 2889 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2890 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2891 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2892 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2893 | num_csums += num_csums_per_leaf - 1; |
2894 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2895 | return num_csums; | |
2896 | } | |
2897 | ||
0a2b2a84 | 2898 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2899 | struct btrfs_fs_info *fs_info) |
1be41b78 JB |
2900 | { |
2901 | struct btrfs_block_rsv *global_rsv; | |
2902 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
1262133b | 2903 | u64 csum_bytes = trans->transaction->delayed_refs.pending_csums; |
165c8b02 | 2904 | unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs; |
cb723e49 | 2905 | u64 num_bytes, num_dirty_bgs_bytes; |
1be41b78 JB |
2906 | int ret = 0; |
2907 | ||
0b246afa | 2908 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
2ff7e61e | 2909 | num_heads = heads_to_leaves(fs_info, num_heads); |
1be41b78 | 2910 | if (num_heads > 1) |
0b246afa | 2911 | num_bytes += (num_heads - 1) * fs_info->nodesize; |
1be41b78 | 2912 | num_bytes <<= 1; |
2ff7e61e JM |
2913 | num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) * |
2914 | fs_info->nodesize; | |
0b246afa | 2915 | num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info, |
cb723e49 | 2916 | num_dirty_bgs); |
0b246afa | 2917 | global_rsv = &fs_info->global_block_rsv; |
1be41b78 JB |
2918 | |
2919 | /* | |
2920 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2921 | * wiggle room since running delayed refs can create more delayed refs. | |
2922 | */ | |
cb723e49 JB |
2923 | if (global_rsv->space_info->full) { |
2924 | num_dirty_bgs_bytes <<= 1; | |
1be41b78 | 2925 | num_bytes <<= 1; |
cb723e49 | 2926 | } |
1be41b78 JB |
2927 | |
2928 | spin_lock(&global_rsv->lock); | |
cb723e49 | 2929 | if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes) |
1be41b78 JB |
2930 | ret = 1; |
2931 | spin_unlock(&global_rsv->lock); | |
2932 | return ret; | |
2933 | } | |
2934 | ||
0a2b2a84 | 2935 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2936 | struct btrfs_fs_info *fs_info) |
0a2b2a84 | 2937 | { |
0a2b2a84 JB |
2938 | u64 num_entries = |
2939 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2940 | u64 avg_runtime; | |
a79b7d4b | 2941 | u64 val; |
0a2b2a84 JB |
2942 | |
2943 | smp_mb(); | |
2944 | avg_runtime = fs_info->avg_delayed_ref_runtime; | |
a79b7d4b | 2945 | val = num_entries * avg_runtime; |
dc1a90c6 | 2946 | if (val >= NSEC_PER_SEC) |
0a2b2a84 | 2947 | return 1; |
a79b7d4b CM |
2948 | if (val >= NSEC_PER_SEC / 2) |
2949 | return 2; | |
0a2b2a84 | 2950 | |
2ff7e61e | 2951 | return btrfs_check_space_for_delayed_refs(trans, fs_info); |
0a2b2a84 JB |
2952 | } |
2953 | ||
a79b7d4b CM |
2954 | struct async_delayed_refs { |
2955 | struct btrfs_root *root; | |
31b9655f | 2956 | u64 transid; |
a79b7d4b CM |
2957 | int count; |
2958 | int error; | |
2959 | int sync; | |
2960 | struct completion wait; | |
2961 | struct btrfs_work work; | |
2962 | }; | |
2963 | ||
2ff7e61e JM |
2964 | static inline struct async_delayed_refs * |
2965 | to_async_delayed_refs(struct btrfs_work *work) | |
2966 | { | |
2967 | return container_of(work, struct async_delayed_refs, work); | |
2968 | } | |
2969 | ||
a79b7d4b CM |
2970 | static void delayed_ref_async_start(struct btrfs_work *work) |
2971 | { | |
2ff7e61e | 2972 | struct async_delayed_refs *async = to_async_delayed_refs(work); |
a79b7d4b | 2973 | struct btrfs_trans_handle *trans; |
2ff7e61e | 2974 | struct btrfs_fs_info *fs_info = async->root->fs_info; |
a79b7d4b CM |
2975 | int ret; |
2976 | ||
0f873eca | 2977 | /* if the commit is already started, we don't need to wait here */ |
2ff7e61e | 2978 | if (btrfs_transaction_blocked(fs_info)) |
31b9655f | 2979 | goto done; |
31b9655f | 2980 | |
0f873eca CM |
2981 | trans = btrfs_join_transaction(async->root); |
2982 | if (IS_ERR(trans)) { | |
2983 | async->error = PTR_ERR(trans); | |
a79b7d4b CM |
2984 | goto done; |
2985 | } | |
2986 | ||
2987 | /* | |
01327610 | 2988 | * trans->sync means that when we call end_transaction, we won't |
a79b7d4b CM |
2989 | * wait on delayed refs |
2990 | */ | |
2991 | trans->sync = true; | |
0f873eca CM |
2992 | |
2993 | /* Don't bother flushing if we got into a different transaction */ | |
2994 | if (trans->transid > async->transid) | |
2995 | goto end; | |
2996 | ||
2ff7e61e | 2997 | ret = btrfs_run_delayed_refs(trans, fs_info, async->count); |
a79b7d4b CM |
2998 | if (ret) |
2999 | async->error = ret; | |
0f873eca | 3000 | end: |
3a45bb20 | 3001 | ret = btrfs_end_transaction(trans); |
a79b7d4b CM |
3002 | if (ret && !async->error) |
3003 | async->error = ret; | |
3004 | done: | |
3005 | if (async->sync) | |
3006 | complete(&async->wait); | |
3007 | else | |
3008 | kfree(async); | |
3009 | } | |
3010 | ||
2ff7e61e | 3011 | int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info, |
31b9655f | 3012 | unsigned long count, u64 transid, int wait) |
a79b7d4b CM |
3013 | { |
3014 | struct async_delayed_refs *async; | |
3015 | int ret; | |
3016 | ||
3017 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
3018 | if (!async) | |
3019 | return -ENOMEM; | |
3020 | ||
0b246afa | 3021 | async->root = fs_info->tree_root; |
a79b7d4b CM |
3022 | async->count = count; |
3023 | async->error = 0; | |
31b9655f | 3024 | async->transid = transid; |
a79b7d4b CM |
3025 | if (wait) |
3026 | async->sync = 1; | |
3027 | else | |
3028 | async->sync = 0; | |
3029 | init_completion(&async->wait); | |
3030 | ||
9e0af237 LB |
3031 | btrfs_init_work(&async->work, btrfs_extent_refs_helper, |
3032 | delayed_ref_async_start, NULL, NULL); | |
a79b7d4b | 3033 | |
0b246afa | 3034 | btrfs_queue_work(fs_info->extent_workers, &async->work); |
a79b7d4b CM |
3035 | |
3036 | if (wait) { | |
3037 | wait_for_completion(&async->wait); | |
3038 | ret = async->error; | |
3039 | kfree(async); | |
3040 | return ret; | |
3041 | } | |
3042 | return 0; | |
3043 | } | |
3044 | ||
c3e69d58 CM |
3045 | /* |
3046 | * this starts processing the delayed reference count updates and | |
3047 | * extent insertions we have queued up so far. count can be | |
3048 | * 0, which means to process everything in the tree at the start | |
3049 | * of the run (but not newly added entries), or it can be some target | |
3050 | * number you'd like to process. | |
79787eaa JM |
3051 | * |
3052 | * Returns 0 on success or if called with an aborted transaction | |
3053 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
3054 | */ |
3055 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3056 | struct btrfs_fs_info *fs_info, unsigned long count) |
c3e69d58 CM |
3057 | { |
3058 | struct rb_node *node; | |
3059 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 3060 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
3061 | int ret; |
3062 | int run_all = count == (unsigned long)-1; | |
d9a0540a | 3063 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
c3e69d58 | 3064 | |
79787eaa JM |
3065 | /* We'll clean this up in btrfs_cleanup_transaction */ |
3066 | if (trans->aborted) | |
3067 | return 0; | |
3068 | ||
0b246afa | 3069 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
3070 | return 0; |
3071 | ||
c3e69d58 | 3072 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 3073 | if (count == 0) |
d7df2c79 | 3074 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 3075 | |
c3e69d58 | 3076 | again: |
709c0486 AJ |
3077 | #ifdef SCRAMBLE_DELAYED_REFS |
3078 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
3079 | #endif | |
d9a0540a | 3080 | trans->can_flush_pending_bgs = false; |
2ff7e61e | 3081 | ret = __btrfs_run_delayed_refs(trans, fs_info, count); |
d7df2c79 | 3082 | if (ret < 0) { |
66642832 | 3083 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 3084 | return ret; |
eb099670 | 3085 | } |
c3e69d58 | 3086 | |
56bec294 | 3087 | if (run_all) { |
d7df2c79 | 3088 | if (!list_empty(&trans->new_bgs)) |
6c686b35 | 3089 | btrfs_create_pending_block_groups(trans); |
ea658bad | 3090 | |
d7df2c79 | 3091 | spin_lock(&delayed_refs->lock); |
c46effa6 | 3092 | node = rb_first(&delayed_refs->href_root); |
d7df2c79 JB |
3093 | if (!node) { |
3094 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 3095 | goto out; |
d7df2c79 | 3096 | } |
d278850e JB |
3097 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
3098 | href_node); | |
3099 | refcount_inc(&head->refs); | |
3100 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 3101 | |
d278850e JB |
3102 | /* Mutex was contended, block until it's released and retry. */ |
3103 | mutex_lock(&head->mutex); | |
3104 | mutex_unlock(&head->mutex); | |
56bec294 | 3105 | |
d278850e | 3106 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 3107 | cond_resched(); |
56bec294 | 3108 | goto again; |
5f39d397 | 3109 | } |
54aa1f4d | 3110 | out: |
d9a0540a | 3111 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
a28ec197 CM |
3112 | return 0; |
3113 | } | |
3114 | ||
5d4f98a2 | 3115 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2ff7e61e | 3116 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 3117 | u64 bytenr, u64 num_bytes, u64 flags, |
b1c79e09 | 3118 | int level, int is_data) |
5d4f98a2 YZ |
3119 | { |
3120 | struct btrfs_delayed_extent_op *extent_op; | |
3121 | int ret; | |
3122 | ||
78a6184a | 3123 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
3124 | if (!extent_op) |
3125 | return -ENOMEM; | |
3126 | ||
3127 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
3128 | extent_op->update_flags = true; |
3129 | extent_op->update_key = false; | |
3130 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 3131 | extent_op->level = level; |
5d4f98a2 | 3132 | |
0b246afa | 3133 | ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr, |
66d7e7f0 | 3134 | num_bytes, extent_op); |
5d4f98a2 | 3135 | if (ret) |
78a6184a | 3136 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
3137 | return ret; |
3138 | } | |
3139 | ||
e4c3b2dc | 3140 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3141 | struct btrfs_path *path, |
3142 | u64 objectid, u64 offset, u64 bytenr) | |
3143 | { | |
3144 | struct btrfs_delayed_ref_head *head; | |
3145 | struct btrfs_delayed_ref_node *ref; | |
3146 | struct btrfs_delayed_data_ref *data_ref; | |
3147 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 3148 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 3149 | struct rb_node *node; |
5d4f98a2 YZ |
3150 | int ret = 0; |
3151 | ||
e4c3b2dc LB |
3152 | cur_trans = root->fs_info->running_transaction; |
3153 | if (!cur_trans) | |
3154 | return 0; | |
3155 | ||
3156 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 3157 | spin_lock(&delayed_refs->lock); |
f72ad18e | 3158 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
3159 | if (!head) { |
3160 | spin_unlock(&delayed_refs->lock); | |
3161 | return 0; | |
3162 | } | |
5d4f98a2 YZ |
3163 | |
3164 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 3165 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
3166 | spin_unlock(&delayed_refs->lock); |
3167 | ||
b3b4aa74 | 3168 | btrfs_release_path(path); |
5d4f98a2 | 3169 | |
8cc33e5c DS |
3170 | /* |
3171 | * Mutex was contended, block until it's released and let | |
3172 | * caller try again | |
3173 | */ | |
5d4f98a2 YZ |
3174 | mutex_lock(&head->mutex); |
3175 | mutex_unlock(&head->mutex); | |
d278850e | 3176 | btrfs_put_delayed_ref_head(head); |
5d4f98a2 YZ |
3177 | return -EAGAIN; |
3178 | } | |
d7df2c79 | 3179 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 3180 | |
d7df2c79 | 3181 | spin_lock(&head->lock); |
0e0adbcf JB |
3182 | /* |
3183 | * XXX: We should replace this with a proper search function in the | |
3184 | * future. | |
3185 | */ | |
3186 | for (node = rb_first(&head->ref_tree); node; node = rb_next(node)) { | |
3187 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); | |
d7df2c79 JB |
3188 | /* If it's a shared ref we know a cross reference exists */ |
3189 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3190 | ret = 1; | |
3191 | break; | |
3192 | } | |
5d4f98a2 | 3193 | |
d7df2c79 | 3194 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 3195 | |
d7df2c79 JB |
3196 | /* |
3197 | * If our ref doesn't match the one we're currently looking at | |
3198 | * then we have a cross reference. | |
3199 | */ | |
3200 | if (data_ref->root != root->root_key.objectid || | |
3201 | data_ref->objectid != objectid || | |
3202 | data_ref->offset != offset) { | |
3203 | ret = 1; | |
3204 | break; | |
3205 | } | |
5d4f98a2 | 3206 | } |
d7df2c79 | 3207 | spin_unlock(&head->lock); |
5d4f98a2 | 3208 | mutex_unlock(&head->mutex); |
5d4f98a2 YZ |
3209 | return ret; |
3210 | } | |
3211 | ||
e4c3b2dc | 3212 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3213 | struct btrfs_path *path, |
3214 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d | 3215 | { |
0b246afa JM |
3216 | struct btrfs_fs_info *fs_info = root->fs_info; |
3217 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 3218 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3219 | struct btrfs_extent_data_ref *ref; |
3220 | struct btrfs_extent_inline_ref *iref; | |
3221 | struct btrfs_extent_item *ei; | |
f321e491 | 3222 | struct btrfs_key key; |
5d4f98a2 | 3223 | u32 item_size; |
3de28d57 | 3224 | int type; |
be20aa9d | 3225 | int ret; |
925baedd | 3226 | |
be20aa9d | 3227 | key.objectid = bytenr; |
31840ae1 | 3228 | key.offset = (u64)-1; |
f321e491 | 3229 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 3230 | |
be20aa9d CM |
3231 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
3232 | if (ret < 0) | |
3233 | goto out; | |
79787eaa | 3234 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
3235 | |
3236 | ret = -ENOENT; | |
3237 | if (path->slots[0] == 0) | |
31840ae1 | 3238 | goto out; |
be20aa9d | 3239 | |
31840ae1 | 3240 | path->slots[0]--; |
f321e491 | 3241 | leaf = path->nodes[0]; |
5d4f98a2 | 3242 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3243 | |
5d4f98a2 | 3244 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3245 | goto out; |
f321e491 | 3246 | |
5d4f98a2 YZ |
3247 | ret = 1; |
3248 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3249 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3250 | if (item_size < sizeof(*ei)) { | |
3251 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
3252 | goto out; | |
3253 | } | |
3254 | #endif | |
3255 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 3256 | |
5d4f98a2 YZ |
3257 | if (item_size != sizeof(*ei) + |
3258 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3259 | goto out; | |
be20aa9d | 3260 | |
5d4f98a2 YZ |
3261 | if (btrfs_extent_generation(leaf, ei) <= |
3262 | btrfs_root_last_snapshot(&root->root_item)) | |
3263 | goto out; | |
3264 | ||
3265 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 LB |
3266 | |
3267 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
3268 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
3269 | goto out; |
3270 | ||
3271 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3272 | if (btrfs_extent_refs(leaf, ei) != | |
3273 | btrfs_extent_data_ref_count(leaf, ref) || | |
3274 | btrfs_extent_data_ref_root(leaf, ref) != | |
3275 | root->root_key.objectid || | |
3276 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3277 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3278 | goto out; | |
3279 | ||
3280 | ret = 0; | |
3281 | out: | |
3282 | return ret; | |
3283 | } | |
3284 | ||
e4c3b2dc LB |
3285 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
3286 | u64 bytenr) | |
5d4f98a2 YZ |
3287 | { |
3288 | struct btrfs_path *path; | |
3289 | int ret; | |
3290 | int ret2; | |
3291 | ||
3292 | path = btrfs_alloc_path(); | |
3293 | if (!path) | |
3294 | return -ENOENT; | |
3295 | ||
3296 | do { | |
e4c3b2dc | 3297 | ret = check_committed_ref(root, path, objectid, |
5d4f98a2 YZ |
3298 | offset, bytenr); |
3299 | if (ret && ret != -ENOENT) | |
f321e491 | 3300 | goto out; |
80ff3856 | 3301 | |
e4c3b2dc | 3302 | ret2 = check_delayed_ref(root, path, objectid, |
5d4f98a2 YZ |
3303 | offset, bytenr); |
3304 | } while (ret2 == -EAGAIN); | |
3305 | ||
3306 | if (ret2 && ret2 != -ENOENT) { | |
3307 | ret = ret2; | |
3308 | goto out; | |
f321e491 | 3309 | } |
5d4f98a2 YZ |
3310 | |
3311 | if (ret != -ENOENT || ret2 != -ENOENT) | |
3312 | ret = 0; | |
be20aa9d | 3313 | out: |
80ff3856 | 3314 | btrfs_free_path(path); |
f0486c68 YZ |
3315 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3316 | WARN_ON(ret > 0); | |
f321e491 | 3317 | return ret; |
be20aa9d | 3318 | } |
c5739bba | 3319 | |
5d4f98a2 | 3320 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3321 | struct btrfs_root *root, |
5d4f98a2 | 3322 | struct extent_buffer *buf, |
e339a6b0 | 3323 | int full_backref, int inc) |
31840ae1 | 3324 | { |
0b246afa | 3325 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 3326 | u64 bytenr; |
5d4f98a2 YZ |
3327 | u64 num_bytes; |
3328 | u64 parent; | |
31840ae1 | 3329 | u64 ref_root; |
31840ae1 | 3330 | u32 nritems; |
31840ae1 ZY |
3331 | struct btrfs_key key; |
3332 | struct btrfs_file_extent_item *fi; | |
3333 | int i; | |
3334 | int level; | |
3335 | int ret = 0; | |
2ff7e61e | 3336 | int (*process_func)(struct btrfs_trans_handle *, |
84f7d8e6 | 3337 | struct btrfs_root *, |
b06c4bf5 | 3338 | u64, u64, u64, u64, u64, u64); |
31840ae1 | 3339 | |
fccb84c9 | 3340 | |
0b246afa | 3341 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3342 | return 0; |
fccb84c9 | 3343 | |
31840ae1 | 3344 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3345 | nritems = btrfs_header_nritems(buf); |
3346 | level = btrfs_header_level(buf); | |
3347 | ||
27cdeb70 | 3348 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3349 | return 0; |
31840ae1 | 3350 | |
5d4f98a2 YZ |
3351 | if (inc) |
3352 | process_func = btrfs_inc_extent_ref; | |
3353 | else | |
3354 | process_func = btrfs_free_extent; | |
31840ae1 | 3355 | |
5d4f98a2 YZ |
3356 | if (full_backref) |
3357 | parent = buf->start; | |
3358 | else | |
3359 | parent = 0; | |
3360 | ||
3361 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3362 | if (level == 0) { |
5d4f98a2 | 3363 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3364 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3365 | continue; |
5d4f98a2 | 3366 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3367 | struct btrfs_file_extent_item); |
3368 | if (btrfs_file_extent_type(buf, fi) == | |
3369 | BTRFS_FILE_EXTENT_INLINE) | |
3370 | continue; | |
3371 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3372 | if (bytenr == 0) | |
3373 | continue; | |
5d4f98a2 YZ |
3374 | |
3375 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3376 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
84f7d8e6 | 3377 | ret = process_func(trans, root, bytenr, num_bytes, |
5d4f98a2 | 3378 | parent, ref_root, key.objectid, |
b06c4bf5 | 3379 | key.offset); |
31840ae1 ZY |
3380 | if (ret) |
3381 | goto fail; | |
3382 | } else { | |
5d4f98a2 | 3383 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 3384 | num_bytes = fs_info->nodesize; |
84f7d8e6 | 3385 | ret = process_func(trans, root, bytenr, num_bytes, |
b06c4bf5 | 3386 | parent, ref_root, level - 1, 0); |
31840ae1 ZY |
3387 | if (ret) |
3388 | goto fail; | |
3389 | } | |
3390 | } | |
3391 | return 0; | |
3392 | fail: | |
5d4f98a2 YZ |
3393 | return ret; |
3394 | } | |
3395 | ||
3396 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3397 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3398 | { |
e339a6b0 | 3399 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3400 | } |
3401 | ||
3402 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3403 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3404 | { |
e339a6b0 | 3405 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3406 | } |
3407 | ||
9078a3e1 | 3408 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 3409 | struct btrfs_fs_info *fs_info, |
9078a3e1 CM |
3410 | struct btrfs_path *path, |
3411 | struct btrfs_block_group_cache *cache) | |
3412 | { | |
3413 | int ret; | |
0b246afa | 3414 | struct btrfs_root *extent_root = fs_info->extent_root; |
5f39d397 CM |
3415 | unsigned long bi; |
3416 | struct extent_buffer *leaf; | |
9078a3e1 | 3417 | |
9078a3e1 | 3418 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3419 | if (ret) { |
3420 | if (ret > 0) | |
3421 | ret = -ENOENT; | |
54aa1f4d | 3422 | goto fail; |
df95e7f0 | 3423 | } |
5f39d397 CM |
3424 | |
3425 | leaf = path->nodes[0]; | |
3426 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3427 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3428 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3429 | fail: |
24b89d08 | 3430 | btrfs_release_path(path); |
df95e7f0 | 3431 | return ret; |
9078a3e1 CM |
3432 | |
3433 | } | |
3434 | ||
4a8c9a62 | 3435 | static struct btrfs_block_group_cache * |
2ff7e61e | 3436 | next_block_group(struct btrfs_fs_info *fs_info, |
4a8c9a62 YZ |
3437 | struct btrfs_block_group_cache *cache) |
3438 | { | |
3439 | struct rb_node *node; | |
292cbd51 | 3440 | |
0b246afa | 3441 | spin_lock(&fs_info->block_group_cache_lock); |
292cbd51 FM |
3442 | |
3443 | /* If our block group was removed, we need a full search. */ | |
3444 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3445 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3446 | ||
0b246afa | 3447 | spin_unlock(&fs_info->block_group_cache_lock); |
292cbd51 | 3448 | btrfs_put_block_group(cache); |
0b246afa | 3449 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; |
292cbd51 | 3450 | } |
4a8c9a62 YZ |
3451 | node = rb_next(&cache->cache_node); |
3452 | btrfs_put_block_group(cache); | |
3453 | if (node) { | |
3454 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3455 | cache_node); | |
11dfe35a | 3456 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3457 | } else |
3458 | cache = NULL; | |
0b246afa | 3459 | spin_unlock(&fs_info->block_group_cache_lock); |
4a8c9a62 YZ |
3460 | return cache; |
3461 | } | |
3462 | ||
0af3d00b JB |
3463 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3464 | struct btrfs_trans_handle *trans, | |
3465 | struct btrfs_path *path) | |
3466 | { | |
0b246afa JM |
3467 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
3468 | struct btrfs_root *root = fs_info->tree_root; | |
0af3d00b | 3469 | struct inode *inode = NULL; |
364ecf36 | 3470 | struct extent_changeset *data_reserved = NULL; |
0af3d00b | 3471 | u64 alloc_hint = 0; |
2b20982e | 3472 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3473 | u64 num_pages = 0; |
0af3d00b JB |
3474 | int retries = 0; |
3475 | int ret = 0; | |
3476 | ||
3477 | /* | |
3478 | * If this block group is smaller than 100 megs don't bother caching the | |
3479 | * block group. | |
3480 | */ | |
ee22184b | 3481 | if (block_group->key.offset < (100 * SZ_1M)) { |
0af3d00b JB |
3482 | spin_lock(&block_group->lock); |
3483 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3484 | spin_unlock(&block_group->lock); | |
3485 | return 0; | |
3486 | } | |
3487 | ||
0c0ef4bc JB |
3488 | if (trans->aborted) |
3489 | return 0; | |
0af3d00b | 3490 | again: |
77ab86bf | 3491 | inode = lookup_free_space_inode(fs_info, block_group, path); |
0af3d00b JB |
3492 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { |
3493 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3494 | btrfs_release_path(path); |
0af3d00b JB |
3495 | goto out; |
3496 | } | |
3497 | ||
3498 | if (IS_ERR(inode)) { | |
3499 | BUG_ON(retries); | |
3500 | retries++; | |
3501 | ||
3502 | if (block_group->ro) | |
3503 | goto out_free; | |
3504 | ||
77ab86bf JM |
3505 | ret = create_free_space_inode(fs_info, trans, block_group, |
3506 | path); | |
0af3d00b JB |
3507 | if (ret) |
3508 | goto out_free; | |
3509 | goto again; | |
3510 | } | |
3511 | ||
3512 | /* | |
3513 | * We want to set the generation to 0, that way if anything goes wrong | |
3514 | * from here on out we know not to trust this cache when we load up next | |
3515 | * time. | |
3516 | */ | |
3517 | BTRFS_I(inode)->generation = 0; | |
3518 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3519 | if (ret) { |
3520 | /* | |
3521 | * So theoretically we could recover from this, simply set the | |
3522 | * super cache generation to 0 so we know to invalidate the | |
3523 | * cache, but then we'd have to keep track of the block groups | |
3524 | * that fail this way so we know we _have_ to reset this cache | |
3525 | * before the next commit or risk reading stale cache. So to | |
3526 | * limit our exposure to horrible edge cases lets just abort the | |
3527 | * transaction, this only happens in really bad situations | |
3528 | * anyway. | |
3529 | */ | |
66642832 | 3530 | btrfs_abort_transaction(trans, ret); |
0c0ef4bc JB |
3531 | goto out_put; |
3532 | } | |
0af3d00b JB |
3533 | WARN_ON(ret); |
3534 | ||
8e138e0d JB |
3535 | /* We've already setup this transaction, go ahead and exit */ |
3536 | if (block_group->cache_generation == trans->transid && | |
3537 | i_size_read(inode)) { | |
3538 | dcs = BTRFS_DC_SETUP; | |
3539 | goto out_put; | |
3540 | } | |
3541 | ||
0af3d00b | 3542 | if (i_size_read(inode) > 0) { |
2ff7e61e | 3543 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
0b246afa | 3544 | &fs_info->global_block_rsv); |
7b61cd92 MX |
3545 | if (ret) |
3546 | goto out_put; | |
3547 | ||
77ab86bf | 3548 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); |
0af3d00b JB |
3549 | if (ret) |
3550 | goto out_put; | |
3551 | } | |
3552 | ||
3553 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3554 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
0b246afa | 3555 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cf7c1ef6 LB |
3556 | /* |
3557 | * don't bother trying to write stuff out _if_ | |
3558 | * a) we're not cached, | |
1a79c1f2 LB |
3559 | * b) we're with nospace_cache mount option, |
3560 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
cf7c1ef6 | 3561 | */ |
2b20982e | 3562 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3563 | spin_unlock(&block_group->lock); |
3564 | goto out_put; | |
3565 | } | |
3566 | spin_unlock(&block_group->lock); | |
3567 | ||
2968b1f4 JB |
3568 | /* |
3569 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3570 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3571 | */ | |
3572 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3573 | ret = -ENOSPC; | |
3574 | goto out_put; | |
3575 | } | |
3576 | ||
6fc823b1 JB |
3577 | /* |
3578 | * Try to preallocate enough space based on how big the block group is. | |
3579 | * Keep in mind this has to include any pinned space which could end up | |
3580 | * taking up quite a bit since it's not folded into the other space | |
3581 | * cache. | |
3582 | */ | |
ee22184b | 3583 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
0af3d00b JB |
3584 | if (!num_pages) |
3585 | num_pages = 1; | |
3586 | ||
0af3d00b | 3587 | num_pages *= 16; |
09cbfeaf | 3588 | num_pages *= PAGE_SIZE; |
0af3d00b | 3589 | |
364ecf36 | 3590 | ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages); |
0af3d00b JB |
3591 | if (ret) |
3592 | goto out_put; | |
3593 | ||
3594 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3595 | num_pages, num_pages, | |
3596 | &alloc_hint); | |
2968b1f4 JB |
3597 | /* |
3598 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3599 | * of metadata or split extents when writing the cache out, which means | |
3600 | * we can enospc if we are heavily fragmented in addition to just normal | |
3601 | * out of space conditions. So if we hit this just skip setting up any | |
3602 | * other block groups for this transaction, maybe we'll unpin enough | |
3603 | * space the next time around. | |
3604 | */ | |
2b20982e JB |
3605 | if (!ret) |
3606 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3607 | else if (ret == -ENOSPC) |
3608 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
c09544e0 | 3609 | |
0af3d00b JB |
3610 | out_put: |
3611 | iput(inode); | |
3612 | out_free: | |
b3b4aa74 | 3613 | btrfs_release_path(path); |
0af3d00b JB |
3614 | out: |
3615 | spin_lock(&block_group->lock); | |
e65cbb94 | 3616 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3617 | block_group->cache_generation = trans->transid; |
2b20982e | 3618 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3619 | spin_unlock(&block_group->lock); |
3620 | ||
364ecf36 | 3621 | extent_changeset_free(data_reserved); |
0af3d00b JB |
3622 | return ret; |
3623 | } | |
3624 | ||
dcdf7f6d | 3625 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans, |
2ff7e61e | 3626 | struct btrfs_fs_info *fs_info) |
dcdf7f6d JB |
3627 | { |
3628 | struct btrfs_block_group_cache *cache, *tmp; | |
3629 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3630 | struct btrfs_path *path; | |
3631 | ||
3632 | if (list_empty(&cur_trans->dirty_bgs) || | |
0b246afa | 3633 | !btrfs_test_opt(fs_info, SPACE_CACHE)) |
dcdf7f6d JB |
3634 | return 0; |
3635 | ||
3636 | path = btrfs_alloc_path(); | |
3637 | if (!path) | |
3638 | return -ENOMEM; | |
3639 | ||
3640 | /* Could add new block groups, use _safe just in case */ | |
3641 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3642 | dirty_list) { | |
3643 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3644 | cache_save_setup(cache, trans, path); | |
3645 | } | |
3646 | ||
3647 | btrfs_free_path(path); | |
3648 | return 0; | |
3649 | } | |
3650 | ||
1bbc621e CM |
3651 | /* |
3652 | * transaction commit does final block group cache writeback during a | |
3653 | * critical section where nothing is allowed to change the FS. This is | |
3654 | * required in order for the cache to actually match the block group, | |
3655 | * but can introduce a lot of latency into the commit. | |
3656 | * | |
3657 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3658 | * cache IO. There's a chance we'll have to redo some of it if the | |
3659 | * block group changes again during the commit, but it greatly reduces | |
3660 | * the commit latency by getting rid of the easy block groups while | |
3661 | * we're still allowing others to join the commit. | |
3662 | */ | |
21217054 | 3663 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) |
9078a3e1 | 3664 | { |
21217054 | 3665 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4a8c9a62 | 3666 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3667 | struct btrfs_transaction *cur_trans = trans->transaction; |
3668 | int ret = 0; | |
c9dc4c65 | 3669 | int should_put; |
1bbc621e CM |
3670 | struct btrfs_path *path = NULL; |
3671 | LIST_HEAD(dirty); | |
3672 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3673 | int num_started = 0; |
1bbc621e CM |
3674 | int loops = 0; |
3675 | ||
3676 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3677 | if (list_empty(&cur_trans->dirty_bgs)) { |
3678 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3679 | return 0; | |
1bbc621e | 3680 | } |
b58d1a9e | 3681 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3682 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3683 | |
1bbc621e | 3684 | again: |
1bbc621e CM |
3685 | /* |
3686 | * make sure all the block groups on our dirty list actually | |
3687 | * exist | |
3688 | */ | |
6c686b35 | 3689 | btrfs_create_pending_block_groups(trans); |
1bbc621e CM |
3690 | |
3691 | if (!path) { | |
3692 | path = btrfs_alloc_path(); | |
3693 | if (!path) | |
3694 | return -ENOMEM; | |
3695 | } | |
3696 | ||
b58d1a9e FM |
3697 | /* |
3698 | * cache_write_mutex is here only to save us from balance or automatic | |
3699 | * removal of empty block groups deleting this block group while we are | |
3700 | * writing out the cache | |
3701 | */ | |
3702 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e CM |
3703 | while (!list_empty(&dirty)) { |
3704 | cache = list_first_entry(&dirty, | |
3705 | struct btrfs_block_group_cache, | |
3706 | dirty_list); | |
1bbc621e CM |
3707 | /* |
3708 | * this can happen if something re-dirties a block | |
3709 | * group that is already under IO. Just wait for it to | |
3710 | * finish and then do it all again | |
3711 | */ | |
3712 | if (!list_empty(&cache->io_list)) { | |
3713 | list_del_init(&cache->io_list); | |
afdb5718 | 3714 | btrfs_wait_cache_io(trans, cache, path); |
1bbc621e CM |
3715 | btrfs_put_block_group(cache); |
3716 | } | |
3717 | ||
3718 | ||
3719 | /* | |
3720 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3721 | * if it should update the cache_state. Don't delete | |
3722 | * until after we wait. | |
3723 | * | |
3724 | * Since we're not running in the commit critical section | |
3725 | * we need the dirty_bgs_lock to protect from update_block_group | |
3726 | */ | |
3727 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3728 | list_del_init(&cache->dirty_list); | |
3729 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3730 | ||
3731 | should_put = 1; | |
3732 | ||
3733 | cache_save_setup(cache, trans, path); | |
3734 | ||
3735 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3736 | cache->io_ctl.inode = NULL; | |
0b246afa | 3737 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3738 | cache, path); |
1bbc621e CM |
3739 | if (ret == 0 && cache->io_ctl.inode) { |
3740 | num_started++; | |
3741 | should_put = 0; | |
3742 | ||
3743 | /* | |
45ae2c18 NB |
3744 | * The cache_write_mutex is protecting the |
3745 | * io_list, also refer to the definition of | |
3746 | * btrfs_transaction::io_bgs for more details | |
1bbc621e CM |
3747 | */ |
3748 | list_add_tail(&cache->io_list, io); | |
3749 | } else { | |
3750 | /* | |
3751 | * if we failed to write the cache, the | |
3752 | * generation will be bad and life goes on | |
3753 | */ | |
3754 | ret = 0; | |
3755 | } | |
3756 | } | |
ff1f8250 | 3757 | if (!ret) { |
2ff7e61e JM |
3758 | ret = write_one_cache_group(trans, fs_info, |
3759 | path, cache); | |
ff1f8250 FM |
3760 | /* |
3761 | * Our block group might still be attached to the list | |
3762 | * of new block groups in the transaction handle of some | |
3763 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3764 | * means its block group item isn't yet in the extent | |
3765 | * tree. If this happens ignore the error, as we will | |
3766 | * try again later in the critical section of the | |
3767 | * transaction commit. | |
3768 | */ | |
3769 | if (ret == -ENOENT) { | |
3770 | ret = 0; | |
3771 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3772 | if (list_empty(&cache->dirty_list)) { | |
3773 | list_add_tail(&cache->dirty_list, | |
3774 | &cur_trans->dirty_bgs); | |
3775 | btrfs_get_block_group(cache); | |
3776 | } | |
3777 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3778 | } else if (ret) { | |
66642832 | 3779 | btrfs_abort_transaction(trans, ret); |
ff1f8250 FM |
3780 | } |
3781 | } | |
1bbc621e CM |
3782 | |
3783 | /* if its not on the io list, we need to put the block group */ | |
3784 | if (should_put) | |
3785 | btrfs_put_block_group(cache); | |
3786 | ||
3787 | if (ret) | |
3788 | break; | |
b58d1a9e FM |
3789 | |
3790 | /* | |
3791 | * Avoid blocking other tasks for too long. It might even save | |
3792 | * us from writing caches for block groups that are going to be | |
3793 | * removed. | |
3794 | */ | |
3795 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3796 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3797 | } |
b58d1a9e | 3798 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3799 | |
3800 | /* | |
3801 | * go through delayed refs for all the stuff we've just kicked off | |
3802 | * and then loop back (just once) | |
3803 | */ | |
2ff7e61e | 3804 | ret = btrfs_run_delayed_refs(trans, fs_info, 0); |
1bbc621e CM |
3805 | if (!ret && loops == 0) { |
3806 | loops++; | |
3807 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3808 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3809 | /* |
3810 | * dirty_bgs_lock protects us from concurrent block group | |
3811 | * deletes too (not just cache_write_mutex). | |
3812 | */ | |
3813 | if (!list_empty(&dirty)) { | |
3814 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3815 | goto again; | |
3816 | } | |
1bbc621e | 3817 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c79a1751 | 3818 | } else if (ret < 0) { |
2ff7e61e | 3819 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
1bbc621e CM |
3820 | } |
3821 | ||
3822 | btrfs_free_path(path); | |
3823 | return ret; | |
3824 | } | |
3825 | ||
3826 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3827 | struct btrfs_fs_info *fs_info) |
1bbc621e CM |
3828 | { |
3829 | struct btrfs_block_group_cache *cache; | |
3830 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3831 | int ret = 0; | |
3832 | int should_put; | |
3833 | struct btrfs_path *path; | |
3834 | struct list_head *io = &cur_trans->io_bgs; | |
3835 | int num_started = 0; | |
9078a3e1 CM |
3836 | |
3837 | path = btrfs_alloc_path(); | |
3838 | if (!path) | |
3839 | return -ENOMEM; | |
3840 | ||
ce93ec54 | 3841 | /* |
e44081ef FM |
3842 | * Even though we are in the critical section of the transaction commit, |
3843 | * we can still have concurrent tasks adding elements to this | |
3844 | * transaction's list of dirty block groups. These tasks correspond to | |
3845 | * endio free space workers started when writeback finishes for a | |
3846 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3847 | * allocate new block groups as a result of COWing nodes of the root | |
3848 | * tree when updating the free space inode. The writeback for the space | |
3849 | * caches is triggered by an earlier call to | |
3850 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3851 | * loop. | |
3852 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3853 | * delayed refs to make sure we have the best chance at doing this all |
3854 | * in one shot. | |
3855 | */ | |
e44081ef | 3856 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3857 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3858 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3859 | struct btrfs_block_group_cache, | |
3860 | dirty_list); | |
c9dc4c65 CM |
3861 | |
3862 | /* | |
3863 | * this can happen if cache_save_setup re-dirties a block | |
3864 | * group that is already under IO. Just wait for it to | |
3865 | * finish and then do it all again | |
3866 | */ | |
3867 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3868 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3869 | list_del_init(&cache->io_list); |
afdb5718 | 3870 | btrfs_wait_cache_io(trans, cache, path); |
c9dc4c65 | 3871 | btrfs_put_block_group(cache); |
e44081ef | 3872 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3873 | } |
3874 | ||
1bbc621e CM |
3875 | /* |
3876 | * don't remove from the dirty list until after we've waited | |
3877 | * on any pending IO | |
3878 | */ | |
ce93ec54 | 3879 | list_del_init(&cache->dirty_list); |
e44081ef | 3880 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3881 | should_put = 1; |
3882 | ||
1bbc621e | 3883 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3884 | |
ce93ec54 | 3885 | if (!ret) |
2ff7e61e JM |
3886 | ret = btrfs_run_delayed_refs(trans, fs_info, |
3887 | (unsigned long) -1); | |
c9dc4c65 CM |
3888 | |
3889 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3890 | cache->io_ctl.inode = NULL; | |
0b246afa | 3891 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3892 | cache, path); |
c9dc4c65 CM |
3893 | if (ret == 0 && cache->io_ctl.inode) { |
3894 | num_started++; | |
3895 | should_put = 0; | |
1bbc621e | 3896 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3897 | } else { |
3898 | /* | |
3899 | * if we failed to write the cache, the | |
3900 | * generation will be bad and life goes on | |
3901 | */ | |
3902 | ret = 0; | |
3903 | } | |
3904 | } | |
ff1f8250 | 3905 | if (!ret) { |
2ff7e61e JM |
3906 | ret = write_one_cache_group(trans, fs_info, |
3907 | path, cache); | |
2bc0bb5f FM |
3908 | /* |
3909 | * One of the free space endio workers might have | |
3910 | * created a new block group while updating a free space | |
3911 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3912 | * and hasn't released its transaction handle yet, in | |
3913 | * which case the new block group is still attached to | |
3914 | * its transaction handle and its creation has not | |
3915 | * finished yet (no block group item in the extent tree | |
3916 | * yet, etc). If this is the case, wait for all free | |
3917 | * space endio workers to finish and retry. This is a | |
3918 | * a very rare case so no need for a more efficient and | |
3919 | * complex approach. | |
3920 | */ | |
3921 | if (ret == -ENOENT) { | |
3922 | wait_event(cur_trans->writer_wait, | |
3923 | atomic_read(&cur_trans->num_writers) == 1); | |
2ff7e61e JM |
3924 | ret = write_one_cache_group(trans, fs_info, |
3925 | path, cache); | |
2bc0bb5f | 3926 | } |
ff1f8250 | 3927 | if (ret) |
66642832 | 3928 | btrfs_abort_transaction(trans, ret); |
ff1f8250 | 3929 | } |
c9dc4c65 CM |
3930 | |
3931 | /* if its not on the io list, we need to put the block group */ | |
3932 | if (should_put) | |
3933 | btrfs_put_block_group(cache); | |
e44081ef | 3934 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3935 | } |
e44081ef | 3936 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3937 | |
45ae2c18 NB |
3938 | /* |
3939 | * Refer to the definition of io_bgs member for details why it's safe | |
3940 | * to use it without any locking | |
3941 | */ | |
1bbc621e CM |
3942 | while (!list_empty(io)) { |
3943 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3944 | io_list); |
3945 | list_del_init(&cache->io_list); | |
afdb5718 | 3946 | btrfs_wait_cache_io(trans, cache, path); |
0cb59c99 JB |
3947 | btrfs_put_block_group(cache); |
3948 | } | |
3949 | ||
9078a3e1 | 3950 | btrfs_free_path(path); |
ce93ec54 | 3951 | return ret; |
9078a3e1 CM |
3952 | } |
3953 | ||
2ff7e61e | 3954 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 YZ |
3955 | { |
3956 | struct btrfs_block_group_cache *block_group; | |
3957 | int readonly = 0; | |
3958 | ||
0b246afa | 3959 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
3960 | if (!block_group || block_group->ro) |
3961 | readonly = 1; | |
3962 | if (block_group) | |
fa9c0d79 | 3963 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3964 | return readonly; |
3965 | } | |
3966 | ||
f78c436c FM |
3967 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
3968 | { | |
3969 | struct btrfs_block_group_cache *bg; | |
3970 | bool ret = true; | |
3971 | ||
3972 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3973 | if (!bg) | |
3974 | return false; | |
3975 | ||
3976 | spin_lock(&bg->lock); | |
3977 | if (bg->ro) | |
3978 | ret = false; | |
3979 | else | |
3980 | atomic_inc(&bg->nocow_writers); | |
3981 | spin_unlock(&bg->lock); | |
3982 | ||
3983 | /* no put on block group, done by btrfs_dec_nocow_writers */ | |
3984 | if (!ret) | |
3985 | btrfs_put_block_group(bg); | |
3986 | ||
3987 | return ret; | |
3988 | ||
3989 | } | |
3990 | ||
3991 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
3992 | { | |
3993 | struct btrfs_block_group_cache *bg; | |
3994 | ||
3995 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3996 | ASSERT(bg); | |
3997 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
3998 | wake_up_atomic_t(&bg->nocow_writers); | |
3999 | /* | |
4000 | * Once for our lookup and once for the lookup done by a previous call | |
4001 | * to btrfs_inc_nocow_writers() | |
4002 | */ | |
4003 | btrfs_put_block_group(bg); | |
4004 | btrfs_put_block_group(bg); | |
4005 | } | |
4006 | ||
f78c436c FM |
4007 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) |
4008 | { | |
5e4def20 | 4009 | wait_on_atomic_t(&bg->nocow_writers, atomic_t_wait, |
f78c436c FM |
4010 | TASK_UNINTERRUPTIBLE); |
4011 | } | |
4012 | ||
6ab0a202 JM |
4013 | static const char *alloc_name(u64 flags) |
4014 | { | |
4015 | switch (flags) { | |
4016 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
4017 | return "mixed"; | |
4018 | case BTRFS_BLOCK_GROUP_METADATA: | |
4019 | return "metadata"; | |
4020 | case BTRFS_BLOCK_GROUP_DATA: | |
4021 | return "data"; | |
4022 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
4023 | return "system"; | |
4024 | default: | |
4025 | WARN_ON(1); | |
4026 | return "invalid-combination"; | |
4027 | }; | |
4028 | } | |
4029 | ||
2be12ef7 NB |
4030 | static int create_space_info(struct btrfs_fs_info *info, u64 flags, |
4031 | struct btrfs_space_info **new) | |
4032 | { | |
4033 | ||
4034 | struct btrfs_space_info *space_info; | |
4035 | int i; | |
4036 | int ret; | |
4037 | ||
4038 | space_info = kzalloc(sizeof(*space_info), GFP_NOFS); | |
4039 | if (!space_info) | |
4040 | return -ENOMEM; | |
4041 | ||
4042 | ret = percpu_counter_init(&space_info->total_bytes_pinned, 0, | |
4043 | GFP_KERNEL); | |
4044 | if (ret) { | |
4045 | kfree(space_info); | |
4046 | return ret; | |
4047 | } | |
4048 | ||
4049 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
4050 | INIT_LIST_HEAD(&space_info->block_groups[i]); | |
4051 | init_rwsem(&space_info->groups_sem); | |
4052 | spin_lock_init(&space_info->lock); | |
4053 | space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; | |
4054 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
4055 | init_waitqueue_head(&space_info->wait); | |
4056 | INIT_LIST_HEAD(&space_info->ro_bgs); | |
4057 | INIT_LIST_HEAD(&space_info->tickets); | |
4058 | INIT_LIST_HEAD(&space_info->priority_tickets); | |
4059 | ||
4060 | ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype, | |
4061 | info->space_info_kobj, "%s", | |
4062 | alloc_name(space_info->flags)); | |
4063 | if (ret) { | |
4064 | percpu_counter_destroy(&space_info->total_bytes_pinned); | |
4065 | kfree(space_info); | |
4066 | return ret; | |
4067 | } | |
4068 | ||
4069 | *new = space_info; | |
4070 | list_add_rcu(&space_info->list, &info->space_info); | |
4071 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
4072 | info->data_sinfo = space_info; | |
4073 | ||
4074 | return ret; | |
4075 | } | |
4076 | ||
d2006e6d | 4077 | static void update_space_info(struct btrfs_fs_info *info, u64 flags, |
593060d7 | 4078 | u64 total_bytes, u64 bytes_used, |
e40edf2d | 4079 | u64 bytes_readonly, |
593060d7 CM |
4080 | struct btrfs_space_info **space_info) |
4081 | { | |
4082 | struct btrfs_space_info *found; | |
b742bb82 YZ |
4083 | int factor; |
4084 | ||
4085 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
4086 | BTRFS_BLOCK_GROUP_RAID10)) | |
4087 | factor = 2; | |
4088 | else | |
4089 | factor = 1; | |
593060d7 CM |
4090 | |
4091 | found = __find_space_info(info, flags); | |
d2006e6d NB |
4092 | ASSERT(found); |
4093 | spin_lock(&found->lock); | |
4094 | found->total_bytes += total_bytes; | |
4095 | found->disk_total += total_bytes * factor; | |
4096 | found->bytes_used += bytes_used; | |
4097 | found->disk_used += bytes_used * factor; | |
4098 | found->bytes_readonly += bytes_readonly; | |
4099 | if (total_bytes > 0) | |
4100 | found->full = 0; | |
4101 | space_info_add_new_bytes(info, found, total_bytes - | |
4102 | bytes_used - bytes_readonly); | |
4103 | spin_unlock(&found->lock); | |
4104 | *space_info = found; | |
593060d7 CM |
4105 | } |
4106 | ||
8790d502 CM |
4107 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
4108 | { | |
899c81ea ID |
4109 | u64 extra_flags = chunk_to_extended(flags) & |
4110 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 4111 | |
de98ced9 | 4112 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
4113 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
4114 | fs_info->avail_data_alloc_bits |= extra_flags; | |
4115 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
4116 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
4117 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
4118 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 4119 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 4120 | } |
593060d7 | 4121 | |
fc67c450 ID |
4122 | /* |
4123 | * returns target flags in extended format or 0 if restripe for this | |
4124 | * chunk_type is not in progress | |
c6664b42 ID |
4125 | * |
4126 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
4127 | */ |
4128 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
4129 | { | |
4130 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4131 | u64 target = 0; | |
4132 | ||
fc67c450 ID |
4133 | if (!bctl) |
4134 | return 0; | |
4135 | ||
4136 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
4137 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4138 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
4139 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
4140 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4141 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
4142 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
4143 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4144 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
4145 | } | |
4146 | ||
4147 | return target; | |
4148 | } | |
4149 | ||
a46d11a8 ID |
4150 | /* |
4151 | * @flags: available profiles in extended format (see ctree.h) | |
4152 | * | |
e4d8ec0f ID |
4153 | * Returns reduced profile in chunk format. If profile changing is in |
4154 | * progress (either running or paused) picks the target profile (if it's | |
4155 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 4156 | */ |
2ff7e61e | 4157 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c | 4158 | { |
0b246afa | 4159 | u64 num_devices = fs_info->fs_devices->rw_devices; |
fc67c450 | 4160 | u64 target; |
9c170b26 ZL |
4161 | u64 raid_type; |
4162 | u64 allowed = 0; | |
a061fc8d | 4163 | |
fc67c450 ID |
4164 | /* |
4165 | * see if restripe for this chunk_type is in progress, if so | |
4166 | * try to reduce to the target profile | |
4167 | */ | |
0b246afa JM |
4168 | spin_lock(&fs_info->balance_lock); |
4169 | target = get_restripe_target(fs_info, flags); | |
fc67c450 ID |
4170 | if (target) { |
4171 | /* pick target profile only if it's already available */ | |
4172 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
0b246afa | 4173 | spin_unlock(&fs_info->balance_lock); |
fc67c450 | 4174 | return extended_to_chunk(target); |
e4d8ec0f ID |
4175 | } |
4176 | } | |
0b246afa | 4177 | spin_unlock(&fs_info->balance_lock); |
e4d8ec0f | 4178 | |
53b381b3 | 4179 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
4180 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
4181 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
4182 | allowed |= btrfs_raid_group[raid_type]; | |
4183 | } | |
4184 | allowed &= flags; | |
4185 | ||
4186 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
4187 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
4188 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
4189 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
4190 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
4191 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
4192 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
4193 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
4194 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
4195 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
4196 | ||
4197 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
4198 | ||
4199 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
4200 | } |
4201 | ||
2ff7e61e | 4202 | static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
6a63209f | 4203 | { |
de98ced9 | 4204 | unsigned seq; |
f8213bdc | 4205 | u64 flags; |
de98ced9 MX |
4206 | |
4207 | do { | |
f8213bdc | 4208 | flags = orig_flags; |
0b246afa | 4209 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 MX |
4210 | |
4211 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
0b246afa | 4212 | flags |= fs_info->avail_data_alloc_bits; |
de98ced9 | 4213 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
0b246afa | 4214 | flags |= fs_info->avail_system_alloc_bits; |
de98ced9 | 4215 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
0b246afa JM |
4216 | flags |= fs_info->avail_metadata_alloc_bits; |
4217 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
6fef8df1 | 4218 | |
2ff7e61e | 4219 | return btrfs_reduce_alloc_profile(fs_info, flags); |
6a63209f JB |
4220 | } |
4221 | ||
1b86826d | 4222 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 4223 | { |
0b246afa | 4224 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 4225 | u64 flags; |
53b381b3 | 4226 | u64 ret; |
9ed74f2d | 4227 | |
b742bb82 YZ |
4228 | if (data) |
4229 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 4230 | else if (root == fs_info->chunk_root) |
b742bb82 | 4231 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 4232 | else |
b742bb82 | 4233 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 4234 | |
2ff7e61e | 4235 | ret = get_alloc_profile(fs_info, flags); |
53b381b3 | 4236 | return ret; |
6a63209f | 4237 | } |
9ed74f2d | 4238 | |
1b86826d JM |
4239 | u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info) |
4240 | { | |
4241 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA); | |
4242 | } | |
4243 | ||
4244 | u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info) | |
4245 | { | |
4246 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4247 | } | |
4248 | ||
4249 | u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info) | |
4250 | { | |
4251 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
4252 | } | |
4253 | ||
4136135b LB |
4254 | static u64 btrfs_space_info_used(struct btrfs_space_info *s_info, |
4255 | bool may_use_included) | |
4256 | { | |
4257 | ASSERT(s_info); | |
4258 | return s_info->bytes_used + s_info->bytes_reserved + | |
4259 | s_info->bytes_pinned + s_info->bytes_readonly + | |
4260 | (may_use_included ? s_info->bytes_may_use : 0); | |
4261 | } | |
4262 | ||
04f4f916 | 4263 | int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes) |
6a63209f | 4264 | { |
04f4f916 | 4265 | struct btrfs_root *root = inode->root; |
b4d7c3c9 | 4266 | struct btrfs_fs_info *fs_info = root->fs_info; |
1174cade | 4267 | struct btrfs_space_info *data_sinfo = fs_info->data_sinfo; |
ab6e2410 | 4268 | u64 used; |
94b947b2 | 4269 | int ret = 0; |
c99f1b0c ZL |
4270 | int need_commit = 2; |
4271 | int have_pinned_space; | |
6a63209f | 4272 | |
6a63209f | 4273 | /* make sure bytes are sectorsize aligned */ |
0b246afa | 4274 | bytes = ALIGN(bytes, fs_info->sectorsize); |
6a63209f | 4275 | |
9dced186 | 4276 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 4277 | need_commit = 0; |
9dced186 | 4278 | ASSERT(current->journal_info); |
0af3d00b JB |
4279 | } |
4280 | ||
6a63209f JB |
4281 | again: |
4282 | /* make sure we have enough space to handle the data first */ | |
4283 | spin_lock(&data_sinfo->lock); | |
4136135b | 4284 | used = btrfs_space_info_used(data_sinfo, true); |
ab6e2410 JB |
4285 | |
4286 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 4287 | struct btrfs_trans_handle *trans; |
9ed74f2d | 4288 | |
6a63209f JB |
4289 | /* |
4290 | * if we don't have enough free bytes in this space then we need | |
4291 | * to alloc a new chunk. | |
4292 | */ | |
b9fd47cd | 4293 | if (!data_sinfo->full) { |
6a63209f | 4294 | u64 alloc_target; |
9ed74f2d | 4295 | |
0e4f8f88 | 4296 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 4297 | spin_unlock(&data_sinfo->lock); |
1174cade | 4298 | |
1b86826d | 4299 | alloc_target = btrfs_data_alloc_profile(fs_info); |
9dced186 MX |
4300 | /* |
4301 | * It is ugly that we don't call nolock join | |
4302 | * transaction for the free space inode case here. | |
4303 | * But it is safe because we only do the data space | |
4304 | * reservation for the free space cache in the | |
4305 | * transaction context, the common join transaction | |
4306 | * just increase the counter of the current transaction | |
4307 | * handler, doesn't try to acquire the trans_lock of | |
4308 | * the fs. | |
4309 | */ | |
7a7eaa40 | 4310 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4311 | if (IS_ERR(trans)) |
4312 | return PTR_ERR(trans); | |
9ed74f2d | 4313 | |
2ff7e61e | 4314 | ret = do_chunk_alloc(trans, fs_info, alloc_target, |
0e4f8f88 | 4315 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 4316 | btrfs_end_transaction(trans); |
d52a5b5f MX |
4317 | if (ret < 0) { |
4318 | if (ret != -ENOSPC) | |
4319 | return ret; | |
c99f1b0c ZL |
4320 | else { |
4321 | have_pinned_space = 1; | |
d52a5b5f | 4322 | goto commit_trans; |
c99f1b0c | 4323 | } |
d52a5b5f | 4324 | } |
9ed74f2d | 4325 | |
6a63209f JB |
4326 | goto again; |
4327 | } | |
f2bb8f5c JB |
4328 | |
4329 | /* | |
b150a4f1 | 4330 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
4331 | * allocation, and no removed chunk in current transaction, |
4332 | * don't bother committing the transaction. | |
f2bb8f5c | 4333 | */ |
c99f1b0c ZL |
4334 | have_pinned_space = percpu_counter_compare( |
4335 | &data_sinfo->total_bytes_pinned, | |
4336 | used + bytes - data_sinfo->total_bytes); | |
6a63209f | 4337 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4338 | |
4e06bdd6 | 4339 | /* commit the current transaction and try again */ |
d52a5b5f | 4340 | commit_trans: |
92e2f7e3 | 4341 | if (need_commit) { |
c99f1b0c | 4342 | need_commit--; |
b150a4f1 | 4343 | |
e1746e83 ZL |
4344 | if (need_commit > 0) { |
4345 | btrfs_start_delalloc_roots(fs_info, 0, -1); | |
6374e57a | 4346 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, |
0b246afa | 4347 | (u64)-1); |
e1746e83 | 4348 | } |
9a4e7276 | 4349 | |
7a7eaa40 | 4350 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4351 | if (IS_ERR(trans)) |
4352 | return PTR_ERR(trans); | |
c99f1b0c | 4353 | if (have_pinned_space >= 0 || |
3204d33c JB |
4354 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4355 | &trans->transaction->flags) || | |
c99f1b0c | 4356 | need_commit > 0) { |
3a45bb20 | 4357 | ret = btrfs_commit_transaction(trans); |
94b947b2 ZL |
4358 | if (ret) |
4359 | return ret; | |
d7c15171 | 4360 | /* |
c2d6cb16 FM |
4361 | * The cleaner kthread might still be doing iput |
4362 | * operations. Wait for it to finish so that | |
4363 | * more space is released. | |
d7c15171 | 4364 | */ |
0b246afa JM |
4365 | mutex_lock(&fs_info->cleaner_delayed_iput_mutex); |
4366 | mutex_unlock(&fs_info->cleaner_delayed_iput_mutex); | |
94b947b2 ZL |
4367 | goto again; |
4368 | } else { | |
3a45bb20 | 4369 | btrfs_end_transaction(trans); |
94b947b2 | 4370 | } |
4e06bdd6 | 4371 | } |
9ed74f2d | 4372 | |
0b246afa | 4373 | trace_btrfs_space_reservation(fs_info, |
cab45e22 JM |
4374 | "space_info:enospc", |
4375 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4376 | return -ENOSPC; |
4377 | } | |
4378 | data_sinfo->bytes_may_use += bytes; | |
0b246afa | 4379 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4380 | data_sinfo->flags, bytes, 1); |
6a63209f | 4381 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4382 | |
237c0e9f | 4383 | return ret; |
9ed74f2d | 4384 | } |
6a63209f | 4385 | |
364ecf36 QW |
4386 | int btrfs_check_data_free_space(struct inode *inode, |
4387 | struct extent_changeset **reserved, u64 start, u64 len) | |
4ceff079 | 4388 | { |
0b246afa | 4389 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4390 | int ret; |
4391 | ||
4392 | /* align the range */ | |
0b246afa JM |
4393 | len = round_up(start + len, fs_info->sectorsize) - |
4394 | round_down(start, fs_info->sectorsize); | |
4395 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4396 | |
04f4f916 | 4397 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len); |
4ceff079 QW |
4398 | if (ret < 0) |
4399 | return ret; | |
4400 | ||
1e5ec2e7 | 4401 | /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */ |
364ecf36 | 4402 | ret = btrfs_qgroup_reserve_data(inode, reserved, start, len); |
7bc329c1 | 4403 | if (ret < 0) |
1e5ec2e7 | 4404 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
364ecf36 QW |
4405 | else |
4406 | ret = 0; | |
4ceff079 QW |
4407 | return ret; |
4408 | } | |
4409 | ||
4ceff079 QW |
4410 | /* |
4411 | * Called if we need to clear a data reservation for this inode | |
4412 | * Normally in a error case. | |
4413 | * | |
51773bec QW |
4414 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4415 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4416 | * Like clear_bit_hook(). | |
4ceff079 | 4417 | */ |
51773bec QW |
4418 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4419 | u64 len) | |
4ceff079 | 4420 | { |
0b246afa | 4421 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4422 | struct btrfs_space_info *data_sinfo; |
4423 | ||
4424 | /* Make sure the range is aligned to sectorsize */ | |
0b246afa JM |
4425 | len = round_up(start + len, fs_info->sectorsize) - |
4426 | round_down(start, fs_info->sectorsize); | |
4427 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4428 | |
0b246afa | 4429 | data_sinfo = fs_info->data_sinfo; |
4ceff079 QW |
4430 | spin_lock(&data_sinfo->lock); |
4431 | if (WARN_ON(data_sinfo->bytes_may_use < len)) | |
4432 | data_sinfo->bytes_may_use = 0; | |
4433 | else | |
4434 | data_sinfo->bytes_may_use -= len; | |
0b246afa | 4435 | trace_btrfs_space_reservation(fs_info, "space_info", |
4ceff079 QW |
4436 | data_sinfo->flags, len, 0); |
4437 | spin_unlock(&data_sinfo->lock); | |
4438 | } | |
4439 | ||
51773bec QW |
4440 | /* |
4441 | * Called if we need to clear a data reservation for this inode | |
4442 | * Normally in a error case. | |
4443 | * | |
01327610 | 4444 | * This one will handle the per-inode data rsv map for accurate reserved |
51773bec QW |
4445 | * space framework. |
4446 | */ | |
bc42bda2 QW |
4447 | void btrfs_free_reserved_data_space(struct inode *inode, |
4448 | struct extent_changeset *reserved, u64 start, u64 len) | |
51773bec | 4449 | { |
0c476a5d JM |
4450 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4451 | ||
4452 | /* Make sure the range is aligned to sectorsize */ | |
da17066c JM |
4453 | len = round_up(start + len, root->fs_info->sectorsize) - |
4454 | round_down(start, root->fs_info->sectorsize); | |
4455 | start = round_down(start, root->fs_info->sectorsize); | |
0c476a5d | 4456 | |
51773bec | 4457 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
bc42bda2 | 4458 | btrfs_qgroup_free_data(inode, reserved, start, len); |
51773bec QW |
4459 | } |
4460 | ||
97e728d4 | 4461 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4462 | { |
97e728d4 JB |
4463 | struct list_head *head = &info->space_info; |
4464 | struct btrfs_space_info *found; | |
e3ccfa98 | 4465 | |
97e728d4 JB |
4466 | rcu_read_lock(); |
4467 | list_for_each_entry_rcu(found, head, list) { | |
4468 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4469 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4470 | } |
97e728d4 | 4471 | rcu_read_unlock(); |
e3ccfa98 JB |
4472 | } |
4473 | ||
3c76cd84 MX |
4474 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
4475 | { | |
4476 | return (global->size << 1); | |
4477 | } | |
4478 | ||
2ff7e61e | 4479 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, |
698d0082 | 4480 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4481 | { |
0b246afa | 4482 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
8d8aafee | 4483 | u64 bytes_used = btrfs_space_info_used(sinfo, false); |
e5bc2458 | 4484 | u64 thresh; |
e3ccfa98 | 4485 | |
0e4f8f88 CM |
4486 | if (force == CHUNK_ALLOC_FORCE) |
4487 | return 1; | |
4488 | ||
fb25e914 JB |
4489 | /* |
4490 | * We need to take into account the global rsv because for all intents | |
4491 | * and purposes it's used space. Don't worry about locking the | |
4492 | * global_rsv, it doesn't change except when the transaction commits. | |
4493 | */ | |
54338b5c | 4494 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
8d8aafee | 4495 | bytes_used += calc_global_rsv_need_space(global_rsv); |
fb25e914 | 4496 | |
0e4f8f88 CM |
4497 | /* |
4498 | * in limited mode, we want to have some free space up to | |
4499 | * about 1% of the FS size. | |
4500 | */ | |
4501 | if (force == CHUNK_ALLOC_LIMITED) { | |
0b246afa | 4502 | thresh = btrfs_super_total_bytes(fs_info->super_copy); |
ee22184b | 4503 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
0e4f8f88 | 4504 | |
8d8aafee | 4505 | if (sinfo->total_bytes - bytes_used < thresh) |
0e4f8f88 CM |
4506 | return 1; |
4507 | } | |
0e4f8f88 | 4508 | |
8d8aafee | 4509 | if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) |
14ed0ca6 | 4510 | return 0; |
424499db | 4511 | return 1; |
32c00aff JB |
4512 | } |
4513 | ||
2ff7e61e | 4514 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4515 | { |
4516 | u64 num_dev; | |
4517 | ||
53b381b3 DW |
4518 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
4519 | BTRFS_BLOCK_GROUP_RAID0 | | |
4520 | BTRFS_BLOCK_GROUP_RAID5 | | |
4521 | BTRFS_BLOCK_GROUP_RAID6)) | |
0b246afa | 4522 | num_dev = fs_info->fs_devices->rw_devices; |
15d1ff81 LB |
4523 | else if (type & BTRFS_BLOCK_GROUP_RAID1) |
4524 | num_dev = 2; | |
4525 | else | |
4526 | num_dev = 1; /* DUP or single */ | |
4527 | ||
39c2d7fa | 4528 | return num_dev; |
15d1ff81 LB |
4529 | } |
4530 | ||
39c2d7fa FM |
4531 | /* |
4532 | * If @is_allocation is true, reserve space in the system space info necessary | |
4533 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4534 | * removing a chunk. | |
4535 | */ | |
4536 | void check_system_chunk(struct btrfs_trans_handle *trans, | |
2ff7e61e | 4537 | struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4538 | { |
4539 | struct btrfs_space_info *info; | |
4540 | u64 left; | |
4541 | u64 thresh; | |
4fbcdf66 | 4542 | int ret = 0; |
39c2d7fa | 4543 | u64 num_devs; |
4fbcdf66 FM |
4544 | |
4545 | /* | |
4546 | * Needed because we can end up allocating a system chunk and for an | |
4547 | * atomic and race free space reservation in the chunk block reserve. | |
4548 | */ | |
0b246afa | 4549 | ASSERT(mutex_is_locked(&fs_info->chunk_mutex)); |
15d1ff81 | 4550 | |
0b246afa | 4551 | info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
15d1ff81 | 4552 | spin_lock(&info->lock); |
4136135b | 4553 | left = info->total_bytes - btrfs_space_info_used(info, true); |
15d1ff81 LB |
4554 | spin_unlock(&info->lock); |
4555 | ||
2ff7e61e | 4556 | num_devs = get_profile_num_devs(fs_info, type); |
39c2d7fa FM |
4557 | |
4558 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
0b246afa JM |
4559 | thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) + |
4560 | btrfs_calc_trans_metadata_size(fs_info, 1); | |
39c2d7fa | 4561 | |
0b246afa JM |
4562 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
4563 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
4564 | left, thresh, type); | |
4565 | dump_space_info(fs_info, info, 0, 0); | |
15d1ff81 LB |
4566 | } |
4567 | ||
4568 | if (left < thresh) { | |
1b86826d | 4569 | u64 flags = btrfs_system_alloc_profile(fs_info); |
15d1ff81 | 4570 | |
4fbcdf66 FM |
4571 | /* |
4572 | * Ignore failure to create system chunk. We might end up not | |
4573 | * needing it, as we might not need to COW all nodes/leafs from | |
4574 | * the paths we visit in the chunk tree (they were already COWed | |
4575 | * or created in the current transaction for example). | |
4576 | */ | |
2ff7e61e | 4577 | ret = btrfs_alloc_chunk(trans, fs_info, flags); |
4fbcdf66 FM |
4578 | } |
4579 | ||
4580 | if (!ret) { | |
0b246afa JM |
4581 | ret = btrfs_block_rsv_add(fs_info->chunk_root, |
4582 | &fs_info->chunk_block_rsv, | |
4fbcdf66 FM |
4583 | thresh, BTRFS_RESERVE_NO_FLUSH); |
4584 | if (!ret) | |
4585 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4586 | } |
4587 | } | |
4588 | ||
28b737f6 LB |
4589 | /* |
4590 | * If force is CHUNK_ALLOC_FORCE: | |
4591 | * - return 1 if it successfully allocates a chunk, | |
4592 | * - return errors including -ENOSPC otherwise. | |
4593 | * If force is NOT CHUNK_ALLOC_FORCE: | |
4594 | * - return 0 if it doesn't need to allocate a new chunk, | |
4595 | * - return 1 if it successfully allocates a chunk, | |
4596 | * - return errors including -ENOSPC otherwise. | |
4597 | */ | |
6324fbf3 | 4598 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 4599 | struct btrfs_fs_info *fs_info, u64 flags, int force) |
9ed74f2d | 4600 | { |
6324fbf3 | 4601 | struct btrfs_space_info *space_info; |
6d74119f | 4602 | int wait_for_alloc = 0; |
9ed74f2d | 4603 | int ret = 0; |
9ed74f2d | 4604 | |
c6b305a8 JB |
4605 | /* Don't re-enter if we're already allocating a chunk */ |
4606 | if (trans->allocating_chunk) | |
4607 | return -ENOSPC; | |
4608 | ||
0b246afa | 4609 | space_info = __find_space_info(fs_info, flags); |
593060d7 | 4610 | if (!space_info) { |
2be12ef7 NB |
4611 | ret = create_space_info(fs_info, flags, &space_info); |
4612 | if (ret) | |
4613 | return ret; | |
9ed74f2d JB |
4614 | } |
4615 | ||
6d74119f | 4616 | again: |
25179201 | 4617 | spin_lock(&space_info->lock); |
9e622d6b | 4618 | if (force < space_info->force_alloc) |
0e4f8f88 | 4619 | force = space_info->force_alloc; |
25179201 | 4620 | if (space_info->full) { |
2ff7e61e | 4621 | if (should_alloc_chunk(fs_info, space_info, force)) |
09fb99a6 FDBM |
4622 | ret = -ENOSPC; |
4623 | else | |
4624 | ret = 0; | |
25179201 | 4625 | spin_unlock(&space_info->lock); |
09fb99a6 | 4626 | return ret; |
9ed74f2d JB |
4627 | } |
4628 | ||
2ff7e61e | 4629 | if (!should_alloc_chunk(fs_info, space_info, force)) { |
25179201 | 4630 | spin_unlock(&space_info->lock); |
6d74119f JB |
4631 | return 0; |
4632 | } else if (space_info->chunk_alloc) { | |
4633 | wait_for_alloc = 1; | |
4634 | } else { | |
4635 | space_info->chunk_alloc = 1; | |
9ed74f2d | 4636 | } |
0e4f8f88 | 4637 | |
25179201 | 4638 | spin_unlock(&space_info->lock); |
9ed74f2d | 4639 | |
6d74119f JB |
4640 | mutex_lock(&fs_info->chunk_mutex); |
4641 | ||
4642 | /* | |
4643 | * The chunk_mutex is held throughout the entirety of a chunk | |
4644 | * allocation, so once we've acquired the chunk_mutex we know that the | |
4645 | * other guy is done and we need to recheck and see if we should | |
4646 | * allocate. | |
4647 | */ | |
4648 | if (wait_for_alloc) { | |
4649 | mutex_unlock(&fs_info->chunk_mutex); | |
4650 | wait_for_alloc = 0; | |
4651 | goto again; | |
4652 | } | |
4653 | ||
c6b305a8 JB |
4654 | trans->allocating_chunk = true; |
4655 | ||
67377734 JB |
4656 | /* |
4657 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4658 | * allocating mixed chunks instead of individual chunks. | |
4659 | */ | |
4660 | if (btrfs_mixed_space_info(space_info)) | |
4661 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4662 | ||
97e728d4 JB |
4663 | /* |
4664 | * if we're doing a data chunk, go ahead and make sure that | |
4665 | * we keep a reasonable number of metadata chunks allocated in the | |
4666 | * FS as well. | |
4667 | */ | |
9ed74f2d | 4668 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4669 | fs_info->data_chunk_allocations++; |
4670 | if (!(fs_info->data_chunk_allocations % | |
4671 | fs_info->metadata_ratio)) | |
4672 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4673 | } |
4674 | ||
15d1ff81 LB |
4675 | /* |
4676 | * Check if we have enough space in SYSTEM chunk because we may need | |
4677 | * to update devices. | |
4678 | */ | |
2ff7e61e | 4679 | check_system_chunk(trans, fs_info, flags); |
15d1ff81 | 4680 | |
2ff7e61e | 4681 | ret = btrfs_alloc_chunk(trans, fs_info, flags); |
c6b305a8 | 4682 | trans->allocating_chunk = false; |
92b8e897 | 4683 | |
9ed74f2d | 4684 | spin_lock(&space_info->lock); |
a81cb9a2 AO |
4685 | if (ret < 0 && ret != -ENOSPC) |
4686 | goto out; | |
9ed74f2d | 4687 | if (ret) |
6324fbf3 | 4688 | space_info->full = 1; |
424499db YZ |
4689 | else |
4690 | ret = 1; | |
6d74119f | 4691 | |
0e4f8f88 | 4692 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4693 | out: |
6d74119f | 4694 | space_info->chunk_alloc = 0; |
9ed74f2d | 4695 | spin_unlock(&space_info->lock); |
a25c75d5 | 4696 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4697 | /* |
4698 | * When we allocate a new chunk we reserve space in the chunk block | |
4699 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4700 | * add new nodes/leafs to it if we end up needing to do it when | |
4701 | * inserting the chunk item and updating device items as part of the | |
4702 | * second phase of chunk allocation, performed by | |
4703 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4704 | * large number of new block groups to create in our transaction | |
4705 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4706 | * in extreme cases - like having a single transaction create many new | |
4707 | * block groups when starting to write out the free space caches of all | |
4708 | * the block groups that were made dirty during the lifetime of the | |
4709 | * transaction. | |
4710 | */ | |
d9a0540a | 4711 | if (trans->can_flush_pending_bgs && |
ee22184b | 4712 | trans->chunk_bytes_reserved >= (u64)SZ_2M) { |
6c686b35 | 4713 | btrfs_create_pending_block_groups(trans); |
00d80e34 FM |
4714 | btrfs_trans_release_chunk_metadata(trans); |
4715 | } | |
0f9dd46c | 4716 | return ret; |
6324fbf3 | 4717 | } |
9ed74f2d | 4718 | |
c1c4919b | 4719 | static int can_overcommit(struct btrfs_fs_info *fs_info, |
a80c8dcf | 4720 | struct btrfs_space_info *space_info, u64 bytes, |
c1c4919b JM |
4721 | enum btrfs_reserve_flush_enum flush, |
4722 | bool system_chunk) | |
a80c8dcf | 4723 | { |
0b246afa | 4724 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
957780eb | 4725 | u64 profile; |
3c76cd84 | 4726 | u64 space_size; |
a80c8dcf JB |
4727 | u64 avail; |
4728 | u64 used; | |
4729 | ||
957780eb JB |
4730 | /* Don't overcommit when in mixed mode. */ |
4731 | if (space_info->flags & BTRFS_BLOCK_GROUP_DATA) | |
4732 | return 0; | |
4733 | ||
c1c4919b JM |
4734 | if (system_chunk) |
4735 | profile = btrfs_system_alloc_profile(fs_info); | |
4736 | else | |
4737 | profile = btrfs_metadata_alloc_profile(fs_info); | |
4738 | ||
4136135b | 4739 | used = btrfs_space_info_used(space_info, false); |
96f1bb57 | 4740 | |
96f1bb57 JB |
4741 | /* |
4742 | * We only want to allow over committing if we have lots of actual space | |
4743 | * free, but if we don't have enough space to handle the global reserve | |
4744 | * space then we could end up having a real enospc problem when trying | |
4745 | * to allocate a chunk or some other such important allocation. | |
4746 | */ | |
3c76cd84 MX |
4747 | spin_lock(&global_rsv->lock); |
4748 | space_size = calc_global_rsv_need_space(global_rsv); | |
4749 | spin_unlock(&global_rsv->lock); | |
4750 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
4751 | return 0; |
4752 | ||
4753 | used += space_info->bytes_may_use; | |
a80c8dcf | 4754 | |
a5ed45f8 | 4755 | avail = atomic64_read(&fs_info->free_chunk_space); |
a80c8dcf JB |
4756 | |
4757 | /* | |
4758 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
4759 | * space is actually useable. For raid56, the space info used |
4760 | * doesn't include the parity drive, so we don't have to | |
4761 | * change the math | |
a80c8dcf JB |
4762 | */ |
4763 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
4764 | BTRFS_BLOCK_GROUP_RAID1 | | |
4765 | BTRFS_BLOCK_GROUP_RAID10)) | |
4766 | avail >>= 1; | |
4767 | ||
4768 | /* | |
561c294d MX |
4769 | * If we aren't flushing all things, let us overcommit up to |
4770 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4771 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 4772 | */ |
08e007d2 | 4773 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 4774 | avail >>= 3; |
a80c8dcf | 4775 | else |
14575aef | 4776 | avail >>= 1; |
a80c8dcf | 4777 | |
14575aef | 4778 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
4779 | return 1; |
4780 | return 0; | |
4781 | } | |
4782 | ||
2ff7e61e | 4783 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info, |
6c255e67 | 4784 | unsigned long nr_pages, int nr_items) |
da633a42 | 4785 | { |
0b246afa | 4786 | struct super_block *sb = fs_info->sb; |
da633a42 | 4787 | |
925a6efb JB |
4788 | if (down_read_trylock(&sb->s_umount)) { |
4789 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4790 | up_read(&sb->s_umount); | |
4791 | } else { | |
da633a42 MX |
4792 | /* |
4793 | * We needn't worry the filesystem going from r/w to r/o though | |
4794 | * we don't acquire ->s_umount mutex, because the filesystem | |
4795 | * should guarantee the delalloc inodes list be empty after | |
4796 | * the filesystem is readonly(all dirty pages are written to | |
4797 | * the disk). | |
4798 | */ | |
0b246afa | 4799 | btrfs_start_delalloc_roots(fs_info, 0, nr_items); |
98ad69cf | 4800 | if (!current->journal_info) |
0b246afa | 4801 | btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1); |
da633a42 MX |
4802 | } |
4803 | } | |
4804 | ||
6374e57a | 4805 | static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info, |
2ff7e61e | 4806 | u64 to_reclaim) |
18cd8ea6 MX |
4807 | { |
4808 | u64 bytes; | |
6374e57a | 4809 | u64 nr; |
18cd8ea6 | 4810 | |
2ff7e61e | 4811 | bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
6374e57a | 4812 | nr = div64_u64(to_reclaim, bytes); |
18cd8ea6 MX |
4813 | if (!nr) |
4814 | nr = 1; | |
4815 | return nr; | |
4816 | } | |
4817 | ||
ee22184b | 4818 | #define EXTENT_SIZE_PER_ITEM SZ_256K |
c61a16a7 | 4819 | |
9ed74f2d | 4820 | /* |
5da9d01b | 4821 | * shrink metadata reservation for delalloc |
9ed74f2d | 4822 | */ |
c1c4919b JM |
4823 | static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim, |
4824 | u64 orig, bool wait_ordered) | |
5da9d01b | 4825 | { |
0019f10d | 4826 | struct btrfs_space_info *space_info; |
663350ac | 4827 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4828 | u64 delalloc_bytes; |
5da9d01b | 4829 | u64 max_reclaim; |
6374e57a | 4830 | u64 items; |
b1953bce | 4831 | long time_left; |
d3ee29e3 MX |
4832 | unsigned long nr_pages; |
4833 | int loops; | |
08e007d2 | 4834 | enum btrfs_reserve_flush_enum flush; |
5da9d01b | 4835 | |
c61a16a7 | 4836 | /* Calc the number of the pages we need flush for space reservation */ |
2ff7e61e | 4837 | items = calc_reclaim_items_nr(fs_info, to_reclaim); |
6374e57a | 4838 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; |
c61a16a7 | 4839 | |
663350ac | 4840 | trans = (struct btrfs_trans_handle *)current->journal_info; |
69fe2d75 | 4841 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
bf9022e0 | 4842 | |
963d678b | 4843 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4844 | &fs_info->delalloc_bytes); |
f4c738c2 | 4845 | if (delalloc_bytes == 0) { |
fdb5effd | 4846 | if (trans) |
f4c738c2 | 4847 | return; |
38c135af | 4848 | if (wait_ordered) |
0b246afa | 4849 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f4c738c2 | 4850 | return; |
fdb5effd JB |
4851 | } |
4852 | ||
d3ee29e3 | 4853 | loops = 0; |
f4c738c2 JB |
4854 | while (delalloc_bytes && loops < 3) { |
4855 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
09cbfeaf | 4856 | nr_pages = max_reclaim >> PAGE_SHIFT; |
2ff7e61e | 4857 | btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items); |
dea31f52 JB |
4858 | /* |
4859 | * We need to wait for the async pages to actually start before | |
4860 | * we do anything. | |
4861 | */ | |
0b246afa | 4862 | max_reclaim = atomic_read(&fs_info->async_delalloc_pages); |
9f3a074d MX |
4863 | if (!max_reclaim) |
4864 | goto skip_async; | |
4865 | ||
4866 | if (max_reclaim <= nr_pages) | |
4867 | max_reclaim = 0; | |
4868 | else | |
4869 | max_reclaim -= nr_pages; | |
dea31f52 | 4870 | |
0b246afa JM |
4871 | wait_event(fs_info->async_submit_wait, |
4872 | atomic_read(&fs_info->async_delalloc_pages) <= | |
9f3a074d MX |
4873 | (int)max_reclaim); |
4874 | skip_async: | |
08e007d2 MX |
4875 | if (!trans) |
4876 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
4877 | else | |
4878 | flush = BTRFS_RESERVE_NO_FLUSH; | |
0019f10d | 4879 | spin_lock(&space_info->lock); |
957780eb JB |
4880 | if (list_empty(&space_info->tickets) && |
4881 | list_empty(&space_info->priority_tickets)) { | |
4882 | spin_unlock(&space_info->lock); | |
4883 | break; | |
4884 | } | |
0019f10d | 4885 | spin_unlock(&space_info->lock); |
5da9d01b | 4886 | |
36e39c40 | 4887 | loops++; |
f104d044 | 4888 | if (wait_ordered && !trans) { |
0b246afa | 4889 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f104d044 | 4890 | } else { |
f4c738c2 | 4891 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4892 | if (time_left) |
4893 | break; | |
4894 | } | |
963d678b | 4895 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4896 | &fs_info->delalloc_bytes); |
5da9d01b | 4897 | } |
5da9d01b YZ |
4898 | } |
4899 | ||
996478ca JB |
4900 | struct reserve_ticket { |
4901 | u64 bytes; | |
4902 | int error; | |
4903 | struct list_head list; | |
4904 | wait_queue_head_t wait; | |
4905 | }; | |
4906 | ||
663350ac JB |
4907 | /** |
4908 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4909 | * @root - the root we're allocating for | |
4910 | * @bytes - the number of bytes we want to reserve | |
4911 | * @force - force the commit | |
8bb8ab2e | 4912 | * |
663350ac JB |
4913 | * This will check to make sure that committing the transaction will actually |
4914 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4915 | * will return -ENOSPC. | |
8bb8ab2e | 4916 | */ |
0c9ab349 | 4917 | static int may_commit_transaction(struct btrfs_fs_info *fs_info, |
996478ca | 4918 | struct btrfs_space_info *space_info) |
663350ac | 4919 | { |
996478ca | 4920 | struct reserve_ticket *ticket = NULL; |
0b246afa | 4921 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv; |
663350ac | 4922 | struct btrfs_trans_handle *trans; |
996478ca | 4923 | u64 bytes; |
663350ac JB |
4924 | |
4925 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4926 | if (trans) | |
4927 | return -EAGAIN; | |
4928 | ||
996478ca JB |
4929 | spin_lock(&space_info->lock); |
4930 | if (!list_empty(&space_info->priority_tickets)) | |
4931 | ticket = list_first_entry(&space_info->priority_tickets, | |
4932 | struct reserve_ticket, list); | |
4933 | else if (!list_empty(&space_info->tickets)) | |
4934 | ticket = list_first_entry(&space_info->tickets, | |
4935 | struct reserve_ticket, list); | |
4936 | bytes = (ticket) ? ticket->bytes : 0; | |
4937 | spin_unlock(&space_info->lock); | |
4938 | ||
4939 | if (!bytes) | |
4940 | return 0; | |
663350ac JB |
4941 | |
4942 | /* See if there is enough pinned space to make this reservation */ | |
b150a4f1 | 4943 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
0424c548 | 4944 | bytes) >= 0) |
663350ac | 4945 | goto commit; |
663350ac JB |
4946 | |
4947 | /* | |
4948 | * See if there is some space in the delayed insertion reservation for | |
4949 | * this reservation. | |
4950 | */ | |
4951 | if (space_info != delayed_rsv->space_info) | |
4952 | return -ENOSPC; | |
4953 | ||
4954 | spin_lock(&delayed_rsv->lock); | |
996478ca JB |
4955 | if (delayed_rsv->size > bytes) |
4956 | bytes = 0; | |
4957 | else | |
4958 | bytes -= delayed_rsv->size; | |
057aac3e NB |
4959 | spin_unlock(&delayed_rsv->lock); |
4960 | ||
b150a4f1 | 4961 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
996478ca | 4962 | bytes) < 0) { |
663350ac JB |
4963 | return -ENOSPC; |
4964 | } | |
663350ac JB |
4965 | |
4966 | commit: | |
a9b3311e | 4967 | trans = btrfs_join_transaction(fs_info->extent_root); |
663350ac JB |
4968 | if (IS_ERR(trans)) |
4969 | return -ENOSPC; | |
4970 | ||
3a45bb20 | 4971 | return btrfs_commit_transaction(trans); |
663350ac JB |
4972 | } |
4973 | ||
e38ae7a0 NB |
4974 | /* |
4975 | * Try to flush some data based on policy set by @state. This is only advisory | |
4976 | * and may fail for various reasons. The caller is supposed to examine the | |
4977 | * state of @space_info to detect the outcome. | |
4978 | */ | |
4979 | static void flush_space(struct btrfs_fs_info *fs_info, | |
96c3f433 | 4980 | struct btrfs_space_info *space_info, u64 num_bytes, |
7bdd6277 | 4981 | int state) |
96c3f433 | 4982 | { |
a9b3311e | 4983 | struct btrfs_root *root = fs_info->extent_root; |
96c3f433 JB |
4984 | struct btrfs_trans_handle *trans; |
4985 | int nr; | |
f4c738c2 | 4986 | int ret = 0; |
96c3f433 JB |
4987 | |
4988 | switch (state) { | |
96c3f433 JB |
4989 | case FLUSH_DELAYED_ITEMS_NR: |
4990 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 | 4991 | if (state == FLUSH_DELAYED_ITEMS_NR) |
2ff7e61e | 4992 | nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2; |
18cd8ea6 | 4993 | else |
96c3f433 | 4994 | nr = -1; |
18cd8ea6 | 4995 | |
96c3f433 JB |
4996 | trans = btrfs_join_transaction(root); |
4997 | if (IS_ERR(trans)) { | |
4998 | ret = PTR_ERR(trans); | |
4999 | break; | |
5000 | } | |
e5c304e6 | 5001 | ret = btrfs_run_delayed_items_nr(trans, nr); |
3a45bb20 | 5002 | btrfs_end_transaction(trans); |
96c3f433 | 5003 | break; |
67b0fd63 JB |
5004 | case FLUSH_DELALLOC: |
5005 | case FLUSH_DELALLOC_WAIT: | |
7bdd6277 | 5006 | shrink_delalloc(fs_info, num_bytes * 2, num_bytes, |
67b0fd63 JB |
5007 | state == FLUSH_DELALLOC_WAIT); |
5008 | break; | |
ea658bad JB |
5009 | case ALLOC_CHUNK: |
5010 | trans = btrfs_join_transaction(root); | |
5011 | if (IS_ERR(trans)) { | |
5012 | ret = PTR_ERR(trans); | |
5013 | break; | |
5014 | } | |
2ff7e61e | 5015 | ret = do_chunk_alloc(trans, fs_info, |
1b86826d | 5016 | btrfs_metadata_alloc_profile(fs_info), |
ea658bad | 5017 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 5018 | btrfs_end_transaction(trans); |
eecba891 | 5019 | if (ret > 0 || ret == -ENOSPC) |
ea658bad JB |
5020 | ret = 0; |
5021 | break; | |
96c3f433 | 5022 | case COMMIT_TRANS: |
996478ca | 5023 | ret = may_commit_transaction(fs_info, space_info); |
96c3f433 JB |
5024 | break; |
5025 | default: | |
5026 | ret = -ENOSPC; | |
5027 | break; | |
5028 | } | |
5029 | ||
7bdd6277 NB |
5030 | trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state, |
5031 | ret); | |
e38ae7a0 | 5032 | return; |
96c3f433 | 5033 | } |
21c7e756 MX |
5034 | |
5035 | static inline u64 | |
c1c4919b JM |
5036 | btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info, |
5037 | struct btrfs_space_info *space_info, | |
5038 | bool system_chunk) | |
21c7e756 | 5039 | { |
957780eb | 5040 | struct reserve_ticket *ticket; |
21c7e756 MX |
5041 | u64 used; |
5042 | u64 expected; | |
957780eb | 5043 | u64 to_reclaim = 0; |
21c7e756 | 5044 | |
957780eb JB |
5045 | list_for_each_entry(ticket, &space_info->tickets, list) |
5046 | to_reclaim += ticket->bytes; | |
5047 | list_for_each_entry(ticket, &space_info->priority_tickets, list) | |
5048 | to_reclaim += ticket->bytes; | |
5049 | if (to_reclaim) | |
5050 | return to_reclaim; | |
21c7e756 | 5051 | |
e0af2484 | 5052 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); |
c1c4919b JM |
5053 | if (can_overcommit(fs_info, space_info, to_reclaim, |
5054 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
e0af2484 WX |
5055 | return 0; |
5056 | ||
0eee8a49 NB |
5057 | used = btrfs_space_info_used(space_info, true); |
5058 | ||
c1c4919b JM |
5059 | if (can_overcommit(fs_info, space_info, SZ_1M, |
5060 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
21c7e756 MX |
5061 | expected = div_factor_fine(space_info->total_bytes, 95); |
5062 | else | |
5063 | expected = div_factor_fine(space_info->total_bytes, 90); | |
5064 | ||
5065 | if (used > expected) | |
5066 | to_reclaim = used - expected; | |
5067 | else | |
5068 | to_reclaim = 0; | |
5069 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
5070 | space_info->bytes_reserved); | |
21c7e756 MX |
5071 | return to_reclaim; |
5072 | } | |
5073 | ||
c1c4919b JM |
5074 | static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info, |
5075 | struct btrfs_space_info *space_info, | |
5076 | u64 used, bool system_chunk) | |
21c7e756 | 5077 | { |
365c5313 JB |
5078 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
5079 | ||
5080 | /* If we're just plain full then async reclaim just slows us down. */ | |
baee8790 | 5081 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) |
365c5313 JB |
5082 | return 0; |
5083 | ||
c1c4919b JM |
5084 | if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5085 | system_chunk)) | |
d38b349c JB |
5086 | return 0; |
5087 | ||
0b246afa JM |
5088 | return (used >= thresh && !btrfs_fs_closing(fs_info) && |
5089 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); | |
21c7e756 MX |
5090 | } |
5091 | ||
957780eb | 5092 | static void wake_all_tickets(struct list_head *head) |
21c7e756 | 5093 | { |
957780eb | 5094 | struct reserve_ticket *ticket; |
25ce459c | 5095 | |
957780eb JB |
5096 | while (!list_empty(head)) { |
5097 | ticket = list_first_entry(head, struct reserve_ticket, list); | |
5098 | list_del_init(&ticket->list); | |
5099 | ticket->error = -ENOSPC; | |
5100 | wake_up(&ticket->wait); | |
21c7e756 | 5101 | } |
21c7e756 MX |
5102 | } |
5103 | ||
957780eb JB |
5104 | /* |
5105 | * This is for normal flushers, we can wait all goddamned day if we want to. We | |
5106 | * will loop and continuously try to flush as long as we are making progress. | |
5107 | * We count progress as clearing off tickets each time we have to loop. | |
5108 | */ | |
21c7e756 MX |
5109 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) |
5110 | { | |
5111 | struct btrfs_fs_info *fs_info; | |
5112 | struct btrfs_space_info *space_info; | |
5113 | u64 to_reclaim; | |
5114 | int flush_state; | |
957780eb | 5115 | int commit_cycles = 0; |
ce129655 | 5116 | u64 last_tickets_id; |
21c7e756 MX |
5117 | |
5118 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
5119 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
5120 | ||
957780eb | 5121 | spin_lock(&space_info->lock); |
c1c4919b JM |
5122 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5123 | false); | |
957780eb JB |
5124 | if (!to_reclaim) { |
5125 | space_info->flush = 0; | |
5126 | spin_unlock(&space_info->lock); | |
21c7e756 | 5127 | return; |
957780eb | 5128 | } |
ce129655 | 5129 | last_tickets_id = space_info->tickets_id; |
957780eb | 5130 | spin_unlock(&space_info->lock); |
21c7e756 MX |
5131 | |
5132 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
957780eb | 5133 | do { |
e38ae7a0 | 5134 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
957780eb JB |
5135 | spin_lock(&space_info->lock); |
5136 | if (list_empty(&space_info->tickets)) { | |
5137 | space_info->flush = 0; | |
5138 | spin_unlock(&space_info->lock); | |
5139 | return; | |
5140 | } | |
c1c4919b JM |
5141 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, |
5142 | space_info, | |
5143 | false); | |
ce129655 | 5144 | if (last_tickets_id == space_info->tickets_id) { |
957780eb JB |
5145 | flush_state++; |
5146 | } else { | |
ce129655 | 5147 | last_tickets_id = space_info->tickets_id; |
957780eb JB |
5148 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
5149 | if (commit_cycles) | |
5150 | commit_cycles--; | |
5151 | } | |
5152 | ||
5153 | if (flush_state > COMMIT_TRANS) { | |
5154 | commit_cycles++; | |
5155 | if (commit_cycles > 2) { | |
5156 | wake_all_tickets(&space_info->tickets); | |
5157 | space_info->flush = 0; | |
5158 | } else { | |
5159 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5160 | } | |
5161 | } | |
5162 | spin_unlock(&space_info->lock); | |
5163 | } while (flush_state <= COMMIT_TRANS); | |
5164 | } | |
5165 | ||
5166 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
5167 | { | |
5168 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
5169 | } | |
5170 | ||
5171 | static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, | |
5172 | struct btrfs_space_info *space_info, | |
5173 | struct reserve_ticket *ticket) | |
5174 | { | |
5175 | u64 to_reclaim; | |
5176 | int flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5177 | ||
5178 | spin_lock(&space_info->lock); | |
c1c4919b JM |
5179 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5180 | false); | |
957780eb JB |
5181 | if (!to_reclaim) { |
5182 | spin_unlock(&space_info->lock); | |
5183 | return; | |
5184 | } | |
5185 | spin_unlock(&space_info->lock); | |
5186 | ||
21c7e756 | 5187 | do { |
7bdd6277 | 5188 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
21c7e756 | 5189 | flush_state++; |
957780eb JB |
5190 | spin_lock(&space_info->lock); |
5191 | if (ticket->bytes == 0) { | |
5192 | spin_unlock(&space_info->lock); | |
21c7e756 | 5193 | return; |
957780eb JB |
5194 | } |
5195 | spin_unlock(&space_info->lock); | |
5196 | ||
5197 | /* | |
5198 | * Priority flushers can't wait on delalloc without | |
5199 | * deadlocking. | |
5200 | */ | |
5201 | if (flush_state == FLUSH_DELALLOC || | |
5202 | flush_state == FLUSH_DELALLOC_WAIT) | |
5203 | flush_state = ALLOC_CHUNK; | |
365c5313 | 5204 | } while (flush_state < COMMIT_TRANS); |
21c7e756 MX |
5205 | } |
5206 | ||
957780eb JB |
5207 | static int wait_reserve_ticket(struct btrfs_fs_info *fs_info, |
5208 | struct btrfs_space_info *space_info, | |
5209 | struct reserve_ticket *ticket, u64 orig_bytes) | |
5210 | ||
21c7e756 | 5211 | { |
957780eb JB |
5212 | DEFINE_WAIT(wait); |
5213 | int ret = 0; | |
5214 | ||
5215 | spin_lock(&space_info->lock); | |
5216 | while (ticket->bytes > 0 && ticket->error == 0) { | |
5217 | ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE); | |
5218 | if (ret) { | |
5219 | ret = -EINTR; | |
5220 | break; | |
5221 | } | |
5222 | spin_unlock(&space_info->lock); | |
5223 | ||
5224 | schedule(); | |
5225 | ||
5226 | finish_wait(&ticket->wait, &wait); | |
5227 | spin_lock(&space_info->lock); | |
5228 | } | |
5229 | if (!ret) | |
5230 | ret = ticket->error; | |
5231 | if (!list_empty(&ticket->list)) | |
5232 | list_del_init(&ticket->list); | |
5233 | if (ticket->bytes && ticket->bytes < orig_bytes) { | |
5234 | u64 num_bytes = orig_bytes - ticket->bytes; | |
5235 | space_info->bytes_may_use -= num_bytes; | |
5236 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5237 | space_info->flags, num_bytes, 0); | |
5238 | } | |
5239 | spin_unlock(&space_info->lock); | |
5240 | ||
5241 | return ret; | |
21c7e756 MX |
5242 | } |
5243 | ||
4a92b1b8 JB |
5244 | /** |
5245 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5246 | * @root - the root we're allocating for | |
957780eb | 5247 | * @space_info - the space info we want to allocate from |
4a92b1b8 | 5248 | * @orig_bytes - the number of bytes we want |
48fc7f7e | 5249 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 5250 | * |
01327610 | 5251 | * This will reserve orig_bytes number of bytes from the space info associated |
4a92b1b8 JB |
5252 | * with the block_rsv. If there is not enough space it will make an attempt to |
5253 | * flush out space to make room. It will do this by flushing delalloc if | |
5254 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5255 | * regain reservations will be made and this will fail if there is not enough | |
5256 | * space already. | |
8bb8ab2e | 5257 | */ |
c1c4919b | 5258 | static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info, |
957780eb JB |
5259 | struct btrfs_space_info *space_info, |
5260 | u64 orig_bytes, | |
c1c4919b JM |
5261 | enum btrfs_reserve_flush_enum flush, |
5262 | bool system_chunk) | |
9ed74f2d | 5263 | { |
957780eb | 5264 | struct reserve_ticket ticket; |
2bf64758 | 5265 | u64 used; |
8bb8ab2e | 5266 | int ret = 0; |
9ed74f2d | 5267 | |
957780eb | 5268 | ASSERT(orig_bytes); |
8ca17f0f | 5269 | ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL); |
fdb5effd | 5270 | |
8bb8ab2e | 5271 | spin_lock(&space_info->lock); |
fdb5effd | 5272 | ret = -ENOSPC; |
4136135b | 5273 | used = btrfs_space_info_used(space_info, true); |
9ed74f2d | 5274 | |
8bb8ab2e | 5275 | /* |
957780eb JB |
5276 | * If we have enough space then hooray, make our reservation and carry |
5277 | * on. If not see if we can overcommit, and if we can, hooray carry on. | |
5278 | * If not things get more complicated. | |
8bb8ab2e | 5279 | */ |
957780eb JB |
5280 | if (used + orig_bytes <= space_info->total_bytes) { |
5281 | space_info->bytes_may_use += orig_bytes; | |
0b246afa JM |
5282 | trace_btrfs_space_reservation(fs_info, "space_info", |
5283 | space_info->flags, orig_bytes, 1); | |
957780eb | 5284 | ret = 0; |
c1c4919b JM |
5285 | } else if (can_overcommit(fs_info, space_info, orig_bytes, flush, |
5286 | system_chunk)) { | |
44734ed1 | 5287 | space_info->bytes_may_use += orig_bytes; |
0b246afa JM |
5288 | trace_btrfs_space_reservation(fs_info, "space_info", |
5289 | space_info->flags, orig_bytes, 1); | |
44734ed1 | 5290 | ret = 0; |
2bf64758 JB |
5291 | } |
5292 | ||
8bb8ab2e | 5293 | /* |
957780eb JB |
5294 | * If we couldn't make a reservation then setup our reservation ticket |
5295 | * and kick the async worker if it's not already running. | |
08e007d2 | 5296 | * |
957780eb JB |
5297 | * If we are a priority flusher then we just need to add our ticket to |
5298 | * the list and we will do our own flushing further down. | |
8bb8ab2e | 5299 | */ |
72bcd99d | 5300 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
957780eb JB |
5301 | ticket.bytes = orig_bytes; |
5302 | ticket.error = 0; | |
5303 | init_waitqueue_head(&ticket.wait); | |
5304 | if (flush == BTRFS_RESERVE_FLUSH_ALL) { | |
5305 | list_add_tail(&ticket.list, &space_info->tickets); | |
5306 | if (!space_info->flush) { | |
5307 | space_info->flush = 1; | |
0b246afa | 5308 | trace_btrfs_trigger_flush(fs_info, |
f376df2b JB |
5309 | space_info->flags, |
5310 | orig_bytes, flush, | |
5311 | "enospc"); | |
957780eb | 5312 | queue_work(system_unbound_wq, |
c1c4919b | 5313 | &fs_info->async_reclaim_work); |
957780eb JB |
5314 | } |
5315 | } else { | |
5316 | list_add_tail(&ticket.list, | |
5317 | &space_info->priority_tickets); | |
5318 | } | |
21c7e756 MX |
5319 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
5320 | used += orig_bytes; | |
f6acfd50 JB |
5321 | /* |
5322 | * We will do the space reservation dance during log replay, | |
5323 | * which means we won't have fs_info->fs_root set, so don't do | |
5324 | * the async reclaim as we will panic. | |
5325 | */ | |
0b246afa | 5326 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) && |
c1c4919b JM |
5327 | need_do_async_reclaim(fs_info, space_info, |
5328 | used, system_chunk) && | |
0b246afa JM |
5329 | !work_busy(&fs_info->async_reclaim_work)) { |
5330 | trace_btrfs_trigger_flush(fs_info, space_info->flags, | |
5331 | orig_bytes, flush, "preempt"); | |
21c7e756 | 5332 | queue_work(system_unbound_wq, |
0b246afa | 5333 | &fs_info->async_reclaim_work); |
f376df2b | 5334 | } |
8bb8ab2e | 5335 | } |
f0486c68 | 5336 | spin_unlock(&space_info->lock); |
08e007d2 | 5337 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
957780eb | 5338 | return ret; |
f0486c68 | 5339 | |
957780eb | 5340 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
0b246afa | 5341 | return wait_reserve_ticket(fs_info, space_info, &ticket, |
957780eb | 5342 | orig_bytes); |
08e007d2 | 5343 | |
957780eb | 5344 | ret = 0; |
0b246afa | 5345 | priority_reclaim_metadata_space(fs_info, space_info, &ticket); |
957780eb JB |
5346 | spin_lock(&space_info->lock); |
5347 | if (ticket.bytes) { | |
5348 | if (ticket.bytes < orig_bytes) { | |
5349 | u64 num_bytes = orig_bytes - ticket.bytes; | |
5350 | space_info->bytes_may_use -= num_bytes; | |
0b246afa JM |
5351 | trace_btrfs_space_reservation(fs_info, "space_info", |
5352 | space_info->flags, | |
5353 | num_bytes, 0); | |
08e007d2 | 5354 | |
957780eb JB |
5355 | } |
5356 | list_del_init(&ticket.list); | |
5357 | ret = -ENOSPC; | |
5358 | } | |
5359 | spin_unlock(&space_info->lock); | |
5360 | ASSERT(list_empty(&ticket.list)); | |
5361 | return ret; | |
5362 | } | |
8bb8ab2e | 5363 | |
957780eb JB |
5364 | /** |
5365 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5366 | * @root - the root we're allocating for | |
5367 | * @block_rsv - the block_rsv we're allocating for | |
5368 | * @orig_bytes - the number of bytes we want | |
5369 | * @flush - whether or not we can flush to make our reservation | |
5370 | * | |
5371 | * This will reserve orgi_bytes number of bytes from the space info associated | |
5372 | * with the block_rsv. If there is not enough space it will make an attempt to | |
5373 | * flush out space to make room. It will do this by flushing delalloc if | |
5374 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5375 | * regain reservations will be made and this will fail if there is not enough | |
5376 | * space already. | |
5377 | */ | |
5378 | static int reserve_metadata_bytes(struct btrfs_root *root, | |
5379 | struct btrfs_block_rsv *block_rsv, | |
5380 | u64 orig_bytes, | |
5381 | enum btrfs_reserve_flush_enum flush) | |
5382 | { | |
0b246afa JM |
5383 | struct btrfs_fs_info *fs_info = root->fs_info; |
5384 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
957780eb | 5385 | int ret; |
c1c4919b | 5386 | bool system_chunk = (root == fs_info->chunk_root); |
957780eb | 5387 | |
c1c4919b JM |
5388 | ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info, |
5389 | orig_bytes, flush, system_chunk); | |
5d80366e JB |
5390 | if (ret == -ENOSPC && |
5391 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5d80366e JB |
5392 | if (block_rsv != global_rsv && |
5393 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5394 | ret = 0; | |
5395 | } | |
9a3daff3 | 5396 | if (ret == -ENOSPC) { |
0b246afa | 5397 | trace_btrfs_space_reservation(fs_info, "space_info:enospc", |
957780eb JB |
5398 | block_rsv->space_info->flags, |
5399 | orig_bytes, 1); | |
9a3daff3 NB |
5400 | |
5401 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) | |
5402 | dump_space_info(fs_info, block_rsv->space_info, | |
5403 | orig_bytes, 0); | |
5404 | } | |
f0486c68 YZ |
5405 | return ret; |
5406 | } | |
5407 | ||
79787eaa JM |
5408 | static struct btrfs_block_rsv *get_block_rsv( |
5409 | const struct btrfs_trans_handle *trans, | |
5410 | const struct btrfs_root *root) | |
f0486c68 | 5411 | { |
0b246afa | 5412 | struct btrfs_fs_info *fs_info = root->fs_info; |
4c13d758 JB |
5413 | struct btrfs_block_rsv *block_rsv = NULL; |
5414 | ||
e9cf439f | 5415 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa JM |
5416 | (root == fs_info->csum_root && trans->adding_csums) || |
5417 | (root == fs_info->uuid_root)) | |
f7a81ea4 SB |
5418 | block_rsv = trans->block_rsv; |
5419 | ||
4c13d758 | 5420 | if (!block_rsv) |
f0486c68 YZ |
5421 | block_rsv = root->block_rsv; |
5422 | ||
5423 | if (!block_rsv) | |
0b246afa | 5424 | block_rsv = &fs_info->empty_block_rsv; |
f0486c68 YZ |
5425 | |
5426 | return block_rsv; | |
5427 | } | |
5428 | ||
5429 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
5430 | u64 num_bytes) | |
5431 | { | |
5432 | int ret = -ENOSPC; | |
5433 | spin_lock(&block_rsv->lock); | |
5434 | if (block_rsv->reserved >= num_bytes) { | |
5435 | block_rsv->reserved -= num_bytes; | |
5436 | if (block_rsv->reserved < block_rsv->size) | |
5437 | block_rsv->full = 0; | |
5438 | ret = 0; | |
5439 | } | |
5440 | spin_unlock(&block_rsv->lock); | |
5441 | return ret; | |
5442 | } | |
5443 | ||
5444 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
5445 | u64 num_bytes, int update_size) | |
5446 | { | |
5447 | spin_lock(&block_rsv->lock); | |
5448 | block_rsv->reserved += num_bytes; | |
5449 | if (update_size) | |
5450 | block_rsv->size += num_bytes; | |
5451 | else if (block_rsv->reserved >= block_rsv->size) | |
5452 | block_rsv->full = 1; | |
5453 | spin_unlock(&block_rsv->lock); | |
5454 | } | |
5455 | ||
d52be818 JB |
5456 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
5457 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5458 | int min_factor) | |
5459 | { | |
5460 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5461 | u64 min_bytes; | |
5462 | ||
5463 | if (global_rsv->space_info != dest->space_info) | |
5464 | return -ENOSPC; | |
5465 | ||
5466 | spin_lock(&global_rsv->lock); | |
5467 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5468 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5469 | spin_unlock(&global_rsv->lock); | |
5470 | return -ENOSPC; | |
5471 | } | |
5472 | global_rsv->reserved -= num_bytes; | |
5473 | if (global_rsv->reserved < global_rsv->size) | |
5474 | global_rsv->full = 0; | |
5475 | spin_unlock(&global_rsv->lock); | |
5476 | ||
5477 | block_rsv_add_bytes(dest, num_bytes, 1); | |
5478 | return 0; | |
5479 | } | |
5480 | ||
957780eb JB |
5481 | /* |
5482 | * This is for space we already have accounted in space_info->bytes_may_use, so | |
5483 | * basically when we're returning space from block_rsv's. | |
5484 | */ | |
5485 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
5486 | struct btrfs_space_info *space_info, | |
5487 | u64 num_bytes) | |
5488 | { | |
5489 | struct reserve_ticket *ticket; | |
5490 | struct list_head *head; | |
5491 | u64 used; | |
5492 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH; | |
5493 | bool check_overcommit = false; | |
5494 | ||
5495 | spin_lock(&space_info->lock); | |
5496 | head = &space_info->priority_tickets; | |
5497 | ||
5498 | /* | |
5499 | * If we are over our limit then we need to check and see if we can | |
5500 | * overcommit, and if we can't then we just need to free up our space | |
5501 | * and not satisfy any requests. | |
5502 | */ | |
0eee8a49 | 5503 | used = btrfs_space_info_used(space_info, true); |
957780eb JB |
5504 | if (used - num_bytes >= space_info->total_bytes) |
5505 | check_overcommit = true; | |
5506 | again: | |
5507 | while (!list_empty(head) && num_bytes) { | |
5508 | ticket = list_first_entry(head, struct reserve_ticket, | |
5509 | list); | |
5510 | /* | |
5511 | * We use 0 bytes because this space is already reserved, so | |
5512 | * adding the ticket space would be a double count. | |
5513 | */ | |
5514 | if (check_overcommit && | |
c1c4919b | 5515 | !can_overcommit(fs_info, space_info, 0, flush, false)) |
957780eb JB |
5516 | break; |
5517 | if (num_bytes >= ticket->bytes) { | |
5518 | list_del_init(&ticket->list); | |
5519 | num_bytes -= ticket->bytes; | |
5520 | ticket->bytes = 0; | |
ce129655 | 5521 | space_info->tickets_id++; |
957780eb JB |
5522 | wake_up(&ticket->wait); |
5523 | } else { | |
5524 | ticket->bytes -= num_bytes; | |
5525 | num_bytes = 0; | |
5526 | } | |
5527 | } | |
5528 | ||
5529 | if (num_bytes && head == &space_info->priority_tickets) { | |
5530 | head = &space_info->tickets; | |
5531 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
5532 | goto again; | |
5533 | } | |
5534 | space_info->bytes_may_use -= num_bytes; | |
5535 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5536 | space_info->flags, num_bytes, 0); | |
5537 | spin_unlock(&space_info->lock); | |
5538 | } | |
5539 | ||
5540 | /* | |
5541 | * This is for newly allocated space that isn't accounted in | |
5542 | * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent | |
5543 | * we use this helper. | |
5544 | */ | |
5545 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, | |
5546 | struct btrfs_space_info *space_info, | |
5547 | u64 num_bytes) | |
5548 | { | |
5549 | struct reserve_ticket *ticket; | |
5550 | struct list_head *head = &space_info->priority_tickets; | |
5551 | ||
5552 | again: | |
5553 | while (!list_empty(head) && num_bytes) { | |
5554 | ticket = list_first_entry(head, struct reserve_ticket, | |
5555 | list); | |
5556 | if (num_bytes >= ticket->bytes) { | |
5557 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5558 | space_info->flags, | |
5559 | ticket->bytes, 1); | |
5560 | list_del_init(&ticket->list); | |
5561 | num_bytes -= ticket->bytes; | |
5562 | space_info->bytes_may_use += ticket->bytes; | |
5563 | ticket->bytes = 0; | |
ce129655 | 5564 | space_info->tickets_id++; |
957780eb JB |
5565 | wake_up(&ticket->wait); |
5566 | } else { | |
5567 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5568 | space_info->flags, | |
5569 | num_bytes, 1); | |
5570 | space_info->bytes_may_use += num_bytes; | |
5571 | ticket->bytes -= num_bytes; | |
5572 | num_bytes = 0; | |
5573 | } | |
5574 | } | |
5575 | ||
5576 | if (num_bytes && head == &space_info->priority_tickets) { | |
5577 | head = &space_info->tickets; | |
5578 | goto again; | |
5579 | } | |
5580 | } | |
5581 | ||
69fe2d75 | 5582 | static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
8c2a3ca2 | 5583 | struct btrfs_block_rsv *block_rsv, |
62a45b60 | 5584 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
5585 | { |
5586 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
69fe2d75 | 5587 | u64 ret; |
f0486c68 YZ |
5588 | |
5589 | spin_lock(&block_rsv->lock); | |
5590 | if (num_bytes == (u64)-1) | |
5591 | num_bytes = block_rsv->size; | |
5592 | block_rsv->size -= num_bytes; | |
5593 | if (block_rsv->reserved >= block_rsv->size) { | |
5594 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5595 | block_rsv->reserved = block_rsv->size; | |
5596 | block_rsv->full = 1; | |
5597 | } else { | |
5598 | num_bytes = 0; | |
5599 | } | |
5600 | spin_unlock(&block_rsv->lock); | |
5601 | ||
69fe2d75 | 5602 | ret = num_bytes; |
f0486c68 YZ |
5603 | if (num_bytes > 0) { |
5604 | if (dest) { | |
e9e22899 JB |
5605 | spin_lock(&dest->lock); |
5606 | if (!dest->full) { | |
5607 | u64 bytes_to_add; | |
5608 | ||
5609 | bytes_to_add = dest->size - dest->reserved; | |
5610 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5611 | dest->reserved += bytes_to_add; | |
5612 | if (dest->reserved >= dest->size) | |
5613 | dest->full = 1; | |
5614 | num_bytes -= bytes_to_add; | |
5615 | } | |
5616 | spin_unlock(&dest->lock); | |
5617 | } | |
957780eb JB |
5618 | if (num_bytes) |
5619 | space_info_add_old_bytes(fs_info, space_info, | |
5620 | num_bytes); | |
9ed74f2d | 5621 | } |
69fe2d75 | 5622 | return ret; |
f0486c68 | 5623 | } |
4e06bdd6 | 5624 | |
25d609f8 JB |
5625 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src, |
5626 | struct btrfs_block_rsv *dst, u64 num_bytes, | |
5627 | int update_size) | |
f0486c68 YZ |
5628 | { |
5629 | int ret; | |
9ed74f2d | 5630 | |
f0486c68 YZ |
5631 | ret = block_rsv_use_bytes(src, num_bytes); |
5632 | if (ret) | |
5633 | return ret; | |
9ed74f2d | 5634 | |
25d609f8 | 5635 | block_rsv_add_bytes(dst, num_bytes, update_size); |
9ed74f2d JB |
5636 | return 0; |
5637 | } | |
5638 | ||
66d8f3dd | 5639 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 5640 | { |
f0486c68 YZ |
5641 | memset(rsv, 0, sizeof(*rsv)); |
5642 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 5643 | rsv->type = type; |
f0486c68 YZ |
5644 | } |
5645 | ||
69fe2d75 JB |
5646 | void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info, |
5647 | struct btrfs_block_rsv *rsv, | |
5648 | unsigned short type) | |
5649 | { | |
5650 | btrfs_init_block_rsv(rsv, type); | |
5651 | rsv->space_info = __find_space_info(fs_info, | |
5652 | BTRFS_BLOCK_GROUP_METADATA); | |
5653 | } | |
5654 | ||
2ff7e61e | 5655 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info, |
66d8f3dd | 5656 | unsigned short type) |
f0486c68 YZ |
5657 | { |
5658 | struct btrfs_block_rsv *block_rsv; | |
9ed74f2d | 5659 | |
f0486c68 YZ |
5660 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
5661 | if (!block_rsv) | |
5662 | return NULL; | |
9ed74f2d | 5663 | |
69fe2d75 | 5664 | btrfs_init_metadata_block_rsv(fs_info, block_rsv, type); |
f0486c68 YZ |
5665 | return block_rsv; |
5666 | } | |
9ed74f2d | 5667 | |
2ff7e61e | 5668 | void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5669 | struct btrfs_block_rsv *rsv) |
5670 | { | |
2aaa6655 JB |
5671 | if (!rsv) |
5672 | return; | |
2ff7e61e | 5673 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
dabdb640 | 5674 | kfree(rsv); |
9ed74f2d JB |
5675 | } |
5676 | ||
cdfb080e CM |
5677 | void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv) |
5678 | { | |
5679 | kfree(rsv); | |
5680 | } | |
5681 | ||
08e007d2 MX |
5682 | int btrfs_block_rsv_add(struct btrfs_root *root, |
5683 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5684 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5685 | { |
f0486c68 | 5686 | int ret; |
9ed74f2d | 5687 | |
f0486c68 YZ |
5688 | if (num_bytes == 0) |
5689 | return 0; | |
8bb8ab2e | 5690 | |
61b520a9 | 5691 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
5692 | if (!ret) { |
5693 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
5694 | return 0; | |
5695 | } | |
9ed74f2d | 5696 | |
f0486c68 | 5697 | return ret; |
f0486c68 | 5698 | } |
9ed74f2d | 5699 | |
2ff7e61e | 5700 | int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
5701 | { |
5702 | u64 num_bytes = 0; | |
f0486c68 | 5703 | int ret = -ENOSPC; |
9ed74f2d | 5704 | |
f0486c68 YZ |
5705 | if (!block_rsv) |
5706 | return 0; | |
9ed74f2d | 5707 | |
f0486c68 | 5708 | spin_lock(&block_rsv->lock); |
36ba022a JB |
5709 | num_bytes = div_factor(block_rsv->size, min_factor); |
5710 | if (block_rsv->reserved >= num_bytes) | |
5711 | ret = 0; | |
5712 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 5713 | |
36ba022a JB |
5714 | return ret; |
5715 | } | |
5716 | ||
08e007d2 MX |
5717 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
5718 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5719 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
5720 | { |
5721 | u64 num_bytes = 0; | |
5722 | int ret = -ENOSPC; | |
5723 | ||
5724 | if (!block_rsv) | |
5725 | return 0; | |
5726 | ||
5727 | spin_lock(&block_rsv->lock); | |
5728 | num_bytes = min_reserved; | |
13553e52 | 5729 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 5730 | ret = 0; |
13553e52 | 5731 | else |
f0486c68 | 5732 | num_bytes -= block_rsv->reserved; |
f0486c68 | 5733 | spin_unlock(&block_rsv->lock); |
13553e52 | 5734 | |
f0486c68 YZ |
5735 | if (!ret) |
5736 | return 0; | |
5737 | ||
aa38a711 | 5738 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
5739 | if (!ret) { |
5740 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 5741 | return 0; |
6a63209f | 5742 | } |
9ed74f2d | 5743 | |
13553e52 | 5744 | return ret; |
f0486c68 YZ |
5745 | } |
5746 | ||
69fe2d75 JB |
5747 | /** |
5748 | * btrfs_inode_rsv_refill - refill the inode block rsv. | |
5749 | * @inode - the inode we are refilling. | |
5750 | * @flush - the flusing restriction. | |
5751 | * | |
5752 | * Essentially the same as btrfs_block_rsv_refill, except it uses the | |
5753 | * block_rsv->size as the minimum size. We'll either refill the missing amount | |
5754 | * or return if we already have enough space. This will also handle the resreve | |
5755 | * tracepoint for the reserved amount. | |
5756 | */ | |
3f2dd7a0 QW |
5757 | static int btrfs_inode_rsv_refill(struct btrfs_inode *inode, |
5758 | enum btrfs_reserve_flush_enum flush) | |
69fe2d75 JB |
5759 | { |
5760 | struct btrfs_root *root = inode->root; | |
5761 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; | |
5762 | u64 num_bytes = 0; | |
5763 | int ret = -ENOSPC; | |
5764 | ||
5765 | spin_lock(&block_rsv->lock); | |
5766 | if (block_rsv->reserved < block_rsv->size) | |
5767 | num_bytes = block_rsv->size - block_rsv->reserved; | |
5768 | spin_unlock(&block_rsv->lock); | |
5769 | ||
5770 | if (num_bytes == 0) | |
5771 | return 0; | |
5772 | ||
5773 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); | |
5774 | if (!ret) { | |
5775 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
5776 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
5777 | btrfs_ino(inode), num_bytes, 1); | |
5778 | } | |
5779 | return ret; | |
5780 | } | |
5781 | ||
5782 | /** | |
5783 | * btrfs_inode_rsv_release - release any excessive reservation. | |
5784 | * @inode - the inode we need to release from. | |
5785 | * | |
5786 | * This is the same as btrfs_block_rsv_release, except that it handles the | |
5787 | * tracepoint for the reservation. | |
5788 | */ | |
d2560ebd | 5789 | static void btrfs_inode_rsv_release(struct btrfs_inode *inode) |
69fe2d75 JB |
5790 | { |
5791 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
5792 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5793 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; | |
5794 | u64 released = 0; | |
5795 | ||
5796 | /* | |
5797 | * Since we statically set the block_rsv->size we just want to say we | |
5798 | * are releasing 0 bytes, and then we'll just get the reservation over | |
5799 | * the size free'd. | |
5800 | */ | |
5801 | released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0); | |
5802 | if (released > 0) | |
5803 | trace_btrfs_space_reservation(fs_info, "delalloc", | |
5804 | btrfs_ino(inode), released, 0); | |
5805 | } | |
5806 | ||
2ff7e61e | 5807 | void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5808 | struct btrfs_block_rsv *block_rsv, |
5809 | u64 num_bytes) | |
5810 | { | |
0b246afa JM |
5811 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
5812 | ||
17504584 | 5813 | if (global_rsv == block_rsv || |
f0486c68 YZ |
5814 | block_rsv->space_info != global_rsv->space_info) |
5815 | global_rsv = NULL; | |
0b246afa | 5816 | block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes); |
6a63209f JB |
5817 | } |
5818 | ||
8929ecfa YZ |
5819 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5820 | { | |
5821 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5822 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5823 | u64 num_bytes; | |
6a63209f | 5824 | |
ae2e4728 JB |
5825 | /* |
5826 | * The global block rsv is based on the size of the extent tree, the | |
5827 | * checksum tree and the root tree. If the fs is empty we want to set | |
5828 | * it to a minimal amount for safety. | |
5829 | */ | |
5830 | num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) + | |
5831 | btrfs_root_used(&fs_info->csum_root->root_item) + | |
5832 | btrfs_root_used(&fs_info->tree_root->root_item); | |
5833 | num_bytes = max_t(u64, num_bytes, SZ_16M); | |
33b4d47f | 5834 | |
8929ecfa | 5835 | spin_lock(&sinfo->lock); |
1f699d38 | 5836 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5837 | |
ee22184b | 5838 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); |
4e06bdd6 | 5839 | |
fb4b10e5 | 5840 | if (block_rsv->reserved < block_rsv->size) { |
4136135b | 5841 | num_bytes = btrfs_space_info_used(sinfo, true); |
fb4b10e5 JB |
5842 | if (sinfo->total_bytes > num_bytes) { |
5843 | num_bytes = sinfo->total_bytes - num_bytes; | |
5844 | num_bytes = min(num_bytes, | |
5845 | block_rsv->size - block_rsv->reserved); | |
5846 | block_rsv->reserved += num_bytes; | |
5847 | sinfo->bytes_may_use += num_bytes; | |
5848 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5849 | sinfo->flags, num_bytes, | |
5850 | 1); | |
5851 | } | |
5852 | } else if (block_rsv->reserved > block_rsv->size) { | |
8929ecfa | 5853 | num_bytes = block_rsv->reserved - block_rsv->size; |
fb25e914 | 5854 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 5855 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5856 | sinfo->flags, num_bytes, 0); |
8929ecfa | 5857 | block_rsv->reserved = block_rsv->size; |
8929ecfa | 5858 | } |
182608c8 | 5859 | |
fb4b10e5 JB |
5860 | if (block_rsv->reserved == block_rsv->size) |
5861 | block_rsv->full = 1; | |
5862 | else | |
5863 | block_rsv->full = 0; | |
5864 | ||
8929ecfa | 5865 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5866 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5867 | } |
5868 | ||
f0486c68 | 5869 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5870 | { |
f0486c68 | 5871 | struct btrfs_space_info *space_info; |
6a63209f | 5872 | |
f0486c68 YZ |
5873 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
5874 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 5875 | |
f0486c68 | 5876 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5877 | fs_info->global_block_rsv.space_info = space_info; |
f0486c68 YZ |
5878 | fs_info->trans_block_rsv.space_info = space_info; |
5879 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5880 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 5881 | |
8929ecfa YZ |
5882 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
5883 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
5884 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
5885 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5886 | if (fs_info->quota_root) |
5887 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5888 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5889 | |
8929ecfa | 5890 | update_global_block_rsv(fs_info); |
6a63209f JB |
5891 | } |
5892 | ||
8929ecfa | 5893 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5894 | { |
8c2a3ca2 JB |
5895 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
5896 | (u64)-1); | |
8929ecfa YZ |
5897 | WARN_ON(fs_info->trans_block_rsv.size > 0); |
5898 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5899 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5900 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5901 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5902 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
5903 | } |
5904 | ||
6a63209f | 5905 | |
4fbcdf66 FM |
5906 | /* |
5907 | * To be called after all the new block groups attached to the transaction | |
5908 | * handle have been created (btrfs_create_pending_block_groups()). | |
5909 | */ | |
5910 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5911 | { | |
64b63580 | 5912 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4fbcdf66 FM |
5913 | |
5914 | if (!trans->chunk_bytes_reserved) | |
5915 | return; | |
5916 | ||
5917 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5918 | ||
5919 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
5920 | trans->chunk_bytes_reserved); | |
5921 | trans->chunk_bytes_reserved = 0; | |
5922 | } | |
5923 | ||
79787eaa | 5924 | /* Can only return 0 or -ENOSPC */ |
d68fc57b | 5925 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
8ed7a2a0 | 5926 | struct btrfs_inode *inode) |
d68fc57b | 5927 | { |
8ed7a2a0 NB |
5928 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
5929 | struct btrfs_root *root = inode->root; | |
40acc3ee JB |
5930 | /* |
5931 | * We always use trans->block_rsv here as we will have reserved space | |
5932 | * for our orphan when starting the transaction, using get_block_rsv() | |
5933 | * here will sometimes make us choose the wrong block rsv as we could be | |
5934 | * doing a reloc inode for a non refcounted root. | |
5935 | */ | |
5936 | struct btrfs_block_rsv *src_rsv = trans->block_rsv; | |
d68fc57b YZ |
5937 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; |
5938 | ||
5939 | /* | |
fcb80c2a JB |
5940 | * We need to hold space in order to delete our orphan item once we've |
5941 | * added it, so this takes the reservation so we can release it later | |
5942 | * when we are truly done with the orphan item. | |
d68fc57b | 5943 | */ |
0b246afa JM |
5944 | u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
5945 | ||
9678c543 | 5946 | trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode), |
8ed7a2a0 | 5947 | num_bytes, 1); |
25d609f8 | 5948 | return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1); |
6a63209f JB |
5949 | } |
5950 | ||
703b391a | 5951 | void btrfs_orphan_release_metadata(struct btrfs_inode *inode) |
97e728d4 | 5952 | { |
703b391a NB |
5953 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
5954 | struct btrfs_root *root = inode->root; | |
0b246afa JM |
5955 | u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
5956 | ||
703b391a NB |
5957 | trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode), |
5958 | num_bytes, 0); | |
2ff7e61e | 5959 | btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes); |
d68fc57b | 5960 | } |
97e728d4 | 5961 | |
d5c12070 MX |
5962 | /* |
5963 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5964 | * root: the root of the parent directory | |
5965 | * rsv: block reservation | |
5966 | * items: the number of items that we need do reservation | |
5967 | * qgroup_reserved: used to return the reserved size in qgroup | |
5968 | * | |
5969 | * This function is used to reserve the space for snapshot/subvolume | |
5970 | * creation and deletion. Those operations are different with the | |
5971 | * common file/directory operations, they change two fs/file trees | |
5972 | * and root tree, the number of items that the qgroup reserves is | |
5973 | * different with the free space reservation. So we can not use | |
01327610 | 5974 | * the space reservation mechanism in start_transaction(). |
d5c12070 MX |
5975 | */ |
5976 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
5977 | struct btrfs_block_rsv *rsv, | |
5978 | int items, | |
ee3441b4 JM |
5979 | u64 *qgroup_reserved, |
5980 | bool use_global_rsv) | |
a22285a6 | 5981 | { |
d5c12070 MX |
5982 | u64 num_bytes; |
5983 | int ret; | |
0b246afa JM |
5984 | struct btrfs_fs_info *fs_info = root->fs_info; |
5985 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d5c12070 | 5986 | |
0b246afa | 5987 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
d5c12070 | 5988 | /* One for parent inode, two for dir entries */ |
0b246afa | 5989 | num_bytes = 3 * fs_info->nodesize; |
003d7c59 | 5990 | ret = btrfs_qgroup_reserve_meta(root, num_bytes, true); |
d5c12070 MX |
5991 | if (ret) |
5992 | return ret; | |
5993 | } else { | |
5994 | num_bytes = 0; | |
5995 | } | |
5996 | ||
5997 | *qgroup_reserved = num_bytes; | |
5998 | ||
0b246afa JM |
5999 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, items); |
6000 | rsv->space_info = __find_space_info(fs_info, | |
d5c12070 MX |
6001 | BTRFS_BLOCK_GROUP_METADATA); |
6002 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
6003 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
6004 | |
6005 | if (ret == -ENOSPC && use_global_rsv) | |
25d609f8 | 6006 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1); |
ee3441b4 | 6007 | |
7174109c QW |
6008 | if (ret && *qgroup_reserved) |
6009 | btrfs_qgroup_free_meta(root, *qgroup_reserved); | |
d5c12070 MX |
6010 | |
6011 | return ret; | |
6012 | } | |
6013 | ||
2ff7e61e | 6014 | void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info, |
7775c818 | 6015 | struct btrfs_block_rsv *rsv) |
d5c12070 | 6016 | { |
2ff7e61e | 6017 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
97e728d4 JB |
6018 | } |
6019 | ||
69fe2d75 JB |
6020 | static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info, |
6021 | struct btrfs_inode *inode) | |
9e0baf60 | 6022 | { |
69fe2d75 JB |
6023 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
6024 | u64 reserve_size = 0; | |
6025 | u64 csum_leaves; | |
6026 | unsigned outstanding_extents; | |
9e0baf60 | 6027 | |
69fe2d75 JB |
6028 | lockdep_assert_held(&inode->lock); |
6029 | outstanding_extents = inode->outstanding_extents; | |
6030 | if (outstanding_extents) | |
6031 | reserve_size = btrfs_calc_trans_metadata_size(fs_info, | |
6032 | outstanding_extents + 1); | |
6033 | csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, | |
6034 | inode->csum_bytes); | |
6035 | reserve_size += btrfs_calc_trans_metadata_size(fs_info, | |
6036 | csum_leaves); | |
9e0baf60 | 6037 | |
69fe2d75 JB |
6038 | spin_lock(&block_rsv->lock); |
6039 | block_rsv->size = reserve_size; | |
6040 | spin_unlock(&block_rsv->lock); | |
0ca1f7ce | 6041 | } |
c146afad | 6042 | |
9f3db423 | 6043 | int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 6044 | { |
9f3db423 NB |
6045 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
6046 | struct btrfs_root *root = inode->root; | |
69fe2d75 | 6047 | unsigned nr_extents; |
08e007d2 | 6048 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 6049 | int ret = 0; |
c64c2bd8 | 6050 | bool delalloc_lock = true; |
6324fbf3 | 6051 | |
c64c2bd8 JB |
6052 | /* If we are a free space inode we need to not flush since we will be in |
6053 | * the middle of a transaction commit. We also don't need the delalloc | |
6054 | * mutex since we won't race with anybody. We need this mostly to make | |
6055 | * lockdep shut its filthy mouth. | |
bac357dc JB |
6056 | * |
6057 | * If we have a transaction open (can happen if we call truncate_block | |
6058 | * from truncate), then we need FLUSH_LIMIT so we don't deadlock. | |
c64c2bd8 JB |
6059 | */ |
6060 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 6061 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 | 6062 | delalloc_lock = false; |
da07d4ab NB |
6063 | } else { |
6064 | if (current->journal_info) | |
6065 | flush = BTRFS_RESERVE_FLUSH_LIMIT; | |
c09544e0 | 6066 | |
da07d4ab NB |
6067 | if (btrfs_transaction_in_commit(fs_info)) |
6068 | schedule_timeout(1); | |
6069 | } | |
ec44a35c | 6070 | |
c64c2bd8 | 6071 | if (delalloc_lock) |
9f3db423 | 6072 | mutex_lock(&inode->delalloc_mutex); |
c64c2bd8 | 6073 | |
0b246afa | 6074 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
69fe2d75 JB |
6075 | |
6076 | /* Add our new extents and calculate the new rsv size. */ | |
9f3db423 | 6077 | spin_lock(&inode->lock); |
69fe2d75 | 6078 | nr_extents = count_max_extents(num_bytes); |
8b62f87b | 6079 | btrfs_mod_outstanding_extents(inode, nr_extents); |
69fe2d75 JB |
6080 | inode->csum_bytes += num_bytes; |
6081 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 6082 | spin_unlock(&inode->lock); |
57a45ced | 6083 | |
0b246afa | 6084 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
7174109c | 6085 | ret = btrfs_qgroup_reserve_meta(root, |
003d7c59 | 6086 | nr_extents * fs_info->nodesize, true); |
88e081bf WS |
6087 | if (ret) |
6088 | goto out_fail; | |
6089 | } | |
c5567237 | 6090 | |
69fe2d75 | 6091 | ret = btrfs_inode_rsv_refill(inode, flush); |
88e081bf | 6092 | if (unlikely(ret)) { |
da17066c | 6093 | btrfs_qgroup_free_meta(root, |
0b246afa | 6094 | nr_extents * fs_info->nodesize); |
88e081bf | 6095 | goto out_fail; |
9e0baf60 | 6096 | } |
25179201 | 6097 | |
c64c2bd8 | 6098 | if (delalloc_lock) |
9f3db423 | 6099 | mutex_unlock(&inode->delalloc_mutex); |
0ca1f7ce | 6100 | return 0; |
88e081bf WS |
6101 | |
6102 | out_fail: | |
9f3db423 | 6103 | spin_lock(&inode->lock); |
8b62f87b JB |
6104 | nr_extents = count_max_extents(num_bytes); |
6105 | btrfs_mod_outstanding_extents(inode, -nr_extents); | |
69fe2d75 JB |
6106 | inode->csum_bytes -= num_bytes; |
6107 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 6108 | spin_unlock(&inode->lock); |
88e081bf | 6109 | |
69fe2d75 | 6110 | btrfs_inode_rsv_release(inode); |
88e081bf | 6111 | if (delalloc_lock) |
9f3db423 | 6112 | mutex_unlock(&inode->delalloc_mutex); |
88e081bf | 6113 | return ret; |
0ca1f7ce YZ |
6114 | } |
6115 | ||
7709cde3 JB |
6116 | /** |
6117 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
8b62f87b JB |
6118 | * @inode: the inode to release the reservation for. |
6119 | * @num_bytes: the number of bytes we are releasing. | |
7709cde3 JB |
6120 | * |
6121 | * This will release the metadata reservation for an inode. This can be called | |
6122 | * once we complete IO for a given set of bytes to release their metadata | |
8b62f87b | 6123 | * reservations, or on error for the same reason. |
7709cde3 | 6124 | */ |
691fa059 | 6125 | void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 6126 | { |
691fa059 | 6127 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
0ca1f7ce | 6128 | |
0b246afa | 6129 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
691fa059 | 6130 | spin_lock(&inode->lock); |
69fe2d75 JB |
6131 | inode->csum_bytes -= num_bytes; |
6132 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
691fa059 | 6133 | spin_unlock(&inode->lock); |
0ca1f7ce | 6134 | |
0b246afa | 6135 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
6136 | return; |
6137 | ||
69fe2d75 | 6138 | btrfs_inode_rsv_release(inode); |
0ca1f7ce YZ |
6139 | } |
6140 | ||
8b62f87b JB |
6141 | /** |
6142 | * btrfs_delalloc_release_extents - release our outstanding_extents | |
6143 | * @inode: the inode to balance the reservation for. | |
6144 | * @num_bytes: the number of bytes we originally reserved with | |
6145 | * | |
6146 | * When we reserve space we increase outstanding_extents for the extents we may | |
6147 | * add. Once we've set the range as delalloc or created our ordered extents we | |
6148 | * have outstanding_extents to track the real usage, so we use this to free our | |
6149 | * temporarily tracked outstanding_extents. This _must_ be used in conjunction | |
6150 | * with btrfs_delalloc_reserve_metadata. | |
6151 | */ | |
6152 | void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes) | |
6153 | { | |
6154 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); | |
6155 | unsigned num_extents; | |
8b62f87b JB |
6156 | |
6157 | spin_lock(&inode->lock); | |
6158 | num_extents = count_max_extents(num_bytes); | |
6159 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
69fe2d75 | 6160 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); |
8b62f87b JB |
6161 | spin_unlock(&inode->lock); |
6162 | ||
8b62f87b JB |
6163 | if (btrfs_is_testing(fs_info)) |
6164 | return; | |
6165 | ||
69fe2d75 | 6166 | btrfs_inode_rsv_release(inode); |
8b62f87b JB |
6167 | } |
6168 | ||
1ada3a62 | 6169 | /** |
7cf5b976 | 6170 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
6171 | * delalloc |
6172 | * @inode: inode we're writing to | |
6173 | * @start: start range we are writing to | |
6174 | * @len: how long the range we are writing to | |
364ecf36 QW |
6175 | * @reserved: mandatory parameter, record actually reserved qgroup ranges of |
6176 | * current reservation. | |
1ada3a62 | 6177 | * |
1ada3a62 QW |
6178 | * This will do the following things |
6179 | * | |
6180 | * o reserve space in data space info for num bytes | |
6181 | * and reserve precious corresponding qgroup space | |
6182 | * (Done in check_data_free_space) | |
6183 | * | |
6184 | * o reserve space for metadata space, based on the number of outstanding | |
6185 | * extents and how much csums will be needed | |
6186 | * also reserve metadata space in a per root over-reserve method. | |
6187 | * o add to the inodes->delalloc_bytes | |
6188 | * o add it to the fs_info's delalloc inodes list. | |
6189 | * (Above 3 all done in delalloc_reserve_metadata) | |
6190 | * | |
6191 | * Return 0 for success | |
6192 | * Return <0 for error(-ENOSPC or -EQUOT) | |
6193 | */ | |
364ecf36 QW |
6194 | int btrfs_delalloc_reserve_space(struct inode *inode, |
6195 | struct extent_changeset **reserved, u64 start, u64 len) | |
1ada3a62 QW |
6196 | { |
6197 | int ret; | |
6198 | ||
364ecf36 | 6199 | ret = btrfs_check_data_free_space(inode, reserved, start, len); |
1ada3a62 QW |
6200 | if (ret < 0) |
6201 | return ret; | |
9f3db423 | 6202 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len); |
1ada3a62 | 6203 | if (ret < 0) |
bc42bda2 | 6204 | btrfs_free_reserved_data_space(inode, *reserved, start, len); |
1ada3a62 QW |
6205 | return ret; |
6206 | } | |
6207 | ||
7709cde3 | 6208 | /** |
7cf5b976 | 6209 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
6210 | * @inode: inode we're releasing space for |
6211 | * @start: start position of the space already reserved | |
6212 | * @len: the len of the space already reserved | |
8b62f87b | 6213 | * @release_bytes: the len of the space we consumed or didn't use |
1ada3a62 QW |
6214 | * |
6215 | * This function will release the metadata space that was not used and will | |
6216 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
6217 | * list if there are no delalloc bytes left. | |
6218 | * Also it will handle the qgroup reserved space. | |
6219 | */ | |
bc42bda2 | 6220 | void btrfs_delalloc_release_space(struct inode *inode, |
8b62f87b JB |
6221 | struct extent_changeset *reserved, |
6222 | u64 start, u64 len) | |
1ada3a62 | 6223 | { |
691fa059 | 6224 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len); |
bc42bda2 | 6225 | btrfs_free_reserved_data_space(inode, reserved, start, len); |
6324fbf3 CM |
6226 | } |
6227 | ||
ce93ec54 | 6228 | static int update_block_group(struct btrfs_trans_handle *trans, |
6202df69 | 6229 | struct btrfs_fs_info *info, u64 bytenr, |
ce93ec54 | 6230 | u64 num_bytes, int alloc) |
9078a3e1 | 6231 | { |
0af3d00b | 6232 | struct btrfs_block_group_cache *cache = NULL; |
db94535d | 6233 | u64 total = num_bytes; |
9078a3e1 | 6234 | u64 old_val; |
db94535d | 6235 | u64 byte_in_group; |
0af3d00b | 6236 | int factor; |
3e1ad54f | 6237 | |
5d4f98a2 | 6238 | /* block accounting for super block */ |
eb73c1b7 | 6239 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 6240 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
6241 | if (alloc) |
6242 | old_val += num_bytes; | |
6243 | else | |
6244 | old_val -= num_bytes; | |
6c41761f | 6245 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 6246 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 6247 | |
d397712b | 6248 | while (total) { |
db94535d | 6249 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 6250 | if (!cache) |
79787eaa | 6251 | return -ENOENT; |
b742bb82 YZ |
6252 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
6253 | BTRFS_BLOCK_GROUP_RAID1 | | |
6254 | BTRFS_BLOCK_GROUP_RAID10)) | |
6255 | factor = 2; | |
6256 | else | |
6257 | factor = 1; | |
9d66e233 JB |
6258 | /* |
6259 | * If this block group has free space cache written out, we | |
6260 | * need to make sure to load it if we are removing space. This | |
6261 | * is because we need the unpinning stage to actually add the | |
6262 | * space back to the block group, otherwise we will leak space. | |
6263 | */ | |
6264 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 6265 | cache_block_group(cache, 1); |
0af3d00b | 6266 | |
db94535d CM |
6267 | byte_in_group = bytenr - cache->key.objectid; |
6268 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 6269 | |
25179201 | 6270 | spin_lock(&cache->space_info->lock); |
c286ac48 | 6271 | spin_lock(&cache->lock); |
0af3d00b | 6272 | |
6202df69 | 6273 | if (btrfs_test_opt(info, SPACE_CACHE) && |
0af3d00b JB |
6274 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
6275 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
6276 | ||
9078a3e1 | 6277 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 6278 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 6279 | if (alloc) { |
db94535d | 6280 | old_val += num_bytes; |
11833d66 YZ |
6281 | btrfs_set_block_group_used(&cache->item, old_val); |
6282 | cache->reserved -= num_bytes; | |
11833d66 | 6283 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
6284 | cache->space_info->bytes_used += num_bytes; |
6285 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 6286 | spin_unlock(&cache->lock); |
25179201 | 6287 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 6288 | } else { |
db94535d | 6289 | old_val -= num_bytes; |
ae0ab003 FM |
6290 | btrfs_set_block_group_used(&cache->item, old_val); |
6291 | cache->pinned += num_bytes; | |
6292 | cache->space_info->bytes_pinned += num_bytes; | |
6293 | cache->space_info->bytes_used -= num_bytes; | |
6294 | cache->space_info->disk_used -= num_bytes * factor; | |
6295 | spin_unlock(&cache->lock); | |
6296 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 6297 | |
0b246afa | 6298 | trace_btrfs_space_reservation(info, "pinned", |
c51e7bb1 JB |
6299 | cache->space_info->flags, |
6300 | num_bytes, 1); | |
d7eae340 OS |
6301 | percpu_counter_add(&cache->space_info->total_bytes_pinned, |
6302 | num_bytes); | |
ae0ab003 FM |
6303 | set_extent_dirty(info->pinned_extents, |
6304 | bytenr, bytenr + num_bytes - 1, | |
6305 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 6306 | } |
1bbc621e CM |
6307 | |
6308 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
6309 | if (list_empty(&cache->dirty_list)) { | |
6310 | list_add_tail(&cache->dirty_list, | |
6311 | &trans->transaction->dirty_bgs); | |
6312 | trans->transaction->num_dirty_bgs++; | |
6313 | btrfs_get_block_group(cache); | |
6314 | } | |
6315 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
6316 | ||
036a9348 FM |
6317 | /* |
6318 | * No longer have used bytes in this block group, queue it for | |
6319 | * deletion. We do this after adding the block group to the | |
6320 | * dirty list to avoid races between cleaner kthread and space | |
6321 | * cache writeout. | |
6322 | */ | |
6323 | if (!alloc && old_val == 0) { | |
6324 | spin_lock(&info->unused_bgs_lock); | |
6325 | if (list_empty(&cache->bg_list)) { | |
6326 | btrfs_get_block_group(cache); | |
6327 | list_add_tail(&cache->bg_list, | |
6328 | &info->unused_bgs); | |
6329 | } | |
6330 | spin_unlock(&info->unused_bgs_lock); | |
6331 | } | |
6332 | ||
fa9c0d79 | 6333 | btrfs_put_block_group(cache); |
db94535d CM |
6334 | total -= num_bytes; |
6335 | bytenr += num_bytes; | |
9078a3e1 CM |
6336 | } |
6337 | return 0; | |
6338 | } | |
6324fbf3 | 6339 | |
2ff7e61e | 6340 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 6341 | { |
0f9dd46c | 6342 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 6343 | u64 bytenr; |
0f9dd46c | 6344 | |
0b246afa JM |
6345 | spin_lock(&fs_info->block_group_cache_lock); |
6346 | bytenr = fs_info->first_logical_byte; | |
6347 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
6348 | |
6349 | if (bytenr < (u64)-1) | |
6350 | return bytenr; | |
6351 | ||
0b246afa | 6352 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 6353 | if (!cache) |
a061fc8d | 6354 | return 0; |
0f9dd46c | 6355 | |
d2fb3437 | 6356 | bytenr = cache->key.objectid; |
fa9c0d79 | 6357 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
6358 | |
6359 | return bytenr; | |
a061fc8d CM |
6360 | } |
6361 | ||
2ff7e61e | 6362 | static int pin_down_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6363 | struct btrfs_block_group_cache *cache, |
6364 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 6365 | { |
11833d66 YZ |
6366 | spin_lock(&cache->space_info->lock); |
6367 | spin_lock(&cache->lock); | |
6368 | cache->pinned += num_bytes; | |
6369 | cache->space_info->bytes_pinned += num_bytes; | |
6370 | if (reserved) { | |
6371 | cache->reserved -= num_bytes; | |
6372 | cache->space_info->bytes_reserved -= num_bytes; | |
6373 | } | |
6374 | spin_unlock(&cache->lock); | |
6375 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 6376 | |
0b246afa | 6377 | trace_btrfs_space_reservation(fs_info, "pinned", |
c51e7bb1 | 6378 | cache->space_info->flags, num_bytes, 1); |
4da8b76d | 6379 | percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes); |
0b246afa | 6380 | set_extent_dirty(fs_info->pinned_extents, bytenr, |
f0486c68 YZ |
6381 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
6382 | return 0; | |
6383 | } | |
68b38550 | 6384 | |
f0486c68 YZ |
6385 | /* |
6386 | * this function must be called within transaction | |
6387 | */ | |
2ff7e61e | 6388 | int btrfs_pin_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6389 | u64 bytenr, u64 num_bytes, int reserved) |
6390 | { | |
6391 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6392 | |
0b246afa | 6393 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
79787eaa | 6394 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 6395 | |
2ff7e61e | 6396 | pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
6397 | |
6398 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6399 | return 0; |
6400 | } | |
6401 | ||
f0486c68 | 6402 | /* |
e688b725 CM |
6403 | * this function must be called within transaction |
6404 | */ | |
2ff7e61e | 6405 | int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info, |
e688b725 CM |
6406 | u64 bytenr, u64 num_bytes) |
6407 | { | |
6408 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6409 | int ret; |
e688b725 | 6410 | |
0b246afa | 6411 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
b50c6e25 JB |
6412 | if (!cache) |
6413 | return -EINVAL; | |
e688b725 CM |
6414 | |
6415 | /* | |
6416 | * pull in the free space cache (if any) so that our pin | |
6417 | * removes the free space from the cache. We have load_only set | |
6418 | * to one because the slow code to read in the free extents does check | |
6419 | * the pinned extents. | |
6420 | */ | |
f6373bf3 | 6421 | cache_block_group(cache, 1); |
e688b725 | 6422 | |
2ff7e61e | 6423 | pin_down_extent(fs_info, cache, bytenr, num_bytes, 0); |
e688b725 CM |
6424 | |
6425 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6426 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6427 | btrfs_put_block_group(cache); |
b50c6e25 | 6428 | return ret; |
e688b725 CM |
6429 | } |
6430 | ||
2ff7e61e JM |
6431 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
6432 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
6433 | { |
6434 | int ret; | |
6435 | struct btrfs_block_group_cache *block_group; | |
6436 | struct btrfs_caching_control *caching_ctl; | |
6437 | ||
0b246afa | 6438 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
6439 | if (!block_group) |
6440 | return -EINVAL; | |
6441 | ||
6442 | cache_block_group(block_group, 0); | |
6443 | caching_ctl = get_caching_control(block_group); | |
6444 | ||
6445 | if (!caching_ctl) { | |
6446 | /* Logic error */ | |
6447 | BUG_ON(!block_group_cache_done(block_group)); | |
6448 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6449 | } else { | |
6450 | mutex_lock(&caching_ctl->mutex); | |
6451 | ||
6452 | if (start >= caching_ctl->progress) { | |
2ff7e61e | 6453 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6454 | } else if (start + num_bytes <= caching_ctl->progress) { |
6455 | ret = btrfs_remove_free_space(block_group, | |
6456 | start, num_bytes); | |
6457 | } else { | |
6458 | num_bytes = caching_ctl->progress - start; | |
6459 | ret = btrfs_remove_free_space(block_group, | |
6460 | start, num_bytes); | |
6461 | if (ret) | |
6462 | goto out_lock; | |
6463 | ||
6464 | num_bytes = (start + num_bytes) - | |
6465 | caching_ctl->progress; | |
6466 | start = caching_ctl->progress; | |
2ff7e61e | 6467 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6468 | } |
6469 | out_lock: | |
6470 | mutex_unlock(&caching_ctl->mutex); | |
6471 | put_caching_control(caching_ctl); | |
6472 | } | |
6473 | btrfs_put_block_group(block_group); | |
6474 | return ret; | |
6475 | } | |
6476 | ||
2ff7e61e | 6477 | int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info, |
8c2a1a30 JB |
6478 | struct extent_buffer *eb) |
6479 | { | |
6480 | struct btrfs_file_extent_item *item; | |
6481 | struct btrfs_key key; | |
6482 | int found_type; | |
6483 | int i; | |
6484 | ||
2ff7e61e | 6485 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
6486 | return 0; |
6487 | ||
6488 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6489 | btrfs_item_key_to_cpu(eb, &key, i); | |
6490 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6491 | continue; | |
6492 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6493 | found_type = btrfs_file_extent_type(eb, item); | |
6494 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6495 | continue; | |
6496 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6497 | continue; | |
6498 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6499 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
2ff7e61e | 6500 | __exclude_logged_extent(fs_info, key.objectid, key.offset); |
8c2a1a30 JB |
6501 | } |
6502 | ||
6503 | return 0; | |
6504 | } | |
6505 | ||
9cfa3e34 FM |
6506 | static void |
6507 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6508 | { | |
6509 | atomic_inc(&bg->reservations); | |
6510 | } | |
6511 | ||
6512 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
6513 | const u64 start) | |
6514 | { | |
6515 | struct btrfs_block_group_cache *bg; | |
6516 | ||
6517 | bg = btrfs_lookup_block_group(fs_info, start); | |
6518 | ASSERT(bg); | |
6519 | if (atomic_dec_and_test(&bg->reservations)) | |
6520 | wake_up_atomic_t(&bg->reservations); | |
6521 | btrfs_put_block_group(bg); | |
6522 | } | |
6523 | ||
9cfa3e34 FM |
6524 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) |
6525 | { | |
6526 | struct btrfs_space_info *space_info = bg->space_info; | |
6527 | ||
6528 | ASSERT(bg->ro); | |
6529 | ||
6530 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
6531 | return; | |
6532 | ||
6533 | /* | |
6534 | * Our block group is read only but before we set it to read only, | |
6535 | * some task might have had allocated an extent from it already, but it | |
6536 | * has not yet created a respective ordered extent (and added it to a | |
6537 | * root's list of ordered extents). | |
6538 | * Therefore wait for any task currently allocating extents, since the | |
6539 | * block group's reservations counter is incremented while a read lock | |
6540 | * on the groups' semaphore is held and decremented after releasing | |
6541 | * the read access on that semaphore and creating the ordered extent. | |
6542 | */ | |
6543 | down_write(&space_info->groups_sem); | |
6544 | up_write(&space_info->groups_sem); | |
6545 | ||
5e4def20 | 6546 | wait_on_atomic_t(&bg->reservations, atomic_t_wait, |
9cfa3e34 FM |
6547 | TASK_UNINTERRUPTIBLE); |
6548 | } | |
6549 | ||
fb25e914 | 6550 | /** |
4824f1f4 | 6551 | * btrfs_add_reserved_bytes - update the block_group and space info counters |
fb25e914 | 6552 | * @cache: The cache we are manipulating |
18513091 WX |
6553 | * @ram_bytes: The number of bytes of file content, and will be same to |
6554 | * @num_bytes except for the compress path. | |
fb25e914 | 6555 | * @num_bytes: The number of bytes in question |
e570fd27 | 6556 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 | 6557 | * |
745699ef XW |
6558 | * This is called by the allocator when it reserves space. If this is a |
6559 | * reservation and the block group has become read only we cannot make the | |
6560 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
f0486c68 | 6561 | */ |
4824f1f4 | 6562 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 6563 | u64 ram_bytes, u64 num_bytes, int delalloc) |
11833d66 | 6564 | { |
fb25e914 | 6565 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6566 | int ret = 0; |
79787eaa | 6567 | |
fb25e914 JB |
6568 | spin_lock(&space_info->lock); |
6569 | spin_lock(&cache->lock); | |
4824f1f4 WX |
6570 | if (cache->ro) { |
6571 | ret = -EAGAIN; | |
fb25e914 | 6572 | } else { |
4824f1f4 WX |
6573 | cache->reserved += num_bytes; |
6574 | space_info->bytes_reserved += num_bytes; | |
e570fd27 | 6575 | |
18513091 WX |
6576 | trace_btrfs_space_reservation(cache->fs_info, |
6577 | "space_info", space_info->flags, | |
6578 | ram_bytes, 0); | |
6579 | space_info->bytes_may_use -= ram_bytes; | |
e570fd27 | 6580 | if (delalloc) |
4824f1f4 | 6581 | cache->delalloc_bytes += num_bytes; |
324ae4df | 6582 | } |
fb25e914 JB |
6583 | spin_unlock(&cache->lock); |
6584 | spin_unlock(&space_info->lock); | |
f0486c68 | 6585 | return ret; |
324ae4df | 6586 | } |
9078a3e1 | 6587 | |
4824f1f4 WX |
6588 | /** |
6589 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
6590 | * @cache: The cache we are manipulating | |
6591 | * @num_bytes: The number of bytes in question | |
6592 | * @delalloc: The blocks are allocated for the delalloc write | |
6593 | * | |
6594 | * This is called by somebody who is freeing space that was never actually used | |
6595 | * on disk. For example if you reserve some space for a new leaf in transaction | |
6596 | * A and before transaction A commits you free that leaf, you call this with | |
6597 | * reserve set to 0 in order to clear the reservation. | |
6598 | */ | |
6599 | ||
6600 | static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, | |
6601 | u64 num_bytes, int delalloc) | |
6602 | { | |
6603 | struct btrfs_space_info *space_info = cache->space_info; | |
6604 | int ret = 0; | |
6605 | ||
6606 | spin_lock(&space_info->lock); | |
6607 | spin_lock(&cache->lock); | |
6608 | if (cache->ro) | |
6609 | space_info->bytes_readonly += num_bytes; | |
6610 | cache->reserved -= num_bytes; | |
6611 | space_info->bytes_reserved -= num_bytes; | |
6612 | ||
6613 | if (delalloc) | |
6614 | cache->delalloc_bytes -= num_bytes; | |
6615 | spin_unlock(&cache->lock); | |
6616 | spin_unlock(&space_info->lock); | |
6617 | return ret; | |
6618 | } | |
8b74c03e | 6619 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 6620 | { |
11833d66 YZ |
6621 | struct btrfs_caching_control *next; |
6622 | struct btrfs_caching_control *caching_ctl; | |
6623 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6624 | |
9e351cc8 | 6625 | down_write(&fs_info->commit_root_sem); |
25179201 | 6626 | |
11833d66 YZ |
6627 | list_for_each_entry_safe(caching_ctl, next, |
6628 | &fs_info->caching_block_groups, list) { | |
6629 | cache = caching_ctl->block_group; | |
6630 | if (block_group_cache_done(cache)) { | |
6631 | cache->last_byte_to_unpin = (u64)-1; | |
6632 | list_del_init(&caching_ctl->list); | |
6633 | put_caching_control(caching_ctl); | |
e8569813 | 6634 | } else { |
11833d66 | 6635 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6636 | } |
e8569813 | 6637 | } |
11833d66 YZ |
6638 | |
6639 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6640 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6641 | else | |
6642 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6643 | ||
9e351cc8 | 6644 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6645 | |
6646 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6647 | } |
6648 | ||
c759c4e1 JB |
6649 | /* |
6650 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6651 | * what it should be based on the mount options. | |
6652 | */ | |
6653 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
6654 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
6655 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
6656 | { |
6657 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
6658 | |
6659 | *empty_cluster = 0; | |
6660 | if (btrfs_mixed_space_info(space_info)) | |
6661 | return ret; | |
6662 | ||
c759c4e1 | 6663 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 6664 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
6665 | if (btrfs_test_opt(fs_info, SSD)) |
6666 | *empty_cluster = SZ_2M; | |
6667 | else | |
ee22184b | 6668 | *empty_cluster = SZ_64K; |
583b7231 HK |
6669 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
6670 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
6671 | *empty_cluster = SZ_2M; | |
0b246afa | 6672 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
6673 | } |
6674 | ||
6675 | return ret; | |
6676 | } | |
6677 | ||
2ff7e61e JM |
6678 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
6679 | u64 start, u64 end, | |
678886bd | 6680 | const bool return_free_space) |
ccd467d6 | 6681 | { |
11833d66 | 6682 | struct btrfs_block_group_cache *cache = NULL; |
7b398f8e JB |
6683 | struct btrfs_space_info *space_info; |
6684 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6685 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6686 | u64 len; |
c759c4e1 JB |
6687 | u64 total_unpinned = 0; |
6688 | u64 empty_cluster = 0; | |
7b398f8e | 6689 | bool readonly; |
ccd467d6 | 6690 | |
11833d66 | 6691 | while (start <= end) { |
7b398f8e | 6692 | readonly = false; |
11833d66 YZ |
6693 | if (!cache || |
6694 | start >= cache->key.objectid + cache->key.offset) { | |
6695 | if (cache) | |
6696 | btrfs_put_block_group(cache); | |
c759c4e1 | 6697 | total_unpinned = 0; |
11833d66 | 6698 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6699 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 6700 | |
2ff7e61e | 6701 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
6702 | cache->space_info, |
6703 | &empty_cluster); | |
6704 | empty_cluster <<= 1; | |
11833d66 YZ |
6705 | } |
6706 | ||
6707 | len = cache->key.objectid + cache->key.offset - start; | |
6708 | len = min(len, end + 1 - start); | |
6709 | ||
6710 | if (start < cache->last_byte_to_unpin) { | |
6711 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6712 | if (return_free_space) |
6713 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6714 | } |
6715 | ||
f0486c68 | 6716 | start += len; |
c759c4e1 | 6717 | total_unpinned += len; |
7b398f8e | 6718 | space_info = cache->space_info; |
f0486c68 | 6719 | |
c759c4e1 JB |
6720 | /* |
6721 | * If this space cluster has been marked as fragmented and we've | |
6722 | * unpinned enough in this block group to potentially allow a | |
6723 | * cluster to be created inside of it go ahead and clear the | |
6724 | * fragmented check. | |
6725 | */ | |
6726 | if (cluster && cluster->fragmented && | |
6727 | total_unpinned > empty_cluster) { | |
6728 | spin_lock(&cluster->lock); | |
6729 | cluster->fragmented = 0; | |
6730 | spin_unlock(&cluster->lock); | |
6731 | } | |
6732 | ||
7b398f8e | 6733 | spin_lock(&space_info->lock); |
11833d66 YZ |
6734 | spin_lock(&cache->lock); |
6735 | cache->pinned -= len; | |
7b398f8e | 6736 | space_info->bytes_pinned -= len; |
c51e7bb1 JB |
6737 | |
6738 | trace_btrfs_space_reservation(fs_info, "pinned", | |
6739 | space_info->flags, len, 0); | |
4f4db217 | 6740 | space_info->max_extent_size = 0; |
d288db5d | 6741 | percpu_counter_add(&space_info->total_bytes_pinned, -len); |
7b398f8e JB |
6742 | if (cache->ro) { |
6743 | space_info->bytes_readonly += len; | |
6744 | readonly = true; | |
6745 | } | |
11833d66 | 6746 | spin_unlock(&cache->lock); |
957780eb JB |
6747 | if (!readonly && return_free_space && |
6748 | global_rsv->space_info == space_info) { | |
6749 | u64 to_add = len; | |
92ac58ec | 6750 | |
7b398f8e JB |
6751 | spin_lock(&global_rsv->lock); |
6752 | if (!global_rsv->full) { | |
957780eb JB |
6753 | to_add = min(len, global_rsv->size - |
6754 | global_rsv->reserved); | |
6755 | global_rsv->reserved += to_add; | |
6756 | space_info->bytes_may_use += to_add; | |
7b398f8e JB |
6757 | if (global_rsv->reserved >= global_rsv->size) |
6758 | global_rsv->full = 1; | |
957780eb JB |
6759 | trace_btrfs_space_reservation(fs_info, |
6760 | "space_info", | |
6761 | space_info->flags, | |
6762 | to_add, 1); | |
6763 | len -= to_add; | |
7b398f8e JB |
6764 | } |
6765 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
6766 | /* Add to any tickets we may have */ |
6767 | if (len) | |
6768 | space_info_add_new_bytes(fs_info, space_info, | |
6769 | len); | |
7b398f8e JB |
6770 | } |
6771 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6772 | } |
11833d66 YZ |
6773 | |
6774 | if (cache) | |
6775 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6776 | return 0; |
6777 | } | |
6778 | ||
6779 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
2ff7e61e | 6780 | struct btrfs_fs_info *fs_info) |
a28ec197 | 6781 | { |
e33e17ee JM |
6782 | struct btrfs_block_group_cache *block_group, *tmp; |
6783 | struct list_head *deleted_bgs; | |
11833d66 | 6784 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6785 | u64 start; |
6786 | u64 end; | |
a28ec197 | 6787 | int ret; |
a28ec197 | 6788 | |
11833d66 YZ |
6789 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6790 | unpin = &fs_info->freed_extents[1]; | |
6791 | else | |
6792 | unpin = &fs_info->freed_extents[0]; | |
6793 | ||
e33e17ee | 6794 | while (!trans->aborted) { |
d4b450cd | 6795 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6796 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 6797 | EXTENT_DIRTY, NULL); |
d4b450cd FM |
6798 | if (ret) { |
6799 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6800 | break; |
d4b450cd | 6801 | } |
1f3c79a2 | 6802 | |
0b246afa | 6803 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 6804 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 6805 | end + 1 - start, NULL); |
1f3c79a2 | 6806 | |
af6f8f60 | 6807 | clear_extent_dirty(unpin, start, end); |
2ff7e61e | 6808 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 6809 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
b9473439 | 6810 | cond_resched(); |
a28ec197 | 6811 | } |
817d52f8 | 6812 | |
e33e17ee JM |
6813 | /* |
6814 | * Transaction is finished. We don't need the lock anymore. We | |
6815 | * do need to clean up the block groups in case of a transaction | |
6816 | * abort. | |
6817 | */ | |
6818 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6819 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6820 | u64 trimmed = 0; | |
6821 | ||
6822 | ret = -EROFS; | |
6823 | if (!trans->aborted) | |
2ff7e61e | 6824 | ret = btrfs_discard_extent(fs_info, |
e33e17ee JM |
6825 | block_group->key.objectid, |
6826 | block_group->key.offset, | |
6827 | &trimmed); | |
6828 | ||
6829 | list_del_init(&block_group->bg_list); | |
6830 | btrfs_put_block_group_trimming(block_group); | |
6831 | btrfs_put_block_group(block_group); | |
6832 | ||
6833 | if (ret) { | |
6834 | const char *errstr = btrfs_decode_error(ret); | |
6835 | btrfs_warn(fs_info, | |
913e1535 | 6836 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
6837 | ret, errstr); |
6838 | } | |
6839 | } | |
6840 | ||
e20d96d6 CM |
6841 | return 0; |
6842 | } | |
6843 | ||
5d4f98a2 | 6844 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 6845 | struct btrfs_fs_info *info, |
c682f9b3 | 6846 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
6847 | u64 root_objectid, u64 owner_objectid, |
6848 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 6849 | struct btrfs_delayed_extent_op *extent_op) |
a28ec197 | 6850 | { |
e2fa7227 | 6851 | struct btrfs_key key; |
5d4f98a2 | 6852 | struct btrfs_path *path; |
1261ec42 | 6853 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 6854 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6855 | struct btrfs_extent_item *ei; |
6856 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6857 | int ret; |
5d4f98a2 | 6858 | int is_data; |
952fccac CM |
6859 | int extent_slot = 0; |
6860 | int found_extent = 0; | |
6861 | int num_to_del = 1; | |
5d4f98a2 YZ |
6862 | u32 item_size; |
6863 | u64 refs; | |
c682f9b3 QW |
6864 | u64 bytenr = node->bytenr; |
6865 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6866 | int last_ref = 0; |
0b246afa | 6867 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 6868 | |
5caf2a00 | 6869 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6870 | if (!path) |
6871 | return -ENOMEM; | |
5f26f772 | 6872 | |
e4058b54 | 6873 | path->reada = READA_FORWARD; |
b9473439 | 6874 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6875 | |
6876 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6877 | BUG_ON(!is_data && refs_to_drop != 1); | |
6878 | ||
3173a18f | 6879 | if (is_data) |
897ca819 | 6880 | skinny_metadata = false; |
3173a18f | 6881 | |
87bde3cd | 6882 | ret = lookup_extent_backref(trans, info, path, &iref, |
5d4f98a2 YZ |
6883 | bytenr, num_bytes, parent, |
6884 | root_objectid, owner_objectid, | |
6885 | owner_offset); | |
7bb86316 | 6886 | if (ret == 0) { |
952fccac | 6887 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6888 | while (extent_slot >= 0) { |
6889 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6890 | extent_slot); |
5d4f98a2 | 6891 | if (key.objectid != bytenr) |
952fccac | 6892 | break; |
5d4f98a2 YZ |
6893 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6894 | key.offset == num_bytes) { | |
952fccac CM |
6895 | found_extent = 1; |
6896 | break; | |
6897 | } | |
3173a18f JB |
6898 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
6899 | key.offset == owner_objectid) { | |
6900 | found_extent = 1; | |
6901 | break; | |
6902 | } | |
952fccac CM |
6903 | if (path->slots[0] - extent_slot > 5) |
6904 | break; | |
5d4f98a2 | 6905 | extent_slot--; |
952fccac | 6906 | } |
5d4f98a2 YZ |
6907 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
6908 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
6909 | if (found_extent && item_size < sizeof(*ei)) | |
6910 | found_extent = 0; | |
6911 | #endif | |
31840ae1 | 6912 | if (!found_extent) { |
5d4f98a2 | 6913 | BUG_ON(iref); |
87bde3cd JM |
6914 | ret = remove_extent_backref(trans, info, path, NULL, |
6915 | refs_to_drop, | |
fcebe456 | 6916 | is_data, &last_ref); |
005d6427 | 6917 | if (ret) { |
66642832 | 6918 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6919 | goto out; |
6920 | } | |
b3b4aa74 | 6921 | btrfs_release_path(path); |
b9473439 | 6922 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6923 | |
6924 | key.objectid = bytenr; | |
6925 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6926 | key.offset = num_bytes; | |
6927 | ||
3173a18f JB |
6928 | if (!is_data && skinny_metadata) { |
6929 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6930 | key.offset = owner_objectid; | |
6931 | } | |
6932 | ||
31840ae1 ZY |
6933 | ret = btrfs_search_slot(trans, extent_root, |
6934 | &key, path, -1, 1); | |
3173a18f JB |
6935 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
6936 | /* | |
6937 | * Couldn't find our skinny metadata item, | |
6938 | * see if we have ye olde extent item. | |
6939 | */ | |
6940 | path->slots[0]--; | |
6941 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6942 | path->slots[0]); | |
6943 | if (key.objectid == bytenr && | |
6944 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6945 | key.offset == num_bytes) | |
6946 | ret = 0; | |
6947 | } | |
6948 | ||
6949 | if (ret > 0 && skinny_metadata) { | |
6950 | skinny_metadata = false; | |
9ce49a0b | 6951 | key.objectid = bytenr; |
3173a18f JB |
6952 | key.type = BTRFS_EXTENT_ITEM_KEY; |
6953 | key.offset = num_bytes; | |
6954 | btrfs_release_path(path); | |
6955 | ret = btrfs_search_slot(trans, extent_root, | |
6956 | &key, path, -1, 1); | |
6957 | } | |
6958 | ||
f3465ca4 | 6959 | if (ret) { |
5d163e0e JM |
6960 | btrfs_err(info, |
6961 | "umm, got %d back from search, was looking for %llu", | |
6962 | ret, bytenr); | |
b783e62d | 6963 | if (ret > 0) |
a4f78750 | 6964 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 6965 | } |
005d6427 | 6966 | if (ret < 0) { |
66642832 | 6967 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6968 | goto out; |
6969 | } | |
31840ae1 ZY |
6970 | extent_slot = path->slots[0]; |
6971 | } | |
fae7f21c | 6972 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 6973 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
6974 | btrfs_err(info, |
6975 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
6976 | bytenr, parent, root_objectid, owner_objectid, |
6977 | owner_offset); | |
66642832 | 6978 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 6979 | goto out; |
79787eaa | 6980 | } else { |
66642832 | 6981 | btrfs_abort_transaction(trans, ret); |
005d6427 | 6982 | goto out; |
7bb86316 | 6983 | } |
5f39d397 CM |
6984 | |
6985 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
6986 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6987 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
6988 | if (item_size < sizeof(*ei)) { | |
6989 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
87bde3cd JM |
6990 | ret = convert_extent_item_v0(trans, info, path, owner_objectid, |
6991 | 0); | |
005d6427 | 6992 | if (ret < 0) { |
66642832 | 6993 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6994 | goto out; |
6995 | } | |
5d4f98a2 | 6996 | |
b3b4aa74 | 6997 | btrfs_release_path(path); |
5d4f98a2 YZ |
6998 | path->leave_spinning = 1; |
6999 | ||
7000 | key.objectid = bytenr; | |
7001 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
7002 | key.offset = num_bytes; | |
7003 | ||
7004 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
7005 | -1, 1); | |
7006 | if (ret) { | |
5d163e0e JM |
7007 | btrfs_err(info, |
7008 | "umm, got %d back from search, was looking for %llu", | |
c1c9ff7c | 7009 | ret, bytenr); |
a4f78750 | 7010 | btrfs_print_leaf(path->nodes[0]); |
5d4f98a2 | 7011 | } |
005d6427 | 7012 | if (ret < 0) { |
66642832 | 7013 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7014 | goto out; |
7015 | } | |
7016 | ||
5d4f98a2 YZ |
7017 | extent_slot = path->slots[0]; |
7018 | leaf = path->nodes[0]; | |
7019 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
7020 | } | |
7021 | #endif | |
7022 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 7023 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 7024 | struct btrfs_extent_item); |
3173a18f JB |
7025 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
7026 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
7027 | struct btrfs_tree_block_info *bi; |
7028 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
7029 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
7030 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
7031 | } | |
56bec294 | 7032 | |
5d4f98a2 | 7033 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 7034 | if (refs < refs_to_drop) { |
5d163e0e JM |
7035 | btrfs_err(info, |
7036 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
7037 | refs_to_drop, refs, bytenr); | |
32b02538 | 7038 | ret = -EINVAL; |
66642832 | 7039 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
7040 | goto out; |
7041 | } | |
56bec294 | 7042 | refs -= refs_to_drop; |
5f39d397 | 7043 | |
5d4f98a2 YZ |
7044 | if (refs > 0) { |
7045 | if (extent_op) | |
7046 | __run_delayed_extent_op(extent_op, leaf, ei); | |
7047 | /* | |
7048 | * In the case of inline back ref, reference count will | |
7049 | * be updated by remove_extent_backref | |
952fccac | 7050 | */ |
5d4f98a2 YZ |
7051 | if (iref) { |
7052 | BUG_ON(!found_extent); | |
7053 | } else { | |
7054 | btrfs_set_extent_refs(leaf, ei, refs); | |
7055 | btrfs_mark_buffer_dirty(leaf); | |
7056 | } | |
7057 | if (found_extent) { | |
87bde3cd | 7058 | ret = remove_extent_backref(trans, info, path, |
5d4f98a2 | 7059 | iref, refs_to_drop, |
fcebe456 | 7060 | is_data, &last_ref); |
005d6427 | 7061 | if (ret) { |
66642832 | 7062 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7063 | goto out; |
7064 | } | |
952fccac | 7065 | } |
5d4f98a2 | 7066 | } else { |
5d4f98a2 YZ |
7067 | if (found_extent) { |
7068 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 7069 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
7070 | if (iref) { |
7071 | BUG_ON(path->slots[0] != extent_slot); | |
7072 | } else { | |
7073 | BUG_ON(path->slots[0] != extent_slot + 1); | |
7074 | path->slots[0] = extent_slot; | |
7075 | num_to_del = 2; | |
7076 | } | |
78fae27e | 7077 | } |
b9473439 | 7078 | |
fcebe456 | 7079 | last_ref = 1; |
952fccac CM |
7080 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
7081 | num_to_del); | |
005d6427 | 7082 | if (ret) { |
66642832 | 7083 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7084 | goto out; |
7085 | } | |
b3b4aa74 | 7086 | btrfs_release_path(path); |
21af804c | 7087 | |
5d4f98a2 | 7088 | if (is_data) { |
5b4aacef | 7089 | ret = btrfs_del_csums(trans, info, bytenr, num_bytes); |
005d6427 | 7090 | if (ret) { |
66642832 | 7091 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7092 | goto out; |
7093 | } | |
459931ec CM |
7094 | } |
7095 | ||
0b246afa | 7096 | ret = add_to_free_space_tree(trans, info, bytenr, num_bytes); |
1e144fb8 | 7097 | if (ret) { |
66642832 | 7098 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
7099 | goto out; |
7100 | } | |
7101 | ||
0b246afa | 7102 | ret = update_block_group(trans, info, bytenr, num_bytes, 0); |
005d6427 | 7103 | if (ret) { |
66642832 | 7104 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7105 | goto out; |
7106 | } | |
a28ec197 | 7107 | } |
fcebe456 JB |
7108 | btrfs_release_path(path); |
7109 | ||
79787eaa | 7110 | out: |
5caf2a00 | 7111 | btrfs_free_path(path); |
a28ec197 CM |
7112 | return ret; |
7113 | } | |
7114 | ||
1887be66 | 7115 | /* |
f0486c68 | 7116 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
7117 | * delayed ref for that extent as well. This searches the delayed ref tree for |
7118 | * a given extent, and if there are no other delayed refs to be processed, it | |
7119 | * removes it from the tree. | |
7120 | */ | |
7121 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 7122 | u64 bytenr) |
1887be66 CM |
7123 | { |
7124 | struct btrfs_delayed_ref_head *head; | |
7125 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 7126 | int ret = 0; |
1887be66 CM |
7127 | |
7128 | delayed_refs = &trans->transaction->delayed_refs; | |
7129 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 7130 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 7131 | if (!head) |
cf93da7b | 7132 | goto out_delayed_unlock; |
1887be66 | 7133 | |
d7df2c79 | 7134 | spin_lock(&head->lock); |
0e0adbcf | 7135 | if (!RB_EMPTY_ROOT(&head->ref_tree)) |
1887be66 CM |
7136 | goto out; |
7137 | ||
5d4f98a2 YZ |
7138 | if (head->extent_op) { |
7139 | if (!head->must_insert_reserved) | |
7140 | goto out; | |
78a6184a | 7141 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
7142 | head->extent_op = NULL; |
7143 | } | |
7144 | ||
1887be66 CM |
7145 | /* |
7146 | * waiting for the lock here would deadlock. If someone else has it | |
7147 | * locked they are already in the process of dropping it anyway | |
7148 | */ | |
7149 | if (!mutex_trylock(&head->mutex)) | |
7150 | goto out; | |
7151 | ||
7152 | /* | |
7153 | * at this point we have a head with no other entries. Go | |
7154 | * ahead and process it. | |
7155 | */ | |
c46effa6 | 7156 | rb_erase(&head->href_node, &delayed_refs->href_root); |
d278850e | 7157 | RB_CLEAR_NODE(&head->href_node); |
d7df2c79 | 7158 | atomic_dec(&delayed_refs->num_entries); |
1887be66 CM |
7159 | |
7160 | /* | |
7161 | * we don't take a ref on the node because we're removing it from the | |
7162 | * tree, so we just steal the ref the tree was holding. | |
7163 | */ | |
c3e69d58 | 7164 | delayed_refs->num_heads--; |
d7df2c79 | 7165 | if (head->processing == 0) |
c3e69d58 | 7166 | delayed_refs->num_heads_ready--; |
d7df2c79 JB |
7167 | head->processing = 0; |
7168 | spin_unlock(&head->lock); | |
1887be66 CM |
7169 | spin_unlock(&delayed_refs->lock); |
7170 | ||
f0486c68 YZ |
7171 | BUG_ON(head->extent_op); |
7172 | if (head->must_insert_reserved) | |
7173 | ret = 1; | |
7174 | ||
7175 | mutex_unlock(&head->mutex); | |
d278850e | 7176 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 7177 | return ret; |
1887be66 | 7178 | out: |
d7df2c79 | 7179 | spin_unlock(&head->lock); |
cf93da7b CM |
7180 | |
7181 | out_delayed_unlock: | |
1887be66 CM |
7182 | spin_unlock(&delayed_refs->lock); |
7183 | return 0; | |
7184 | } | |
7185 | ||
f0486c68 YZ |
7186 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
7187 | struct btrfs_root *root, | |
7188 | struct extent_buffer *buf, | |
5581a51a | 7189 | u64 parent, int last_ref) |
f0486c68 | 7190 | { |
0b246afa | 7191 | struct btrfs_fs_info *fs_info = root->fs_info; |
b150a4f1 | 7192 | int pin = 1; |
f0486c68 YZ |
7193 | int ret; |
7194 | ||
7195 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
d7eae340 OS |
7196 | int old_ref_mod, new_ref_mod; |
7197 | ||
fd708b81 JB |
7198 | btrfs_ref_tree_mod(root, buf->start, buf->len, parent, |
7199 | root->root_key.objectid, | |
7200 | btrfs_header_level(buf), 0, | |
7201 | BTRFS_DROP_DELAYED_REF); | |
7be07912 OS |
7202 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start, |
7203 | buf->len, parent, | |
0b246afa JM |
7204 | root->root_key.objectid, |
7205 | btrfs_header_level(buf), | |
7be07912 | 7206 | BTRFS_DROP_DELAYED_REF, NULL, |
d7eae340 | 7207 | &old_ref_mod, &new_ref_mod); |
79787eaa | 7208 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 7209 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
7210 | } |
7211 | ||
0a16c7d7 | 7212 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
7213 | struct btrfs_block_group_cache *cache; |
7214 | ||
f0486c68 | 7215 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 7216 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 7217 | if (!ret) |
37be25bc | 7218 | goto out; |
f0486c68 YZ |
7219 | } |
7220 | ||
4da8b76d | 7221 | pin = 0; |
0b246afa | 7222 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 7223 | |
f0486c68 | 7224 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
2ff7e61e JM |
7225 | pin_down_extent(fs_info, cache, buf->start, |
7226 | buf->len, 1); | |
6219872d | 7227 | btrfs_put_block_group(cache); |
37be25bc | 7228 | goto out; |
f0486c68 YZ |
7229 | } |
7230 | ||
7231 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
7232 | ||
7233 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 7234 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 7235 | btrfs_put_block_group(cache); |
71ff6437 | 7236 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
7237 | } |
7238 | out: | |
b150a4f1 | 7239 | if (pin) |
0b246afa | 7240 | add_pinned_bytes(fs_info, buf->len, btrfs_header_level(buf), |
b150a4f1 JB |
7241 | root->root_key.objectid); |
7242 | ||
0a16c7d7 OS |
7243 | if (last_ref) { |
7244 | /* | |
7245 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
7246 | * matter anymore. | |
7247 | */ | |
7248 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
7249 | } | |
f0486c68 YZ |
7250 | } |
7251 | ||
79787eaa | 7252 | /* Can return -ENOMEM */ |
2ff7e61e | 7253 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
84f7d8e6 | 7254 | struct btrfs_root *root, |
66d7e7f0 | 7255 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, |
b06c4bf5 | 7256 | u64 owner, u64 offset) |
925baedd | 7257 | { |
84f7d8e6 | 7258 | struct btrfs_fs_info *fs_info = root->fs_info; |
d7eae340 | 7259 | int old_ref_mod, new_ref_mod; |
925baedd CM |
7260 | int ret; |
7261 | ||
f5ee5c9a | 7262 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 7263 | return 0; |
fccb84c9 | 7264 | |
fd708b81 JB |
7265 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) |
7266 | btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, | |
7267 | root_objectid, owner, offset, | |
7268 | BTRFS_DROP_DELAYED_REF); | |
7269 | ||
56bec294 CM |
7270 | /* |
7271 | * tree log blocks never actually go into the extent allocation | |
7272 | * tree, just update pinning info and exit early. | |
56bec294 | 7273 | */ |
5d4f98a2 YZ |
7274 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
7275 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 7276 | /* unlocks the pinned mutex */ |
2ff7e61e | 7277 | btrfs_pin_extent(fs_info, bytenr, num_bytes, 1); |
d7eae340 | 7278 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 7279 | ret = 0; |
5d4f98a2 | 7280 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 | 7281 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
7be07912 OS |
7282 | num_bytes, parent, |
7283 | root_objectid, (int)owner, | |
7284 | BTRFS_DROP_DELAYED_REF, NULL, | |
d7eae340 | 7285 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 7286 | } else { |
66d7e7f0 | 7287 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
7be07912 OS |
7288 | num_bytes, parent, |
7289 | root_objectid, owner, offset, | |
7290 | 0, BTRFS_DROP_DELAYED_REF, | |
d7eae340 | 7291 | &old_ref_mod, &new_ref_mod); |
56bec294 | 7292 | } |
d7eae340 OS |
7293 | |
7294 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) | |
7295 | add_pinned_bytes(fs_info, num_bytes, owner, root_objectid); | |
7296 | ||
925baedd CM |
7297 | return ret; |
7298 | } | |
7299 | ||
817d52f8 JB |
7300 | /* |
7301 | * when we wait for progress in the block group caching, its because | |
7302 | * our allocation attempt failed at least once. So, we must sleep | |
7303 | * and let some progress happen before we try again. | |
7304 | * | |
7305 | * This function will sleep at least once waiting for new free space to | |
7306 | * show up, and then it will check the block group free space numbers | |
7307 | * for our min num_bytes. Another option is to have it go ahead | |
7308 | * and look in the rbtree for a free extent of a given size, but this | |
7309 | * is a good start. | |
36cce922 JB |
7310 | * |
7311 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
7312 | * any of the information in this block group. | |
817d52f8 | 7313 | */ |
36cce922 | 7314 | static noinline void |
817d52f8 JB |
7315 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
7316 | u64 num_bytes) | |
7317 | { | |
11833d66 | 7318 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 7319 | |
11833d66 YZ |
7320 | caching_ctl = get_caching_control(cache); |
7321 | if (!caching_ctl) | |
36cce922 | 7322 | return; |
817d52f8 | 7323 | |
11833d66 | 7324 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 7325 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
7326 | |
7327 | put_caching_control(caching_ctl); | |
11833d66 YZ |
7328 | } |
7329 | ||
7330 | static noinline int | |
7331 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
7332 | { | |
7333 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 7334 | int ret = 0; |
11833d66 YZ |
7335 | |
7336 | caching_ctl = get_caching_control(cache); | |
7337 | if (!caching_ctl) | |
36cce922 | 7338 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
7339 | |
7340 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
7341 | if (cache->cached == BTRFS_CACHE_ERROR) |
7342 | ret = -EIO; | |
11833d66 | 7343 | put_caching_control(caching_ctl); |
36cce922 | 7344 | return ret; |
817d52f8 JB |
7345 | } |
7346 | ||
6ab0a202 JM |
7347 | static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = { |
7348 | [BTRFS_RAID_RAID10] = "raid10", | |
7349 | [BTRFS_RAID_RAID1] = "raid1", | |
7350 | [BTRFS_RAID_DUP] = "dup", | |
7351 | [BTRFS_RAID_RAID0] = "raid0", | |
7352 | [BTRFS_RAID_SINGLE] = "single", | |
7353 | [BTRFS_RAID_RAID5] = "raid5", | |
7354 | [BTRFS_RAID_RAID6] = "raid6", | |
7355 | }; | |
7356 | ||
1b8e5df6 | 7357 | static const char *get_raid_name(enum btrfs_raid_types type) |
6ab0a202 JM |
7358 | { |
7359 | if (type >= BTRFS_NR_RAID_TYPES) | |
7360 | return NULL; | |
7361 | ||
7362 | return btrfs_raid_type_names[type]; | |
7363 | } | |
7364 | ||
817d52f8 | 7365 | enum btrfs_loop_type { |
285ff5af JB |
7366 | LOOP_CACHING_NOWAIT = 0, |
7367 | LOOP_CACHING_WAIT = 1, | |
7368 | LOOP_ALLOC_CHUNK = 2, | |
7369 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
7370 | }; |
7371 | ||
e570fd27 MX |
7372 | static inline void |
7373 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
7374 | int delalloc) | |
7375 | { | |
7376 | if (delalloc) | |
7377 | down_read(&cache->data_rwsem); | |
7378 | } | |
7379 | ||
7380 | static inline void | |
7381 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
7382 | int delalloc) | |
7383 | { | |
7384 | btrfs_get_block_group(cache); | |
7385 | if (delalloc) | |
7386 | down_read(&cache->data_rwsem); | |
7387 | } | |
7388 | ||
7389 | static struct btrfs_block_group_cache * | |
7390 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
7391 | struct btrfs_free_cluster *cluster, | |
7392 | int delalloc) | |
7393 | { | |
89771cc9 | 7394 | struct btrfs_block_group_cache *used_bg = NULL; |
6719afdc | 7395 | |
e570fd27 | 7396 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
7397 | while (1) { |
7398 | used_bg = cluster->block_group; | |
7399 | if (!used_bg) | |
7400 | return NULL; | |
7401 | ||
7402 | if (used_bg == block_group) | |
e570fd27 MX |
7403 | return used_bg; |
7404 | ||
6719afdc | 7405 | btrfs_get_block_group(used_bg); |
e570fd27 | 7406 | |
6719afdc GU |
7407 | if (!delalloc) |
7408 | return used_bg; | |
e570fd27 | 7409 | |
6719afdc GU |
7410 | if (down_read_trylock(&used_bg->data_rwsem)) |
7411 | return used_bg; | |
e570fd27 | 7412 | |
6719afdc | 7413 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 7414 | |
e321f8a8 LB |
7415 | /* We should only have one-level nested. */ |
7416 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 7417 | |
6719afdc GU |
7418 | spin_lock(&cluster->refill_lock); |
7419 | if (used_bg == cluster->block_group) | |
7420 | return used_bg; | |
e570fd27 | 7421 | |
6719afdc GU |
7422 | up_read(&used_bg->data_rwsem); |
7423 | btrfs_put_block_group(used_bg); | |
7424 | } | |
e570fd27 MX |
7425 | } |
7426 | ||
7427 | static inline void | |
7428 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7429 | int delalloc) | |
7430 | { | |
7431 | if (delalloc) | |
7432 | up_read(&cache->data_rwsem); | |
7433 | btrfs_put_block_group(cache); | |
7434 | } | |
7435 | ||
fec577fb CM |
7436 | /* |
7437 | * walks the btree of allocated extents and find a hole of a given size. | |
7438 | * The key ins is changed to record the hole: | |
a4820398 | 7439 | * ins->objectid == start position |
62e2749e | 7440 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7441 | * ins->offset == the size of the hole. |
fec577fb | 7442 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7443 | * |
7444 | * If there is no suitable free space, we will record the max size of | |
7445 | * the free space extent currently. | |
fec577fb | 7446 | */ |
87bde3cd | 7447 | static noinline int find_free_extent(struct btrfs_fs_info *fs_info, |
18513091 WX |
7448 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
7449 | u64 hint_byte, struct btrfs_key *ins, | |
7450 | u64 flags, int delalloc) | |
fec577fb | 7451 | { |
80eb234a | 7452 | int ret = 0; |
0b246afa | 7453 | struct btrfs_root *root = fs_info->extent_root; |
fa9c0d79 | 7454 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7455 | struct btrfs_block_group_cache *block_group = NULL; |
81c9ad23 | 7456 | u64 search_start = 0; |
a4820398 | 7457 | u64 max_extent_size = 0; |
c759c4e1 | 7458 | u64 empty_cluster = 0; |
80eb234a | 7459 | struct btrfs_space_info *space_info; |
fa9c0d79 | 7460 | int loop = 0; |
3e72ee88 | 7461 | int index = btrfs_bg_flags_to_raid_index(flags); |
0a24325e | 7462 | bool failed_cluster_refill = false; |
1cdda9b8 | 7463 | bool failed_alloc = false; |
67377734 | 7464 | bool use_cluster = true; |
60d2adbb | 7465 | bool have_caching_bg = false; |
13a0db5a | 7466 | bool orig_have_caching_bg = false; |
a5e681d9 | 7467 | bool full_search = false; |
fec577fb | 7468 | |
0b246afa | 7469 | WARN_ON(num_bytes < fs_info->sectorsize); |
962a298f | 7470 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7471 | ins->objectid = 0; |
7472 | ins->offset = 0; | |
b1a4d965 | 7473 | |
71ff6437 | 7474 | trace_find_free_extent(fs_info, num_bytes, empty_size, flags); |
3f7de037 | 7475 | |
0b246afa | 7476 | space_info = __find_space_info(fs_info, flags); |
1b1d1f66 | 7477 | if (!space_info) { |
0b246afa | 7478 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7479 | return -ENOSPC; |
7480 | } | |
2552d17e | 7481 | |
67377734 | 7482 | /* |
4f4db217 JB |
7483 | * If our free space is heavily fragmented we may not be able to make |
7484 | * big contiguous allocations, so instead of doing the expensive search | |
7485 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7486 | * can go ahead and search for a more manageable chunk. | |
7487 | * | |
7488 | * If our max_extent_size is large enough for our allocation simply | |
7489 | * disable clustering since we will likely not be able to find enough | |
7490 | * space to create a cluster and induce latency trying. | |
67377734 | 7491 | */ |
4f4db217 JB |
7492 | if (unlikely(space_info->max_extent_size)) { |
7493 | spin_lock(&space_info->lock); | |
7494 | if (space_info->max_extent_size && | |
7495 | num_bytes > space_info->max_extent_size) { | |
7496 | ins->offset = space_info->max_extent_size; | |
7497 | spin_unlock(&space_info->lock); | |
7498 | return -ENOSPC; | |
7499 | } else if (space_info->max_extent_size) { | |
7500 | use_cluster = false; | |
7501 | } | |
7502 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7503 | } |
0f9dd46c | 7504 | |
2ff7e61e | 7505 | last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster); |
239b14b3 | 7506 | if (last_ptr) { |
fa9c0d79 CM |
7507 | spin_lock(&last_ptr->lock); |
7508 | if (last_ptr->block_group) | |
7509 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7510 | if (last_ptr->fragmented) { |
7511 | /* | |
7512 | * We still set window_start so we can keep track of the | |
7513 | * last place we found an allocation to try and save | |
7514 | * some time. | |
7515 | */ | |
7516 | hint_byte = last_ptr->window_start; | |
7517 | use_cluster = false; | |
7518 | } | |
fa9c0d79 | 7519 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7520 | } |
fa9c0d79 | 7521 | |
2ff7e61e | 7522 | search_start = max(search_start, first_logical_byte(fs_info, 0)); |
239b14b3 | 7523 | search_start = max(search_start, hint_byte); |
2552d17e | 7524 | if (search_start == hint_byte) { |
0b246afa | 7525 | block_group = btrfs_lookup_block_group(fs_info, search_start); |
817d52f8 JB |
7526 | /* |
7527 | * we don't want to use the block group if it doesn't match our | |
7528 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7529 | * |
7530 | * However if we are re-searching with an ideal block group | |
7531 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7532 | */ |
b6919a58 | 7533 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7534 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7535 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7536 | if (list_empty(&block_group->list) || |
7537 | block_group->ro) { | |
7538 | /* | |
7539 | * someone is removing this block group, | |
7540 | * we can't jump into the have_block_group | |
7541 | * target because our list pointers are not | |
7542 | * valid | |
7543 | */ | |
7544 | btrfs_put_block_group(block_group); | |
7545 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7546 | } else { |
3e72ee88 QW |
7547 | index = btrfs_bg_flags_to_raid_index( |
7548 | block_group->flags); | |
e570fd27 | 7549 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7550 | goto have_block_group; |
ccf0e725 | 7551 | } |
2552d17e | 7552 | } else if (block_group) { |
fa9c0d79 | 7553 | btrfs_put_block_group(block_group); |
2552d17e | 7554 | } |
42e70e7a | 7555 | } |
2552d17e | 7556 | search: |
60d2adbb | 7557 | have_caching_bg = false; |
3e72ee88 | 7558 | if (index == 0 || index == btrfs_bg_flags_to_raid_index(flags)) |
a5e681d9 | 7559 | full_search = true; |
80eb234a | 7560 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
7561 | list_for_each_entry(block_group, &space_info->block_groups[index], |
7562 | list) { | |
6226cb0a | 7563 | u64 offset; |
817d52f8 | 7564 | int cached; |
8a1413a2 | 7565 | |
14443937 JM |
7566 | /* If the block group is read-only, we can skip it entirely. */ |
7567 | if (unlikely(block_group->ro)) | |
7568 | continue; | |
7569 | ||
e570fd27 | 7570 | btrfs_grab_block_group(block_group, delalloc); |
2552d17e | 7571 | search_start = block_group->key.objectid; |
42e70e7a | 7572 | |
83a50de9 CM |
7573 | /* |
7574 | * this can happen if we end up cycling through all the | |
7575 | * raid types, but we want to make sure we only allocate | |
7576 | * for the proper type. | |
7577 | */ | |
b6919a58 | 7578 | if (!block_group_bits(block_group, flags)) { |
83a50de9 CM |
7579 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
7580 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
7581 | BTRFS_BLOCK_GROUP_RAID5 | |
7582 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
7583 | BTRFS_BLOCK_GROUP_RAID10; |
7584 | ||
7585 | /* | |
7586 | * if they asked for extra copies and this block group | |
7587 | * doesn't provide them, bail. This does allow us to | |
7588 | * fill raid0 from raid1. | |
7589 | */ | |
b6919a58 | 7590 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
7591 | goto loop; |
7592 | } | |
7593 | ||
2552d17e | 7594 | have_block_group: |
291c7d2f JB |
7595 | cached = block_group_cache_done(block_group); |
7596 | if (unlikely(!cached)) { | |
a5e681d9 | 7597 | have_caching_bg = true; |
f6373bf3 | 7598 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7599 | BUG_ON(ret < 0); |
7600 | ret = 0; | |
817d52f8 JB |
7601 | } |
7602 | ||
36cce922 JB |
7603 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7604 | goto loop; | |
0f9dd46c | 7605 | |
0a24325e | 7606 | /* |
062c05c4 AO |
7607 | * Ok we want to try and use the cluster allocator, so |
7608 | * lets look there | |
0a24325e | 7609 | */ |
c759c4e1 | 7610 | if (last_ptr && use_cluster) { |
215a63d1 | 7611 | struct btrfs_block_group_cache *used_block_group; |
8de972b4 | 7612 | unsigned long aligned_cluster; |
fa9c0d79 CM |
7613 | /* |
7614 | * the refill lock keeps out other | |
7615 | * people trying to start a new cluster | |
7616 | */ | |
e570fd27 MX |
7617 | used_block_group = btrfs_lock_cluster(block_group, |
7618 | last_ptr, | |
7619 | delalloc); | |
7620 | if (!used_block_group) | |
44fb5511 | 7621 | goto refill_cluster; |
274bd4fb | 7622 | |
e570fd27 MX |
7623 | if (used_block_group != block_group && |
7624 | (used_block_group->ro || | |
7625 | !block_group_bits(used_block_group, flags))) | |
7626 | goto release_cluster; | |
44fb5511 | 7627 | |
274bd4fb | 7628 | offset = btrfs_alloc_from_cluster(used_block_group, |
a4820398 MX |
7629 | last_ptr, |
7630 | num_bytes, | |
7631 | used_block_group->key.objectid, | |
7632 | &max_extent_size); | |
fa9c0d79 CM |
7633 | if (offset) { |
7634 | /* we have a block, we're done */ | |
7635 | spin_unlock(&last_ptr->refill_lock); | |
71ff6437 | 7636 | trace_btrfs_reserve_extent_cluster(fs_info, |
89d4346a MX |
7637 | used_block_group, |
7638 | search_start, num_bytes); | |
215a63d1 | 7639 | if (used_block_group != block_group) { |
e570fd27 MX |
7640 | btrfs_release_block_group(block_group, |
7641 | delalloc); | |
215a63d1 MX |
7642 | block_group = used_block_group; |
7643 | } | |
fa9c0d79 CM |
7644 | goto checks; |
7645 | } | |
7646 | ||
274bd4fb | 7647 | WARN_ON(last_ptr->block_group != used_block_group); |
e570fd27 | 7648 | release_cluster: |
062c05c4 AO |
7649 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
7650 | * set up a new clusters, so lets just skip it | |
7651 | * and let the allocator find whatever block | |
7652 | * it can find. If we reach this point, we | |
7653 | * will have tried the cluster allocator | |
7654 | * plenty of times and not have found | |
7655 | * anything, so we are likely way too | |
7656 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
7657 | * anything. |
7658 | * | |
7659 | * However, if the cluster is taken from the | |
7660 | * current block group, release the cluster | |
7661 | * first, so that we stand a better chance of | |
7662 | * succeeding in the unclustered | |
7663 | * allocation. */ | |
7664 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
e570fd27 | 7665 | used_block_group != block_group) { |
062c05c4 | 7666 | spin_unlock(&last_ptr->refill_lock); |
e570fd27 MX |
7667 | btrfs_release_block_group(used_block_group, |
7668 | delalloc); | |
062c05c4 AO |
7669 | goto unclustered_alloc; |
7670 | } | |
7671 | ||
fa9c0d79 CM |
7672 | /* |
7673 | * this cluster didn't work out, free it and | |
7674 | * start over | |
7675 | */ | |
7676 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7677 | ||
e570fd27 MX |
7678 | if (used_block_group != block_group) |
7679 | btrfs_release_block_group(used_block_group, | |
7680 | delalloc); | |
7681 | refill_cluster: | |
a5f6f719 AO |
7682 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
7683 | spin_unlock(&last_ptr->refill_lock); | |
7684 | goto unclustered_alloc; | |
7685 | } | |
7686 | ||
8de972b4 CM |
7687 | aligned_cluster = max_t(unsigned long, |
7688 | empty_cluster + empty_size, | |
7689 | block_group->full_stripe_len); | |
7690 | ||
fa9c0d79 | 7691 | /* allocate a cluster in this block group */ |
2ff7e61e | 7692 | ret = btrfs_find_space_cluster(fs_info, block_group, |
00361589 JB |
7693 | last_ptr, search_start, |
7694 | num_bytes, | |
7695 | aligned_cluster); | |
fa9c0d79 CM |
7696 | if (ret == 0) { |
7697 | /* | |
7698 | * now pull our allocation out of this | |
7699 | * cluster | |
7700 | */ | |
7701 | offset = btrfs_alloc_from_cluster(block_group, | |
a4820398 MX |
7702 | last_ptr, |
7703 | num_bytes, | |
7704 | search_start, | |
7705 | &max_extent_size); | |
fa9c0d79 CM |
7706 | if (offset) { |
7707 | /* we found one, proceed */ | |
7708 | spin_unlock(&last_ptr->refill_lock); | |
71ff6437 | 7709 | trace_btrfs_reserve_extent_cluster(fs_info, |
3f7de037 JB |
7710 | block_group, search_start, |
7711 | num_bytes); | |
fa9c0d79 CM |
7712 | goto checks; |
7713 | } | |
0a24325e JB |
7714 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
7715 | && !failed_cluster_refill) { | |
817d52f8 JB |
7716 | spin_unlock(&last_ptr->refill_lock); |
7717 | ||
0a24325e | 7718 | failed_cluster_refill = true; |
817d52f8 JB |
7719 | wait_block_group_cache_progress(block_group, |
7720 | num_bytes + empty_cluster + empty_size); | |
7721 | goto have_block_group; | |
fa9c0d79 | 7722 | } |
817d52f8 | 7723 | |
fa9c0d79 CM |
7724 | /* |
7725 | * at this point we either didn't find a cluster | |
7726 | * or we weren't able to allocate a block from our | |
7727 | * cluster. Free the cluster we've been trying | |
7728 | * to use, and go to the next block group | |
7729 | */ | |
0a24325e | 7730 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 7731 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 7732 | goto loop; |
fa9c0d79 CM |
7733 | } |
7734 | ||
062c05c4 | 7735 | unclustered_alloc: |
c759c4e1 JB |
7736 | /* |
7737 | * We are doing an unclustered alloc, set the fragmented flag so | |
7738 | * we don't bother trying to setup a cluster again until we get | |
7739 | * more space. | |
7740 | */ | |
7741 | if (unlikely(last_ptr)) { | |
7742 | spin_lock(&last_ptr->lock); | |
7743 | last_ptr->fragmented = 1; | |
7744 | spin_unlock(&last_ptr->lock); | |
7745 | } | |
0c9b36e0 LB |
7746 | if (cached) { |
7747 | struct btrfs_free_space_ctl *ctl = | |
7748 | block_group->free_space_ctl; | |
7749 | ||
7750 | spin_lock(&ctl->tree_lock); | |
7751 | if (ctl->free_space < | |
7752 | num_bytes + empty_cluster + empty_size) { | |
7753 | if (ctl->free_space > max_extent_size) | |
7754 | max_extent_size = ctl->free_space; | |
7755 | spin_unlock(&ctl->tree_lock); | |
7756 | goto loop; | |
7757 | } | |
7758 | spin_unlock(&ctl->tree_lock); | |
a5f6f719 | 7759 | } |
a5f6f719 | 7760 | |
6226cb0a | 7761 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
a4820398 MX |
7762 | num_bytes, empty_size, |
7763 | &max_extent_size); | |
1cdda9b8 JB |
7764 | /* |
7765 | * If we didn't find a chunk, and we haven't failed on this | |
7766 | * block group before, and this block group is in the middle of | |
7767 | * caching and we are ok with waiting, then go ahead and wait | |
7768 | * for progress to be made, and set failed_alloc to true. | |
7769 | * | |
7770 | * If failed_alloc is true then we've already waited on this | |
7771 | * block group once and should move on to the next block group. | |
7772 | */ | |
7773 | if (!offset && !failed_alloc && !cached && | |
7774 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 7775 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
7776 | num_bytes + empty_size); |
7777 | failed_alloc = true; | |
817d52f8 | 7778 | goto have_block_group; |
1cdda9b8 JB |
7779 | } else if (!offset) { |
7780 | goto loop; | |
817d52f8 | 7781 | } |
fa9c0d79 | 7782 | checks: |
0b246afa | 7783 | search_start = ALIGN(offset, fs_info->stripesize); |
25179201 | 7784 | |
2552d17e JB |
7785 | /* move on to the next group */ |
7786 | if (search_start + num_bytes > | |
215a63d1 MX |
7787 | block_group->key.objectid + block_group->key.offset) { |
7788 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 7789 | goto loop; |
6226cb0a | 7790 | } |
f5a31e16 | 7791 | |
f0486c68 | 7792 | if (offset < search_start) |
215a63d1 | 7793 | btrfs_add_free_space(block_group, offset, |
f0486c68 YZ |
7794 | search_start - offset); |
7795 | BUG_ON(offset > search_start); | |
2552d17e | 7796 | |
18513091 WX |
7797 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
7798 | num_bytes, delalloc); | |
f0486c68 | 7799 | if (ret == -EAGAIN) { |
215a63d1 | 7800 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 7801 | goto loop; |
0f9dd46c | 7802 | } |
9cfa3e34 | 7803 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 7804 | |
f0486c68 | 7805 | /* we are all good, lets return */ |
2552d17e JB |
7806 | ins->objectid = search_start; |
7807 | ins->offset = num_bytes; | |
d2fb3437 | 7808 | |
71ff6437 | 7809 | trace_btrfs_reserve_extent(fs_info, block_group, |
3f7de037 | 7810 | search_start, num_bytes); |
e570fd27 | 7811 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7812 | break; |
7813 | loop: | |
0a24325e | 7814 | failed_cluster_refill = false; |
1cdda9b8 | 7815 | failed_alloc = false; |
3e72ee88 QW |
7816 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != |
7817 | index); | |
e570fd27 | 7818 | btrfs_release_block_group(block_group, delalloc); |
14443937 | 7819 | cond_resched(); |
2552d17e JB |
7820 | } |
7821 | up_read(&space_info->groups_sem); | |
7822 | ||
13a0db5a | 7823 | if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg |
7824 | && !orig_have_caching_bg) | |
7825 | orig_have_caching_bg = true; | |
7826 | ||
60d2adbb MX |
7827 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
7828 | goto search; | |
7829 | ||
b742bb82 YZ |
7830 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
7831 | goto search; | |
7832 | ||
285ff5af | 7833 | /* |
ccf0e725 JB |
7834 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
7835 | * caching kthreads as we move along | |
817d52f8 JB |
7836 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
7837 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7838 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7839 | * again | |
fa9c0d79 | 7840 | */ |
723bda20 | 7841 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 7842 | index = 0; |
a5e681d9 JB |
7843 | if (loop == LOOP_CACHING_NOWAIT) { |
7844 | /* | |
7845 | * We want to skip the LOOP_CACHING_WAIT step if we | |
01327610 | 7846 | * don't have any uncached bgs and we've already done a |
a5e681d9 JB |
7847 | * full search through. |
7848 | */ | |
13a0db5a | 7849 | if (orig_have_caching_bg || !full_search) |
a5e681d9 JB |
7850 | loop = LOOP_CACHING_WAIT; |
7851 | else | |
7852 | loop = LOOP_ALLOC_CHUNK; | |
7853 | } else { | |
7854 | loop++; | |
7855 | } | |
7856 | ||
817d52f8 | 7857 | if (loop == LOOP_ALLOC_CHUNK) { |
00361589 | 7858 | struct btrfs_trans_handle *trans; |
f017f15f WS |
7859 | int exist = 0; |
7860 | ||
7861 | trans = current->journal_info; | |
7862 | if (trans) | |
7863 | exist = 1; | |
7864 | else | |
7865 | trans = btrfs_join_transaction(root); | |
00361589 | 7866 | |
00361589 JB |
7867 | if (IS_ERR(trans)) { |
7868 | ret = PTR_ERR(trans); | |
7869 | goto out; | |
7870 | } | |
7871 | ||
2ff7e61e | 7872 | ret = do_chunk_alloc(trans, fs_info, flags, |
ea658bad | 7873 | CHUNK_ALLOC_FORCE); |
a5e681d9 JB |
7874 | |
7875 | /* | |
7876 | * If we can't allocate a new chunk we've already looped | |
7877 | * through at least once, move on to the NO_EMPTY_SIZE | |
7878 | * case. | |
7879 | */ | |
7880 | if (ret == -ENOSPC) | |
7881 | loop = LOOP_NO_EMPTY_SIZE; | |
7882 | ||
ea658bad JB |
7883 | /* |
7884 | * Do not bail out on ENOSPC since we | |
7885 | * can do more things. | |
7886 | */ | |
00361589 | 7887 | if (ret < 0 && ret != -ENOSPC) |
66642832 | 7888 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
7889 | else |
7890 | ret = 0; | |
f017f15f | 7891 | if (!exist) |
3a45bb20 | 7892 | btrfs_end_transaction(trans); |
00361589 | 7893 | if (ret) |
ea658bad | 7894 | goto out; |
2552d17e JB |
7895 | } |
7896 | ||
723bda20 | 7897 | if (loop == LOOP_NO_EMPTY_SIZE) { |
a5e681d9 JB |
7898 | /* |
7899 | * Don't loop again if we already have no empty_size and | |
7900 | * no empty_cluster. | |
7901 | */ | |
7902 | if (empty_size == 0 && | |
7903 | empty_cluster == 0) { | |
7904 | ret = -ENOSPC; | |
7905 | goto out; | |
7906 | } | |
723bda20 JB |
7907 | empty_size = 0; |
7908 | empty_cluster = 0; | |
fa9c0d79 | 7909 | } |
723bda20 JB |
7910 | |
7911 | goto search; | |
2552d17e JB |
7912 | } else if (!ins->objectid) { |
7913 | ret = -ENOSPC; | |
d82a6f1d | 7914 | } else if (ins->objectid) { |
c759c4e1 JB |
7915 | if (!use_cluster && last_ptr) { |
7916 | spin_lock(&last_ptr->lock); | |
7917 | last_ptr->window_start = ins->objectid; | |
7918 | spin_unlock(&last_ptr->lock); | |
7919 | } | |
80eb234a | 7920 | ret = 0; |
be744175 | 7921 | } |
79787eaa | 7922 | out: |
4f4db217 JB |
7923 | if (ret == -ENOSPC) { |
7924 | spin_lock(&space_info->lock); | |
7925 | space_info->max_extent_size = max_extent_size; | |
7926 | spin_unlock(&space_info->lock); | |
a4820398 | 7927 | ins->offset = max_extent_size; |
4f4db217 | 7928 | } |
0f70abe2 | 7929 | return ret; |
fec577fb | 7930 | } |
ec44a35c | 7931 | |
ab8d0fc4 JM |
7932 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
7933 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 7934 | int dump_block_groups) |
0f9dd46c JB |
7935 | { |
7936 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 7937 | int index = 0; |
0f9dd46c | 7938 | |
9ed74f2d | 7939 | spin_lock(&info->lock); |
ab8d0fc4 JM |
7940 | btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull", |
7941 | info->flags, | |
4136135b LB |
7942 | info->total_bytes - btrfs_space_info_used(info, true), |
7943 | info->full ? "" : "not "); | |
ab8d0fc4 JM |
7944 | btrfs_info(fs_info, |
7945 | "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu", | |
7946 | info->total_bytes, info->bytes_used, info->bytes_pinned, | |
7947 | info->bytes_reserved, info->bytes_may_use, | |
7948 | info->bytes_readonly); | |
9ed74f2d JB |
7949 | spin_unlock(&info->lock); |
7950 | ||
7951 | if (!dump_block_groups) | |
7952 | return; | |
0f9dd46c | 7953 | |
80eb234a | 7954 | down_read(&info->groups_sem); |
b742bb82 YZ |
7955 | again: |
7956 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 7957 | spin_lock(&cache->lock); |
ab8d0fc4 JM |
7958 | btrfs_info(fs_info, |
7959 | "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s", | |
7960 | cache->key.objectid, cache->key.offset, | |
7961 | btrfs_block_group_used(&cache->item), cache->pinned, | |
7962 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
7963 | btrfs_dump_free_space(cache, bytes); |
7964 | spin_unlock(&cache->lock); | |
7965 | } | |
b742bb82 YZ |
7966 | if (++index < BTRFS_NR_RAID_TYPES) |
7967 | goto again; | |
80eb234a | 7968 | up_read(&info->groups_sem); |
0f9dd46c | 7969 | } |
e8569813 | 7970 | |
18513091 | 7971 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
7972 | u64 num_bytes, u64 min_alloc_size, |
7973 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 7974 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 7975 | { |
ab8d0fc4 | 7976 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 7977 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 7978 | u64 flags; |
fec577fb | 7979 | int ret; |
925baedd | 7980 | |
1b86826d | 7981 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 7982 | again: |
0b246afa | 7983 | WARN_ON(num_bytes < fs_info->sectorsize); |
87bde3cd | 7984 | ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size, |
18513091 | 7985 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 7986 | if (!ret && !is_data) { |
ab8d0fc4 | 7987 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 7988 | } else if (ret == -ENOSPC) { |
a4820398 MX |
7989 | if (!final_tried && ins->offset) { |
7990 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 7991 | num_bytes = round_down(num_bytes, |
0b246afa | 7992 | fs_info->sectorsize); |
9e622d6b | 7993 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 7994 | ram_bytes = num_bytes; |
9e622d6b MX |
7995 | if (num_bytes == min_alloc_size) |
7996 | final_tried = true; | |
7997 | goto again; | |
ab8d0fc4 | 7998 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
7999 | struct btrfs_space_info *sinfo; |
8000 | ||
ab8d0fc4 | 8001 | sinfo = __find_space_info(fs_info, flags); |
0b246afa | 8002 | btrfs_err(fs_info, |
5d163e0e JM |
8003 | "allocation failed flags %llu, wanted %llu", |
8004 | flags, num_bytes); | |
53804280 | 8005 | if (sinfo) |
ab8d0fc4 | 8006 | dump_space_info(fs_info, sinfo, num_bytes, 1); |
9e622d6b | 8007 | } |
925baedd | 8008 | } |
0f9dd46c JB |
8009 | |
8010 | return ret; | |
e6dcd2dc CM |
8011 | } |
8012 | ||
2ff7e61e | 8013 | static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
8014 | u64 start, u64 len, |
8015 | int pin, int delalloc) | |
65b51a00 | 8016 | { |
0f9dd46c | 8017 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 8018 | int ret = 0; |
0f9dd46c | 8019 | |
0b246afa | 8020 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 8021 | if (!cache) { |
0b246afa JM |
8022 | btrfs_err(fs_info, "Unable to find block group for %llu", |
8023 | start); | |
0f9dd46c JB |
8024 | return -ENOSPC; |
8025 | } | |
1f3c79a2 | 8026 | |
e688b725 | 8027 | if (pin) |
2ff7e61e | 8028 | pin_down_extent(fs_info, cache, start, len, 1); |
e688b725 | 8029 | else { |
0b246afa | 8030 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 8031 | ret = btrfs_discard_extent(fs_info, start, len, NULL); |
e688b725 | 8032 | btrfs_add_free_space(cache, start, len); |
4824f1f4 | 8033 | btrfs_free_reserved_bytes(cache, len, delalloc); |
71ff6437 | 8034 | trace_btrfs_reserved_extent_free(fs_info, start, len); |
e688b725 | 8035 | } |
31193213 | 8036 | |
fa9c0d79 | 8037 | btrfs_put_block_group(cache); |
e6dcd2dc CM |
8038 | return ret; |
8039 | } | |
8040 | ||
2ff7e61e | 8041 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 | 8042 | u64 start, u64 len, int delalloc) |
e688b725 | 8043 | { |
2ff7e61e | 8044 | return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc); |
e688b725 CM |
8045 | } |
8046 | ||
2ff7e61e | 8047 | int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info, |
e688b725 CM |
8048 | u64 start, u64 len) |
8049 | { | |
2ff7e61e | 8050 | return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0); |
e688b725 CM |
8051 | } |
8052 | ||
5d4f98a2 | 8053 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 8054 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8055 | u64 parent, u64 root_objectid, |
8056 | u64 flags, u64 owner, u64 offset, | |
8057 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
8058 | { |
8059 | int ret; | |
e6dcd2dc | 8060 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 8061 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 8062 | struct btrfs_path *path; |
5d4f98a2 YZ |
8063 | struct extent_buffer *leaf; |
8064 | int type; | |
8065 | u32 size; | |
26b8003f | 8066 | |
5d4f98a2 YZ |
8067 | if (parent > 0) |
8068 | type = BTRFS_SHARED_DATA_REF_KEY; | |
8069 | else | |
8070 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 8071 | |
5d4f98a2 | 8072 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
8073 | |
8074 | path = btrfs_alloc_path(); | |
db5b493a TI |
8075 | if (!path) |
8076 | return -ENOMEM; | |
47e4bb98 | 8077 | |
b9473439 | 8078 | path->leave_spinning = 1; |
5d4f98a2 YZ |
8079 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
8080 | ins, size); | |
79787eaa JM |
8081 | if (ret) { |
8082 | btrfs_free_path(path); | |
8083 | return ret; | |
8084 | } | |
0f9dd46c | 8085 | |
5d4f98a2 YZ |
8086 | leaf = path->nodes[0]; |
8087 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 8088 | struct btrfs_extent_item); |
5d4f98a2 YZ |
8089 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
8090 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8091 | btrfs_set_extent_flags(leaf, extent_item, | |
8092 | flags | BTRFS_EXTENT_FLAG_DATA); | |
8093 | ||
8094 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8095 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
8096 | if (parent > 0) { | |
8097 | struct btrfs_shared_data_ref *ref; | |
8098 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
8099 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8100 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
8101 | } else { | |
8102 | struct btrfs_extent_data_ref *ref; | |
8103 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
8104 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
8105 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
8106 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
8107 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
8108 | } | |
47e4bb98 CM |
8109 | |
8110 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 8111 | btrfs_free_path(path); |
f510cfec | 8112 | |
1e144fb8 OS |
8113 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
8114 | ins->offset); | |
8115 | if (ret) | |
8116 | return ret; | |
8117 | ||
6202df69 | 8118 | ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1); |
79787eaa | 8119 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8120 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8121 | ins->objectid, ins->offset); |
f5947066 CM |
8122 | BUG(); |
8123 | } | |
71ff6437 | 8124 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
8125 | return ret; |
8126 | } | |
8127 | ||
5d4f98a2 | 8128 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
2ff7e61e | 8129 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8130 | u64 parent, u64 root_objectid, |
8131 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 8132 | int level, struct btrfs_key *ins) |
e6dcd2dc CM |
8133 | { |
8134 | int ret; | |
5d4f98a2 YZ |
8135 | struct btrfs_extent_item *extent_item; |
8136 | struct btrfs_tree_block_info *block_info; | |
8137 | struct btrfs_extent_inline_ref *iref; | |
8138 | struct btrfs_path *path; | |
8139 | struct extent_buffer *leaf; | |
3173a18f | 8140 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
fcebe456 | 8141 | u64 num_bytes = ins->offset; |
0b246afa | 8142 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f JB |
8143 | |
8144 | if (!skinny_metadata) | |
8145 | size += sizeof(*block_info); | |
1c2308f8 | 8146 | |
5d4f98a2 | 8147 | path = btrfs_alloc_path(); |
857cc2fc | 8148 | if (!path) { |
2ff7e61e | 8149 | btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid, |
0b246afa | 8150 | fs_info->nodesize); |
d8926bb3 | 8151 | return -ENOMEM; |
857cc2fc | 8152 | } |
56bec294 | 8153 | |
5d4f98a2 YZ |
8154 | path->leave_spinning = 1; |
8155 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
8156 | ins, size); | |
79787eaa | 8157 | if (ret) { |
dd825259 | 8158 | btrfs_free_path(path); |
2ff7e61e | 8159 | btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid, |
0b246afa | 8160 | fs_info->nodesize); |
79787eaa JM |
8161 | return ret; |
8162 | } | |
5d4f98a2 YZ |
8163 | |
8164 | leaf = path->nodes[0]; | |
8165 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
8166 | struct btrfs_extent_item); | |
8167 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
8168 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8169 | btrfs_set_extent_flags(leaf, extent_item, | |
8170 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 8171 | |
3173a18f JB |
8172 | if (skinny_metadata) { |
8173 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
0b246afa | 8174 | num_bytes = fs_info->nodesize; |
3173a18f JB |
8175 | } else { |
8176 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
8177 | btrfs_set_tree_block_key(leaf, block_info, key); | |
8178 | btrfs_set_tree_block_level(leaf, block_info, level); | |
8179 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
8180 | } | |
5d4f98a2 | 8181 | |
5d4f98a2 YZ |
8182 | if (parent > 0) { |
8183 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
8184 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8185 | BTRFS_SHARED_BLOCK_REF_KEY); | |
8186 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8187 | } else { | |
8188 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8189 | BTRFS_TREE_BLOCK_REF_KEY); | |
8190 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
8191 | } | |
8192 | ||
8193 | btrfs_mark_buffer_dirty(leaf); | |
8194 | btrfs_free_path(path); | |
8195 | ||
1e144fb8 OS |
8196 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
8197 | num_bytes); | |
8198 | if (ret) | |
8199 | return ret; | |
8200 | ||
6202df69 JM |
8201 | ret = update_block_group(trans, fs_info, ins->objectid, |
8202 | fs_info->nodesize, 1); | |
79787eaa | 8203 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8204 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8205 | ins->objectid, ins->offset); |
5d4f98a2 YZ |
8206 | BUG(); |
8207 | } | |
0be5dc67 | 8208 | |
71ff6437 | 8209 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, |
0b246afa | 8210 | fs_info->nodesize); |
5d4f98a2 YZ |
8211 | return ret; |
8212 | } | |
8213 | ||
8214 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 8215 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
8216 | u64 offset, u64 ram_bytes, |
8217 | struct btrfs_key *ins) | |
5d4f98a2 | 8218 | { |
84f7d8e6 | 8219 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
8220 | int ret; |
8221 | ||
84f7d8e6 | 8222 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 8223 | |
fd708b81 JB |
8224 | btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0, |
8225 | root->root_key.objectid, owner, offset, | |
8226 | BTRFS_ADD_DELAYED_EXTENT); | |
8227 | ||
0b246afa | 8228 | ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid, |
84f7d8e6 JB |
8229 | ins->offset, 0, |
8230 | root->root_key.objectid, owner, | |
7be07912 OS |
8231 | offset, ram_bytes, |
8232 | BTRFS_ADD_DELAYED_EXTENT, NULL, NULL); | |
e6dcd2dc CM |
8233 | return ret; |
8234 | } | |
e02119d5 CM |
8235 | |
8236 | /* | |
8237 | * this is used by the tree logging recovery code. It records that | |
8238 | * an extent has been allocated and makes sure to clear the free | |
8239 | * space cache bits as well | |
8240 | */ | |
5d4f98a2 | 8241 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 8242 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8243 | u64 root_objectid, u64 owner, u64 offset, |
8244 | struct btrfs_key *ins) | |
e02119d5 CM |
8245 | { |
8246 | int ret; | |
8247 | struct btrfs_block_group_cache *block_group; | |
ed7a6948 | 8248 | struct btrfs_space_info *space_info; |
11833d66 | 8249 | |
8c2a1a30 JB |
8250 | /* |
8251 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 8252 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 8253 | */ |
0b246afa | 8254 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
8255 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
8256 | ins->offset); | |
b50c6e25 | 8257 | if (ret) |
8c2a1a30 | 8258 | return ret; |
11833d66 YZ |
8259 | } |
8260 | ||
0b246afa | 8261 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
8262 | if (!block_group) |
8263 | return -EINVAL; | |
8264 | ||
ed7a6948 WX |
8265 | space_info = block_group->space_info; |
8266 | spin_lock(&space_info->lock); | |
8267 | spin_lock(&block_group->lock); | |
8268 | space_info->bytes_reserved += ins->offset; | |
8269 | block_group->reserved += ins->offset; | |
8270 | spin_unlock(&block_group->lock); | |
8271 | spin_unlock(&space_info->lock); | |
8272 | ||
2ff7e61e | 8273 | ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid, |
5d4f98a2 | 8274 | 0, owner, offset, ins, 1); |
b50c6e25 | 8275 | btrfs_put_block_group(block_group); |
e02119d5 CM |
8276 | return ret; |
8277 | } | |
8278 | ||
48a3b636 ES |
8279 | static struct extent_buffer * |
8280 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
fe864576 | 8281 | u64 bytenr, int level) |
65b51a00 | 8282 | { |
0b246afa | 8283 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
8284 | struct extent_buffer *buf; |
8285 | ||
2ff7e61e | 8286 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8287 | if (IS_ERR(buf)) |
8288 | return buf; | |
8289 | ||
65b51a00 | 8290 | btrfs_set_header_generation(buf, trans->transid); |
85d4e461 | 8291 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 8292 | btrfs_tree_lock(buf); |
7c302b49 | 8293 | clean_tree_block(fs_info, buf); |
3083ee2e | 8294 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
8295 | |
8296 | btrfs_set_lock_blocking(buf); | |
4db8c528 | 8297 | set_extent_buffer_uptodate(buf); |
b4ce94de | 8298 | |
d0c803c4 | 8299 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 8300 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
8301 | /* |
8302 | * we allow two log transactions at a time, use different | |
8303 | * EXENT bit to differentiate dirty pages. | |
8304 | */ | |
656f30db | 8305 | if (buf->log_index == 0) |
8cef4e16 YZ |
8306 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
8307 | buf->start + buf->len - 1, GFP_NOFS); | |
8308 | else | |
8309 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 8310 | buf->start + buf->len - 1); |
d0c803c4 | 8311 | } else { |
656f30db | 8312 | buf->log_index = -1; |
d0c803c4 | 8313 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 8314 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 8315 | } |
64c12921 | 8316 | trans->dirty = true; |
b4ce94de | 8317 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
8318 | return buf; |
8319 | } | |
8320 | ||
f0486c68 YZ |
8321 | static struct btrfs_block_rsv * |
8322 | use_block_rsv(struct btrfs_trans_handle *trans, | |
8323 | struct btrfs_root *root, u32 blocksize) | |
8324 | { | |
0b246afa | 8325 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8326 | struct btrfs_block_rsv *block_rsv; |
0b246afa | 8327 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
f0486c68 | 8328 | int ret; |
d88033db | 8329 | bool global_updated = false; |
f0486c68 YZ |
8330 | |
8331 | block_rsv = get_block_rsv(trans, root); | |
8332 | ||
b586b323 MX |
8333 | if (unlikely(block_rsv->size == 0)) |
8334 | goto try_reserve; | |
d88033db | 8335 | again: |
f0486c68 YZ |
8336 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
8337 | if (!ret) | |
8338 | return block_rsv; | |
8339 | ||
b586b323 MX |
8340 | if (block_rsv->failfast) |
8341 | return ERR_PTR(ret); | |
8342 | ||
d88033db MX |
8343 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
8344 | global_updated = true; | |
0b246afa | 8345 | update_global_block_rsv(fs_info); |
d88033db MX |
8346 | goto again; |
8347 | } | |
8348 | ||
0b246afa | 8349 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
b586b323 MX |
8350 | static DEFINE_RATELIMIT_STATE(_rs, |
8351 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
8352 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
8353 | if (__ratelimit(&_rs)) | |
8354 | WARN(1, KERN_DEBUG | |
efe120a0 | 8355 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
8356 | } |
8357 | try_reserve: | |
8358 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
8359 | BTRFS_RESERVE_NO_FLUSH); | |
8360 | if (!ret) | |
8361 | return block_rsv; | |
8362 | /* | |
8363 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
8364 | * the global reserve if its space type is the same as the global |
8365 | * reservation. | |
b586b323 | 8366 | */ |
5881cfc9 MX |
8367 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
8368 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
8369 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
8370 | if (!ret) | |
8371 | return global_rsv; | |
8372 | } | |
8373 | return ERR_PTR(ret); | |
f0486c68 YZ |
8374 | } |
8375 | ||
8c2a3ca2 JB |
8376 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
8377 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
8378 | { |
8379 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 8380 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
8381 | } |
8382 | ||
fec577fb | 8383 | /* |
f0486c68 | 8384 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 8385 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 8386 | */ |
4d75f8a9 | 8387 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
8388 | struct btrfs_root *root, |
8389 | u64 parent, u64 root_objectid, | |
8390 | const struct btrfs_disk_key *key, | |
8391 | int level, u64 hint, | |
8392 | u64 empty_size) | |
fec577fb | 8393 | { |
0b246afa | 8394 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 8395 | struct btrfs_key ins; |
f0486c68 | 8396 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 8397 | struct extent_buffer *buf; |
67b7859e | 8398 | struct btrfs_delayed_extent_op *extent_op; |
f0486c68 YZ |
8399 | u64 flags = 0; |
8400 | int ret; | |
0b246afa JM |
8401 | u32 blocksize = fs_info->nodesize; |
8402 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 8403 | |
05653ef3 | 8404 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 8405 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 8406 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
fe864576 | 8407 | level); |
faa2dbf0 JB |
8408 | if (!IS_ERR(buf)) |
8409 | root->alloc_bytenr += blocksize; | |
8410 | return buf; | |
8411 | } | |
05653ef3 | 8412 | #endif |
fccb84c9 | 8413 | |
f0486c68 YZ |
8414 | block_rsv = use_block_rsv(trans, root, blocksize); |
8415 | if (IS_ERR(block_rsv)) | |
8416 | return ERR_CAST(block_rsv); | |
8417 | ||
18513091 | 8418 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 8419 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
8420 | if (ret) |
8421 | goto out_unuse; | |
55c69072 | 8422 | |
fe864576 | 8423 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level); |
67b7859e OS |
8424 | if (IS_ERR(buf)) { |
8425 | ret = PTR_ERR(buf); | |
8426 | goto out_free_reserved; | |
8427 | } | |
f0486c68 YZ |
8428 | |
8429 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8430 | if (parent == 0) | |
8431 | parent = ins.objectid; | |
8432 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8433 | } else | |
8434 | BUG_ON(parent > 0); | |
8435 | ||
8436 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 8437 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
8438 | if (!extent_op) { |
8439 | ret = -ENOMEM; | |
8440 | goto out_free_buf; | |
8441 | } | |
f0486c68 YZ |
8442 | if (key) |
8443 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8444 | else | |
8445 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8446 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
8447 | extent_op->update_key = skinny_metadata ? false : true; |
8448 | extent_op->update_flags = true; | |
8449 | extent_op->is_data = false; | |
b1c79e09 | 8450 | extent_op->level = level; |
f0486c68 | 8451 | |
fd708b81 JB |
8452 | btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent, |
8453 | root_objectid, level, 0, | |
8454 | BTRFS_ADD_DELAYED_EXTENT); | |
7be07912 OS |
8455 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid, |
8456 | ins.offset, parent, | |
8457 | root_objectid, level, | |
67b7859e | 8458 | BTRFS_ADD_DELAYED_EXTENT, |
7be07912 | 8459 | extent_op, NULL, NULL); |
67b7859e OS |
8460 | if (ret) |
8461 | goto out_free_delayed; | |
f0486c68 | 8462 | } |
fec577fb | 8463 | return buf; |
67b7859e OS |
8464 | |
8465 | out_free_delayed: | |
8466 | btrfs_free_delayed_extent_op(extent_op); | |
8467 | out_free_buf: | |
8468 | free_extent_buffer(buf); | |
8469 | out_free_reserved: | |
2ff7e61e | 8470 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 8471 | out_unuse: |
0b246afa | 8472 | unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 8473 | return ERR_PTR(ret); |
fec577fb | 8474 | } |
a28ec197 | 8475 | |
2c47e605 YZ |
8476 | struct walk_control { |
8477 | u64 refs[BTRFS_MAX_LEVEL]; | |
8478 | u64 flags[BTRFS_MAX_LEVEL]; | |
8479 | struct btrfs_key update_progress; | |
8480 | int stage; | |
8481 | int level; | |
8482 | int shared_level; | |
8483 | int update_ref; | |
8484 | int keep_locks; | |
1c4850e2 YZ |
8485 | int reada_slot; |
8486 | int reada_count; | |
66d7e7f0 | 8487 | int for_reloc; |
2c47e605 YZ |
8488 | }; |
8489 | ||
8490 | #define DROP_REFERENCE 1 | |
8491 | #define UPDATE_BACKREF 2 | |
8492 | ||
1c4850e2 YZ |
8493 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8494 | struct btrfs_root *root, | |
8495 | struct walk_control *wc, | |
8496 | struct btrfs_path *path) | |
6407bf6d | 8497 | { |
0b246afa | 8498 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8499 | u64 bytenr; |
8500 | u64 generation; | |
8501 | u64 refs; | |
94fcca9f | 8502 | u64 flags; |
5d4f98a2 | 8503 | u32 nritems; |
1c4850e2 YZ |
8504 | struct btrfs_key key; |
8505 | struct extent_buffer *eb; | |
6407bf6d | 8506 | int ret; |
1c4850e2 YZ |
8507 | int slot; |
8508 | int nread = 0; | |
6407bf6d | 8509 | |
1c4850e2 YZ |
8510 | if (path->slots[wc->level] < wc->reada_slot) { |
8511 | wc->reada_count = wc->reada_count * 2 / 3; | |
8512 | wc->reada_count = max(wc->reada_count, 2); | |
8513 | } else { | |
8514 | wc->reada_count = wc->reada_count * 3 / 2; | |
8515 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 8516 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 8517 | } |
7bb86316 | 8518 | |
1c4850e2 YZ |
8519 | eb = path->nodes[wc->level]; |
8520 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 8521 | |
1c4850e2 YZ |
8522 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8523 | if (nread >= wc->reada_count) | |
8524 | break; | |
bd56b302 | 8525 | |
2dd3e67b | 8526 | cond_resched(); |
1c4850e2 YZ |
8527 | bytenr = btrfs_node_blockptr(eb, slot); |
8528 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8529 | |
1c4850e2 YZ |
8530 | if (slot == path->slots[wc->level]) |
8531 | goto reada; | |
5d4f98a2 | 8532 | |
1c4850e2 YZ |
8533 | if (wc->stage == UPDATE_BACKREF && |
8534 | generation <= root->root_key.offset) | |
bd56b302 CM |
8535 | continue; |
8536 | ||
94fcca9f | 8537 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 8538 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
8539 | wc->level - 1, 1, &refs, |
8540 | &flags); | |
79787eaa JM |
8541 | /* We don't care about errors in readahead. */ |
8542 | if (ret < 0) | |
8543 | continue; | |
94fcca9f YZ |
8544 | BUG_ON(refs == 0); |
8545 | ||
1c4850e2 | 8546 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8547 | if (refs == 1) |
8548 | goto reada; | |
bd56b302 | 8549 | |
94fcca9f YZ |
8550 | if (wc->level == 1 && |
8551 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8552 | continue; | |
1c4850e2 YZ |
8553 | if (!wc->update_ref || |
8554 | generation <= root->root_key.offset) | |
8555 | continue; | |
8556 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8557 | ret = btrfs_comp_cpu_keys(&key, | |
8558 | &wc->update_progress); | |
8559 | if (ret < 0) | |
8560 | continue; | |
94fcca9f YZ |
8561 | } else { |
8562 | if (wc->level == 1 && | |
8563 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8564 | continue; | |
6407bf6d | 8565 | } |
1c4850e2 | 8566 | reada: |
2ff7e61e | 8567 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 8568 | nread++; |
20524f02 | 8569 | } |
1c4850e2 | 8570 | wc->reada_slot = slot; |
20524f02 | 8571 | } |
2c47e605 | 8572 | |
f82d02d9 | 8573 | /* |
2c016dc2 | 8574 | * helper to process tree block while walking down the tree. |
2c47e605 | 8575 | * |
2c47e605 YZ |
8576 | * when wc->stage == UPDATE_BACKREF, this function updates |
8577 | * back refs for pointers in the block. | |
8578 | * | |
8579 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8580 | */ |
2c47e605 | 8581 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8582 | struct btrfs_root *root, |
2c47e605 | 8583 | struct btrfs_path *path, |
94fcca9f | 8584 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8585 | { |
2ff7e61e | 8586 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
8587 | int level = wc->level; |
8588 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 8589 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
8590 | int ret; |
8591 | ||
2c47e605 YZ |
8592 | if (wc->stage == UPDATE_BACKREF && |
8593 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8594 | return 1; | |
f82d02d9 | 8595 | |
2c47e605 YZ |
8596 | /* |
8597 | * when reference count of tree block is 1, it won't increase | |
8598 | * again. once full backref flag is set, we never clear it. | |
8599 | */ | |
94fcca9f YZ |
8600 | if (lookup_info && |
8601 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8602 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 8603 | BUG_ON(!path->locks[level]); |
2ff7e61e | 8604 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8605 | eb->start, level, 1, |
2c47e605 YZ |
8606 | &wc->refs[level], |
8607 | &wc->flags[level]); | |
79787eaa JM |
8608 | BUG_ON(ret == -ENOMEM); |
8609 | if (ret) | |
8610 | return ret; | |
2c47e605 YZ |
8611 | BUG_ON(wc->refs[level] == 0); |
8612 | } | |
5d4f98a2 | 8613 | |
2c47e605 YZ |
8614 | if (wc->stage == DROP_REFERENCE) { |
8615 | if (wc->refs[level] > 1) | |
8616 | return 1; | |
f82d02d9 | 8617 | |
2c47e605 | 8618 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 8619 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8620 | path->locks[level] = 0; |
8621 | } | |
8622 | return 0; | |
8623 | } | |
f82d02d9 | 8624 | |
2c47e605 YZ |
8625 | /* wc->stage == UPDATE_BACKREF */ |
8626 | if (!(wc->flags[level] & flag)) { | |
8627 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 8628 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 8629 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 8630 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8631 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8632 | ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start, |
b1c79e09 JB |
8633 | eb->len, flag, |
8634 | btrfs_header_level(eb), 0); | |
79787eaa | 8635 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
8636 | wc->flags[level] |= flag; |
8637 | } | |
8638 | ||
8639 | /* | |
8640 | * the block is shared by multiple trees, so it's not good to | |
8641 | * keep the tree lock | |
8642 | */ | |
8643 | if (path->locks[level] && level > 0) { | |
bd681513 | 8644 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8645 | path->locks[level] = 0; |
8646 | } | |
8647 | return 0; | |
8648 | } | |
8649 | ||
1c4850e2 | 8650 | /* |
2c016dc2 | 8651 | * helper to process tree block pointer. |
1c4850e2 YZ |
8652 | * |
8653 | * when wc->stage == DROP_REFERENCE, this function checks | |
8654 | * reference count of the block pointed to. if the block | |
8655 | * is shared and we need update back refs for the subtree | |
8656 | * rooted at the block, this function changes wc->stage to | |
8657 | * UPDATE_BACKREF. if the block is shared and there is no | |
8658 | * need to update back, this function drops the reference | |
8659 | * to the block. | |
8660 | * | |
8661 | * NOTE: return value 1 means we should stop walking down. | |
8662 | */ | |
8663 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8664 | struct btrfs_root *root, | |
8665 | struct btrfs_path *path, | |
94fcca9f | 8666 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 8667 | { |
0b246afa | 8668 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8669 | u64 bytenr; |
8670 | u64 generation; | |
8671 | u64 parent; | |
8672 | u32 blocksize; | |
8673 | struct btrfs_key key; | |
8674 | struct extent_buffer *next; | |
8675 | int level = wc->level; | |
8676 | int reada = 0; | |
8677 | int ret = 0; | |
1152651a | 8678 | bool need_account = false; |
1c4850e2 YZ |
8679 | |
8680 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8681 | path->slots[level]); | |
8682 | /* | |
8683 | * if the lower level block was created before the snapshot | |
8684 | * was created, we know there is no need to update back refs | |
8685 | * for the subtree | |
8686 | */ | |
8687 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8688 | generation <= root->root_key.offset) { |
8689 | *lookup_info = 1; | |
1c4850e2 | 8690 | return 1; |
94fcca9f | 8691 | } |
1c4850e2 YZ |
8692 | |
8693 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
0b246afa | 8694 | blocksize = fs_info->nodesize; |
1c4850e2 | 8695 | |
0b246afa | 8696 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 8697 | if (!next) { |
2ff7e61e | 8698 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8699 | if (IS_ERR(next)) |
8700 | return PTR_ERR(next); | |
8701 | ||
b2aaaa3b JB |
8702 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8703 | level - 1); | |
1c4850e2 YZ |
8704 | reada = 1; |
8705 | } | |
8706 | btrfs_tree_lock(next); | |
8707 | btrfs_set_lock_blocking(next); | |
8708 | ||
2ff7e61e | 8709 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
8710 | &wc->refs[level - 1], |
8711 | &wc->flags[level - 1]); | |
4867268c JB |
8712 | if (ret < 0) |
8713 | goto out_unlock; | |
79787eaa | 8714 | |
c2cf52eb | 8715 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 8716 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
8717 | ret = -EIO; |
8718 | goto out_unlock; | |
c2cf52eb | 8719 | } |
94fcca9f | 8720 | *lookup_info = 0; |
1c4850e2 | 8721 | |
94fcca9f | 8722 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8723 | if (wc->refs[level - 1] > 1) { |
1152651a | 8724 | need_account = true; |
94fcca9f YZ |
8725 | if (level == 1 && |
8726 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8727 | goto skip; | |
8728 | ||
1c4850e2 YZ |
8729 | if (!wc->update_ref || |
8730 | generation <= root->root_key.offset) | |
8731 | goto skip; | |
8732 | ||
8733 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8734 | path->slots[level]); | |
8735 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8736 | if (ret < 0) | |
8737 | goto skip; | |
8738 | ||
8739 | wc->stage = UPDATE_BACKREF; | |
8740 | wc->shared_level = level - 1; | |
8741 | } | |
94fcca9f YZ |
8742 | } else { |
8743 | if (level == 1 && | |
8744 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8745 | goto skip; | |
1c4850e2 YZ |
8746 | } |
8747 | ||
b9fab919 | 8748 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8749 | btrfs_tree_unlock(next); |
8750 | free_extent_buffer(next); | |
8751 | next = NULL; | |
94fcca9f | 8752 | *lookup_info = 1; |
1c4850e2 YZ |
8753 | } |
8754 | ||
8755 | if (!next) { | |
8756 | if (reada && level == 1) | |
8757 | reada_walk_down(trans, root, wc, path); | |
2ff7e61e | 8758 | next = read_tree_block(fs_info, bytenr, generation); |
64c043de LB |
8759 | if (IS_ERR(next)) { |
8760 | return PTR_ERR(next); | |
8761 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8762 | free_extent_buffer(next); |
97d9a8a4 | 8763 | return -EIO; |
416bc658 | 8764 | } |
1c4850e2 YZ |
8765 | btrfs_tree_lock(next); |
8766 | btrfs_set_lock_blocking(next); | |
8767 | } | |
8768 | ||
8769 | level--; | |
4867268c JB |
8770 | ASSERT(level == btrfs_header_level(next)); |
8771 | if (level != btrfs_header_level(next)) { | |
8772 | btrfs_err(root->fs_info, "mismatched level"); | |
8773 | ret = -EIO; | |
8774 | goto out_unlock; | |
8775 | } | |
1c4850e2 YZ |
8776 | path->nodes[level] = next; |
8777 | path->slots[level] = 0; | |
bd681513 | 8778 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8779 | wc->level = level; |
8780 | if (wc->level == 1) | |
8781 | wc->reada_slot = 0; | |
8782 | return 0; | |
8783 | skip: | |
8784 | wc->refs[level - 1] = 0; | |
8785 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
8786 | if (wc->stage == DROP_REFERENCE) { |
8787 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8788 | parent = path->nodes[level]->start; | |
8789 | } else { | |
4867268c | 8790 | ASSERT(root->root_key.objectid == |
94fcca9f | 8791 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
8792 | if (root->root_key.objectid != |
8793 | btrfs_header_owner(path->nodes[level])) { | |
8794 | btrfs_err(root->fs_info, | |
8795 | "mismatched block owner"); | |
8796 | ret = -EIO; | |
8797 | goto out_unlock; | |
8798 | } | |
94fcca9f YZ |
8799 | parent = 0; |
8800 | } | |
1c4850e2 | 8801 | |
1152651a | 8802 | if (need_account) { |
33d1f05c QW |
8803 | ret = btrfs_qgroup_trace_subtree(trans, root, next, |
8804 | generation, level - 1); | |
1152651a | 8805 | if (ret) { |
0b246afa | 8806 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8807 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
8808 | ret); | |
1152651a MF |
8809 | } |
8810 | } | |
84f7d8e6 | 8811 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, |
2ff7e61e JM |
8812 | parent, root->root_key.objectid, |
8813 | level - 1, 0); | |
4867268c JB |
8814 | if (ret) |
8815 | goto out_unlock; | |
1c4850e2 | 8816 | } |
4867268c JB |
8817 | |
8818 | *lookup_info = 1; | |
8819 | ret = 1; | |
8820 | ||
8821 | out_unlock: | |
1c4850e2 YZ |
8822 | btrfs_tree_unlock(next); |
8823 | free_extent_buffer(next); | |
4867268c JB |
8824 | |
8825 | return ret; | |
1c4850e2 YZ |
8826 | } |
8827 | ||
2c47e605 | 8828 | /* |
2c016dc2 | 8829 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
8830 | * |
8831 | * when wc->stage == DROP_REFERENCE, this function drops | |
8832 | * reference count on the block. | |
8833 | * | |
8834 | * when wc->stage == UPDATE_BACKREF, this function changes | |
8835 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
8836 | * to UPDATE_BACKREF previously while processing the block. | |
8837 | * | |
8838 | * NOTE: return value 1 means we should stop walking up. | |
8839 | */ | |
8840 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
8841 | struct btrfs_root *root, | |
8842 | struct btrfs_path *path, | |
8843 | struct walk_control *wc) | |
8844 | { | |
0b246afa | 8845 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8846 | int ret; |
2c47e605 YZ |
8847 | int level = wc->level; |
8848 | struct extent_buffer *eb = path->nodes[level]; | |
8849 | u64 parent = 0; | |
8850 | ||
8851 | if (wc->stage == UPDATE_BACKREF) { | |
8852 | BUG_ON(wc->shared_level < level); | |
8853 | if (level < wc->shared_level) | |
8854 | goto out; | |
8855 | ||
2c47e605 YZ |
8856 | ret = find_next_key(path, level + 1, &wc->update_progress); |
8857 | if (ret > 0) | |
8858 | wc->update_ref = 0; | |
8859 | ||
8860 | wc->stage = DROP_REFERENCE; | |
8861 | wc->shared_level = -1; | |
8862 | path->slots[level] = 0; | |
8863 | ||
8864 | /* | |
8865 | * check reference count again if the block isn't locked. | |
8866 | * we should start walking down the tree again if reference | |
8867 | * count is one. | |
8868 | */ | |
8869 | if (!path->locks[level]) { | |
8870 | BUG_ON(level == 0); | |
8871 | btrfs_tree_lock(eb); | |
8872 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8873 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8874 | |
2ff7e61e | 8875 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8876 | eb->start, level, 1, |
2c47e605 YZ |
8877 | &wc->refs[level], |
8878 | &wc->flags[level]); | |
79787eaa JM |
8879 | if (ret < 0) { |
8880 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 8881 | path->locks[level] = 0; |
79787eaa JM |
8882 | return ret; |
8883 | } | |
2c47e605 YZ |
8884 | BUG_ON(wc->refs[level] == 0); |
8885 | if (wc->refs[level] == 1) { | |
bd681513 | 8886 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 8887 | path->locks[level] = 0; |
2c47e605 YZ |
8888 | return 1; |
8889 | } | |
f82d02d9 | 8890 | } |
2c47e605 | 8891 | } |
f82d02d9 | 8892 | |
2c47e605 YZ |
8893 | /* wc->stage == DROP_REFERENCE */ |
8894 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 8895 | |
2c47e605 YZ |
8896 | if (wc->refs[level] == 1) { |
8897 | if (level == 0) { | |
8898 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 8899 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 8900 | else |
e339a6b0 | 8901 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8902 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8903 | ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb); |
1152651a | 8904 | if (ret) { |
0b246afa | 8905 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8906 | "error %d accounting leaf items. Quota is out of sync, rescan required.", |
8907 | ret); | |
1152651a | 8908 | } |
2c47e605 YZ |
8909 | } |
8910 | /* make block locked assertion in clean_tree_block happy */ | |
8911 | if (!path->locks[level] && | |
8912 | btrfs_header_generation(eb) == trans->transid) { | |
8913 | btrfs_tree_lock(eb); | |
8914 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8915 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8916 | } |
7c302b49 | 8917 | clean_tree_block(fs_info, eb); |
2c47e605 YZ |
8918 | } |
8919 | ||
8920 | if (eb == root->node) { | |
8921 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8922 | parent = eb->start; | |
8923 | else | |
8924 | BUG_ON(root->root_key.objectid != | |
8925 | btrfs_header_owner(eb)); | |
8926 | } else { | |
8927 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8928 | parent = path->nodes[level + 1]->start; | |
8929 | else | |
8930 | BUG_ON(root->root_key.objectid != | |
8931 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 8932 | } |
f82d02d9 | 8933 | |
5581a51a | 8934 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
8935 | out: |
8936 | wc->refs[level] = 0; | |
8937 | wc->flags[level] = 0; | |
f0486c68 | 8938 | return 0; |
2c47e605 YZ |
8939 | } |
8940 | ||
8941 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
8942 | struct btrfs_root *root, | |
8943 | struct btrfs_path *path, | |
8944 | struct walk_control *wc) | |
8945 | { | |
2c47e605 | 8946 | int level = wc->level; |
94fcca9f | 8947 | int lookup_info = 1; |
2c47e605 YZ |
8948 | int ret; |
8949 | ||
8950 | while (level >= 0) { | |
94fcca9f | 8951 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
8952 | if (ret > 0) |
8953 | break; | |
8954 | ||
8955 | if (level == 0) | |
8956 | break; | |
8957 | ||
7a7965f8 YZ |
8958 | if (path->slots[level] >= |
8959 | btrfs_header_nritems(path->nodes[level])) | |
8960 | break; | |
8961 | ||
94fcca9f | 8962 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
8963 | if (ret > 0) { |
8964 | path->slots[level]++; | |
8965 | continue; | |
90d2c51d MX |
8966 | } else if (ret < 0) |
8967 | return ret; | |
1c4850e2 | 8968 | level = wc->level; |
f82d02d9 | 8969 | } |
f82d02d9 YZ |
8970 | return 0; |
8971 | } | |
8972 | ||
d397712b | 8973 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 8974 | struct btrfs_root *root, |
f82d02d9 | 8975 | struct btrfs_path *path, |
2c47e605 | 8976 | struct walk_control *wc, int max_level) |
20524f02 | 8977 | { |
2c47e605 | 8978 | int level = wc->level; |
20524f02 | 8979 | int ret; |
9f3a7427 | 8980 | |
2c47e605 YZ |
8981 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
8982 | while (level < max_level && path->nodes[level]) { | |
8983 | wc->level = level; | |
8984 | if (path->slots[level] + 1 < | |
8985 | btrfs_header_nritems(path->nodes[level])) { | |
8986 | path->slots[level]++; | |
20524f02 CM |
8987 | return 0; |
8988 | } else { | |
2c47e605 YZ |
8989 | ret = walk_up_proc(trans, root, path, wc); |
8990 | if (ret > 0) | |
8991 | return 0; | |
bd56b302 | 8992 | |
2c47e605 | 8993 | if (path->locks[level]) { |
bd681513 CM |
8994 | btrfs_tree_unlock_rw(path->nodes[level], |
8995 | path->locks[level]); | |
2c47e605 | 8996 | path->locks[level] = 0; |
f82d02d9 | 8997 | } |
2c47e605 YZ |
8998 | free_extent_buffer(path->nodes[level]); |
8999 | path->nodes[level] = NULL; | |
9000 | level++; | |
20524f02 CM |
9001 | } |
9002 | } | |
9003 | return 1; | |
9004 | } | |
9005 | ||
9aca1d51 | 9006 | /* |
2c47e605 YZ |
9007 | * drop a subvolume tree. |
9008 | * | |
9009 | * this function traverses the tree freeing any blocks that only | |
9010 | * referenced by the tree. | |
9011 | * | |
9012 | * when a shared tree block is found. this function decreases its | |
9013 | * reference count by one. if update_ref is true, this function | |
9014 | * also make sure backrefs for the shared block and all lower level | |
9015 | * blocks are properly updated. | |
9d1a2a3a DS |
9016 | * |
9017 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 9018 | */ |
2c536799 | 9019 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
9020 | struct btrfs_block_rsv *block_rsv, int update_ref, |
9021 | int for_reloc) | |
20524f02 | 9022 | { |
ab8d0fc4 | 9023 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 9024 | struct btrfs_path *path; |
2c47e605 | 9025 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 9026 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 9027 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
9028 | struct walk_control *wc; |
9029 | struct btrfs_key key; | |
9030 | int err = 0; | |
9031 | int ret; | |
9032 | int level; | |
d29a9f62 | 9033 | bool root_dropped = false; |
20524f02 | 9034 | |
ab8d0fc4 | 9035 | btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid); |
1152651a | 9036 | |
5caf2a00 | 9037 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
9038 | if (!path) { |
9039 | err = -ENOMEM; | |
9040 | goto out; | |
9041 | } | |
20524f02 | 9042 | |
2c47e605 | 9043 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
9044 | if (!wc) { |
9045 | btrfs_free_path(path); | |
cb1b69f4 TI |
9046 | err = -ENOMEM; |
9047 | goto out; | |
38a1a919 | 9048 | } |
2c47e605 | 9049 | |
a22285a6 | 9050 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9051 | if (IS_ERR(trans)) { |
9052 | err = PTR_ERR(trans); | |
9053 | goto out_free; | |
9054 | } | |
98d5dc13 | 9055 | |
3fd0a558 YZ |
9056 | if (block_rsv) |
9057 | trans->block_rsv = block_rsv; | |
2c47e605 | 9058 | |
9f3a7427 | 9059 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 9060 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
9061 | path->nodes[level] = btrfs_lock_root_node(root); |
9062 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 9063 | path->slots[level] = 0; |
bd681513 | 9064 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9065 | memset(&wc->update_progress, 0, |
9066 | sizeof(wc->update_progress)); | |
9f3a7427 | 9067 | } else { |
9f3a7427 | 9068 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
9069 | memcpy(&wc->update_progress, &key, |
9070 | sizeof(wc->update_progress)); | |
9071 | ||
6702ed49 | 9072 | level = root_item->drop_level; |
2c47e605 | 9073 | BUG_ON(level == 0); |
6702ed49 | 9074 | path->lowest_level = level; |
2c47e605 YZ |
9075 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
9076 | path->lowest_level = 0; | |
9077 | if (ret < 0) { | |
9078 | err = ret; | |
79787eaa | 9079 | goto out_end_trans; |
9f3a7427 | 9080 | } |
1c4850e2 | 9081 | WARN_ON(ret > 0); |
2c47e605 | 9082 | |
7d9eb12c CM |
9083 | /* |
9084 | * unlock our path, this is safe because only this | |
9085 | * function is allowed to delete this snapshot | |
9086 | */ | |
5d4f98a2 | 9087 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
9088 | |
9089 | level = btrfs_header_level(root->node); | |
9090 | while (1) { | |
9091 | btrfs_tree_lock(path->nodes[level]); | |
9092 | btrfs_set_lock_blocking(path->nodes[level]); | |
fec386ac | 9093 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9094 | |
2ff7e61e | 9095 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 9096 | path->nodes[level]->start, |
3173a18f | 9097 | level, 1, &wc->refs[level], |
2c47e605 | 9098 | &wc->flags[level]); |
79787eaa JM |
9099 | if (ret < 0) { |
9100 | err = ret; | |
9101 | goto out_end_trans; | |
9102 | } | |
2c47e605 YZ |
9103 | BUG_ON(wc->refs[level] == 0); |
9104 | ||
9105 | if (level == root_item->drop_level) | |
9106 | break; | |
9107 | ||
9108 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 9109 | path->locks[level] = 0; |
2c47e605 YZ |
9110 | WARN_ON(wc->refs[level] != 1); |
9111 | level--; | |
9112 | } | |
9f3a7427 | 9113 | } |
2c47e605 YZ |
9114 | |
9115 | wc->level = level; | |
9116 | wc->shared_level = -1; | |
9117 | wc->stage = DROP_REFERENCE; | |
9118 | wc->update_ref = update_ref; | |
9119 | wc->keep_locks = 0; | |
66d7e7f0 | 9120 | wc->for_reloc = for_reloc; |
0b246afa | 9121 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 9122 | |
d397712b | 9123 | while (1) { |
9d1a2a3a | 9124 | |
2c47e605 YZ |
9125 | ret = walk_down_tree(trans, root, path, wc); |
9126 | if (ret < 0) { | |
9127 | err = ret; | |
20524f02 | 9128 | break; |
2c47e605 | 9129 | } |
9aca1d51 | 9130 | |
2c47e605 YZ |
9131 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
9132 | if (ret < 0) { | |
9133 | err = ret; | |
20524f02 | 9134 | break; |
2c47e605 YZ |
9135 | } |
9136 | ||
9137 | if (ret > 0) { | |
9138 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
9139 | break; |
9140 | } | |
2c47e605 YZ |
9141 | |
9142 | if (wc->stage == DROP_REFERENCE) { | |
9143 | level = wc->level; | |
9144 | btrfs_node_key(path->nodes[level], | |
9145 | &root_item->drop_progress, | |
9146 | path->slots[level]); | |
9147 | root_item->drop_level = level; | |
9148 | } | |
9149 | ||
9150 | BUG_ON(wc->level == 0); | |
3a45bb20 | 9151 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 9152 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
9153 | ret = btrfs_update_root(trans, tree_root, |
9154 | &root->root_key, | |
9155 | root_item); | |
79787eaa | 9156 | if (ret) { |
66642832 | 9157 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9158 | err = ret; |
9159 | goto out_end_trans; | |
9160 | } | |
2c47e605 | 9161 | |
3a45bb20 | 9162 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 9163 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
9164 | btrfs_debug(fs_info, |
9165 | "drop snapshot early exit"); | |
3c8f2422 JB |
9166 | err = -EAGAIN; |
9167 | goto out_free; | |
9168 | } | |
9169 | ||
a22285a6 | 9170 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9171 | if (IS_ERR(trans)) { |
9172 | err = PTR_ERR(trans); | |
9173 | goto out_free; | |
9174 | } | |
3fd0a558 YZ |
9175 | if (block_rsv) |
9176 | trans->block_rsv = block_rsv; | |
c3e69d58 | 9177 | } |
20524f02 | 9178 | } |
b3b4aa74 | 9179 | btrfs_release_path(path); |
79787eaa JM |
9180 | if (err) |
9181 | goto out_end_trans; | |
2c47e605 | 9182 | |
1cd5447e | 9183 | ret = btrfs_del_root(trans, fs_info, &root->root_key); |
79787eaa | 9184 | if (ret) { |
66642832 | 9185 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 9186 | err = ret; |
79787eaa JM |
9187 | goto out_end_trans; |
9188 | } | |
2c47e605 | 9189 | |
76dda93c | 9190 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
9191 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
9192 | NULL, NULL); | |
79787eaa | 9193 | if (ret < 0) { |
66642832 | 9194 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9195 | err = ret; |
9196 | goto out_end_trans; | |
9197 | } else if (ret > 0) { | |
84cd948c JB |
9198 | /* if we fail to delete the orphan item this time |
9199 | * around, it'll get picked up the next time. | |
9200 | * | |
9201 | * The most common failure here is just -ENOENT. | |
9202 | */ | |
9203 | btrfs_del_orphan_item(trans, tree_root, | |
9204 | root->root_key.objectid); | |
76dda93c YZ |
9205 | } |
9206 | } | |
9207 | ||
27cdeb70 | 9208 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 9209 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
9210 | } else { |
9211 | free_extent_buffer(root->node); | |
9212 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 9213 | btrfs_put_fs_root(root); |
76dda93c | 9214 | } |
d29a9f62 | 9215 | root_dropped = true; |
79787eaa | 9216 | out_end_trans: |
3a45bb20 | 9217 | btrfs_end_transaction_throttle(trans); |
79787eaa | 9218 | out_free: |
2c47e605 | 9219 | kfree(wc); |
5caf2a00 | 9220 | btrfs_free_path(path); |
cb1b69f4 | 9221 | out: |
d29a9f62 JB |
9222 | /* |
9223 | * So if we need to stop dropping the snapshot for whatever reason we | |
9224 | * need to make sure to add it back to the dead root list so that we | |
9225 | * keep trying to do the work later. This also cleans up roots if we | |
9226 | * don't have it in the radix (like when we recover after a power fail | |
9227 | * or unmount) so we don't leak memory. | |
9228 | */ | |
897ca819 | 9229 | if (!for_reloc && !root_dropped) |
d29a9f62 | 9230 | btrfs_add_dead_root(root); |
90515e7f | 9231 | if (err && err != -EAGAIN) |
ab8d0fc4 | 9232 | btrfs_handle_fs_error(fs_info, err, NULL); |
2c536799 | 9233 | return err; |
20524f02 | 9234 | } |
9078a3e1 | 9235 | |
2c47e605 YZ |
9236 | /* |
9237 | * drop subtree rooted at tree block 'node'. | |
9238 | * | |
9239 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 9240 | * only used by relocation code |
2c47e605 | 9241 | */ |
f82d02d9 YZ |
9242 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
9243 | struct btrfs_root *root, | |
9244 | struct extent_buffer *node, | |
9245 | struct extent_buffer *parent) | |
9246 | { | |
0b246afa | 9247 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 9248 | struct btrfs_path *path; |
2c47e605 | 9249 | struct walk_control *wc; |
f82d02d9 YZ |
9250 | int level; |
9251 | int parent_level; | |
9252 | int ret = 0; | |
9253 | int wret; | |
9254 | ||
2c47e605 YZ |
9255 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
9256 | ||
f82d02d9 | 9257 | path = btrfs_alloc_path(); |
db5b493a TI |
9258 | if (!path) |
9259 | return -ENOMEM; | |
f82d02d9 | 9260 | |
2c47e605 | 9261 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
9262 | if (!wc) { |
9263 | btrfs_free_path(path); | |
9264 | return -ENOMEM; | |
9265 | } | |
2c47e605 | 9266 | |
b9447ef8 | 9267 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
9268 | parent_level = btrfs_header_level(parent); |
9269 | extent_buffer_get(parent); | |
9270 | path->nodes[parent_level] = parent; | |
9271 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9272 | ||
b9447ef8 | 9273 | btrfs_assert_tree_locked(node); |
f82d02d9 | 9274 | level = btrfs_header_level(node); |
f82d02d9 YZ |
9275 | path->nodes[level] = node; |
9276 | path->slots[level] = 0; | |
bd681513 | 9277 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9278 | |
9279 | wc->refs[parent_level] = 1; | |
9280 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9281 | wc->level = level; | |
9282 | wc->shared_level = -1; | |
9283 | wc->stage = DROP_REFERENCE; | |
9284 | wc->update_ref = 0; | |
9285 | wc->keep_locks = 1; | |
66d7e7f0 | 9286 | wc->for_reloc = 1; |
0b246afa | 9287 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
9288 | |
9289 | while (1) { | |
2c47e605 YZ |
9290 | wret = walk_down_tree(trans, root, path, wc); |
9291 | if (wret < 0) { | |
f82d02d9 | 9292 | ret = wret; |
f82d02d9 | 9293 | break; |
2c47e605 | 9294 | } |
f82d02d9 | 9295 | |
2c47e605 | 9296 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9297 | if (wret < 0) |
9298 | ret = wret; | |
9299 | if (wret != 0) | |
9300 | break; | |
9301 | } | |
9302 | ||
2c47e605 | 9303 | kfree(wc); |
f82d02d9 YZ |
9304 | btrfs_free_path(path); |
9305 | return ret; | |
9306 | } | |
9307 | ||
6202df69 | 9308 | static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c CM |
9309 | { |
9310 | u64 num_devices; | |
fc67c450 | 9311 | u64 stripped; |
e4d8ec0f | 9312 | |
fc67c450 ID |
9313 | /* |
9314 | * if restripe for this chunk_type is on pick target profile and | |
9315 | * return, otherwise do the usual balance | |
9316 | */ | |
6202df69 | 9317 | stripped = get_restripe_target(fs_info, flags); |
fc67c450 ID |
9318 | if (stripped) |
9319 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9320 | |
6202df69 | 9321 | num_devices = fs_info->fs_devices->rw_devices; |
cd02dca5 | 9322 | |
fc67c450 | 9323 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 9324 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
9325 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
9326 | ||
ec44a35c CM |
9327 | if (num_devices == 1) { |
9328 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9329 | stripped = flags & ~stripped; | |
9330 | ||
9331 | /* turn raid0 into single device chunks */ | |
9332 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9333 | return stripped; | |
9334 | ||
9335 | /* turn mirroring into duplication */ | |
9336 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9337 | BTRFS_BLOCK_GROUP_RAID10)) | |
9338 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9339 | } else { |
9340 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9341 | if (flags & stripped) |
9342 | return flags; | |
9343 | ||
9344 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9345 | stripped = flags & ~stripped; | |
9346 | ||
9347 | /* switch duplicated blocks with raid1 */ | |
9348 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9349 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9350 | ||
e3176ca2 | 9351 | /* this is drive concat, leave it alone */ |
ec44a35c | 9352 | } |
e3176ca2 | 9353 | |
ec44a35c CM |
9354 | return flags; |
9355 | } | |
9356 | ||
868f401a | 9357 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9358 | { |
f0486c68 YZ |
9359 | struct btrfs_space_info *sinfo = cache->space_info; |
9360 | u64 num_bytes; | |
199c36ea | 9361 | u64 min_allocable_bytes; |
f0486c68 | 9362 | int ret = -ENOSPC; |
0ef3e66b | 9363 | |
199c36ea MX |
9364 | /* |
9365 | * We need some metadata space and system metadata space for | |
9366 | * allocating chunks in some corner cases until we force to set | |
9367 | * it to be readonly. | |
9368 | */ | |
9369 | if ((sinfo->flags & | |
9370 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9371 | !force) | |
ee22184b | 9372 | min_allocable_bytes = SZ_1M; |
199c36ea MX |
9373 | else |
9374 | min_allocable_bytes = 0; | |
9375 | ||
f0486c68 YZ |
9376 | spin_lock(&sinfo->lock); |
9377 | spin_lock(&cache->lock); | |
61cfea9b W |
9378 | |
9379 | if (cache->ro) { | |
868f401a | 9380 | cache->ro++; |
61cfea9b W |
9381 | ret = 0; |
9382 | goto out; | |
9383 | } | |
9384 | ||
f0486c68 YZ |
9385 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9386 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
9387 | ||
4136135b | 9388 | if (btrfs_space_info_used(sinfo, true) + num_bytes + |
37be25bc | 9389 | min_allocable_bytes <= sinfo->total_bytes) { |
f0486c68 | 9390 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9391 | cache->ro++; |
633c0aad | 9392 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9393 | ret = 0; |
9394 | } | |
61cfea9b | 9395 | out: |
f0486c68 YZ |
9396 | spin_unlock(&cache->lock); |
9397 | spin_unlock(&sinfo->lock); | |
9398 | return ret; | |
9399 | } | |
7d9eb12c | 9400 | |
5e00f193 | 9401 | int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info, |
f0486c68 | 9402 | struct btrfs_block_group_cache *cache) |
c286ac48 | 9403 | |
f0486c68 YZ |
9404 | { |
9405 | struct btrfs_trans_handle *trans; | |
9406 | u64 alloc_flags; | |
9407 | int ret; | |
7d9eb12c | 9408 | |
1bbc621e | 9409 | again: |
5e00f193 | 9410 | trans = btrfs_join_transaction(fs_info->extent_root); |
79787eaa JM |
9411 | if (IS_ERR(trans)) |
9412 | return PTR_ERR(trans); | |
5d4f98a2 | 9413 | |
1bbc621e CM |
9414 | /* |
9415 | * we're not allowed to set block groups readonly after the dirty | |
9416 | * block groups cache has started writing. If it already started, | |
9417 | * back off and let this transaction commit | |
9418 | */ | |
0b246afa | 9419 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c | 9420 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9421 | u64 transid = trans->transid; |
9422 | ||
0b246afa | 9423 | mutex_unlock(&fs_info->ro_block_group_mutex); |
3a45bb20 | 9424 | btrfs_end_transaction(trans); |
1bbc621e | 9425 | |
2ff7e61e | 9426 | ret = btrfs_wait_for_commit(fs_info, transid); |
1bbc621e CM |
9427 | if (ret) |
9428 | return ret; | |
9429 | goto again; | |
9430 | } | |
9431 | ||
153c35b6 CM |
9432 | /* |
9433 | * if we are changing raid levels, try to allocate a corresponding | |
9434 | * block group with the new raid level. | |
9435 | */ | |
0b246afa | 9436 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
153c35b6 | 9437 | if (alloc_flags != cache->flags) { |
2ff7e61e | 9438 | ret = do_chunk_alloc(trans, fs_info, alloc_flags, |
153c35b6 CM |
9439 | CHUNK_ALLOC_FORCE); |
9440 | /* | |
9441 | * ENOSPC is allowed here, we may have enough space | |
9442 | * already allocated at the new raid level to | |
9443 | * carry on | |
9444 | */ | |
9445 | if (ret == -ENOSPC) | |
9446 | ret = 0; | |
9447 | if (ret < 0) | |
9448 | goto out; | |
9449 | } | |
1bbc621e | 9450 | |
868f401a | 9451 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9452 | if (!ret) |
9453 | goto out; | |
2ff7e61e JM |
9454 | alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags); |
9455 | ret = do_chunk_alloc(trans, fs_info, alloc_flags, | |
0e4f8f88 | 9456 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9457 | if (ret < 0) |
9458 | goto out; | |
868f401a | 9459 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9460 | out: |
2f081088 | 9461 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 9462 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
34441361 | 9463 | mutex_lock(&fs_info->chunk_mutex); |
2ff7e61e | 9464 | check_system_chunk(trans, fs_info, alloc_flags); |
34441361 | 9465 | mutex_unlock(&fs_info->chunk_mutex); |
2f081088 | 9466 | } |
0b246afa | 9467 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2f081088 | 9468 | |
3a45bb20 | 9469 | btrfs_end_transaction(trans); |
f0486c68 YZ |
9470 | return ret; |
9471 | } | |
5d4f98a2 | 9472 | |
c87f08ca | 9473 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 9474 | struct btrfs_fs_info *fs_info, u64 type) |
c87f08ca | 9475 | { |
2ff7e61e JM |
9476 | u64 alloc_flags = get_alloc_profile(fs_info, type); |
9477 | ||
9478 | return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE); | |
c87f08ca CM |
9479 | } |
9480 | ||
6d07bcec MX |
9481 | /* |
9482 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9483 | * space_info. takes mirrors into account. |
6d07bcec | 9484 | */ |
633c0aad | 9485 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9486 | { |
9487 | struct btrfs_block_group_cache *block_group; | |
9488 | u64 free_bytes = 0; | |
9489 | int factor; | |
9490 | ||
01327610 | 9491 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
9492 | if (list_empty(&sinfo->ro_bgs)) |
9493 | return 0; | |
9494 | ||
9495 | spin_lock(&sinfo->lock); | |
9496 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9497 | spin_lock(&block_group->lock); |
9498 | ||
9499 | if (!block_group->ro) { | |
9500 | spin_unlock(&block_group->lock); | |
9501 | continue; | |
9502 | } | |
9503 | ||
9504 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9505 | BTRFS_BLOCK_GROUP_RAID10 | | |
9506 | BTRFS_BLOCK_GROUP_DUP)) | |
9507 | factor = 2; | |
9508 | else | |
9509 | factor = 1; | |
9510 | ||
9511 | free_bytes += (block_group->key.offset - | |
9512 | btrfs_block_group_used(&block_group->item)) * | |
9513 | factor; | |
9514 | ||
9515 | spin_unlock(&block_group->lock); | |
9516 | } | |
6d07bcec MX |
9517 | spin_unlock(&sinfo->lock); |
9518 | ||
9519 | return free_bytes; | |
9520 | } | |
9521 | ||
2ff7e61e | 9522 | void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9523 | { |
f0486c68 YZ |
9524 | struct btrfs_space_info *sinfo = cache->space_info; |
9525 | u64 num_bytes; | |
9526 | ||
9527 | BUG_ON(!cache->ro); | |
9528 | ||
9529 | spin_lock(&sinfo->lock); | |
9530 | spin_lock(&cache->lock); | |
868f401a Z |
9531 | if (!--cache->ro) { |
9532 | num_bytes = cache->key.offset - cache->reserved - | |
9533 | cache->pinned - cache->bytes_super - | |
9534 | btrfs_block_group_used(&cache->item); | |
9535 | sinfo->bytes_readonly -= num_bytes; | |
9536 | list_del_init(&cache->ro_list); | |
9537 | } | |
f0486c68 YZ |
9538 | spin_unlock(&cache->lock); |
9539 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9540 | } |
9541 | ||
ba1bf481 JB |
9542 | /* |
9543 | * checks to see if its even possible to relocate this block group. | |
9544 | * | |
9545 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9546 | * ok to go ahead and try. | |
9547 | */ | |
6bccf3ab | 9548 | int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr) |
1a40e23b | 9549 | { |
6bccf3ab | 9550 | struct btrfs_root *root = fs_info->extent_root; |
ba1bf481 JB |
9551 | struct btrfs_block_group_cache *block_group; |
9552 | struct btrfs_space_info *space_info; | |
0b246afa | 9553 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
ba1bf481 | 9554 | struct btrfs_device *device; |
6df9a95e | 9555 | struct btrfs_trans_handle *trans; |
cdcb725c | 9556 | u64 min_free; |
6719db6a JB |
9557 | u64 dev_min = 1; |
9558 | u64 dev_nr = 0; | |
4a5e98f5 | 9559 | u64 target; |
0305bc27 | 9560 | int debug; |
cdcb725c | 9561 | int index; |
ba1bf481 JB |
9562 | int full = 0; |
9563 | int ret = 0; | |
1a40e23b | 9564 | |
0b246afa | 9565 | debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG); |
0305bc27 | 9566 | |
0b246afa | 9567 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
1a40e23b | 9568 | |
ba1bf481 | 9569 | /* odd, couldn't find the block group, leave it alone */ |
0305bc27 QW |
9570 | if (!block_group) { |
9571 | if (debug) | |
0b246afa | 9572 | btrfs_warn(fs_info, |
0305bc27 QW |
9573 | "can't find block group for bytenr %llu", |
9574 | bytenr); | |
ba1bf481 | 9575 | return -1; |
0305bc27 | 9576 | } |
1a40e23b | 9577 | |
cdcb725c | 9578 | min_free = btrfs_block_group_used(&block_group->item); |
9579 | ||
ba1bf481 | 9580 | /* no bytes used, we're good */ |
cdcb725c | 9581 | if (!min_free) |
1a40e23b ZY |
9582 | goto out; |
9583 | ||
ba1bf481 JB |
9584 | space_info = block_group->space_info; |
9585 | spin_lock(&space_info->lock); | |
17d217fe | 9586 | |
ba1bf481 | 9587 | full = space_info->full; |
17d217fe | 9588 | |
ba1bf481 JB |
9589 | /* |
9590 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9591 | * relocate it unless we're able to allocate a new chunk below. |
9592 | * | |
9593 | * Otherwise, we need to make sure we have room in the space to handle | |
9594 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9595 | */ |
7ce618db | 9596 | if ((space_info->total_bytes != block_group->key.offset) && |
4136135b LB |
9597 | (btrfs_space_info_used(space_info, false) + min_free < |
9598 | space_info->total_bytes)) { | |
ba1bf481 JB |
9599 | spin_unlock(&space_info->lock); |
9600 | goto out; | |
17d217fe | 9601 | } |
ba1bf481 | 9602 | spin_unlock(&space_info->lock); |
ea8c2819 | 9603 | |
ba1bf481 JB |
9604 | /* |
9605 | * ok we don't have enough space, but maybe we have free space on our | |
9606 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9607 | * alloc devices and guess if we have enough space. if this block |
9608 | * group is going to be restriped, run checks against the target | |
9609 | * profile instead of the current one. | |
ba1bf481 JB |
9610 | */ |
9611 | ret = -1; | |
ea8c2819 | 9612 | |
cdcb725c | 9613 | /* |
9614 | * index: | |
9615 | * 0: raid10 | |
9616 | * 1: raid1 | |
9617 | * 2: dup | |
9618 | * 3: raid0 | |
9619 | * 4: single | |
9620 | */ | |
0b246afa | 9621 | target = get_restripe_target(fs_info, block_group->flags); |
4a5e98f5 | 9622 | if (target) { |
3e72ee88 | 9623 | index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9624 | } else { |
9625 | /* | |
9626 | * this is just a balance, so if we were marked as full | |
9627 | * we know there is no space for a new chunk | |
9628 | */ | |
0305bc27 QW |
9629 | if (full) { |
9630 | if (debug) | |
0b246afa JM |
9631 | btrfs_warn(fs_info, |
9632 | "no space to alloc new chunk for block group %llu", | |
9633 | block_group->key.objectid); | |
4a5e98f5 | 9634 | goto out; |
0305bc27 | 9635 | } |
4a5e98f5 | 9636 | |
3e72ee88 | 9637 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
4a5e98f5 ID |
9638 | } |
9639 | ||
e6ec716f | 9640 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9641 | dev_min = 4; |
6719db6a JB |
9642 | /* Divide by 2 */ |
9643 | min_free >>= 1; | |
e6ec716f | 9644 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9645 | dev_min = 2; |
e6ec716f | 9646 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9647 | /* Multiply by 2 */ |
9648 | min_free <<= 1; | |
e6ec716f | 9649 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9650 | dev_min = fs_devices->rw_devices; |
47c5713f | 9651 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9652 | } |
9653 | ||
6df9a95e JB |
9654 | /* We need to do this so that we can look at pending chunks */ |
9655 | trans = btrfs_join_transaction(root); | |
9656 | if (IS_ERR(trans)) { | |
9657 | ret = PTR_ERR(trans); | |
9658 | goto out; | |
9659 | } | |
9660 | ||
0b246afa | 9661 | mutex_lock(&fs_info->chunk_mutex); |
ba1bf481 | 9662 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
7bfc837d | 9663 | u64 dev_offset; |
56bec294 | 9664 | |
ba1bf481 JB |
9665 | /* |
9666 | * check to make sure we can actually find a chunk with enough | |
9667 | * space to fit our block group in. | |
9668 | */ | |
63a212ab | 9669 | if (device->total_bytes > device->bytes_used + min_free && |
401e29c1 | 9670 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
6df9a95e | 9671 | ret = find_free_dev_extent(trans, device, min_free, |
7bfc837d | 9672 | &dev_offset, NULL); |
ba1bf481 | 9673 | if (!ret) |
cdcb725c | 9674 | dev_nr++; |
9675 | ||
9676 | if (dev_nr >= dev_min) | |
73e48b27 | 9677 | break; |
cdcb725c | 9678 | |
ba1bf481 | 9679 | ret = -1; |
725c8463 | 9680 | } |
edbd8d4e | 9681 | } |
0305bc27 | 9682 | if (debug && ret == -1) |
0b246afa JM |
9683 | btrfs_warn(fs_info, |
9684 | "no space to allocate a new chunk for block group %llu", | |
9685 | block_group->key.objectid); | |
9686 | mutex_unlock(&fs_info->chunk_mutex); | |
3a45bb20 | 9687 | btrfs_end_transaction(trans); |
edbd8d4e | 9688 | out: |
ba1bf481 | 9689 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9690 | return ret; |
9691 | } | |
9692 | ||
6bccf3ab JM |
9693 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
9694 | struct btrfs_path *path, | |
9695 | struct btrfs_key *key) | |
0b86a832 | 9696 | { |
6bccf3ab | 9697 | struct btrfs_root *root = fs_info->extent_root; |
925baedd | 9698 | int ret = 0; |
0b86a832 CM |
9699 | struct btrfs_key found_key; |
9700 | struct extent_buffer *leaf; | |
9701 | int slot; | |
edbd8d4e | 9702 | |
0b86a832 CM |
9703 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9704 | if (ret < 0) | |
925baedd CM |
9705 | goto out; |
9706 | ||
d397712b | 9707 | while (1) { |
0b86a832 | 9708 | slot = path->slots[0]; |
edbd8d4e | 9709 | leaf = path->nodes[0]; |
0b86a832 CM |
9710 | if (slot >= btrfs_header_nritems(leaf)) { |
9711 | ret = btrfs_next_leaf(root, path); | |
9712 | if (ret == 0) | |
9713 | continue; | |
9714 | if (ret < 0) | |
925baedd | 9715 | goto out; |
0b86a832 | 9716 | break; |
edbd8d4e | 9717 | } |
0b86a832 | 9718 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9719 | |
0b86a832 | 9720 | if (found_key.objectid >= key->objectid && |
925baedd | 9721 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6fb37b75 LB |
9722 | struct extent_map_tree *em_tree; |
9723 | struct extent_map *em; | |
9724 | ||
9725 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
9726 | read_lock(&em_tree->lock); | |
9727 | em = lookup_extent_mapping(em_tree, found_key.objectid, | |
9728 | found_key.offset); | |
9729 | read_unlock(&em_tree->lock); | |
9730 | if (!em) { | |
0b246afa | 9731 | btrfs_err(fs_info, |
6fb37b75 LB |
9732 | "logical %llu len %llu found bg but no related chunk", |
9733 | found_key.objectid, found_key.offset); | |
9734 | ret = -ENOENT; | |
9735 | } else { | |
9736 | ret = 0; | |
9737 | } | |
187ee58c | 9738 | free_extent_map(em); |
925baedd CM |
9739 | goto out; |
9740 | } | |
0b86a832 | 9741 | path->slots[0]++; |
edbd8d4e | 9742 | } |
925baedd | 9743 | out: |
0b86a832 | 9744 | return ret; |
edbd8d4e CM |
9745 | } |
9746 | ||
0af3d00b JB |
9747 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
9748 | { | |
9749 | struct btrfs_block_group_cache *block_group; | |
9750 | u64 last = 0; | |
9751 | ||
9752 | while (1) { | |
9753 | struct inode *inode; | |
9754 | ||
9755 | block_group = btrfs_lookup_first_block_group(info, last); | |
9756 | while (block_group) { | |
9757 | spin_lock(&block_group->lock); | |
9758 | if (block_group->iref) | |
9759 | break; | |
9760 | spin_unlock(&block_group->lock); | |
2ff7e61e | 9761 | block_group = next_block_group(info, block_group); |
0af3d00b JB |
9762 | } |
9763 | if (!block_group) { | |
9764 | if (last == 0) | |
9765 | break; | |
9766 | last = 0; | |
9767 | continue; | |
9768 | } | |
9769 | ||
9770 | inode = block_group->inode; | |
9771 | block_group->iref = 0; | |
9772 | block_group->inode = NULL; | |
9773 | spin_unlock(&block_group->lock); | |
f3bca802 | 9774 | ASSERT(block_group->io_ctl.inode == NULL); |
0af3d00b JB |
9775 | iput(inode); |
9776 | last = block_group->key.objectid + block_group->key.offset; | |
9777 | btrfs_put_block_group(block_group); | |
9778 | } | |
9779 | } | |
9780 | ||
5cdd7db6 FM |
9781 | /* |
9782 | * Must be called only after stopping all workers, since we could have block | |
9783 | * group caching kthreads running, and therefore they could race with us if we | |
9784 | * freed the block groups before stopping them. | |
9785 | */ | |
1a40e23b ZY |
9786 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9787 | { | |
9788 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 9789 | struct btrfs_space_info *space_info; |
11833d66 | 9790 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
9791 | struct rb_node *n; |
9792 | ||
9e351cc8 | 9793 | down_write(&info->commit_root_sem); |
11833d66 YZ |
9794 | while (!list_empty(&info->caching_block_groups)) { |
9795 | caching_ctl = list_entry(info->caching_block_groups.next, | |
9796 | struct btrfs_caching_control, list); | |
9797 | list_del(&caching_ctl->list); | |
9798 | put_caching_control(caching_ctl); | |
9799 | } | |
9e351cc8 | 9800 | up_write(&info->commit_root_sem); |
11833d66 | 9801 | |
47ab2a6c JB |
9802 | spin_lock(&info->unused_bgs_lock); |
9803 | while (!list_empty(&info->unused_bgs)) { | |
9804 | block_group = list_first_entry(&info->unused_bgs, | |
9805 | struct btrfs_block_group_cache, | |
9806 | bg_list); | |
9807 | list_del_init(&block_group->bg_list); | |
9808 | btrfs_put_block_group(block_group); | |
9809 | } | |
9810 | spin_unlock(&info->unused_bgs_lock); | |
9811 | ||
1a40e23b ZY |
9812 | spin_lock(&info->block_group_cache_lock); |
9813 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
9814 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
9815 | cache_node); | |
1a40e23b ZY |
9816 | rb_erase(&block_group->cache_node, |
9817 | &info->block_group_cache_tree); | |
01eacb27 | 9818 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
9819 | spin_unlock(&info->block_group_cache_lock); |
9820 | ||
80eb234a | 9821 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 9822 | list_del(&block_group->list); |
80eb234a | 9823 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 9824 | |
3c14874a JB |
9825 | /* |
9826 | * We haven't cached this block group, which means we could | |
9827 | * possibly have excluded extents on this block group. | |
9828 | */ | |
36cce922 JB |
9829 | if (block_group->cached == BTRFS_CACHE_NO || |
9830 | block_group->cached == BTRFS_CACHE_ERROR) | |
2ff7e61e | 9831 | free_excluded_extents(info, block_group); |
3c14874a | 9832 | |
817d52f8 | 9833 | btrfs_remove_free_space_cache(block_group); |
5cdd7db6 | 9834 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); |
f3bca802 LB |
9835 | ASSERT(list_empty(&block_group->dirty_list)); |
9836 | ASSERT(list_empty(&block_group->io_list)); | |
9837 | ASSERT(list_empty(&block_group->bg_list)); | |
9838 | ASSERT(atomic_read(&block_group->count) == 1); | |
11dfe35a | 9839 | btrfs_put_block_group(block_group); |
d899e052 YZ |
9840 | |
9841 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
9842 | } |
9843 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
9844 | |
9845 | /* now that all the block groups are freed, go through and | |
9846 | * free all the space_info structs. This is only called during | |
9847 | * the final stages of unmount, and so we know nobody is | |
9848 | * using them. We call synchronize_rcu() once before we start, | |
9849 | * just to be on the safe side. | |
9850 | */ | |
9851 | synchronize_rcu(); | |
9852 | ||
8929ecfa YZ |
9853 | release_global_block_rsv(info); |
9854 | ||
67871254 | 9855 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
9856 | int i; |
9857 | ||
4184ea7f CM |
9858 | space_info = list_entry(info->space_info.next, |
9859 | struct btrfs_space_info, | |
9860 | list); | |
d555b6c3 JB |
9861 | |
9862 | /* | |
9863 | * Do not hide this behind enospc_debug, this is actually | |
9864 | * important and indicates a real bug if this happens. | |
9865 | */ | |
9866 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
b069e0c3 | 9867 | space_info->bytes_reserved > 0 || |
d555b6c3 | 9868 | space_info->bytes_may_use > 0)) |
ab8d0fc4 | 9869 | dump_space_info(info, space_info, 0, 0); |
4184ea7f | 9870 | list_del(&space_info->list); |
6ab0a202 JM |
9871 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
9872 | struct kobject *kobj; | |
c1895442 JM |
9873 | kobj = space_info->block_group_kobjs[i]; |
9874 | space_info->block_group_kobjs[i] = NULL; | |
9875 | if (kobj) { | |
6ab0a202 JM |
9876 | kobject_del(kobj); |
9877 | kobject_put(kobj); | |
9878 | } | |
9879 | } | |
9880 | kobject_del(&space_info->kobj); | |
9881 | kobject_put(&space_info->kobj); | |
4184ea7f | 9882 | } |
1a40e23b ZY |
9883 | return 0; |
9884 | } | |
9885 | ||
c434d21c | 9886 | static void link_block_group(struct btrfs_block_group_cache *cache) |
b742bb82 | 9887 | { |
c434d21c | 9888 | struct btrfs_space_info *space_info = cache->space_info; |
3e72ee88 | 9889 | int index = btrfs_bg_flags_to_raid_index(cache->flags); |
ed55b6ac | 9890 | bool first = false; |
b742bb82 YZ |
9891 | |
9892 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
9893 | if (list_empty(&space_info->block_groups[index])) |
9894 | first = true; | |
9895 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
9896 | up_write(&space_info->groups_sem); | |
9897 | ||
9898 | if (first) { | |
c1895442 | 9899 | struct raid_kobject *rkobj; |
6ab0a202 JM |
9900 | int ret; |
9901 | ||
c1895442 JM |
9902 | rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
9903 | if (!rkobj) | |
9904 | goto out_err; | |
9905 | rkobj->raid_type = index; | |
9906 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
9907 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
9908 | "%s", get_raid_name(index)); | |
6ab0a202 | 9909 | if (ret) { |
c1895442 JM |
9910 | kobject_put(&rkobj->kobj); |
9911 | goto out_err; | |
6ab0a202 | 9912 | } |
c1895442 | 9913 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 9914 | } |
c1895442 JM |
9915 | |
9916 | return; | |
9917 | out_err: | |
ab8d0fc4 JM |
9918 | btrfs_warn(cache->fs_info, |
9919 | "failed to add kobject for block cache, ignoring"); | |
b742bb82 YZ |
9920 | } |
9921 | ||
920e4a58 | 9922 | static struct btrfs_block_group_cache * |
2ff7e61e JM |
9923 | btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info, |
9924 | u64 start, u64 size) | |
920e4a58 MX |
9925 | { |
9926 | struct btrfs_block_group_cache *cache; | |
9927 | ||
9928 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
9929 | if (!cache) | |
9930 | return NULL; | |
9931 | ||
9932 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
9933 | GFP_NOFS); | |
9934 | if (!cache->free_space_ctl) { | |
9935 | kfree(cache); | |
9936 | return NULL; | |
9937 | } | |
9938 | ||
9939 | cache->key.objectid = start; | |
9940 | cache->key.offset = size; | |
9941 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
9942 | ||
0b246afa | 9943 | cache->fs_info = fs_info; |
e4ff5fb5 | 9944 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); |
1e144fb8 OS |
9945 | set_free_space_tree_thresholds(cache); |
9946 | ||
920e4a58 MX |
9947 | atomic_set(&cache->count, 1); |
9948 | spin_lock_init(&cache->lock); | |
e570fd27 | 9949 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
9950 | INIT_LIST_HEAD(&cache->list); |
9951 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 9952 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 9953 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 9954 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 9955 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 9956 | btrfs_init_free_space_ctl(cache); |
04216820 | 9957 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 9958 | mutex_init(&cache->free_space_lock); |
0966a7b1 | 9959 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); |
920e4a58 MX |
9960 | |
9961 | return cache; | |
9962 | } | |
9963 | ||
5b4aacef | 9964 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
9078a3e1 CM |
9965 | { |
9966 | struct btrfs_path *path; | |
9967 | int ret; | |
9078a3e1 | 9968 | struct btrfs_block_group_cache *cache; |
6324fbf3 | 9969 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
9970 | struct btrfs_key key; |
9971 | struct btrfs_key found_key; | |
5f39d397 | 9972 | struct extent_buffer *leaf; |
0af3d00b JB |
9973 | int need_clear = 0; |
9974 | u64 cache_gen; | |
49303381 LB |
9975 | u64 feature; |
9976 | int mixed; | |
9977 | ||
9978 | feature = btrfs_super_incompat_flags(info->super_copy); | |
9979 | mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS); | |
96b5179d | 9980 | |
9078a3e1 | 9981 | key.objectid = 0; |
0b86a832 | 9982 | key.offset = 0; |
962a298f | 9983 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
9984 | path = btrfs_alloc_path(); |
9985 | if (!path) | |
9986 | return -ENOMEM; | |
e4058b54 | 9987 | path->reada = READA_FORWARD; |
9078a3e1 | 9988 | |
0b246afa JM |
9989 | cache_gen = btrfs_super_cache_generation(info->super_copy); |
9990 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
9991 | btrfs_super_generation(info->super_copy) != cache_gen) | |
0af3d00b | 9992 | need_clear = 1; |
0b246afa | 9993 | if (btrfs_test_opt(info, CLEAR_CACHE)) |
88c2ba3b | 9994 | need_clear = 1; |
0af3d00b | 9995 | |
d397712b | 9996 | while (1) { |
6bccf3ab | 9997 | ret = find_first_block_group(info, path, &key); |
b742bb82 YZ |
9998 | if (ret > 0) |
9999 | break; | |
0b86a832 CM |
10000 | if (ret != 0) |
10001 | goto error; | |
920e4a58 | 10002 | |
5f39d397 CM |
10003 | leaf = path->nodes[0]; |
10004 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 | 10005 | |
2ff7e61e | 10006 | cache = btrfs_create_block_group_cache(info, found_key.objectid, |
920e4a58 | 10007 | found_key.offset); |
9078a3e1 | 10008 | if (!cache) { |
0b86a832 | 10009 | ret = -ENOMEM; |
f0486c68 | 10010 | goto error; |
9078a3e1 | 10011 | } |
96303081 | 10012 | |
cf7c1ef6 LB |
10013 | if (need_clear) { |
10014 | /* | |
10015 | * When we mount with old space cache, we need to | |
10016 | * set BTRFS_DC_CLEAR and set dirty flag. | |
10017 | * | |
10018 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
10019 | * truncate the old free space cache inode and | |
10020 | * setup a new one. | |
10021 | * b) Setting 'dirty flag' makes sure that we flush | |
10022 | * the new space cache info onto disk. | |
10023 | */ | |
0b246afa | 10024 | if (btrfs_test_opt(info, SPACE_CACHE)) |
ce93ec54 | 10025 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 10026 | } |
0af3d00b | 10027 | |
5f39d397 CM |
10028 | read_extent_buffer(leaf, &cache->item, |
10029 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
10030 | sizeof(cache->item)); | |
920e4a58 | 10031 | cache->flags = btrfs_block_group_flags(&cache->item); |
49303381 LB |
10032 | if (!mixed && |
10033 | ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && | |
10034 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
10035 | btrfs_err(info, | |
10036 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
10037 | cache->key.objectid); | |
10038 | ret = -EINVAL; | |
10039 | goto error; | |
10040 | } | |
0b86a832 | 10041 | |
9078a3e1 | 10042 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 10043 | btrfs_release_path(path); |
34d52cb6 | 10044 | |
3c14874a JB |
10045 | /* |
10046 | * We need to exclude the super stripes now so that the space | |
10047 | * info has super bytes accounted for, otherwise we'll think | |
10048 | * we have more space than we actually do. | |
10049 | */ | |
2ff7e61e | 10050 | ret = exclude_super_stripes(info, cache); |
835d974f JB |
10051 | if (ret) { |
10052 | /* | |
10053 | * We may have excluded something, so call this just in | |
10054 | * case. | |
10055 | */ | |
2ff7e61e | 10056 | free_excluded_extents(info, cache); |
920e4a58 | 10057 | btrfs_put_block_group(cache); |
835d974f JB |
10058 | goto error; |
10059 | } | |
3c14874a | 10060 | |
817d52f8 JB |
10061 | /* |
10062 | * check for two cases, either we are full, and therefore | |
10063 | * don't need to bother with the caching work since we won't | |
10064 | * find any space, or we are empty, and we can just add all | |
10065 | * the space in and be done with it. This saves us _alot_ of | |
10066 | * time, particularly in the full case. | |
10067 | */ | |
10068 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 10069 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10070 | cache->cached = BTRFS_CACHE_FINISHED; |
2ff7e61e | 10071 | free_excluded_extents(info, cache); |
817d52f8 | 10072 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 10073 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10074 | cache->cached = BTRFS_CACHE_FINISHED; |
0b246afa | 10075 | add_new_free_space(cache, info, |
817d52f8 JB |
10076 | found_key.objectid, |
10077 | found_key.objectid + | |
10078 | found_key.offset); | |
2ff7e61e | 10079 | free_excluded_extents(info, cache); |
817d52f8 | 10080 | } |
96b5179d | 10081 | |
0b246afa | 10082 | ret = btrfs_add_block_group_cache(info, cache); |
8c579fe7 JB |
10083 | if (ret) { |
10084 | btrfs_remove_free_space_cache(cache); | |
10085 | btrfs_put_block_group(cache); | |
10086 | goto error; | |
10087 | } | |
10088 | ||
0b246afa | 10089 | trace_btrfs_add_block_group(info, cache, 0); |
d2006e6d NB |
10090 | update_space_info(info, cache->flags, found_key.offset, |
10091 | btrfs_block_group_used(&cache->item), | |
10092 | cache->bytes_super, &space_info); | |
8c579fe7 | 10093 | |
6324fbf3 | 10094 | cache->space_info = space_info; |
1b2da372 | 10095 | |
c434d21c | 10096 | link_block_group(cache); |
0f9dd46c | 10097 | |
0b246afa | 10098 | set_avail_alloc_bits(info, cache->flags); |
2ff7e61e | 10099 | if (btrfs_chunk_readonly(info, cache->key.objectid)) { |
868f401a | 10100 | inc_block_group_ro(cache, 1); |
47ab2a6c JB |
10101 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
10102 | spin_lock(&info->unused_bgs_lock); | |
10103 | /* Should always be true but just in case. */ | |
10104 | if (list_empty(&cache->bg_list)) { | |
10105 | btrfs_get_block_group(cache); | |
10106 | list_add_tail(&cache->bg_list, | |
10107 | &info->unused_bgs); | |
10108 | } | |
10109 | spin_unlock(&info->unused_bgs_lock); | |
10110 | } | |
9078a3e1 | 10111 | } |
b742bb82 | 10112 | |
0b246afa | 10113 | list_for_each_entry_rcu(space_info, &info->space_info, list) { |
2ff7e61e | 10114 | if (!(get_alloc_profile(info, space_info->flags) & |
b742bb82 YZ |
10115 | (BTRFS_BLOCK_GROUP_RAID10 | |
10116 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
10117 | BTRFS_BLOCK_GROUP_RAID5 | |
10118 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
10119 | BTRFS_BLOCK_GROUP_DUP))) |
10120 | continue; | |
10121 | /* | |
10122 | * avoid allocating from un-mirrored block group if there are | |
10123 | * mirrored block groups. | |
10124 | */ | |
1095cc0d | 10125 | list_for_each_entry(cache, |
10126 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
10127 | list) | |
868f401a | 10128 | inc_block_group_ro(cache, 1); |
1095cc0d | 10129 | list_for_each_entry(cache, |
10130 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
10131 | list) | |
868f401a | 10132 | inc_block_group_ro(cache, 1); |
9078a3e1 | 10133 | } |
f0486c68 YZ |
10134 | |
10135 | init_global_block_rsv(info); | |
0b86a832 CM |
10136 | ret = 0; |
10137 | error: | |
9078a3e1 | 10138 | btrfs_free_path(path); |
0b86a832 | 10139 | return ret; |
9078a3e1 | 10140 | } |
6324fbf3 | 10141 | |
6c686b35 | 10142 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) |
ea658bad | 10143 | { |
6c686b35 | 10144 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ea658bad | 10145 | struct btrfs_block_group_cache *block_group, *tmp; |
0b246afa | 10146 | struct btrfs_root *extent_root = fs_info->extent_root; |
ea658bad JB |
10147 | struct btrfs_block_group_item item; |
10148 | struct btrfs_key key; | |
10149 | int ret = 0; | |
d9a0540a | 10150 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
ea658bad | 10151 | |
d9a0540a | 10152 | trans->can_flush_pending_bgs = false; |
47ab2a6c | 10153 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { |
ea658bad | 10154 | if (ret) |
c92f6be3 | 10155 | goto next; |
ea658bad JB |
10156 | |
10157 | spin_lock(&block_group->lock); | |
10158 | memcpy(&item, &block_group->item, sizeof(item)); | |
10159 | memcpy(&key, &block_group->key, sizeof(key)); | |
10160 | spin_unlock(&block_group->lock); | |
10161 | ||
10162 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
10163 | sizeof(item)); | |
10164 | if (ret) | |
66642832 | 10165 | btrfs_abort_transaction(trans, ret); |
0b246afa JM |
10166 | ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid, |
10167 | key.offset); | |
6df9a95e | 10168 | if (ret) |
66642832 | 10169 | btrfs_abort_transaction(trans, ret); |
0b246afa | 10170 | add_block_group_free_space(trans, fs_info, block_group); |
1e144fb8 | 10171 | /* already aborted the transaction if it failed. */ |
c92f6be3 FM |
10172 | next: |
10173 | list_del_init(&block_group->bg_list); | |
ea658bad | 10174 | } |
d9a0540a | 10175 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
ea658bad JB |
10176 | } |
10177 | ||
6324fbf3 | 10178 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 10179 | struct btrfs_fs_info *fs_info, u64 bytes_used, |
0174484d | 10180 | u64 type, u64 chunk_offset, u64 size) |
6324fbf3 | 10181 | { |
6324fbf3 | 10182 | struct btrfs_block_group_cache *cache; |
0b246afa | 10183 | int ret; |
6324fbf3 | 10184 | |
0b246afa | 10185 | btrfs_set_log_full_commit(fs_info, trans); |
e02119d5 | 10186 | |
2ff7e61e | 10187 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size); |
0f9dd46c JB |
10188 | if (!cache) |
10189 | return -ENOMEM; | |
34d52cb6 | 10190 | |
6324fbf3 | 10191 | btrfs_set_block_group_used(&cache->item, bytes_used); |
0174484d NB |
10192 | btrfs_set_block_group_chunk_objectid(&cache->item, |
10193 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
6324fbf3 CM |
10194 | btrfs_set_block_group_flags(&cache->item, type); |
10195 | ||
920e4a58 | 10196 | cache->flags = type; |
11833d66 | 10197 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10198 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 10199 | cache->needs_free_space = 1; |
2ff7e61e | 10200 | ret = exclude_super_stripes(fs_info, cache); |
835d974f JB |
10201 | if (ret) { |
10202 | /* | |
10203 | * We may have excluded something, so call this just in | |
10204 | * case. | |
10205 | */ | |
2ff7e61e | 10206 | free_excluded_extents(fs_info, cache); |
920e4a58 | 10207 | btrfs_put_block_group(cache); |
835d974f JB |
10208 | return ret; |
10209 | } | |
96303081 | 10210 | |
0b246afa | 10211 | add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size); |
817d52f8 | 10212 | |
2ff7e61e | 10213 | free_excluded_extents(fs_info, cache); |
11833d66 | 10214 | |
d0bd4560 | 10215 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 10216 | if (btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
10217 | u64 new_bytes_used = size - bytes_used; |
10218 | ||
10219 | bytes_used += new_bytes_used >> 1; | |
2ff7e61e | 10220 | fragment_free_space(cache); |
d0bd4560 JB |
10221 | } |
10222 | #endif | |
2e6e5183 | 10223 | /* |
2be12ef7 NB |
10224 | * Ensure the corresponding space_info object is created and |
10225 | * assigned to our block group. We want our bg to be added to the rbtree | |
10226 | * with its ->space_info set. | |
2e6e5183 | 10227 | */ |
2be12ef7 NB |
10228 | cache->space_info = __find_space_info(fs_info, cache->flags); |
10229 | if (!cache->space_info) { | |
10230 | ret = create_space_info(fs_info, cache->flags, | |
10231 | &cache->space_info); | |
10232 | if (ret) { | |
10233 | btrfs_remove_free_space_cache(cache); | |
10234 | btrfs_put_block_group(cache); | |
10235 | return ret; | |
10236 | } | |
2e6e5183 FM |
10237 | } |
10238 | ||
0b246afa | 10239 | ret = btrfs_add_block_group_cache(fs_info, cache); |
8c579fe7 JB |
10240 | if (ret) { |
10241 | btrfs_remove_free_space_cache(cache); | |
10242 | btrfs_put_block_group(cache); | |
10243 | return ret; | |
10244 | } | |
10245 | ||
2e6e5183 FM |
10246 | /* |
10247 | * Now that our block group has its ->space_info set and is inserted in | |
10248 | * the rbtree, update the space info's counters. | |
10249 | */ | |
0b246afa | 10250 | trace_btrfs_add_block_group(fs_info, cache, 1); |
d2006e6d | 10251 | update_space_info(fs_info, cache->flags, size, bytes_used, |
e40edf2d | 10252 | cache->bytes_super, &cache->space_info); |
0b246afa | 10253 | update_global_block_rsv(fs_info); |
1b2da372 | 10254 | |
c434d21c | 10255 | link_block_group(cache); |
6324fbf3 | 10256 | |
47ab2a6c | 10257 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
6324fbf3 | 10258 | |
0b246afa | 10259 | set_avail_alloc_bits(fs_info, type); |
6324fbf3 CM |
10260 | return 0; |
10261 | } | |
1a40e23b | 10262 | |
10ea00f5 ID |
10263 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
10264 | { | |
899c81ea ID |
10265 | u64 extra_flags = chunk_to_extended(flags) & |
10266 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 10267 | |
de98ced9 | 10268 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10269 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
10270 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10271 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10272 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10273 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10274 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 10275 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10276 | } |
10277 | ||
1a40e23b | 10278 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
6bccf3ab | 10279 | struct btrfs_fs_info *fs_info, u64 group_start, |
04216820 | 10280 | struct extent_map *em) |
1a40e23b | 10281 | { |
6bccf3ab | 10282 | struct btrfs_root *root = fs_info->extent_root; |
1a40e23b ZY |
10283 | struct btrfs_path *path; |
10284 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 10285 | struct btrfs_free_cluster *cluster; |
0b246afa | 10286 | struct btrfs_root *tree_root = fs_info->tree_root; |
1a40e23b | 10287 | struct btrfs_key key; |
0af3d00b | 10288 | struct inode *inode; |
c1895442 | 10289 | struct kobject *kobj = NULL; |
1a40e23b | 10290 | int ret; |
10ea00f5 | 10291 | int index; |
89a55897 | 10292 | int factor; |
4f69cb98 | 10293 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 10294 | bool remove_em; |
1a40e23b | 10295 | |
6bccf3ab | 10296 | block_group = btrfs_lookup_block_group(fs_info, group_start); |
1a40e23b | 10297 | BUG_ON(!block_group); |
c146afad | 10298 | BUG_ON(!block_group->ro); |
1a40e23b | 10299 | |
9f7c43c9 | 10300 | /* |
10301 | * Free the reserved super bytes from this block group before | |
10302 | * remove it. | |
10303 | */ | |
2ff7e61e | 10304 | free_excluded_extents(fs_info, block_group); |
fd708b81 JB |
10305 | btrfs_free_ref_tree_range(fs_info, block_group->key.objectid, |
10306 | block_group->key.offset); | |
9f7c43c9 | 10307 | |
1a40e23b | 10308 | memcpy(&key, &block_group->key, sizeof(key)); |
3e72ee88 | 10309 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
89a55897 JB |
10310 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
10311 | BTRFS_BLOCK_GROUP_RAID1 | | |
10312 | BTRFS_BLOCK_GROUP_RAID10)) | |
10313 | factor = 2; | |
10314 | else | |
10315 | factor = 1; | |
1a40e23b | 10316 | |
44fb5511 | 10317 | /* make sure this block group isn't part of an allocation cluster */ |
0b246afa | 10318 | cluster = &fs_info->data_alloc_cluster; |
44fb5511 CM |
10319 | spin_lock(&cluster->refill_lock); |
10320 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10321 | spin_unlock(&cluster->refill_lock); | |
10322 | ||
10323 | /* | |
10324 | * make sure this block group isn't part of a metadata | |
10325 | * allocation cluster | |
10326 | */ | |
0b246afa | 10327 | cluster = &fs_info->meta_alloc_cluster; |
44fb5511 CM |
10328 | spin_lock(&cluster->refill_lock); |
10329 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10330 | spin_unlock(&cluster->refill_lock); | |
10331 | ||
1a40e23b | 10332 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10333 | if (!path) { |
10334 | ret = -ENOMEM; | |
10335 | goto out; | |
10336 | } | |
1a40e23b | 10337 | |
1bbc621e CM |
10338 | /* |
10339 | * get the inode first so any iput calls done for the io_list | |
10340 | * aren't the final iput (no unlinks allowed now) | |
10341 | */ | |
77ab86bf | 10342 | inode = lookup_free_space_inode(fs_info, block_group, path); |
1bbc621e CM |
10343 | |
10344 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10345 | /* | |
10346 | * make sure our free spache cache IO is done before remove the | |
10347 | * free space inode | |
10348 | */ | |
10349 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10350 | if (!list_empty(&block_group->io_list)) { | |
10351 | list_del_init(&block_group->io_list); | |
10352 | ||
10353 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10354 | ||
10355 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
afdb5718 | 10356 | btrfs_wait_cache_io(trans, block_group, path); |
1bbc621e CM |
10357 | btrfs_put_block_group(block_group); |
10358 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10359 | } | |
10360 | ||
10361 | if (!list_empty(&block_group->dirty_list)) { | |
10362 | list_del_init(&block_group->dirty_list); | |
10363 | btrfs_put_block_group(block_group); | |
10364 | } | |
10365 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10366 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10367 | ||
0af3d00b | 10368 | if (!IS_ERR(inode)) { |
73f2e545 | 10369 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
79787eaa JM |
10370 | if (ret) { |
10371 | btrfs_add_delayed_iput(inode); | |
10372 | goto out; | |
10373 | } | |
0af3d00b JB |
10374 | clear_nlink(inode); |
10375 | /* One for the block groups ref */ | |
10376 | spin_lock(&block_group->lock); | |
10377 | if (block_group->iref) { | |
10378 | block_group->iref = 0; | |
10379 | block_group->inode = NULL; | |
10380 | spin_unlock(&block_group->lock); | |
10381 | iput(inode); | |
10382 | } else { | |
10383 | spin_unlock(&block_group->lock); | |
10384 | } | |
10385 | /* One for our lookup ref */ | |
455757c3 | 10386 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10387 | } |
10388 | ||
10389 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10390 | key.offset = block_group->key.objectid; | |
10391 | key.type = 0; | |
10392 | ||
10393 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10394 | if (ret < 0) | |
10395 | goto out; | |
10396 | if (ret > 0) | |
b3b4aa74 | 10397 | btrfs_release_path(path); |
0af3d00b JB |
10398 | if (ret == 0) { |
10399 | ret = btrfs_del_item(trans, tree_root, path); | |
10400 | if (ret) | |
10401 | goto out; | |
b3b4aa74 | 10402 | btrfs_release_path(path); |
0af3d00b JB |
10403 | } |
10404 | ||
0b246afa | 10405 | spin_lock(&fs_info->block_group_cache_lock); |
1a40e23b | 10406 | rb_erase(&block_group->cache_node, |
0b246afa | 10407 | &fs_info->block_group_cache_tree); |
292cbd51 | 10408 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd | 10409 | |
0b246afa JM |
10410 | if (fs_info->first_logical_byte == block_group->key.objectid) |
10411 | fs_info->first_logical_byte = (u64)-1; | |
10412 | spin_unlock(&fs_info->block_group_cache_lock); | |
817d52f8 | 10413 | |
80eb234a | 10414 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10415 | /* |
10416 | * we must use list_del_init so people can check to see if they | |
10417 | * are still on the list after taking the semaphore | |
10418 | */ | |
10419 | list_del_init(&block_group->list); | |
6ab0a202 | 10420 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10421 | kobj = block_group->space_info->block_group_kobjs[index]; |
10422 | block_group->space_info->block_group_kobjs[index] = NULL; | |
0b246afa | 10423 | clear_avail_alloc_bits(fs_info, block_group->flags); |
6ab0a202 | 10424 | } |
80eb234a | 10425 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
10426 | if (kobj) { |
10427 | kobject_del(kobj); | |
10428 | kobject_put(kobj); | |
10429 | } | |
1a40e23b | 10430 | |
4f69cb98 FM |
10431 | if (block_group->has_caching_ctl) |
10432 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10433 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10434 | wait_block_group_cache_done(block_group); |
4f69cb98 | 10435 | if (block_group->has_caching_ctl) { |
0b246afa | 10436 | down_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10437 | if (!caching_ctl) { |
10438 | struct btrfs_caching_control *ctl; | |
10439 | ||
10440 | list_for_each_entry(ctl, | |
0b246afa | 10441 | &fs_info->caching_block_groups, list) |
4f69cb98 FM |
10442 | if (ctl->block_group == block_group) { |
10443 | caching_ctl = ctl; | |
1e4f4714 | 10444 | refcount_inc(&caching_ctl->count); |
4f69cb98 FM |
10445 | break; |
10446 | } | |
10447 | } | |
10448 | if (caching_ctl) | |
10449 | list_del_init(&caching_ctl->list); | |
0b246afa | 10450 | up_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10451 | if (caching_ctl) { |
10452 | /* Once for the caching bgs list and once for us. */ | |
10453 | put_caching_control(caching_ctl); | |
10454 | put_caching_control(caching_ctl); | |
10455 | } | |
10456 | } | |
817d52f8 | 10457 | |
ce93ec54 JB |
10458 | spin_lock(&trans->transaction->dirty_bgs_lock); |
10459 | if (!list_empty(&block_group->dirty_list)) { | |
1bbc621e CM |
10460 | WARN_ON(1); |
10461 | } | |
10462 | if (!list_empty(&block_group->io_list)) { | |
10463 | WARN_ON(1); | |
ce93ec54 JB |
10464 | } |
10465 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
817d52f8 JB |
10466 | btrfs_remove_free_space_cache(block_group); |
10467 | ||
c146afad | 10468 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10469 | list_del_init(&block_group->ro_list); |
18d018ad | 10470 | |
0b246afa | 10471 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
18d018ad ZL |
10472 | WARN_ON(block_group->space_info->total_bytes |
10473 | < block_group->key.offset); | |
10474 | WARN_ON(block_group->space_info->bytes_readonly | |
10475 | < block_group->key.offset); | |
10476 | WARN_ON(block_group->space_info->disk_total | |
10477 | < block_group->key.offset * factor); | |
10478 | } | |
c146afad YZ |
10479 | block_group->space_info->total_bytes -= block_group->key.offset; |
10480 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10481 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10482 | |
c146afad | 10483 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10484 | |
0af3d00b JB |
10485 | memcpy(&key, &block_group->key, sizeof(key)); |
10486 | ||
34441361 | 10487 | mutex_lock(&fs_info->chunk_mutex); |
495e64f4 FM |
10488 | if (!list_empty(&em->list)) { |
10489 | /* We're in the transaction->pending_chunks list. */ | |
10490 | free_extent_map(em); | |
10491 | } | |
04216820 FM |
10492 | spin_lock(&block_group->lock); |
10493 | block_group->removed = 1; | |
10494 | /* | |
10495 | * At this point trimming can't start on this block group, because we | |
10496 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10497 | * so no one can't find it anymore and even if someone already got this | |
10498 | * block group before we removed it from the rbtree, they have already | |
10499 | * incremented block_group->trimming - if they didn't, they won't find | |
10500 | * any free space entries because we already removed them all when we | |
10501 | * called btrfs_remove_free_space_cache(). | |
10502 | * | |
10503 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10504 | * to prevent the same logical address range and physical device space | |
10505 | * ranges from being reused for a new block group. This is because our | |
10506 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10507 | * completely transactionless, so while it is trimming a range the | |
10508 | * currently running transaction might finish and a new one start, | |
10509 | * allowing for new block groups to be created that can reuse the same | |
10510 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10511 | * |
10512 | * There may also be an implicit trim operation if the file system | |
10513 | * is mounted with -odiscard. The same protections must remain | |
10514 | * in place until the extents have been discarded completely when | |
10515 | * the transaction commit has completed. | |
04216820 FM |
10516 | */ |
10517 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
10518 | /* | |
10519 | * Make sure a trimmer task always sees the em in the pinned_chunks list | |
10520 | * if it sees block_group->removed == 1 (needs to lock block_group->lock | |
10521 | * before checking block_group->removed). | |
10522 | */ | |
10523 | if (!remove_em) { | |
10524 | /* | |
10525 | * Our em might be in trans->transaction->pending_chunks which | |
10526 | * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks), | |
10527 | * and so is the fs_info->pinned_chunks list. | |
10528 | * | |
10529 | * So at this point we must be holding the chunk_mutex to avoid | |
10530 | * any races with chunk allocation (more specifically at | |
10531 | * volumes.c:contains_pending_extent()), to ensure it always | |
10532 | * sees the em, either in the pending_chunks list or in the | |
10533 | * pinned_chunks list. | |
10534 | */ | |
0b246afa | 10535 | list_move_tail(&em->list, &fs_info->pinned_chunks); |
04216820 FM |
10536 | } |
10537 | spin_unlock(&block_group->lock); | |
04216820 FM |
10538 | |
10539 | if (remove_em) { | |
10540 | struct extent_map_tree *em_tree; | |
10541 | ||
0b246afa | 10542 | em_tree = &fs_info->mapping_tree.map_tree; |
04216820 | 10543 | write_lock(&em_tree->lock); |
8dbcd10f FM |
10544 | /* |
10545 | * The em might be in the pending_chunks list, so make sure the | |
10546 | * chunk mutex is locked, since remove_extent_mapping() will | |
10547 | * delete us from that list. | |
10548 | */ | |
04216820 FM |
10549 | remove_extent_mapping(em_tree, em); |
10550 | write_unlock(&em_tree->lock); | |
10551 | /* once for the tree */ | |
10552 | free_extent_map(em); | |
10553 | } | |
10554 | ||
34441361 | 10555 | mutex_unlock(&fs_info->chunk_mutex); |
8dbcd10f | 10556 | |
0b246afa | 10557 | ret = remove_block_group_free_space(trans, fs_info, block_group); |
1e144fb8 OS |
10558 | if (ret) |
10559 | goto out; | |
10560 | ||
fa9c0d79 CM |
10561 | btrfs_put_block_group(block_group); |
10562 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10563 | |
10564 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10565 | if (ret > 0) | |
10566 | ret = -EIO; | |
10567 | if (ret < 0) | |
10568 | goto out; | |
10569 | ||
10570 | ret = btrfs_del_item(trans, root, path); | |
10571 | out: | |
10572 | btrfs_free_path(path); | |
10573 | return ret; | |
10574 | } | |
acce952b | 10575 | |
8eab77ff | 10576 | struct btrfs_trans_handle * |
7fd01182 FM |
10577 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10578 | const u64 chunk_offset) | |
8eab77ff | 10579 | { |
7fd01182 FM |
10580 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; |
10581 | struct extent_map *em; | |
10582 | struct map_lookup *map; | |
10583 | unsigned int num_items; | |
10584 | ||
10585 | read_lock(&em_tree->lock); | |
10586 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10587 | read_unlock(&em_tree->lock); | |
10588 | ASSERT(em && em->start == chunk_offset); | |
10589 | ||
8eab77ff | 10590 | /* |
7fd01182 FM |
10591 | * We need to reserve 3 + N units from the metadata space info in order |
10592 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10593 | * btrfs_remove_block_group()), which are used for: | |
10594 | * | |
8eab77ff FM |
10595 | * 1 unit for adding the free space inode's orphan (located in the tree |
10596 | * of tree roots). | |
7fd01182 FM |
10597 | * 1 unit for deleting the block group item (located in the extent |
10598 | * tree). | |
10599 | * 1 unit for deleting the free space item (located in tree of tree | |
10600 | * roots). | |
10601 | * N units for deleting N device extent items corresponding to each | |
10602 | * stripe (located in the device tree). | |
10603 | * | |
10604 | * In order to remove a block group we also need to reserve units in the | |
10605 | * system space info in order to update the chunk tree (update one or | |
10606 | * more device items and remove one chunk item), but this is done at | |
10607 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10608 | */ |
95617d69 | 10609 | map = em->map_lookup; |
7fd01182 FM |
10610 | num_items = 3 + map->num_stripes; |
10611 | free_extent_map(em); | |
10612 | ||
8eab77ff | 10613 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10614 | num_items, 1); |
8eab77ff FM |
10615 | } |
10616 | ||
47ab2a6c JB |
10617 | /* |
10618 | * Process the unused_bgs list and remove any that don't have any allocated | |
10619 | * space inside of them. | |
10620 | */ | |
10621 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10622 | { | |
10623 | struct btrfs_block_group_cache *block_group; | |
10624 | struct btrfs_space_info *space_info; | |
47ab2a6c JB |
10625 | struct btrfs_trans_handle *trans; |
10626 | int ret = 0; | |
10627 | ||
afcdd129 | 10628 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) |
47ab2a6c JB |
10629 | return; |
10630 | ||
10631 | spin_lock(&fs_info->unused_bgs_lock); | |
10632 | while (!list_empty(&fs_info->unused_bgs)) { | |
10633 | u64 start, end; | |
e33e17ee | 10634 | int trimming; |
47ab2a6c JB |
10635 | |
10636 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10637 | struct btrfs_block_group_cache, | |
10638 | bg_list); | |
47ab2a6c | 10639 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10640 | |
10641 | space_info = block_group->space_info; | |
10642 | ||
47ab2a6c JB |
10643 | if (ret || btrfs_mixed_space_info(space_info)) { |
10644 | btrfs_put_block_group(block_group); | |
10645 | continue; | |
10646 | } | |
10647 | spin_unlock(&fs_info->unused_bgs_lock); | |
10648 | ||
d5f2e33b | 10649 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10650 | |
47ab2a6c JB |
10651 | /* Don't want to race with allocators so take the groups_sem */ |
10652 | down_write(&space_info->groups_sem); | |
10653 | spin_lock(&block_group->lock); | |
10654 | if (block_group->reserved || | |
10655 | btrfs_block_group_used(&block_group->item) || | |
19c4d2f9 | 10656 | block_group->ro || |
aefbe9a6 | 10657 | list_is_singular(&block_group->list)) { |
47ab2a6c JB |
10658 | /* |
10659 | * We want to bail if we made new allocations or have | |
10660 | * outstanding allocations in this block group. We do | |
10661 | * the ro check in case balance is currently acting on | |
10662 | * this block group. | |
10663 | */ | |
10664 | spin_unlock(&block_group->lock); | |
10665 | up_write(&space_info->groups_sem); | |
10666 | goto next; | |
10667 | } | |
10668 | spin_unlock(&block_group->lock); | |
10669 | ||
10670 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 10671 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
10672 | up_write(&space_info->groups_sem); |
10673 | if (ret < 0) { | |
10674 | ret = 0; | |
10675 | goto next; | |
10676 | } | |
10677 | ||
10678 | /* | |
10679 | * Want to do this before we do anything else so we can recover | |
10680 | * properly if we fail to join the transaction. | |
10681 | */ | |
7fd01182 FM |
10682 | trans = btrfs_start_trans_remove_block_group(fs_info, |
10683 | block_group->key.objectid); | |
47ab2a6c | 10684 | if (IS_ERR(trans)) { |
2ff7e61e | 10685 | btrfs_dec_block_group_ro(block_group); |
47ab2a6c JB |
10686 | ret = PTR_ERR(trans); |
10687 | goto next; | |
10688 | } | |
10689 | ||
10690 | /* | |
10691 | * We could have pending pinned extents for this block group, | |
10692 | * just delete them, we don't care about them anymore. | |
10693 | */ | |
10694 | start = block_group->key.objectid; | |
10695 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
10696 | /* |
10697 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
10698 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
10699 | * another task might be running finish_extent_commit() for the | |
10700 | * previous transaction N - 1, and have seen a range belonging | |
10701 | * to the block group in freed_extents[] before we were able to | |
10702 | * clear the whole block group range from freed_extents[]. This | |
10703 | * means that task can lookup for the block group after we | |
10704 | * unpinned it from freed_extents[] and removed it, leading to | |
10705 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
10706 | */ | |
10707 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 10708 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
91166212 | 10709 | EXTENT_DIRTY); |
758eb51e | 10710 | if (ret) { |
d4b450cd | 10711 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10712 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10713 | goto end_trans; |
10714 | } | |
10715 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
91166212 | 10716 | EXTENT_DIRTY); |
758eb51e | 10717 | if (ret) { |
d4b450cd | 10718 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10719 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10720 | goto end_trans; |
10721 | } | |
d4b450cd | 10722 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
10723 | |
10724 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
10725 | spin_lock(&space_info->lock); |
10726 | spin_lock(&block_group->lock); | |
10727 | ||
10728 | space_info->bytes_pinned -= block_group->pinned; | |
10729 | space_info->bytes_readonly += block_group->pinned; | |
10730 | percpu_counter_add(&space_info->total_bytes_pinned, | |
10731 | -block_group->pinned); | |
47ab2a6c JB |
10732 | block_group->pinned = 0; |
10733 | ||
c30666d4 ZL |
10734 | spin_unlock(&block_group->lock); |
10735 | spin_unlock(&space_info->lock); | |
10736 | ||
e33e17ee | 10737 | /* DISCARD can flip during remount */ |
0b246afa | 10738 | trimming = btrfs_test_opt(fs_info, DISCARD); |
e33e17ee JM |
10739 | |
10740 | /* Implicit trim during transaction commit. */ | |
10741 | if (trimming) | |
10742 | btrfs_get_block_group_trimming(block_group); | |
10743 | ||
47ab2a6c JB |
10744 | /* |
10745 | * Btrfs_remove_chunk will abort the transaction if things go | |
10746 | * horribly wrong. | |
10747 | */ | |
5b4aacef | 10748 | ret = btrfs_remove_chunk(trans, fs_info, |
47ab2a6c | 10749 | block_group->key.objectid); |
e33e17ee JM |
10750 | |
10751 | if (ret) { | |
10752 | if (trimming) | |
10753 | btrfs_put_block_group_trimming(block_group); | |
10754 | goto end_trans; | |
10755 | } | |
10756 | ||
10757 | /* | |
10758 | * If we're not mounted with -odiscard, we can just forget | |
10759 | * about this block group. Otherwise we'll need to wait | |
10760 | * until transaction commit to do the actual discard. | |
10761 | */ | |
10762 | if (trimming) { | |
348a0013 FM |
10763 | spin_lock(&fs_info->unused_bgs_lock); |
10764 | /* | |
10765 | * A concurrent scrub might have added us to the list | |
10766 | * fs_info->unused_bgs, so use a list_move operation | |
10767 | * to add the block group to the deleted_bgs list. | |
10768 | */ | |
e33e17ee JM |
10769 | list_move(&block_group->bg_list, |
10770 | &trans->transaction->deleted_bgs); | |
348a0013 | 10771 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
10772 | btrfs_get_block_group(block_group); |
10773 | } | |
758eb51e | 10774 | end_trans: |
3a45bb20 | 10775 | btrfs_end_transaction(trans); |
47ab2a6c | 10776 | next: |
d5f2e33b | 10777 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
10778 | btrfs_put_block_group(block_group); |
10779 | spin_lock(&fs_info->unused_bgs_lock); | |
10780 | } | |
10781 | spin_unlock(&fs_info->unused_bgs_lock); | |
10782 | } | |
10783 | ||
c59021f8 | 10784 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
10785 | { | |
10786 | struct btrfs_space_info *space_info; | |
1aba86d6 | 10787 | struct btrfs_super_block *disk_super; |
10788 | u64 features; | |
10789 | u64 flags; | |
10790 | int mixed = 0; | |
c59021f8 | 10791 | int ret; |
10792 | ||
6c41761f | 10793 | disk_super = fs_info->super_copy; |
1aba86d6 | 10794 | if (!btrfs_super_root(disk_super)) |
0dc924c5 | 10795 | return -EINVAL; |
c59021f8 | 10796 | |
1aba86d6 | 10797 | features = btrfs_super_incompat_flags(disk_super); |
10798 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
10799 | mixed = 1; | |
c59021f8 | 10800 | |
1aba86d6 | 10801 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
2be12ef7 | 10802 | ret = create_space_info(fs_info, flags, &space_info); |
c59021f8 | 10803 | if (ret) |
1aba86d6 | 10804 | goto out; |
c59021f8 | 10805 | |
1aba86d6 | 10806 | if (mixed) { |
10807 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
2be12ef7 | 10808 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 10809 | } else { |
10810 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
2be12ef7 | 10811 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 10812 | if (ret) |
10813 | goto out; | |
10814 | ||
10815 | flags = BTRFS_BLOCK_GROUP_DATA; | |
2be12ef7 | 10816 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 10817 | } |
10818 | out: | |
c59021f8 | 10819 | return ret; |
10820 | } | |
10821 | ||
2ff7e61e JM |
10822 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
10823 | u64 start, u64 end) | |
acce952b | 10824 | { |
2ff7e61e | 10825 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 10826 | } |
10827 | ||
499f377f JM |
10828 | /* |
10829 | * It used to be that old block groups would be left around forever. | |
10830 | * Iterating over them would be enough to trim unused space. Since we | |
10831 | * now automatically remove them, we also need to iterate over unallocated | |
10832 | * space. | |
10833 | * | |
10834 | * We don't want a transaction for this since the discard may take a | |
10835 | * substantial amount of time. We don't require that a transaction be | |
10836 | * running, but we do need to take a running transaction into account | |
10837 | * to ensure that we're not discarding chunks that were released in | |
10838 | * the current transaction. | |
10839 | * | |
10840 | * Holding the chunks lock will prevent other threads from allocating | |
10841 | * or releasing chunks, but it won't prevent a running transaction | |
10842 | * from committing and releasing the memory that the pending chunks | |
10843 | * list head uses. For that, we need to take a reference to the | |
10844 | * transaction. | |
10845 | */ | |
10846 | static int btrfs_trim_free_extents(struct btrfs_device *device, | |
10847 | u64 minlen, u64 *trimmed) | |
10848 | { | |
10849 | u64 start = 0, len = 0; | |
10850 | int ret; | |
10851 | ||
10852 | *trimmed = 0; | |
10853 | ||
10854 | /* Not writeable = nothing to do. */ | |
ebbede42 | 10855 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
10856 | return 0; |
10857 | ||
10858 | /* No free space = nothing to do. */ | |
10859 | if (device->total_bytes <= device->bytes_used) | |
10860 | return 0; | |
10861 | ||
10862 | ret = 0; | |
10863 | ||
10864 | while (1) { | |
fb456252 | 10865 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
10866 | struct btrfs_transaction *trans; |
10867 | u64 bytes; | |
10868 | ||
10869 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
10870 | if (ret) | |
10871 | return ret; | |
10872 | ||
10873 | down_read(&fs_info->commit_root_sem); | |
10874 | ||
10875 | spin_lock(&fs_info->trans_lock); | |
10876 | trans = fs_info->running_transaction; | |
10877 | if (trans) | |
9b64f57d | 10878 | refcount_inc(&trans->use_count); |
499f377f JM |
10879 | spin_unlock(&fs_info->trans_lock); |
10880 | ||
10881 | ret = find_free_dev_extent_start(trans, device, minlen, start, | |
10882 | &start, &len); | |
10883 | if (trans) | |
10884 | btrfs_put_transaction(trans); | |
10885 | ||
10886 | if (ret) { | |
10887 | up_read(&fs_info->commit_root_sem); | |
10888 | mutex_unlock(&fs_info->chunk_mutex); | |
10889 | if (ret == -ENOSPC) | |
10890 | ret = 0; | |
10891 | break; | |
10892 | } | |
10893 | ||
10894 | ret = btrfs_issue_discard(device->bdev, start, len, &bytes); | |
10895 | up_read(&fs_info->commit_root_sem); | |
10896 | mutex_unlock(&fs_info->chunk_mutex); | |
10897 | ||
10898 | if (ret) | |
10899 | break; | |
10900 | ||
10901 | start += len; | |
10902 | *trimmed += bytes; | |
10903 | ||
10904 | if (fatal_signal_pending(current)) { | |
10905 | ret = -ERESTARTSYS; | |
10906 | break; | |
10907 | } | |
10908 | ||
10909 | cond_resched(); | |
10910 | } | |
10911 | ||
10912 | return ret; | |
10913 | } | |
10914 | ||
2ff7e61e | 10915 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 10916 | { |
f7039b1d | 10917 | struct btrfs_block_group_cache *cache = NULL; |
499f377f JM |
10918 | struct btrfs_device *device; |
10919 | struct list_head *devices; | |
f7039b1d LD |
10920 | u64 group_trimmed; |
10921 | u64 start; | |
10922 | u64 end; | |
10923 | u64 trimmed = 0; | |
2cac13e4 | 10924 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
10925 | int ret = 0; |
10926 | ||
2cac13e4 LB |
10927 | /* |
10928 | * try to trim all FS space, our block group may start from non-zero. | |
10929 | */ | |
10930 | if (range->len == total_bytes) | |
10931 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
10932 | else | |
10933 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
10934 | |
10935 | while (cache) { | |
10936 | if (cache->key.objectid >= (range->start + range->len)) { | |
10937 | btrfs_put_block_group(cache); | |
10938 | break; | |
10939 | } | |
10940 | ||
10941 | start = max(range->start, cache->key.objectid); | |
10942 | end = min(range->start + range->len, | |
10943 | cache->key.objectid + cache->key.offset); | |
10944 | ||
10945 | if (end - start >= range->minlen) { | |
10946 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 10947 | ret = cache_block_group(cache, 0); |
1be41b78 JB |
10948 | if (ret) { |
10949 | btrfs_put_block_group(cache); | |
10950 | break; | |
10951 | } | |
10952 | ret = wait_block_group_cache_done(cache); | |
10953 | if (ret) { | |
10954 | btrfs_put_block_group(cache); | |
10955 | break; | |
10956 | } | |
f7039b1d LD |
10957 | } |
10958 | ret = btrfs_trim_block_group(cache, | |
10959 | &group_trimmed, | |
10960 | start, | |
10961 | end, | |
10962 | range->minlen); | |
10963 | ||
10964 | trimmed += group_trimmed; | |
10965 | if (ret) { | |
10966 | btrfs_put_block_group(cache); | |
10967 | break; | |
10968 | } | |
10969 | } | |
10970 | ||
2ff7e61e | 10971 | cache = next_block_group(fs_info, cache); |
f7039b1d LD |
10972 | } |
10973 | ||
0b246afa JM |
10974 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
10975 | devices = &fs_info->fs_devices->alloc_list; | |
499f377f JM |
10976 | list_for_each_entry(device, devices, dev_alloc_list) { |
10977 | ret = btrfs_trim_free_extents(device, range->minlen, | |
10978 | &group_trimmed); | |
10979 | if (ret) | |
10980 | break; | |
10981 | ||
10982 | trimmed += group_trimmed; | |
10983 | } | |
0b246afa | 10984 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 10985 | |
f7039b1d LD |
10986 | range->len = trimmed; |
10987 | return ret; | |
10988 | } | |
8257b2dc MX |
10989 | |
10990 | /* | |
ea14b57f | 10991 | * btrfs_{start,end}_write_no_snapshotting() are similar to |
9ea24bbe FM |
10992 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing |
10993 | * data into the page cache through nocow before the subvolume is snapshoted, | |
10994 | * but flush the data into disk after the snapshot creation, or to prevent | |
ea14b57f | 10995 | * operations while snapshotting is ongoing and that cause the snapshot to be |
9ea24bbe | 10996 | * inconsistent (writes followed by expanding truncates for example). |
8257b2dc | 10997 | */ |
ea14b57f | 10998 | void btrfs_end_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc MX |
10999 | { |
11000 | percpu_counter_dec(&root->subv_writers->counter); | |
11001 | /* | |
a83342aa | 11002 | * Make sure counter is updated before we wake up waiters. |
8257b2dc MX |
11003 | */ |
11004 | smp_mb(); | |
11005 | if (waitqueue_active(&root->subv_writers->wait)) | |
11006 | wake_up(&root->subv_writers->wait); | |
11007 | } | |
11008 | ||
ea14b57f | 11009 | int btrfs_start_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc | 11010 | { |
ea14b57f | 11011 | if (atomic_read(&root->will_be_snapshotted)) |
8257b2dc MX |
11012 | return 0; |
11013 | ||
11014 | percpu_counter_inc(&root->subv_writers->counter); | |
11015 | /* | |
11016 | * Make sure counter is updated before we check for snapshot creation. | |
11017 | */ | |
11018 | smp_mb(); | |
ea14b57f DS |
11019 | if (atomic_read(&root->will_be_snapshotted)) { |
11020 | btrfs_end_write_no_snapshotting(root); | |
8257b2dc MX |
11021 | return 0; |
11022 | } | |
11023 | return 1; | |
11024 | } | |
0bc19f90 | 11025 | |
0bc19f90 ZL |
11026 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) |
11027 | { | |
11028 | while (true) { | |
11029 | int ret; | |
11030 | ||
ea14b57f | 11031 | ret = btrfs_start_write_no_snapshotting(root); |
0bc19f90 ZL |
11032 | if (ret) |
11033 | break; | |
5e4def20 | 11034 | wait_on_atomic_t(&root->will_be_snapshotted, atomic_t_wait, |
0bc19f90 ZL |
11035 | TASK_UNINTERRUPTIBLE); |
11036 | } | |
11037 | } |