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> |
74493f7a | 30 | #include "hash.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); | |
14a958e6 | 1206 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1207 | lenum = cpu_to_le64(owner); |
14a958e6 | 1208 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1209 | lenum = cpu_to_le64(offset); |
14a958e6 | 1210 | low_crc = btrfs_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; |
d5e2003c JB |
2148 | if (!stripe->dev->can_discard) |
2149 | continue; | |
2150 | ||
5378e607 LD |
2151 | ret = btrfs_issue_discard(stripe->dev->bdev, |
2152 | stripe->physical, | |
d04c6b88 JM |
2153 | stripe->length, |
2154 | &bytes); | |
5378e607 | 2155 | if (!ret) |
d04c6b88 | 2156 | discarded_bytes += bytes; |
5378e607 | 2157 | else if (ret != -EOPNOTSUPP) |
79787eaa | 2158 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
2159 | |
2160 | /* | |
2161 | * Just in case we get back EOPNOTSUPP for some reason, | |
2162 | * just ignore the return value so we don't screw up | |
2163 | * people calling discard_extent. | |
2164 | */ | |
2165 | ret = 0; | |
5d4f98a2 | 2166 | } |
6e9606d2 | 2167 | btrfs_put_bbio(bbio); |
5d4f98a2 | 2168 | } |
0b246afa | 2169 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
2170 | |
2171 | if (actual_bytes) | |
2172 | *actual_bytes = discarded_bytes; | |
2173 | ||
5d4f98a2 | 2174 | |
53b381b3 DW |
2175 | if (ret == -EOPNOTSUPP) |
2176 | ret = 0; | |
5d4f98a2 | 2177 | return ret; |
5d4f98a2 YZ |
2178 | } |
2179 | ||
79787eaa | 2180 | /* Can return -ENOMEM */ |
5d4f98a2 | 2181 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
84f7d8e6 | 2182 | struct btrfs_root *root, |
5d4f98a2 | 2183 | u64 bytenr, u64 num_bytes, u64 parent, |
b06c4bf5 | 2184 | u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 | 2185 | { |
84f7d8e6 | 2186 | struct btrfs_fs_info *fs_info = root->fs_info; |
d7eae340 | 2187 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 2188 | int ret; |
66d7e7f0 | 2189 | |
5d4f98a2 YZ |
2190 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
2191 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
2192 | ||
fd708b81 JB |
2193 | btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid, |
2194 | owner, offset, BTRFS_ADD_DELAYED_REF); | |
2195 | ||
5d4f98a2 | 2196 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 | 2197 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
7be07912 OS |
2198 | num_bytes, parent, |
2199 | root_objectid, (int)owner, | |
2200 | BTRFS_ADD_DELAYED_REF, NULL, | |
d7eae340 | 2201 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 2202 | } else { |
66d7e7f0 | 2203 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
7be07912 OS |
2204 | num_bytes, parent, |
2205 | root_objectid, owner, offset, | |
d7eae340 OS |
2206 | 0, BTRFS_ADD_DELAYED_REF, |
2207 | &old_ref_mod, &new_ref_mod); | |
5d4f98a2 | 2208 | } |
d7eae340 OS |
2209 | |
2210 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) | |
2211 | add_pinned_bytes(fs_info, -num_bytes, owner, root_objectid); | |
2212 | ||
5d4f98a2 YZ |
2213 | return ret; |
2214 | } | |
2215 | ||
2216 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2217 | struct btrfs_fs_info *fs_info, |
c682f9b3 | 2218 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2219 | u64 parent, u64 root_objectid, |
2220 | u64 owner, u64 offset, int refs_to_add, | |
2221 | struct btrfs_delayed_extent_op *extent_op) | |
2222 | { | |
2223 | struct btrfs_path *path; | |
2224 | struct extent_buffer *leaf; | |
2225 | struct btrfs_extent_item *item; | |
fcebe456 | 2226 | struct btrfs_key key; |
c682f9b3 QW |
2227 | u64 bytenr = node->bytenr; |
2228 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2229 | u64 refs; |
2230 | int ret; | |
5d4f98a2 YZ |
2231 | |
2232 | path = btrfs_alloc_path(); | |
2233 | if (!path) | |
2234 | return -ENOMEM; | |
2235 | ||
e4058b54 | 2236 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2237 | path->leave_spinning = 1; |
2238 | /* this will setup the path even if it fails to insert the back ref */ | |
87bde3cd JM |
2239 | ret = insert_inline_extent_backref(trans, fs_info, path, bytenr, |
2240 | num_bytes, parent, root_objectid, | |
2241 | owner, offset, | |
5d4f98a2 | 2242 | refs_to_add, extent_op); |
0ed4792a | 2243 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2244 | goto out; |
fcebe456 JB |
2245 | |
2246 | /* | |
2247 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2248 | * inline extent ref, so just update the reference count and add a | |
2249 | * normal backref. | |
2250 | */ | |
5d4f98a2 | 2251 | leaf = path->nodes[0]; |
fcebe456 | 2252 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2253 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2254 | refs = btrfs_extent_refs(leaf, item); | |
2255 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2256 | if (extent_op) | |
2257 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2258 | |
5d4f98a2 | 2259 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2260 | btrfs_release_path(path); |
56bec294 | 2261 | |
e4058b54 | 2262 | path->reada = READA_FORWARD; |
b9473439 | 2263 | path->leave_spinning = 1; |
56bec294 | 2264 | /* now insert the actual backref */ |
87bde3cd JM |
2265 | ret = insert_extent_backref(trans, fs_info, path, bytenr, parent, |
2266 | root_objectid, owner, offset, refs_to_add); | |
79787eaa | 2267 | if (ret) |
66642832 | 2268 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 2269 | out: |
56bec294 | 2270 | btrfs_free_path(path); |
30d133fc | 2271 | return ret; |
56bec294 CM |
2272 | } |
2273 | ||
5d4f98a2 | 2274 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2275 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2276 | struct btrfs_delayed_ref_node *node, |
2277 | struct btrfs_delayed_extent_op *extent_op, | |
2278 | int insert_reserved) | |
56bec294 | 2279 | { |
5d4f98a2 YZ |
2280 | int ret = 0; |
2281 | struct btrfs_delayed_data_ref *ref; | |
2282 | struct btrfs_key ins; | |
2283 | u64 parent = 0; | |
2284 | u64 ref_root = 0; | |
2285 | u64 flags = 0; | |
2286 | ||
2287 | ins.objectid = node->bytenr; | |
2288 | ins.offset = node->num_bytes; | |
2289 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2290 | ||
2291 | ref = btrfs_delayed_node_to_data_ref(node); | |
0b246afa | 2292 | trace_run_delayed_data_ref(fs_info, node, ref, node->action); |
599c75ec | 2293 | |
5d4f98a2 YZ |
2294 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2295 | parent = ref->parent; | |
fcebe456 | 2296 | ref_root = ref->root; |
5d4f98a2 YZ |
2297 | |
2298 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2299 | if (extent_op) |
5d4f98a2 | 2300 | flags |= extent_op->flags_to_set; |
2ff7e61e | 2301 | ret = alloc_reserved_file_extent(trans, fs_info, |
5d4f98a2 YZ |
2302 | parent, ref_root, flags, |
2303 | ref->objectid, ref->offset, | |
2304 | &ins, node->ref_mod); | |
5d4f98a2 | 2305 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2ff7e61e | 2306 | ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent, |
5d4f98a2 YZ |
2307 | ref_root, ref->objectid, |
2308 | ref->offset, node->ref_mod, | |
c682f9b3 | 2309 | extent_op); |
5d4f98a2 | 2310 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
2ff7e61e | 2311 | ret = __btrfs_free_extent(trans, fs_info, node, parent, |
5d4f98a2 YZ |
2312 | ref_root, ref->objectid, |
2313 | ref->offset, node->ref_mod, | |
c682f9b3 | 2314 | extent_op); |
5d4f98a2 YZ |
2315 | } else { |
2316 | BUG(); | |
2317 | } | |
2318 | return ret; | |
2319 | } | |
2320 | ||
2321 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2322 | struct extent_buffer *leaf, | |
2323 | struct btrfs_extent_item *ei) | |
2324 | { | |
2325 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2326 | if (extent_op->update_flags) { | |
2327 | flags |= extent_op->flags_to_set; | |
2328 | btrfs_set_extent_flags(leaf, ei, flags); | |
2329 | } | |
2330 | ||
2331 | if (extent_op->update_key) { | |
2332 | struct btrfs_tree_block_info *bi; | |
2333 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2334 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2335 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2336 | } | |
2337 | } | |
2338 | ||
2339 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2340 | struct btrfs_fs_info *fs_info, |
d278850e | 2341 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
2342 | struct btrfs_delayed_extent_op *extent_op) |
2343 | { | |
2344 | struct btrfs_key key; | |
2345 | struct btrfs_path *path; | |
2346 | struct btrfs_extent_item *ei; | |
2347 | struct extent_buffer *leaf; | |
2348 | u32 item_size; | |
56bec294 | 2349 | int ret; |
5d4f98a2 | 2350 | int err = 0; |
b1c79e09 | 2351 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2352 | |
79787eaa JM |
2353 | if (trans->aborted) |
2354 | return 0; | |
2355 | ||
0b246afa | 2356 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
2357 | metadata = 0; |
2358 | ||
5d4f98a2 YZ |
2359 | path = btrfs_alloc_path(); |
2360 | if (!path) | |
2361 | return -ENOMEM; | |
2362 | ||
d278850e | 2363 | key.objectid = head->bytenr; |
5d4f98a2 | 2364 | |
3173a18f | 2365 | if (metadata) { |
3173a18f | 2366 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2367 | key.offset = extent_op->level; |
3173a18f JB |
2368 | } else { |
2369 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 2370 | key.offset = head->num_bytes; |
3173a18f JB |
2371 | } |
2372 | ||
2373 | again: | |
e4058b54 | 2374 | path->reada = READA_FORWARD; |
5d4f98a2 | 2375 | path->leave_spinning = 1; |
0b246afa | 2376 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
2377 | if (ret < 0) { |
2378 | err = ret; | |
2379 | goto out; | |
2380 | } | |
2381 | if (ret > 0) { | |
3173a18f | 2382 | if (metadata) { |
55994887 FDBM |
2383 | if (path->slots[0] > 0) { |
2384 | path->slots[0]--; | |
2385 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2386 | path->slots[0]); | |
d278850e | 2387 | if (key.objectid == head->bytenr && |
55994887 | 2388 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 2389 | key.offset == head->num_bytes) |
55994887 FDBM |
2390 | ret = 0; |
2391 | } | |
2392 | if (ret > 0) { | |
2393 | btrfs_release_path(path); | |
2394 | metadata = 0; | |
3173a18f | 2395 | |
d278850e JB |
2396 | key.objectid = head->bytenr; |
2397 | key.offset = head->num_bytes; | |
55994887 FDBM |
2398 | key.type = BTRFS_EXTENT_ITEM_KEY; |
2399 | goto again; | |
2400 | } | |
2401 | } else { | |
2402 | err = -EIO; | |
2403 | goto out; | |
3173a18f | 2404 | } |
5d4f98a2 YZ |
2405 | } |
2406 | ||
2407 | leaf = path->nodes[0]; | |
2408 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2409 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2410 | if (item_size < sizeof(*ei)) { | |
87bde3cd | 2411 | ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0); |
5d4f98a2 YZ |
2412 | if (ret < 0) { |
2413 | err = ret; | |
2414 | goto out; | |
2415 | } | |
2416 | leaf = path->nodes[0]; | |
2417 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2418 | } | |
2419 | #endif | |
2420 | BUG_ON(item_size < sizeof(*ei)); | |
2421 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2422 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2423 | |
5d4f98a2 YZ |
2424 | btrfs_mark_buffer_dirty(leaf); |
2425 | out: | |
2426 | btrfs_free_path(path); | |
2427 | return err; | |
56bec294 CM |
2428 | } |
2429 | ||
5d4f98a2 | 2430 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2431 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2432 | struct btrfs_delayed_ref_node *node, |
2433 | struct btrfs_delayed_extent_op *extent_op, | |
2434 | int insert_reserved) | |
56bec294 CM |
2435 | { |
2436 | int ret = 0; | |
5d4f98a2 YZ |
2437 | struct btrfs_delayed_tree_ref *ref; |
2438 | struct btrfs_key ins; | |
2439 | u64 parent = 0; | |
2440 | u64 ref_root = 0; | |
0b246afa | 2441 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
56bec294 | 2442 | |
5d4f98a2 | 2443 | ref = btrfs_delayed_node_to_tree_ref(node); |
0b246afa | 2444 | trace_run_delayed_tree_ref(fs_info, node, ref, node->action); |
599c75ec | 2445 | |
5d4f98a2 YZ |
2446 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2447 | parent = ref->parent; | |
fcebe456 | 2448 | ref_root = ref->root; |
5d4f98a2 | 2449 | |
3173a18f JB |
2450 | ins.objectid = node->bytenr; |
2451 | if (skinny_metadata) { | |
2452 | ins.offset = ref->level; | |
2453 | ins.type = BTRFS_METADATA_ITEM_KEY; | |
2454 | } else { | |
2455 | ins.offset = node->num_bytes; | |
2456 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2457 | } | |
2458 | ||
02794222 | 2459 | if (node->ref_mod != 1) { |
2ff7e61e | 2460 | btrfs_err(fs_info, |
02794222 LB |
2461 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
2462 | node->bytenr, node->ref_mod, node->action, ref_root, | |
2463 | parent); | |
2464 | return -EIO; | |
2465 | } | |
5d4f98a2 | 2466 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 2467 | BUG_ON(!extent_op || !extent_op->update_flags); |
2ff7e61e | 2468 | ret = alloc_reserved_tree_block(trans, fs_info, |
5d4f98a2 YZ |
2469 | parent, ref_root, |
2470 | extent_op->flags_to_set, | |
2471 | &extent_op->key, | |
b06c4bf5 | 2472 | ref->level, &ins); |
5d4f98a2 | 2473 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2ff7e61e | 2474 | ret = __btrfs_inc_extent_ref(trans, fs_info, node, |
c682f9b3 QW |
2475 | parent, ref_root, |
2476 | ref->level, 0, 1, | |
fcebe456 | 2477 | extent_op); |
5d4f98a2 | 2478 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
2ff7e61e | 2479 | ret = __btrfs_free_extent(trans, fs_info, node, |
c682f9b3 QW |
2480 | parent, ref_root, |
2481 | ref->level, 0, 1, extent_op); | |
5d4f98a2 YZ |
2482 | } else { |
2483 | BUG(); | |
2484 | } | |
56bec294 CM |
2485 | return ret; |
2486 | } | |
2487 | ||
2488 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 2489 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2490 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2491 | struct btrfs_delayed_ref_node *node, |
2492 | struct btrfs_delayed_extent_op *extent_op, | |
2493 | int insert_reserved) | |
56bec294 | 2494 | { |
79787eaa JM |
2495 | int ret = 0; |
2496 | ||
857cc2fc JB |
2497 | if (trans->aborted) { |
2498 | if (insert_reserved) | |
2ff7e61e | 2499 | btrfs_pin_extent(fs_info, node->bytenr, |
857cc2fc | 2500 | node->num_bytes, 1); |
79787eaa | 2501 | return 0; |
857cc2fc | 2502 | } |
79787eaa | 2503 | |
5d4f98a2 YZ |
2504 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2505 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2ff7e61e | 2506 | ret = run_delayed_tree_ref(trans, fs_info, node, extent_op, |
5d4f98a2 YZ |
2507 | insert_reserved); |
2508 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2509 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2ff7e61e | 2510 | ret = run_delayed_data_ref(trans, fs_info, node, extent_op, |
5d4f98a2 YZ |
2511 | insert_reserved); |
2512 | else | |
2513 | BUG(); | |
2514 | return ret; | |
56bec294 CM |
2515 | } |
2516 | ||
c6fc2454 | 2517 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2518 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2519 | { | |
cffc3374 FM |
2520 | struct btrfs_delayed_ref_node *ref; |
2521 | ||
0e0adbcf | 2522 | if (RB_EMPTY_ROOT(&head->ref_tree)) |
c6fc2454 | 2523 | return NULL; |
d7df2c79 | 2524 | |
cffc3374 FM |
2525 | /* |
2526 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2527 | * This is to prevent a ref count from going down to zero, which deletes | |
2528 | * the extent item from the extent tree, when there still are references | |
2529 | * to add, which would fail because they would not find the extent item. | |
2530 | */ | |
1d57ee94 WX |
2531 | if (!list_empty(&head->ref_add_list)) |
2532 | return list_first_entry(&head->ref_add_list, | |
2533 | struct btrfs_delayed_ref_node, add_list); | |
2534 | ||
0e0adbcf JB |
2535 | ref = rb_entry(rb_first(&head->ref_tree), |
2536 | struct btrfs_delayed_ref_node, ref_node); | |
1d57ee94 WX |
2537 | ASSERT(list_empty(&ref->add_list)); |
2538 | return ref; | |
56bec294 CM |
2539 | } |
2540 | ||
2eadaa22 JB |
2541 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
2542 | struct btrfs_delayed_ref_head *head) | |
2543 | { | |
2544 | spin_lock(&delayed_refs->lock); | |
2545 | head->processing = 0; | |
2546 | delayed_refs->num_heads_ready++; | |
2547 | spin_unlock(&delayed_refs->lock); | |
2548 | btrfs_delayed_ref_unlock(head); | |
2549 | } | |
2550 | ||
b00e6250 JB |
2551 | static int cleanup_extent_op(struct btrfs_trans_handle *trans, |
2552 | struct btrfs_fs_info *fs_info, | |
2553 | struct btrfs_delayed_ref_head *head) | |
2554 | { | |
2555 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
2556 | int ret; | |
2557 | ||
2558 | if (!extent_op) | |
2559 | return 0; | |
2560 | head->extent_op = NULL; | |
2561 | if (head->must_insert_reserved) { | |
2562 | btrfs_free_delayed_extent_op(extent_op); | |
2563 | return 0; | |
2564 | } | |
2565 | spin_unlock(&head->lock); | |
d278850e | 2566 | ret = run_delayed_extent_op(trans, fs_info, head, extent_op); |
b00e6250 JB |
2567 | btrfs_free_delayed_extent_op(extent_op); |
2568 | return ret ? ret : 1; | |
2569 | } | |
2570 | ||
194ab0bc JB |
2571 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
2572 | struct btrfs_fs_info *fs_info, | |
2573 | struct btrfs_delayed_ref_head *head) | |
2574 | { | |
2575 | struct btrfs_delayed_ref_root *delayed_refs; | |
2576 | int ret; | |
2577 | ||
2578 | delayed_refs = &trans->transaction->delayed_refs; | |
2579 | ||
2580 | ret = cleanup_extent_op(trans, fs_info, head); | |
2581 | if (ret < 0) { | |
2582 | unselect_delayed_ref_head(delayed_refs, head); | |
2583 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
2584 | return ret; | |
2585 | } else if (ret) { | |
2586 | return ret; | |
2587 | } | |
2588 | ||
2589 | /* | |
2590 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
2591 | * and then re-check to make sure nobody got added. | |
2592 | */ | |
2593 | spin_unlock(&head->lock); | |
2594 | spin_lock(&delayed_refs->lock); | |
2595 | spin_lock(&head->lock); | |
0e0adbcf | 2596 | if (!RB_EMPTY_ROOT(&head->ref_tree) || head->extent_op) { |
194ab0bc JB |
2597 | spin_unlock(&head->lock); |
2598 | spin_unlock(&delayed_refs->lock); | |
2599 | return 1; | |
2600 | } | |
194ab0bc JB |
2601 | delayed_refs->num_heads--; |
2602 | rb_erase(&head->href_node, &delayed_refs->href_root); | |
d278850e | 2603 | RB_CLEAR_NODE(&head->href_node); |
194ab0bc | 2604 | spin_unlock(&delayed_refs->lock); |
c1103f7a JB |
2605 | spin_unlock(&head->lock); |
2606 | atomic_dec(&delayed_refs->num_entries); | |
2607 | ||
d278850e | 2608 | trace_run_delayed_ref_head(fs_info, head, 0); |
c1103f7a JB |
2609 | |
2610 | if (head->total_ref_mod < 0) { | |
2611 | struct btrfs_block_group_cache *cache; | |
2612 | ||
d278850e | 2613 | cache = btrfs_lookup_block_group(fs_info, head->bytenr); |
c1103f7a JB |
2614 | ASSERT(cache); |
2615 | percpu_counter_add(&cache->space_info->total_bytes_pinned, | |
d278850e | 2616 | -head->num_bytes); |
c1103f7a JB |
2617 | btrfs_put_block_group(cache); |
2618 | ||
2619 | if (head->is_data) { | |
2620 | spin_lock(&delayed_refs->lock); | |
d278850e | 2621 | delayed_refs->pending_csums -= head->num_bytes; |
c1103f7a JB |
2622 | spin_unlock(&delayed_refs->lock); |
2623 | } | |
2624 | } | |
2625 | ||
2626 | if (head->must_insert_reserved) { | |
d278850e JB |
2627 | btrfs_pin_extent(fs_info, head->bytenr, |
2628 | head->num_bytes, 1); | |
c1103f7a | 2629 | if (head->is_data) { |
d278850e JB |
2630 | ret = btrfs_del_csums(trans, fs_info, head->bytenr, |
2631 | head->num_bytes); | |
c1103f7a JB |
2632 | } |
2633 | } | |
2634 | ||
2635 | /* Also free its reserved qgroup space */ | |
2636 | btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root, | |
2637 | head->qgroup_reserved); | |
2638 | btrfs_delayed_ref_unlock(head); | |
d278850e | 2639 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
2640 | return 0; |
2641 | } | |
2642 | ||
79787eaa JM |
2643 | /* |
2644 | * Returns 0 on success or if called with an already aborted transaction. | |
2645 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2646 | */ | |
d7df2c79 | 2647 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2648 | struct btrfs_fs_info *fs_info, |
d7df2c79 | 2649 | unsigned long nr) |
56bec294 | 2650 | { |
56bec294 CM |
2651 | struct btrfs_delayed_ref_root *delayed_refs; |
2652 | struct btrfs_delayed_ref_node *ref; | |
2653 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2654 | struct btrfs_delayed_extent_op *extent_op; |
0a2b2a84 | 2655 | ktime_t start = ktime_get(); |
56bec294 | 2656 | int ret; |
d7df2c79 | 2657 | unsigned long count = 0; |
0a2b2a84 | 2658 | unsigned long actual_count = 0; |
56bec294 | 2659 | int must_insert_reserved = 0; |
56bec294 CM |
2660 | |
2661 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2662 | while (1) { |
2663 | if (!locked_ref) { | |
d7df2c79 | 2664 | if (count >= nr) |
56bec294 | 2665 | break; |
56bec294 | 2666 | |
d7df2c79 JB |
2667 | spin_lock(&delayed_refs->lock); |
2668 | locked_ref = btrfs_select_ref_head(trans); | |
2669 | if (!locked_ref) { | |
2670 | spin_unlock(&delayed_refs->lock); | |
2671 | break; | |
2672 | } | |
c3e69d58 CM |
2673 | |
2674 | /* grab the lock that says we are going to process | |
2675 | * all the refs for this head */ | |
2676 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
d7df2c79 | 2677 | spin_unlock(&delayed_refs->lock); |
c3e69d58 CM |
2678 | /* |
2679 | * we may have dropped the spin lock to get the head | |
2680 | * mutex lock, and that might have given someone else | |
2681 | * time to free the head. If that's true, it has been | |
2682 | * removed from our list and we can move on. | |
2683 | */ | |
2684 | if (ret == -EAGAIN) { | |
2685 | locked_ref = NULL; | |
2686 | count++; | |
2687 | continue; | |
56bec294 CM |
2688 | } |
2689 | } | |
a28ec197 | 2690 | |
2c3cf7d5 FM |
2691 | /* |
2692 | * We need to try and merge add/drops of the same ref since we | |
2693 | * can run into issues with relocate dropping the implicit ref | |
2694 | * and then it being added back again before the drop can | |
2695 | * finish. If we merged anything we need to re-loop so we can | |
2696 | * get a good ref. | |
2697 | * Or we can get node references of the same type that weren't | |
2698 | * merged when created due to bumps in the tree mod seq, and | |
2699 | * we need to merge them to prevent adding an inline extent | |
2700 | * backref before dropping it (triggering a BUG_ON at | |
2701 | * insert_inline_extent_backref()). | |
2702 | */ | |
d7df2c79 | 2703 | spin_lock(&locked_ref->lock); |
2c3cf7d5 FM |
2704 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, |
2705 | locked_ref); | |
ae1e206b | 2706 | |
d1270cd9 AJ |
2707 | /* |
2708 | * locked_ref is the head node, so we have to go one | |
2709 | * node back for any delayed ref updates | |
2710 | */ | |
2711 | ref = select_delayed_ref(locked_ref); | |
2712 | ||
2713 | if (ref && ref->seq && | |
097b8a7c | 2714 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d7df2c79 | 2715 | spin_unlock(&locked_ref->lock); |
2eadaa22 | 2716 | unselect_delayed_ref_head(delayed_refs, locked_ref); |
d7df2c79 | 2717 | locked_ref = NULL; |
d1270cd9 | 2718 | cond_resched(); |
27a377db | 2719 | count++; |
d1270cd9 AJ |
2720 | continue; |
2721 | } | |
2722 | ||
c1103f7a JB |
2723 | /* |
2724 | * We're done processing refs in this ref_head, clean everything | |
2725 | * up and move on to the next ref_head. | |
2726 | */ | |
56bec294 | 2727 | if (!ref) { |
194ab0bc JB |
2728 | ret = cleanup_ref_head(trans, fs_info, locked_ref); |
2729 | if (ret > 0 ) { | |
b00e6250 JB |
2730 | /* We dropped our lock, we need to loop. */ |
2731 | ret = 0; | |
d7df2c79 | 2732 | continue; |
194ab0bc JB |
2733 | } else if (ret) { |
2734 | return ret; | |
5d4f98a2 | 2735 | } |
c1103f7a JB |
2736 | locked_ref = NULL; |
2737 | count++; | |
2738 | continue; | |
2739 | } | |
02217ed2 | 2740 | |
c1103f7a JB |
2741 | actual_count++; |
2742 | ref->in_tree = 0; | |
0e0adbcf JB |
2743 | rb_erase(&ref->ref_node, &locked_ref->ref_tree); |
2744 | RB_CLEAR_NODE(&ref->ref_node); | |
c1103f7a JB |
2745 | if (!list_empty(&ref->add_list)) |
2746 | list_del(&ref->add_list); | |
2747 | /* | |
2748 | * When we play the delayed ref, also correct the ref_mod on | |
2749 | * head | |
2750 | */ | |
2751 | switch (ref->action) { | |
2752 | case BTRFS_ADD_DELAYED_REF: | |
2753 | case BTRFS_ADD_DELAYED_EXTENT: | |
d278850e | 2754 | locked_ref->ref_mod -= ref->ref_mod; |
c1103f7a JB |
2755 | break; |
2756 | case BTRFS_DROP_DELAYED_REF: | |
d278850e | 2757 | locked_ref->ref_mod += ref->ref_mod; |
c1103f7a JB |
2758 | break; |
2759 | default: | |
2760 | WARN_ON(1); | |
22cd2e7d | 2761 | } |
1ce7a5ec JB |
2762 | atomic_dec(&delayed_refs->num_entries); |
2763 | ||
b00e6250 JB |
2764 | /* |
2765 | * Record the must-insert_reserved flag before we drop the spin | |
2766 | * lock. | |
2767 | */ | |
2768 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2769 | locked_ref->must_insert_reserved = 0; | |
2770 | ||
2771 | extent_op = locked_ref->extent_op; | |
2772 | locked_ref->extent_op = NULL; | |
d7df2c79 | 2773 | spin_unlock(&locked_ref->lock); |
925baedd | 2774 | |
2ff7e61e | 2775 | ret = run_one_delayed_ref(trans, fs_info, ref, extent_op, |
56bec294 | 2776 | must_insert_reserved); |
eb099670 | 2777 | |
78a6184a | 2778 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2779 | if (ret) { |
2eadaa22 | 2780 | unselect_delayed_ref_head(delayed_refs, locked_ref); |
093486c4 | 2781 | btrfs_put_delayed_ref(ref); |
5d163e0e JM |
2782 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", |
2783 | ret); | |
79787eaa JM |
2784 | return ret; |
2785 | } | |
2786 | ||
093486c4 MX |
2787 | btrfs_put_delayed_ref(ref); |
2788 | count++; | |
c3e69d58 | 2789 | cond_resched(); |
c3e69d58 | 2790 | } |
0a2b2a84 JB |
2791 | |
2792 | /* | |
2793 | * We don't want to include ref heads since we can have empty ref heads | |
2794 | * and those will drastically skew our runtime down since we just do | |
2795 | * accounting, no actual extent tree updates. | |
2796 | */ | |
2797 | if (actual_count > 0) { | |
2798 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2799 | u64 avg; | |
2800 | ||
2801 | /* | |
2802 | * We weigh the current average higher than our current runtime | |
2803 | * to avoid large swings in the average. | |
2804 | */ | |
2805 | spin_lock(&delayed_refs->lock); | |
2806 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2807 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2808 | spin_unlock(&delayed_refs->lock); |
2809 | } | |
d7df2c79 | 2810 | return 0; |
c3e69d58 CM |
2811 | } |
2812 | ||
709c0486 AJ |
2813 | #ifdef SCRAMBLE_DELAYED_REFS |
2814 | /* | |
2815 | * Normally delayed refs get processed in ascending bytenr order. This | |
2816 | * correlates in most cases to the order added. To expose dependencies on this | |
2817 | * order, we start to process the tree in the middle instead of the beginning | |
2818 | */ | |
2819 | static u64 find_middle(struct rb_root *root) | |
2820 | { | |
2821 | struct rb_node *n = root->rb_node; | |
2822 | struct btrfs_delayed_ref_node *entry; | |
2823 | int alt = 1; | |
2824 | u64 middle; | |
2825 | u64 first = 0, last = 0; | |
2826 | ||
2827 | n = rb_first(root); | |
2828 | if (n) { | |
2829 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2830 | first = entry->bytenr; | |
2831 | } | |
2832 | n = rb_last(root); | |
2833 | if (n) { | |
2834 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2835 | last = entry->bytenr; | |
2836 | } | |
2837 | n = root->rb_node; | |
2838 | ||
2839 | while (n) { | |
2840 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2841 | WARN_ON(!entry->in_tree); | |
2842 | ||
2843 | middle = entry->bytenr; | |
2844 | ||
2845 | if (alt) | |
2846 | n = n->rb_left; | |
2847 | else | |
2848 | n = n->rb_right; | |
2849 | ||
2850 | alt = 1 - alt; | |
2851 | } | |
2852 | return middle; | |
2853 | } | |
2854 | #endif | |
2855 | ||
2ff7e61e | 2856 | static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads) |
1be41b78 JB |
2857 | { |
2858 | u64 num_bytes; | |
2859 | ||
2860 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2861 | sizeof(struct btrfs_extent_inline_ref)); | |
0b246afa | 2862 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1be41b78 JB |
2863 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
2864 | ||
2865 | /* | |
2866 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2867 | * closer to what we're really going to want to use. |
1be41b78 | 2868 | */ |
0b246afa | 2869 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info)); |
1be41b78 JB |
2870 | } |
2871 | ||
1262133b JB |
2872 | /* |
2873 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2874 | * would require to store the csums for that many bytes. | |
2875 | */ | |
2ff7e61e | 2876 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2877 | { |
2878 | u64 csum_size; | |
2879 | u64 num_csums_per_leaf; | |
2880 | u64 num_csums; | |
2881 | ||
0b246afa | 2882 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2883 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2884 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2885 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2886 | num_csums += num_csums_per_leaf - 1; |
2887 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2888 | return num_csums; | |
2889 | } | |
2890 | ||
0a2b2a84 | 2891 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2892 | struct btrfs_fs_info *fs_info) |
1be41b78 JB |
2893 | { |
2894 | struct btrfs_block_rsv *global_rsv; | |
2895 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
1262133b | 2896 | u64 csum_bytes = trans->transaction->delayed_refs.pending_csums; |
cb723e49 JB |
2897 | u64 num_dirty_bgs = trans->transaction->num_dirty_bgs; |
2898 | u64 num_bytes, num_dirty_bgs_bytes; | |
1be41b78 JB |
2899 | int ret = 0; |
2900 | ||
0b246afa | 2901 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
2ff7e61e | 2902 | num_heads = heads_to_leaves(fs_info, num_heads); |
1be41b78 | 2903 | if (num_heads > 1) |
0b246afa | 2904 | num_bytes += (num_heads - 1) * fs_info->nodesize; |
1be41b78 | 2905 | num_bytes <<= 1; |
2ff7e61e JM |
2906 | num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) * |
2907 | fs_info->nodesize; | |
0b246afa | 2908 | num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info, |
cb723e49 | 2909 | num_dirty_bgs); |
0b246afa | 2910 | global_rsv = &fs_info->global_block_rsv; |
1be41b78 JB |
2911 | |
2912 | /* | |
2913 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2914 | * wiggle room since running delayed refs can create more delayed refs. | |
2915 | */ | |
cb723e49 JB |
2916 | if (global_rsv->space_info->full) { |
2917 | num_dirty_bgs_bytes <<= 1; | |
1be41b78 | 2918 | num_bytes <<= 1; |
cb723e49 | 2919 | } |
1be41b78 JB |
2920 | |
2921 | spin_lock(&global_rsv->lock); | |
cb723e49 | 2922 | if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes) |
1be41b78 JB |
2923 | ret = 1; |
2924 | spin_unlock(&global_rsv->lock); | |
2925 | return ret; | |
2926 | } | |
2927 | ||
0a2b2a84 | 2928 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2929 | struct btrfs_fs_info *fs_info) |
0a2b2a84 | 2930 | { |
0a2b2a84 JB |
2931 | u64 num_entries = |
2932 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2933 | u64 avg_runtime; | |
a79b7d4b | 2934 | u64 val; |
0a2b2a84 JB |
2935 | |
2936 | smp_mb(); | |
2937 | avg_runtime = fs_info->avg_delayed_ref_runtime; | |
a79b7d4b | 2938 | val = num_entries * avg_runtime; |
dc1a90c6 | 2939 | if (val >= NSEC_PER_SEC) |
0a2b2a84 | 2940 | return 1; |
a79b7d4b CM |
2941 | if (val >= NSEC_PER_SEC / 2) |
2942 | return 2; | |
0a2b2a84 | 2943 | |
2ff7e61e | 2944 | return btrfs_check_space_for_delayed_refs(trans, fs_info); |
0a2b2a84 JB |
2945 | } |
2946 | ||
a79b7d4b CM |
2947 | struct async_delayed_refs { |
2948 | struct btrfs_root *root; | |
31b9655f | 2949 | u64 transid; |
a79b7d4b CM |
2950 | int count; |
2951 | int error; | |
2952 | int sync; | |
2953 | struct completion wait; | |
2954 | struct btrfs_work work; | |
2955 | }; | |
2956 | ||
2ff7e61e JM |
2957 | static inline struct async_delayed_refs * |
2958 | to_async_delayed_refs(struct btrfs_work *work) | |
2959 | { | |
2960 | return container_of(work, struct async_delayed_refs, work); | |
2961 | } | |
2962 | ||
a79b7d4b CM |
2963 | static void delayed_ref_async_start(struct btrfs_work *work) |
2964 | { | |
2ff7e61e | 2965 | struct async_delayed_refs *async = to_async_delayed_refs(work); |
a79b7d4b | 2966 | struct btrfs_trans_handle *trans; |
2ff7e61e | 2967 | struct btrfs_fs_info *fs_info = async->root->fs_info; |
a79b7d4b CM |
2968 | int ret; |
2969 | ||
0f873eca | 2970 | /* if the commit is already started, we don't need to wait here */ |
2ff7e61e | 2971 | if (btrfs_transaction_blocked(fs_info)) |
31b9655f | 2972 | goto done; |
31b9655f | 2973 | |
0f873eca CM |
2974 | trans = btrfs_join_transaction(async->root); |
2975 | if (IS_ERR(trans)) { | |
2976 | async->error = PTR_ERR(trans); | |
a79b7d4b CM |
2977 | goto done; |
2978 | } | |
2979 | ||
2980 | /* | |
01327610 | 2981 | * trans->sync means that when we call end_transaction, we won't |
a79b7d4b CM |
2982 | * wait on delayed refs |
2983 | */ | |
2984 | trans->sync = true; | |
0f873eca CM |
2985 | |
2986 | /* Don't bother flushing if we got into a different transaction */ | |
2987 | if (trans->transid > async->transid) | |
2988 | goto end; | |
2989 | ||
2ff7e61e | 2990 | ret = btrfs_run_delayed_refs(trans, fs_info, async->count); |
a79b7d4b CM |
2991 | if (ret) |
2992 | async->error = ret; | |
0f873eca | 2993 | end: |
3a45bb20 | 2994 | ret = btrfs_end_transaction(trans); |
a79b7d4b CM |
2995 | if (ret && !async->error) |
2996 | async->error = ret; | |
2997 | done: | |
2998 | if (async->sync) | |
2999 | complete(&async->wait); | |
3000 | else | |
3001 | kfree(async); | |
3002 | } | |
3003 | ||
2ff7e61e | 3004 | int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info, |
31b9655f | 3005 | unsigned long count, u64 transid, int wait) |
a79b7d4b CM |
3006 | { |
3007 | struct async_delayed_refs *async; | |
3008 | int ret; | |
3009 | ||
3010 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
3011 | if (!async) | |
3012 | return -ENOMEM; | |
3013 | ||
0b246afa | 3014 | async->root = fs_info->tree_root; |
a79b7d4b CM |
3015 | async->count = count; |
3016 | async->error = 0; | |
31b9655f | 3017 | async->transid = transid; |
a79b7d4b CM |
3018 | if (wait) |
3019 | async->sync = 1; | |
3020 | else | |
3021 | async->sync = 0; | |
3022 | init_completion(&async->wait); | |
3023 | ||
9e0af237 LB |
3024 | btrfs_init_work(&async->work, btrfs_extent_refs_helper, |
3025 | delayed_ref_async_start, NULL, NULL); | |
a79b7d4b | 3026 | |
0b246afa | 3027 | btrfs_queue_work(fs_info->extent_workers, &async->work); |
a79b7d4b CM |
3028 | |
3029 | if (wait) { | |
3030 | wait_for_completion(&async->wait); | |
3031 | ret = async->error; | |
3032 | kfree(async); | |
3033 | return ret; | |
3034 | } | |
3035 | return 0; | |
3036 | } | |
3037 | ||
c3e69d58 CM |
3038 | /* |
3039 | * this starts processing the delayed reference count updates and | |
3040 | * extent insertions we have queued up so far. count can be | |
3041 | * 0, which means to process everything in the tree at the start | |
3042 | * of the run (but not newly added entries), or it can be some target | |
3043 | * number you'd like to process. | |
79787eaa JM |
3044 | * |
3045 | * Returns 0 on success or if called with an aborted transaction | |
3046 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
3047 | */ |
3048 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3049 | struct btrfs_fs_info *fs_info, unsigned long count) |
c3e69d58 CM |
3050 | { |
3051 | struct rb_node *node; | |
3052 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 3053 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
3054 | int ret; |
3055 | int run_all = count == (unsigned long)-1; | |
d9a0540a | 3056 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
c3e69d58 | 3057 | |
79787eaa JM |
3058 | /* We'll clean this up in btrfs_cleanup_transaction */ |
3059 | if (trans->aborted) | |
3060 | return 0; | |
3061 | ||
0b246afa | 3062 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
3063 | return 0; |
3064 | ||
c3e69d58 | 3065 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 3066 | if (count == 0) |
d7df2c79 | 3067 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 3068 | |
c3e69d58 | 3069 | again: |
709c0486 AJ |
3070 | #ifdef SCRAMBLE_DELAYED_REFS |
3071 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
3072 | #endif | |
d9a0540a | 3073 | trans->can_flush_pending_bgs = false; |
2ff7e61e | 3074 | ret = __btrfs_run_delayed_refs(trans, fs_info, count); |
d7df2c79 | 3075 | if (ret < 0) { |
66642832 | 3076 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 3077 | return ret; |
eb099670 | 3078 | } |
c3e69d58 | 3079 | |
56bec294 | 3080 | if (run_all) { |
d7df2c79 | 3081 | if (!list_empty(&trans->new_bgs)) |
2ff7e61e | 3082 | btrfs_create_pending_block_groups(trans, fs_info); |
ea658bad | 3083 | |
d7df2c79 | 3084 | spin_lock(&delayed_refs->lock); |
c46effa6 | 3085 | node = rb_first(&delayed_refs->href_root); |
d7df2c79 JB |
3086 | if (!node) { |
3087 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 3088 | goto out; |
d7df2c79 | 3089 | } |
d278850e JB |
3090 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
3091 | href_node); | |
3092 | refcount_inc(&head->refs); | |
3093 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 3094 | |
d278850e JB |
3095 | /* Mutex was contended, block until it's released and retry. */ |
3096 | mutex_lock(&head->mutex); | |
3097 | mutex_unlock(&head->mutex); | |
56bec294 | 3098 | |
d278850e | 3099 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 3100 | cond_resched(); |
56bec294 | 3101 | goto again; |
5f39d397 | 3102 | } |
54aa1f4d | 3103 | out: |
d9a0540a | 3104 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
a28ec197 CM |
3105 | return 0; |
3106 | } | |
3107 | ||
5d4f98a2 | 3108 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2ff7e61e | 3109 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 3110 | u64 bytenr, u64 num_bytes, u64 flags, |
b1c79e09 | 3111 | int level, int is_data) |
5d4f98a2 YZ |
3112 | { |
3113 | struct btrfs_delayed_extent_op *extent_op; | |
3114 | int ret; | |
3115 | ||
78a6184a | 3116 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
3117 | if (!extent_op) |
3118 | return -ENOMEM; | |
3119 | ||
3120 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
3121 | extent_op->update_flags = true; |
3122 | extent_op->update_key = false; | |
3123 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 3124 | extent_op->level = level; |
5d4f98a2 | 3125 | |
0b246afa | 3126 | ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr, |
66d7e7f0 | 3127 | num_bytes, extent_op); |
5d4f98a2 | 3128 | if (ret) |
78a6184a | 3129 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
3130 | return ret; |
3131 | } | |
3132 | ||
e4c3b2dc | 3133 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3134 | struct btrfs_path *path, |
3135 | u64 objectid, u64 offset, u64 bytenr) | |
3136 | { | |
3137 | struct btrfs_delayed_ref_head *head; | |
3138 | struct btrfs_delayed_ref_node *ref; | |
3139 | struct btrfs_delayed_data_ref *data_ref; | |
3140 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 3141 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 3142 | struct rb_node *node; |
5d4f98a2 YZ |
3143 | int ret = 0; |
3144 | ||
e4c3b2dc LB |
3145 | cur_trans = root->fs_info->running_transaction; |
3146 | if (!cur_trans) | |
3147 | return 0; | |
3148 | ||
3149 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 3150 | spin_lock(&delayed_refs->lock); |
f72ad18e | 3151 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
3152 | if (!head) { |
3153 | spin_unlock(&delayed_refs->lock); | |
3154 | return 0; | |
3155 | } | |
5d4f98a2 YZ |
3156 | |
3157 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 3158 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
3159 | spin_unlock(&delayed_refs->lock); |
3160 | ||
b3b4aa74 | 3161 | btrfs_release_path(path); |
5d4f98a2 | 3162 | |
8cc33e5c DS |
3163 | /* |
3164 | * Mutex was contended, block until it's released and let | |
3165 | * caller try again | |
3166 | */ | |
5d4f98a2 YZ |
3167 | mutex_lock(&head->mutex); |
3168 | mutex_unlock(&head->mutex); | |
d278850e | 3169 | btrfs_put_delayed_ref_head(head); |
5d4f98a2 YZ |
3170 | return -EAGAIN; |
3171 | } | |
d7df2c79 | 3172 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 3173 | |
d7df2c79 | 3174 | spin_lock(&head->lock); |
0e0adbcf JB |
3175 | /* |
3176 | * XXX: We should replace this with a proper search function in the | |
3177 | * future. | |
3178 | */ | |
3179 | for (node = rb_first(&head->ref_tree); node; node = rb_next(node)) { | |
3180 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); | |
d7df2c79 JB |
3181 | /* If it's a shared ref we know a cross reference exists */ |
3182 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3183 | ret = 1; | |
3184 | break; | |
3185 | } | |
5d4f98a2 | 3186 | |
d7df2c79 | 3187 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 3188 | |
d7df2c79 JB |
3189 | /* |
3190 | * If our ref doesn't match the one we're currently looking at | |
3191 | * then we have a cross reference. | |
3192 | */ | |
3193 | if (data_ref->root != root->root_key.objectid || | |
3194 | data_ref->objectid != objectid || | |
3195 | data_ref->offset != offset) { | |
3196 | ret = 1; | |
3197 | break; | |
3198 | } | |
5d4f98a2 | 3199 | } |
d7df2c79 | 3200 | spin_unlock(&head->lock); |
5d4f98a2 | 3201 | mutex_unlock(&head->mutex); |
5d4f98a2 YZ |
3202 | return ret; |
3203 | } | |
3204 | ||
e4c3b2dc | 3205 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3206 | struct btrfs_path *path, |
3207 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d | 3208 | { |
0b246afa JM |
3209 | struct btrfs_fs_info *fs_info = root->fs_info; |
3210 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 3211 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3212 | struct btrfs_extent_data_ref *ref; |
3213 | struct btrfs_extent_inline_ref *iref; | |
3214 | struct btrfs_extent_item *ei; | |
f321e491 | 3215 | struct btrfs_key key; |
5d4f98a2 | 3216 | u32 item_size; |
3de28d57 | 3217 | int type; |
be20aa9d | 3218 | int ret; |
925baedd | 3219 | |
be20aa9d | 3220 | key.objectid = bytenr; |
31840ae1 | 3221 | key.offset = (u64)-1; |
f321e491 | 3222 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 3223 | |
be20aa9d CM |
3224 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
3225 | if (ret < 0) | |
3226 | goto out; | |
79787eaa | 3227 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
3228 | |
3229 | ret = -ENOENT; | |
3230 | if (path->slots[0] == 0) | |
31840ae1 | 3231 | goto out; |
be20aa9d | 3232 | |
31840ae1 | 3233 | path->slots[0]--; |
f321e491 | 3234 | leaf = path->nodes[0]; |
5d4f98a2 | 3235 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3236 | |
5d4f98a2 | 3237 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3238 | goto out; |
f321e491 | 3239 | |
5d4f98a2 YZ |
3240 | ret = 1; |
3241 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3242 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3243 | if (item_size < sizeof(*ei)) { | |
3244 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
3245 | goto out; | |
3246 | } | |
3247 | #endif | |
3248 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 3249 | |
5d4f98a2 YZ |
3250 | if (item_size != sizeof(*ei) + |
3251 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3252 | goto out; | |
be20aa9d | 3253 | |
5d4f98a2 YZ |
3254 | if (btrfs_extent_generation(leaf, ei) <= |
3255 | btrfs_root_last_snapshot(&root->root_item)) | |
3256 | goto out; | |
3257 | ||
3258 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 LB |
3259 | |
3260 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
3261 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
3262 | goto out; |
3263 | ||
3264 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3265 | if (btrfs_extent_refs(leaf, ei) != | |
3266 | btrfs_extent_data_ref_count(leaf, ref) || | |
3267 | btrfs_extent_data_ref_root(leaf, ref) != | |
3268 | root->root_key.objectid || | |
3269 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3270 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3271 | goto out; | |
3272 | ||
3273 | ret = 0; | |
3274 | out: | |
3275 | return ret; | |
3276 | } | |
3277 | ||
e4c3b2dc LB |
3278 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
3279 | u64 bytenr) | |
5d4f98a2 YZ |
3280 | { |
3281 | struct btrfs_path *path; | |
3282 | int ret; | |
3283 | int ret2; | |
3284 | ||
3285 | path = btrfs_alloc_path(); | |
3286 | if (!path) | |
3287 | return -ENOENT; | |
3288 | ||
3289 | do { | |
e4c3b2dc | 3290 | ret = check_committed_ref(root, path, objectid, |
5d4f98a2 YZ |
3291 | offset, bytenr); |
3292 | if (ret && ret != -ENOENT) | |
f321e491 | 3293 | goto out; |
80ff3856 | 3294 | |
e4c3b2dc | 3295 | ret2 = check_delayed_ref(root, path, objectid, |
5d4f98a2 YZ |
3296 | offset, bytenr); |
3297 | } while (ret2 == -EAGAIN); | |
3298 | ||
3299 | if (ret2 && ret2 != -ENOENT) { | |
3300 | ret = ret2; | |
3301 | goto out; | |
f321e491 | 3302 | } |
5d4f98a2 YZ |
3303 | |
3304 | if (ret != -ENOENT || ret2 != -ENOENT) | |
3305 | ret = 0; | |
be20aa9d | 3306 | out: |
80ff3856 | 3307 | btrfs_free_path(path); |
f0486c68 YZ |
3308 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3309 | WARN_ON(ret > 0); | |
f321e491 | 3310 | return ret; |
be20aa9d | 3311 | } |
c5739bba | 3312 | |
5d4f98a2 | 3313 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3314 | struct btrfs_root *root, |
5d4f98a2 | 3315 | struct extent_buffer *buf, |
e339a6b0 | 3316 | int full_backref, int inc) |
31840ae1 | 3317 | { |
0b246afa | 3318 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 3319 | u64 bytenr; |
5d4f98a2 YZ |
3320 | u64 num_bytes; |
3321 | u64 parent; | |
31840ae1 | 3322 | u64 ref_root; |
31840ae1 | 3323 | u32 nritems; |
31840ae1 ZY |
3324 | struct btrfs_key key; |
3325 | struct btrfs_file_extent_item *fi; | |
3326 | int i; | |
3327 | int level; | |
3328 | int ret = 0; | |
2ff7e61e | 3329 | int (*process_func)(struct btrfs_trans_handle *, |
84f7d8e6 | 3330 | struct btrfs_root *, |
b06c4bf5 | 3331 | u64, u64, u64, u64, u64, u64); |
31840ae1 | 3332 | |
fccb84c9 | 3333 | |
0b246afa | 3334 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3335 | return 0; |
fccb84c9 | 3336 | |
31840ae1 | 3337 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3338 | nritems = btrfs_header_nritems(buf); |
3339 | level = btrfs_header_level(buf); | |
3340 | ||
27cdeb70 | 3341 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3342 | return 0; |
31840ae1 | 3343 | |
5d4f98a2 YZ |
3344 | if (inc) |
3345 | process_func = btrfs_inc_extent_ref; | |
3346 | else | |
3347 | process_func = btrfs_free_extent; | |
31840ae1 | 3348 | |
5d4f98a2 YZ |
3349 | if (full_backref) |
3350 | parent = buf->start; | |
3351 | else | |
3352 | parent = 0; | |
3353 | ||
3354 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3355 | if (level == 0) { |
5d4f98a2 | 3356 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3357 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3358 | continue; |
5d4f98a2 | 3359 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3360 | struct btrfs_file_extent_item); |
3361 | if (btrfs_file_extent_type(buf, fi) == | |
3362 | BTRFS_FILE_EXTENT_INLINE) | |
3363 | continue; | |
3364 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3365 | if (bytenr == 0) | |
3366 | continue; | |
5d4f98a2 YZ |
3367 | |
3368 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3369 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
84f7d8e6 | 3370 | ret = process_func(trans, root, bytenr, num_bytes, |
5d4f98a2 | 3371 | parent, ref_root, key.objectid, |
b06c4bf5 | 3372 | key.offset); |
31840ae1 ZY |
3373 | if (ret) |
3374 | goto fail; | |
3375 | } else { | |
5d4f98a2 | 3376 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 3377 | num_bytes = fs_info->nodesize; |
84f7d8e6 | 3378 | ret = process_func(trans, root, bytenr, num_bytes, |
b06c4bf5 | 3379 | parent, ref_root, level - 1, 0); |
31840ae1 ZY |
3380 | if (ret) |
3381 | goto fail; | |
3382 | } | |
3383 | } | |
3384 | return 0; | |
3385 | fail: | |
5d4f98a2 YZ |
3386 | return ret; |
3387 | } | |
3388 | ||
3389 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3390 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3391 | { |
e339a6b0 | 3392 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3393 | } |
3394 | ||
3395 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3396 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3397 | { |
e339a6b0 | 3398 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3399 | } |
3400 | ||
9078a3e1 | 3401 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 3402 | struct btrfs_fs_info *fs_info, |
9078a3e1 CM |
3403 | struct btrfs_path *path, |
3404 | struct btrfs_block_group_cache *cache) | |
3405 | { | |
3406 | int ret; | |
0b246afa | 3407 | struct btrfs_root *extent_root = fs_info->extent_root; |
5f39d397 CM |
3408 | unsigned long bi; |
3409 | struct extent_buffer *leaf; | |
9078a3e1 | 3410 | |
9078a3e1 | 3411 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3412 | if (ret) { |
3413 | if (ret > 0) | |
3414 | ret = -ENOENT; | |
54aa1f4d | 3415 | goto fail; |
df95e7f0 | 3416 | } |
5f39d397 CM |
3417 | |
3418 | leaf = path->nodes[0]; | |
3419 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3420 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3421 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3422 | fail: |
24b89d08 | 3423 | btrfs_release_path(path); |
df95e7f0 | 3424 | return ret; |
9078a3e1 CM |
3425 | |
3426 | } | |
3427 | ||
4a8c9a62 | 3428 | static struct btrfs_block_group_cache * |
2ff7e61e | 3429 | next_block_group(struct btrfs_fs_info *fs_info, |
4a8c9a62 YZ |
3430 | struct btrfs_block_group_cache *cache) |
3431 | { | |
3432 | struct rb_node *node; | |
292cbd51 | 3433 | |
0b246afa | 3434 | spin_lock(&fs_info->block_group_cache_lock); |
292cbd51 FM |
3435 | |
3436 | /* If our block group was removed, we need a full search. */ | |
3437 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3438 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3439 | ||
0b246afa | 3440 | spin_unlock(&fs_info->block_group_cache_lock); |
292cbd51 | 3441 | btrfs_put_block_group(cache); |
0b246afa | 3442 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; |
292cbd51 | 3443 | } |
4a8c9a62 YZ |
3444 | node = rb_next(&cache->cache_node); |
3445 | btrfs_put_block_group(cache); | |
3446 | if (node) { | |
3447 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3448 | cache_node); | |
11dfe35a | 3449 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3450 | } else |
3451 | cache = NULL; | |
0b246afa | 3452 | spin_unlock(&fs_info->block_group_cache_lock); |
4a8c9a62 YZ |
3453 | return cache; |
3454 | } | |
3455 | ||
0af3d00b JB |
3456 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3457 | struct btrfs_trans_handle *trans, | |
3458 | struct btrfs_path *path) | |
3459 | { | |
0b246afa JM |
3460 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
3461 | struct btrfs_root *root = fs_info->tree_root; | |
0af3d00b | 3462 | struct inode *inode = NULL; |
364ecf36 | 3463 | struct extent_changeset *data_reserved = NULL; |
0af3d00b | 3464 | u64 alloc_hint = 0; |
2b20982e | 3465 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3466 | u64 num_pages = 0; |
0af3d00b JB |
3467 | int retries = 0; |
3468 | int ret = 0; | |
3469 | ||
3470 | /* | |
3471 | * If this block group is smaller than 100 megs don't bother caching the | |
3472 | * block group. | |
3473 | */ | |
ee22184b | 3474 | if (block_group->key.offset < (100 * SZ_1M)) { |
0af3d00b JB |
3475 | spin_lock(&block_group->lock); |
3476 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3477 | spin_unlock(&block_group->lock); | |
3478 | return 0; | |
3479 | } | |
3480 | ||
0c0ef4bc JB |
3481 | if (trans->aborted) |
3482 | return 0; | |
0af3d00b | 3483 | again: |
77ab86bf | 3484 | inode = lookup_free_space_inode(fs_info, block_group, path); |
0af3d00b JB |
3485 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { |
3486 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3487 | btrfs_release_path(path); |
0af3d00b JB |
3488 | goto out; |
3489 | } | |
3490 | ||
3491 | if (IS_ERR(inode)) { | |
3492 | BUG_ON(retries); | |
3493 | retries++; | |
3494 | ||
3495 | if (block_group->ro) | |
3496 | goto out_free; | |
3497 | ||
77ab86bf JM |
3498 | ret = create_free_space_inode(fs_info, trans, block_group, |
3499 | path); | |
0af3d00b JB |
3500 | if (ret) |
3501 | goto out_free; | |
3502 | goto again; | |
3503 | } | |
3504 | ||
5b0e95bf JB |
3505 | /* We've already setup this transaction, go ahead and exit */ |
3506 | if (block_group->cache_generation == trans->transid && | |
3507 | i_size_read(inode)) { | |
3508 | dcs = BTRFS_DC_SETUP; | |
3509 | goto out_put; | |
3510 | } | |
3511 | ||
0af3d00b JB |
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 | ||
3535 | if (i_size_read(inode) > 0) { | |
2ff7e61e | 3536 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
0b246afa | 3537 | &fs_info->global_block_rsv); |
7b61cd92 MX |
3538 | if (ret) |
3539 | goto out_put; | |
3540 | ||
77ab86bf | 3541 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); |
0af3d00b JB |
3542 | if (ret) |
3543 | goto out_put; | |
3544 | } | |
3545 | ||
3546 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3547 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
0b246afa | 3548 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cf7c1ef6 LB |
3549 | /* |
3550 | * don't bother trying to write stuff out _if_ | |
3551 | * a) we're not cached, | |
1a79c1f2 LB |
3552 | * b) we're with nospace_cache mount option, |
3553 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
cf7c1ef6 | 3554 | */ |
2b20982e | 3555 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3556 | spin_unlock(&block_group->lock); |
3557 | goto out_put; | |
3558 | } | |
3559 | spin_unlock(&block_group->lock); | |
3560 | ||
2968b1f4 JB |
3561 | /* |
3562 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3563 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3564 | */ | |
3565 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3566 | ret = -ENOSPC; | |
3567 | goto out_put; | |
3568 | } | |
3569 | ||
6fc823b1 JB |
3570 | /* |
3571 | * Try to preallocate enough space based on how big the block group is. | |
3572 | * Keep in mind this has to include any pinned space which could end up | |
3573 | * taking up quite a bit since it's not folded into the other space | |
3574 | * cache. | |
3575 | */ | |
ee22184b | 3576 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
0af3d00b JB |
3577 | if (!num_pages) |
3578 | num_pages = 1; | |
3579 | ||
0af3d00b | 3580 | num_pages *= 16; |
09cbfeaf | 3581 | num_pages *= PAGE_SIZE; |
0af3d00b | 3582 | |
364ecf36 | 3583 | ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages); |
0af3d00b JB |
3584 | if (ret) |
3585 | goto out_put; | |
3586 | ||
3587 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3588 | num_pages, num_pages, | |
3589 | &alloc_hint); | |
2968b1f4 JB |
3590 | /* |
3591 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3592 | * of metadata or split extents when writing the cache out, which means | |
3593 | * we can enospc if we are heavily fragmented in addition to just normal | |
3594 | * out of space conditions. So if we hit this just skip setting up any | |
3595 | * other block groups for this transaction, maybe we'll unpin enough | |
3596 | * space the next time around. | |
3597 | */ | |
2b20982e JB |
3598 | if (!ret) |
3599 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3600 | else if (ret == -ENOSPC) |
3601 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
c09544e0 | 3602 | |
0af3d00b JB |
3603 | out_put: |
3604 | iput(inode); | |
3605 | out_free: | |
b3b4aa74 | 3606 | btrfs_release_path(path); |
0af3d00b JB |
3607 | out: |
3608 | spin_lock(&block_group->lock); | |
e65cbb94 | 3609 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3610 | block_group->cache_generation = trans->transid; |
2b20982e | 3611 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3612 | spin_unlock(&block_group->lock); |
3613 | ||
364ecf36 | 3614 | extent_changeset_free(data_reserved); |
0af3d00b JB |
3615 | return ret; |
3616 | } | |
3617 | ||
dcdf7f6d | 3618 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans, |
2ff7e61e | 3619 | struct btrfs_fs_info *fs_info) |
dcdf7f6d JB |
3620 | { |
3621 | struct btrfs_block_group_cache *cache, *tmp; | |
3622 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3623 | struct btrfs_path *path; | |
3624 | ||
3625 | if (list_empty(&cur_trans->dirty_bgs) || | |
0b246afa | 3626 | !btrfs_test_opt(fs_info, SPACE_CACHE)) |
dcdf7f6d JB |
3627 | return 0; |
3628 | ||
3629 | path = btrfs_alloc_path(); | |
3630 | if (!path) | |
3631 | return -ENOMEM; | |
3632 | ||
3633 | /* Could add new block groups, use _safe just in case */ | |
3634 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3635 | dirty_list) { | |
3636 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3637 | cache_save_setup(cache, trans, path); | |
3638 | } | |
3639 | ||
3640 | btrfs_free_path(path); | |
3641 | return 0; | |
3642 | } | |
3643 | ||
1bbc621e CM |
3644 | /* |
3645 | * transaction commit does final block group cache writeback during a | |
3646 | * critical section where nothing is allowed to change the FS. This is | |
3647 | * required in order for the cache to actually match the block group, | |
3648 | * but can introduce a lot of latency into the commit. | |
3649 | * | |
3650 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3651 | * cache IO. There's a chance we'll have to redo some of it if the | |
3652 | * block group changes again during the commit, but it greatly reduces | |
3653 | * the commit latency by getting rid of the easy block groups while | |
3654 | * we're still allowing others to join the commit. | |
3655 | */ | |
3656 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3657 | struct btrfs_fs_info *fs_info) |
9078a3e1 | 3658 | { |
4a8c9a62 | 3659 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3660 | struct btrfs_transaction *cur_trans = trans->transaction; |
3661 | int ret = 0; | |
c9dc4c65 | 3662 | int should_put; |
1bbc621e CM |
3663 | struct btrfs_path *path = NULL; |
3664 | LIST_HEAD(dirty); | |
3665 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3666 | int num_started = 0; |
1bbc621e CM |
3667 | int loops = 0; |
3668 | ||
3669 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3670 | if (list_empty(&cur_trans->dirty_bgs)) { |
3671 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3672 | return 0; | |
1bbc621e | 3673 | } |
b58d1a9e | 3674 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3675 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3676 | |
1bbc621e | 3677 | again: |
1bbc621e CM |
3678 | /* |
3679 | * make sure all the block groups on our dirty list actually | |
3680 | * exist | |
3681 | */ | |
2ff7e61e | 3682 | btrfs_create_pending_block_groups(trans, fs_info); |
1bbc621e CM |
3683 | |
3684 | if (!path) { | |
3685 | path = btrfs_alloc_path(); | |
3686 | if (!path) | |
3687 | return -ENOMEM; | |
3688 | } | |
3689 | ||
b58d1a9e FM |
3690 | /* |
3691 | * cache_write_mutex is here only to save us from balance or automatic | |
3692 | * removal of empty block groups deleting this block group while we are | |
3693 | * writing out the cache | |
3694 | */ | |
3695 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e CM |
3696 | while (!list_empty(&dirty)) { |
3697 | cache = list_first_entry(&dirty, | |
3698 | struct btrfs_block_group_cache, | |
3699 | dirty_list); | |
1bbc621e CM |
3700 | /* |
3701 | * this can happen if something re-dirties a block | |
3702 | * group that is already under IO. Just wait for it to | |
3703 | * finish and then do it all again | |
3704 | */ | |
3705 | if (!list_empty(&cache->io_list)) { | |
3706 | list_del_init(&cache->io_list); | |
afdb5718 | 3707 | btrfs_wait_cache_io(trans, cache, path); |
1bbc621e CM |
3708 | btrfs_put_block_group(cache); |
3709 | } | |
3710 | ||
3711 | ||
3712 | /* | |
3713 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3714 | * if it should update the cache_state. Don't delete | |
3715 | * until after we wait. | |
3716 | * | |
3717 | * Since we're not running in the commit critical section | |
3718 | * we need the dirty_bgs_lock to protect from update_block_group | |
3719 | */ | |
3720 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3721 | list_del_init(&cache->dirty_list); | |
3722 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3723 | ||
3724 | should_put = 1; | |
3725 | ||
3726 | cache_save_setup(cache, trans, path); | |
3727 | ||
3728 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3729 | cache->io_ctl.inode = NULL; | |
0b246afa | 3730 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3731 | cache, path); |
1bbc621e CM |
3732 | if (ret == 0 && cache->io_ctl.inode) { |
3733 | num_started++; | |
3734 | should_put = 0; | |
3735 | ||
3736 | /* | |
3737 | * the cache_write_mutex is protecting | |
3738 | * the io_list | |
3739 | */ | |
3740 | list_add_tail(&cache->io_list, io); | |
3741 | } else { | |
3742 | /* | |
3743 | * if we failed to write the cache, the | |
3744 | * generation will be bad and life goes on | |
3745 | */ | |
3746 | ret = 0; | |
3747 | } | |
3748 | } | |
ff1f8250 | 3749 | if (!ret) { |
2ff7e61e JM |
3750 | ret = write_one_cache_group(trans, fs_info, |
3751 | path, cache); | |
ff1f8250 FM |
3752 | /* |
3753 | * Our block group might still be attached to the list | |
3754 | * of new block groups in the transaction handle of some | |
3755 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3756 | * means its block group item isn't yet in the extent | |
3757 | * tree. If this happens ignore the error, as we will | |
3758 | * try again later in the critical section of the | |
3759 | * transaction commit. | |
3760 | */ | |
3761 | if (ret == -ENOENT) { | |
3762 | ret = 0; | |
3763 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3764 | if (list_empty(&cache->dirty_list)) { | |
3765 | list_add_tail(&cache->dirty_list, | |
3766 | &cur_trans->dirty_bgs); | |
3767 | btrfs_get_block_group(cache); | |
3768 | } | |
3769 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3770 | } else if (ret) { | |
66642832 | 3771 | btrfs_abort_transaction(trans, ret); |
ff1f8250 FM |
3772 | } |
3773 | } | |
1bbc621e CM |
3774 | |
3775 | /* if its not on the io list, we need to put the block group */ | |
3776 | if (should_put) | |
3777 | btrfs_put_block_group(cache); | |
3778 | ||
3779 | if (ret) | |
3780 | break; | |
b58d1a9e FM |
3781 | |
3782 | /* | |
3783 | * Avoid blocking other tasks for too long. It might even save | |
3784 | * us from writing caches for block groups that are going to be | |
3785 | * removed. | |
3786 | */ | |
3787 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3788 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3789 | } |
b58d1a9e | 3790 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3791 | |
3792 | /* | |
3793 | * go through delayed refs for all the stuff we've just kicked off | |
3794 | * and then loop back (just once) | |
3795 | */ | |
2ff7e61e | 3796 | ret = btrfs_run_delayed_refs(trans, fs_info, 0); |
1bbc621e CM |
3797 | if (!ret && loops == 0) { |
3798 | loops++; | |
3799 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3800 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3801 | /* |
3802 | * dirty_bgs_lock protects us from concurrent block group | |
3803 | * deletes too (not just cache_write_mutex). | |
3804 | */ | |
3805 | if (!list_empty(&dirty)) { | |
3806 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3807 | goto again; | |
3808 | } | |
1bbc621e | 3809 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c79a1751 | 3810 | } else if (ret < 0) { |
2ff7e61e | 3811 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
1bbc621e CM |
3812 | } |
3813 | ||
3814 | btrfs_free_path(path); | |
3815 | return ret; | |
3816 | } | |
3817 | ||
3818 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3819 | struct btrfs_fs_info *fs_info) |
1bbc621e CM |
3820 | { |
3821 | struct btrfs_block_group_cache *cache; | |
3822 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3823 | int ret = 0; | |
3824 | int should_put; | |
3825 | struct btrfs_path *path; | |
3826 | struct list_head *io = &cur_trans->io_bgs; | |
3827 | int num_started = 0; | |
9078a3e1 CM |
3828 | |
3829 | path = btrfs_alloc_path(); | |
3830 | if (!path) | |
3831 | return -ENOMEM; | |
3832 | ||
ce93ec54 | 3833 | /* |
e44081ef FM |
3834 | * Even though we are in the critical section of the transaction commit, |
3835 | * we can still have concurrent tasks adding elements to this | |
3836 | * transaction's list of dirty block groups. These tasks correspond to | |
3837 | * endio free space workers started when writeback finishes for a | |
3838 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3839 | * allocate new block groups as a result of COWing nodes of the root | |
3840 | * tree when updating the free space inode. The writeback for the space | |
3841 | * caches is triggered by an earlier call to | |
3842 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3843 | * loop. | |
3844 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3845 | * delayed refs to make sure we have the best chance at doing this all |
3846 | * in one shot. | |
3847 | */ | |
e44081ef | 3848 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3849 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3850 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3851 | struct btrfs_block_group_cache, | |
3852 | dirty_list); | |
c9dc4c65 CM |
3853 | |
3854 | /* | |
3855 | * this can happen if cache_save_setup re-dirties a block | |
3856 | * group that is already under IO. Just wait for it to | |
3857 | * finish and then do it all again | |
3858 | */ | |
3859 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3860 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3861 | list_del_init(&cache->io_list); |
afdb5718 | 3862 | btrfs_wait_cache_io(trans, cache, path); |
c9dc4c65 | 3863 | btrfs_put_block_group(cache); |
e44081ef | 3864 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3865 | } |
3866 | ||
1bbc621e CM |
3867 | /* |
3868 | * don't remove from the dirty list until after we've waited | |
3869 | * on any pending IO | |
3870 | */ | |
ce93ec54 | 3871 | list_del_init(&cache->dirty_list); |
e44081ef | 3872 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3873 | should_put = 1; |
3874 | ||
1bbc621e | 3875 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3876 | |
ce93ec54 | 3877 | if (!ret) |
2ff7e61e JM |
3878 | ret = btrfs_run_delayed_refs(trans, fs_info, |
3879 | (unsigned long) -1); | |
c9dc4c65 CM |
3880 | |
3881 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3882 | cache->io_ctl.inode = NULL; | |
0b246afa | 3883 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3884 | cache, path); |
c9dc4c65 CM |
3885 | if (ret == 0 && cache->io_ctl.inode) { |
3886 | num_started++; | |
3887 | should_put = 0; | |
1bbc621e | 3888 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3889 | } else { |
3890 | /* | |
3891 | * if we failed to write the cache, the | |
3892 | * generation will be bad and life goes on | |
3893 | */ | |
3894 | ret = 0; | |
3895 | } | |
3896 | } | |
ff1f8250 | 3897 | if (!ret) { |
2ff7e61e JM |
3898 | ret = write_one_cache_group(trans, fs_info, |
3899 | path, cache); | |
2bc0bb5f FM |
3900 | /* |
3901 | * One of the free space endio workers might have | |
3902 | * created a new block group while updating a free space | |
3903 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3904 | * and hasn't released its transaction handle yet, in | |
3905 | * which case the new block group is still attached to | |
3906 | * its transaction handle and its creation has not | |
3907 | * finished yet (no block group item in the extent tree | |
3908 | * yet, etc). If this is the case, wait for all free | |
3909 | * space endio workers to finish and retry. This is a | |
3910 | * a very rare case so no need for a more efficient and | |
3911 | * complex approach. | |
3912 | */ | |
3913 | if (ret == -ENOENT) { | |
3914 | wait_event(cur_trans->writer_wait, | |
3915 | atomic_read(&cur_trans->num_writers) == 1); | |
2ff7e61e JM |
3916 | ret = write_one_cache_group(trans, fs_info, |
3917 | path, cache); | |
2bc0bb5f | 3918 | } |
ff1f8250 | 3919 | if (ret) |
66642832 | 3920 | btrfs_abort_transaction(trans, ret); |
ff1f8250 | 3921 | } |
c9dc4c65 CM |
3922 | |
3923 | /* if its not on the io list, we need to put the block group */ | |
3924 | if (should_put) | |
3925 | btrfs_put_block_group(cache); | |
e44081ef | 3926 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3927 | } |
e44081ef | 3928 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3929 | |
1bbc621e CM |
3930 | while (!list_empty(io)) { |
3931 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3932 | io_list); |
3933 | list_del_init(&cache->io_list); | |
afdb5718 | 3934 | btrfs_wait_cache_io(trans, cache, path); |
0cb59c99 JB |
3935 | btrfs_put_block_group(cache); |
3936 | } | |
3937 | ||
9078a3e1 | 3938 | btrfs_free_path(path); |
ce93ec54 | 3939 | return ret; |
9078a3e1 CM |
3940 | } |
3941 | ||
2ff7e61e | 3942 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 YZ |
3943 | { |
3944 | struct btrfs_block_group_cache *block_group; | |
3945 | int readonly = 0; | |
3946 | ||
0b246afa | 3947 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
3948 | if (!block_group || block_group->ro) |
3949 | readonly = 1; | |
3950 | if (block_group) | |
fa9c0d79 | 3951 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3952 | return readonly; |
3953 | } | |
3954 | ||
f78c436c FM |
3955 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
3956 | { | |
3957 | struct btrfs_block_group_cache *bg; | |
3958 | bool ret = true; | |
3959 | ||
3960 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3961 | if (!bg) | |
3962 | return false; | |
3963 | ||
3964 | spin_lock(&bg->lock); | |
3965 | if (bg->ro) | |
3966 | ret = false; | |
3967 | else | |
3968 | atomic_inc(&bg->nocow_writers); | |
3969 | spin_unlock(&bg->lock); | |
3970 | ||
3971 | /* no put on block group, done by btrfs_dec_nocow_writers */ | |
3972 | if (!ret) | |
3973 | btrfs_put_block_group(bg); | |
3974 | ||
3975 | return ret; | |
3976 | ||
3977 | } | |
3978 | ||
3979 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
3980 | { | |
3981 | struct btrfs_block_group_cache *bg; | |
3982 | ||
3983 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3984 | ASSERT(bg); | |
3985 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
3986 | wake_up_atomic_t(&bg->nocow_writers); | |
3987 | /* | |
3988 | * Once for our lookup and once for the lookup done by a previous call | |
3989 | * to btrfs_inc_nocow_writers() | |
3990 | */ | |
3991 | btrfs_put_block_group(bg); | |
3992 | btrfs_put_block_group(bg); | |
3993 | } | |
3994 | ||
3995 | static int btrfs_wait_nocow_writers_atomic_t(atomic_t *a) | |
3996 | { | |
3997 | schedule(); | |
3998 | return 0; | |
3999 | } | |
4000 | ||
4001 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) | |
4002 | { | |
4003 | wait_on_atomic_t(&bg->nocow_writers, | |
4004 | btrfs_wait_nocow_writers_atomic_t, | |
4005 | TASK_UNINTERRUPTIBLE); | |
4006 | } | |
4007 | ||
6ab0a202 JM |
4008 | static const char *alloc_name(u64 flags) |
4009 | { | |
4010 | switch (flags) { | |
4011 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
4012 | return "mixed"; | |
4013 | case BTRFS_BLOCK_GROUP_METADATA: | |
4014 | return "metadata"; | |
4015 | case BTRFS_BLOCK_GROUP_DATA: | |
4016 | return "data"; | |
4017 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
4018 | return "system"; | |
4019 | default: | |
4020 | WARN_ON(1); | |
4021 | return "invalid-combination"; | |
4022 | }; | |
4023 | } | |
4024 | ||
2be12ef7 NB |
4025 | static int create_space_info(struct btrfs_fs_info *info, u64 flags, |
4026 | struct btrfs_space_info **new) | |
4027 | { | |
4028 | ||
4029 | struct btrfs_space_info *space_info; | |
4030 | int i; | |
4031 | int ret; | |
4032 | ||
4033 | space_info = kzalloc(sizeof(*space_info), GFP_NOFS); | |
4034 | if (!space_info) | |
4035 | return -ENOMEM; | |
4036 | ||
4037 | ret = percpu_counter_init(&space_info->total_bytes_pinned, 0, | |
4038 | GFP_KERNEL); | |
4039 | if (ret) { | |
4040 | kfree(space_info); | |
4041 | return ret; | |
4042 | } | |
4043 | ||
4044 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
4045 | INIT_LIST_HEAD(&space_info->block_groups[i]); | |
4046 | init_rwsem(&space_info->groups_sem); | |
4047 | spin_lock_init(&space_info->lock); | |
4048 | space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; | |
4049 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
4050 | init_waitqueue_head(&space_info->wait); | |
4051 | INIT_LIST_HEAD(&space_info->ro_bgs); | |
4052 | INIT_LIST_HEAD(&space_info->tickets); | |
4053 | INIT_LIST_HEAD(&space_info->priority_tickets); | |
4054 | ||
4055 | ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype, | |
4056 | info->space_info_kobj, "%s", | |
4057 | alloc_name(space_info->flags)); | |
4058 | if (ret) { | |
4059 | percpu_counter_destroy(&space_info->total_bytes_pinned); | |
4060 | kfree(space_info); | |
4061 | return ret; | |
4062 | } | |
4063 | ||
4064 | *new = space_info; | |
4065 | list_add_rcu(&space_info->list, &info->space_info); | |
4066 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
4067 | info->data_sinfo = space_info; | |
4068 | ||
4069 | return ret; | |
4070 | } | |
4071 | ||
d2006e6d | 4072 | static void update_space_info(struct btrfs_fs_info *info, u64 flags, |
593060d7 | 4073 | u64 total_bytes, u64 bytes_used, |
e40edf2d | 4074 | u64 bytes_readonly, |
593060d7 CM |
4075 | struct btrfs_space_info **space_info) |
4076 | { | |
4077 | struct btrfs_space_info *found; | |
b742bb82 YZ |
4078 | int factor; |
4079 | ||
4080 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
4081 | BTRFS_BLOCK_GROUP_RAID10)) | |
4082 | factor = 2; | |
4083 | else | |
4084 | factor = 1; | |
593060d7 CM |
4085 | |
4086 | found = __find_space_info(info, flags); | |
d2006e6d NB |
4087 | ASSERT(found); |
4088 | spin_lock(&found->lock); | |
4089 | found->total_bytes += total_bytes; | |
4090 | found->disk_total += total_bytes * factor; | |
4091 | found->bytes_used += bytes_used; | |
4092 | found->disk_used += bytes_used * factor; | |
4093 | found->bytes_readonly += bytes_readonly; | |
4094 | if (total_bytes > 0) | |
4095 | found->full = 0; | |
4096 | space_info_add_new_bytes(info, found, total_bytes - | |
4097 | bytes_used - bytes_readonly); | |
4098 | spin_unlock(&found->lock); | |
4099 | *space_info = found; | |
593060d7 CM |
4100 | } |
4101 | ||
8790d502 CM |
4102 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
4103 | { | |
899c81ea ID |
4104 | u64 extra_flags = chunk_to_extended(flags) & |
4105 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 4106 | |
de98ced9 | 4107 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
4108 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
4109 | fs_info->avail_data_alloc_bits |= extra_flags; | |
4110 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
4111 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
4112 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
4113 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 4114 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 4115 | } |
593060d7 | 4116 | |
fc67c450 ID |
4117 | /* |
4118 | * returns target flags in extended format or 0 if restripe for this | |
4119 | * chunk_type is not in progress | |
c6664b42 ID |
4120 | * |
4121 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
4122 | */ |
4123 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
4124 | { | |
4125 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4126 | u64 target = 0; | |
4127 | ||
fc67c450 ID |
4128 | if (!bctl) |
4129 | return 0; | |
4130 | ||
4131 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
4132 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4133 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
4134 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
4135 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4136 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
4137 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
4138 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4139 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
4140 | } | |
4141 | ||
4142 | return target; | |
4143 | } | |
4144 | ||
a46d11a8 ID |
4145 | /* |
4146 | * @flags: available profiles in extended format (see ctree.h) | |
4147 | * | |
e4d8ec0f ID |
4148 | * Returns reduced profile in chunk format. If profile changing is in |
4149 | * progress (either running or paused) picks the target profile (if it's | |
4150 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 4151 | */ |
2ff7e61e | 4152 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c | 4153 | { |
0b246afa | 4154 | u64 num_devices = fs_info->fs_devices->rw_devices; |
fc67c450 | 4155 | u64 target; |
9c170b26 ZL |
4156 | u64 raid_type; |
4157 | u64 allowed = 0; | |
a061fc8d | 4158 | |
fc67c450 ID |
4159 | /* |
4160 | * see if restripe for this chunk_type is in progress, if so | |
4161 | * try to reduce to the target profile | |
4162 | */ | |
0b246afa JM |
4163 | spin_lock(&fs_info->balance_lock); |
4164 | target = get_restripe_target(fs_info, flags); | |
fc67c450 ID |
4165 | if (target) { |
4166 | /* pick target profile only if it's already available */ | |
4167 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
0b246afa | 4168 | spin_unlock(&fs_info->balance_lock); |
fc67c450 | 4169 | return extended_to_chunk(target); |
e4d8ec0f ID |
4170 | } |
4171 | } | |
0b246afa | 4172 | spin_unlock(&fs_info->balance_lock); |
e4d8ec0f | 4173 | |
53b381b3 | 4174 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
4175 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
4176 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
4177 | allowed |= btrfs_raid_group[raid_type]; | |
4178 | } | |
4179 | allowed &= flags; | |
4180 | ||
4181 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
4182 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
4183 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
4184 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
4185 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
4186 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
4187 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
4188 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
4189 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
4190 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
4191 | ||
4192 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
4193 | ||
4194 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
4195 | } |
4196 | ||
2ff7e61e | 4197 | static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
6a63209f | 4198 | { |
de98ced9 | 4199 | unsigned seq; |
f8213bdc | 4200 | u64 flags; |
de98ced9 MX |
4201 | |
4202 | do { | |
f8213bdc | 4203 | flags = orig_flags; |
0b246afa | 4204 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 MX |
4205 | |
4206 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
0b246afa | 4207 | flags |= fs_info->avail_data_alloc_bits; |
de98ced9 | 4208 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
0b246afa | 4209 | flags |= fs_info->avail_system_alloc_bits; |
de98ced9 | 4210 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
0b246afa JM |
4211 | flags |= fs_info->avail_metadata_alloc_bits; |
4212 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
6fef8df1 | 4213 | |
2ff7e61e | 4214 | return btrfs_reduce_alloc_profile(fs_info, flags); |
6a63209f JB |
4215 | } |
4216 | ||
1b86826d | 4217 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 4218 | { |
0b246afa | 4219 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 4220 | u64 flags; |
53b381b3 | 4221 | u64 ret; |
9ed74f2d | 4222 | |
b742bb82 YZ |
4223 | if (data) |
4224 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 4225 | else if (root == fs_info->chunk_root) |
b742bb82 | 4226 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 4227 | else |
b742bb82 | 4228 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 4229 | |
2ff7e61e | 4230 | ret = get_alloc_profile(fs_info, flags); |
53b381b3 | 4231 | return ret; |
6a63209f | 4232 | } |
9ed74f2d | 4233 | |
1b86826d JM |
4234 | u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info) |
4235 | { | |
4236 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA); | |
4237 | } | |
4238 | ||
4239 | u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info) | |
4240 | { | |
4241 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4242 | } | |
4243 | ||
4244 | u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info) | |
4245 | { | |
4246 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
4247 | } | |
4248 | ||
4136135b LB |
4249 | static u64 btrfs_space_info_used(struct btrfs_space_info *s_info, |
4250 | bool may_use_included) | |
4251 | { | |
4252 | ASSERT(s_info); | |
4253 | return s_info->bytes_used + s_info->bytes_reserved + | |
4254 | s_info->bytes_pinned + s_info->bytes_readonly + | |
4255 | (may_use_included ? s_info->bytes_may_use : 0); | |
4256 | } | |
4257 | ||
04f4f916 | 4258 | int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes) |
6a63209f | 4259 | { |
04f4f916 | 4260 | struct btrfs_root *root = inode->root; |
b4d7c3c9 | 4261 | struct btrfs_fs_info *fs_info = root->fs_info; |
1174cade | 4262 | struct btrfs_space_info *data_sinfo = fs_info->data_sinfo; |
ab6e2410 | 4263 | u64 used; |
94b947b2 | 4264 | int ret = 0; |
c99f1b0c ZL |
4265 | int need_commit = 2; |
4266 | int have_pinned_space; | |
6a63209f | 4267 | |
6a63209f | 4268 | /* make sure bytes are sectorsize aligned */ |
0b246afa | 4269 | bytes = ALIGN(bytes, fs_info->sectorsize); |
6a63209f | 4270 | |
9dced186 | 4271 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 4272 | need_commit = 0; |
9dced186 | 4273 | ASSERT(current->journal_info); |
0af3d00b JB |
4274 | } |
4275 | ||
6a63209f JB |
4276 | again: |
4277 | /* make sure we have enough space to handle the data first */ | |
4278 | spin_lock(&data_sinfo->lock); | |
4136135b | 4279 | used = btrfs_space_info_used(data_sinfo, true); |
ab6e2410 JB |
4280 | |
4281 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 4282 | struct btrfs_trans_handle *trans; |
9ed74f2d | 4283 | |
6a63209f JB |
4284 | /* |
4285 | * if we don't have enough free bytes in this space then we need | |
4286 | * to alloc a new chunk. | |
4287 | */ | |
b9fd47cd | 4288 | if (!data_sinfo->full) { |
6a63209f | 4289 | u64 alloc_target; |
9ed74f2d | 4290 | |
0e4f8f88 | 4291 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 4292 | spin_unlock(&data_sinfo->lock); |
1174cade | 4293 | |
1b86826d | 4294 | alloc_target = btrfs_data_alloc_profile(fs_info); |
9dced186 MX |
4295 | /* |
4296 | * It is ugly that we don't call nolock join | |
4297 | * transaction for the free space inode case here. | |
4298 | * But it is safe because we only do the data space | |
4299 | * reservation for the free space cache in the | |
4300 | * transaction context, the common join transaction | |
4301 | * just increase the counter of the current transaction | |
4302 | * handler, doesn't try to acquire the trans_lock of | |
4303 | * the fs. | |
4304 | */ | |
7a7eaa40 | 4305 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4306 | if (IS_ERR(trans)) |
4307 | return PTR_ERR(trans); | |
9ed74f2d | 4308 | |
2ff7e61e | 4309 | ret = do_chunk_alloc(trans, fs_info, alloc_target, |
0e4f8f88 | 4310 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 4311 | btrfs_end_transaction(trans); |
d52a5b5f MX |
4312 | if (ret < 0) { |
4313 | if (ret != -ENOSPC) | |
4314 | return ret; | |
c99f1b0c ZL |
4315 | else { |
4316 | have_pinned_space = 1; | |
d52a5b5f | 4317 | goto commit_trans; |
c99f1b0c | 4318 | } |
d52a5b5f | 4319 | } |
9ed74f2d | 4320 | |
6a63209f JB |
4321 | goto again; |
4322 | } | |
f2bb8f5c JB |
4323 | |
4324 | /* | |
b150a4f1 | 4325 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
4326 | * allocation, and no removed chunk in current transaction, |
4327 | * don't bother committing the transaction. | |
f2bb8f5c | 4328 | */ |
c99f1b0c ZL |
4329 | have_pinned_space = percpu_counter_compare( |
4330 | &data_sinfo->total_bytes_pinned, | |
4331 | used + bytes - data_sinfo->total_bytes); | |
6a63209f | 4332 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4333 | |
4e06bdd6 | 4334 | /* commit the current transaction and try again */ |
d52a5b5f | 4335 | commit_trans: |
c99f1b0c | 4336 | if (need_commit && |
0b246afa | 4337 | !atomic_read(&fs_info->open_ioctl_trans)) { |
c99f1b0c | 4338 | need_commit--; |
b150a4f1 | 4339 | |
e1746e83 ZL |
4340 | if (need_commit > 0) { |
4341 | btrfs_start_delalloc_roots(fs_info, 0, -1); | |
6374e57a | 4342 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, |
0b246afa | 4343 | (u64)-1); |
e1746e83 | 4344 | } |
9a4e7276 | 4345 | |
7a7eaa40 | 4346 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4347 | if (IS_ERR(trans)) |
4348 | return PTR_ERR(trans); | |
c99f1b0c | 4349 | if (have_pinned_space >= 0 || |
3204d33c JB |
4350 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4351 | &trans->transaction->flags) || | |
c99f1b0c | 4352 | need_commit > 0) { |
3a45bb20 | 4353 | ret = btrfs_commit_transaction(trans); |
94b947b2 ZL |
4354 | if (ret) |
4355 | return ret; | |
d7c15171 | 4356 | /* |
c2d6cb16 FM |
4357 | * The cleaner kthread might still be doing iput |
4358 | * operations. Wait for it to finish so that | |
4359 | * more space is released. | |
d7c15171 | 4360 | */ |
0b246afa JM |
4361 | mutex_lock(&fs_info->cleaner_delayed_iput_mutex); |
4362 | mutex_unlock(&fs_info->cleaner_delayed_iput_mutex); | |
94b947b2 ZL |
4363 | goto again; |
4364 | } else { | |
3a45bb20 | 4365 | btrfs_end_transaction(trans); |
94b947b2 | 4366 | } |
4e06bdd6 | 4367 | } |
9ed74f2d | 4368 | |
0b246afa | 4369 | trace_btrfs_space_reservation(fs_info, |
cab45e22 JM |
4370 | "space_info:enospc", |
4371 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4372 | return -ENOSPC; |
4373 | } | |
4374 | data_sinfo->bytes_may_use += bytes; | |
0b246afa | 4375 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4376 | data_sinfo->flags, bytes, 1); |
6a63209f | 4377 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4378 | |
237c0e9f | 4379 | return ret; |
9ed74f2d | 4380 | } |
6a63209f | 4381 | |
364ecf36 QW |
4382 | int btrfs_check_data_free_space(struct inode *inode, |
4383 | struct extent_changeset **reserved, u64 start, u64 len) | |
4ceff079 | 4384 | { |
0b246afa | 4385 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4386 | int ret; |
4387 | ||
4388 | /* align the range */ | |
0b246afa JM |
4389 | len = round_up(start + len, fs_info->sectorsize) - |
4390 | round_down(start, fs_info->sectorsize); | |
4391 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4392 | |
04f4f916 | 4393 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len); |
4ceff079 QW |
4394 | if (ret < 0) |
4395 | return ret; | |
4396 | ||
1e5ec2e7 | 4397 | /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */ |
364ecf36 | 4398 | ret = btrfs_qgroup_reserve_data(inode, reserved, start, len); |
7bc329c1 | 4399 | if (ret < 0) |
1e5ec2e7 | 4400 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
364ecf36 QW |
4401 | else |
4402 | ret = 0; | |
4ceff079 QW |
4403 | return ret; |
4404 | } | |
4405 | ||
4ceff079 QW |
4406 | /* |
4407 | * Called if we need to clear a data reservation for this inode | |
4408 | * Normally in a error case. | |
4409 | * | |
51773bec QW |
4410 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4411 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4412 | * Like clear_bit_hook(). | |
4ceff079 | 4413 | */ |
51773bec QW |
4414 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4415 | u64 len) | |
4ceff079 | 4416 | { |
0b246afa | 4417 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4418 | struct btrfs_space_info *data_sinfo; |
4419 | ||
4420 | /* Make sure the range is aligned to sectorsize */ | |
0b246afa JM |
4421 | len = round_up(start + len, fs_info->sectorsize) - |
4422 | round_down(start, fs_info->sectorsize); | |
4423 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4424 | |
0b246afa | 4425 | data_sinfo = fs_info->data_sinfo; |
4ceff079 QW |
4426 | spin_lock(&data_sinfo->lock); |
4427 | if (WARN_ON(data_sinfo->bytes_may_use < len)) | |
4428 | data_sinfo->bytes_may_use = 0; | |
4429 | else | |
4430 | data_sinfo->bytes_may_use -= len; | |
0b246afa | 4431 | trace_btrfs_space_reservation(fs_info, "space_info", |
4ceff079 QW |
4432 | data_sinfo->flags, len, 0); |
4433 | spin_unlock(&data_sinfo->lock); | |
4434 | } | |
4435 | ||
51773bec QW |
4436 | /* |
4437 | * Called if we need to clear a data reservation for this inode | |
4438 | * Normally in a error case. | |
4439 | * | |
01327610 | 4440 | * This one will handle the per-inode data rsv map for accurate reserved |
51773bec QW |
4441 | * space framework. |
4442 | */ | |
bc42bda2 QW |
4443 | void btrfs_free_reserved_data_space(struct inode *inode, |
4444 | struct extent_changeset *reserved, u64 start, u64 len) | |
51773bec | 4445 | { |
0c476a5d JM |
4446 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4447 | ||
4448 | /* Make sure the range is aligned to sectorsize */ | |
da17066c JM |
4449 | len = round_up(start + len, root->fs_info->sectorsize) - |
4450 | round_down(start, root->fs_info->sectorsize); | |
4451 | start = round_down(start, root->fs_info->sectorsize); | |
0c476a5d | 4452 | |
51773bec | 4453 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
bc42bda2 | 4454 | btrfs_qgroup_free_data(inode, reserved, start, len); |
51773bec QW |
4455 | } |
4456 | ||
97e728d4 | 4457 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4458 | { |
97e728d4 JB |
4459 | struct list_head *head = &info->space_info; |
4460 | struct btrfs_space_info *found; | |
e3ccfa98 | 4461 | |
97e728d4 JB |
4462 | rcu_read_lock(); |
4463 | list_for_each_entry_rcu(found, head, list) { | |
4464 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4465 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4466 | } |
97e728d4 | 4467 | rcu_read_unlock(); |
e3ccfa98 JB |
4468 | } |
4469 | ||
3c76cd84 MX |
4470 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
4471 | { | |
4472 | return (global->size << 1); | |
4473 | } | |
4474 | ||
2ff7e61e | 4475 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, |
698d0082 | 4476 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4477 | { |
0b246afa | 4478 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
8d8aafee | 4479 | u64 bytes_used = btrfs_space_info_used(sinfo, false); |
e5bc2458 | 4480 | u64 thresh; |
e3ccfa98 | 4481 | |
0e4f8f88 CM |
4482 | if (force == CHUNK_ALLOC_FORCE) |
4483 | return 1; | |
4484 | ||
fb25e914 JB |
4485 | /* |
4486 | * We need to take into account the global rsv because for all intents | |
4487 | * and purposes it's used space. Don't worry about locking the | |
4488 | * global_rsv, it doesn't change except when the transaction commits. | |
4489 | */ | |
54338b5c | 4490 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
8d8aafee | 4491 | bytes_used += calc_global_rsv_need_space(global_rsv); |
fb25e914 | 4492 | |
0e4f8f88 CM |
4493 | /* |
4494 | * in limited mode, we want to have some free space up to | |
4495 | * about 1% of the FS size. | |
4496 | */ | |
4497 | if (force == CHUNK_ALLOC_LIMITED) { | |
0b246afa | 4498 | thresh = btrfs_super_total_bytes(fs_info->super_copy); |
ee22184b | 4499 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
0e4f8f88 | 4500 | |
8d8aafee | 4501 | if (sinfo->total_bytes - bytes_used < thresh) |
0e4f8f88 CM |
4502 | return 1; |
4503 | } | |
0e4f8f88 | 4504 | |
8d8aafee | 4505 | if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) |
14ed0ca6 | 4506 | return 0; |
424499db | 4507 | return 1; |
32c00aff JB |
4508 | } |
4509 | ||
2ff7e61e | 4510 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4511 | { |
4512 | u64 num_dev; | |
4513 | ||
53b381b3 DW |
4514 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
4515 | BTRFS_BLOCK_GROUP_RAID0 | | |
4516 | BTRFS_BLOCK_GROUP_RAID5 | | |
4517 | BTRFS_BLOCK_GROUP_RAID6)) | |
0b246afa | 4518 | num_dev = fs_info->fs_devices->rw_devices; |
15d1ff81 LB |
4519 | else if (type & BTRFS_BLOCK_GROUP_RAID1) |
4520 | num_dev = 2; | |
4521 | else | |
4522 | num_dev = 1; /* DUP or single */ | |
4523 | ||
39c2d7fa | 4524 | return num_dev; |
15d1ff81 LB |
4525 | } |
4526 | ||
39c2d7fa FM |
4527 | /* |
4528 | * If @is_allocation is true, reserve space in the system space info necessary | |
4529 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4530 | * removing a chunk. | |
4531 | */ | |
4532 | void check_system_chunk(struct btrfs_trans_handle *trans, | |
2ff7e61e | 4533 | struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4534 | { |
4535 | struct btrfs_space_info *info; | |
4536 | u64 left; | |
4537 | u64 thresh; | |
4fbcdf66 | 4538 | int ret = 0; |
39c2d7fa | 4539 | u64 num_devs; |
4fbcdf66 FM |
4540 | |
4541 | /* | |
4542 | * Needed because we can end up allocating a system chunk and for an | |
4543 | * atomic and race free space reservation in the chunk block reserve. | |
4544 | */ | |
0b246afa | 4545 | ASSERT(mutex_is_locked(&fs_info->chunk_mutex)); |
15d1ff81 | 4546 | |
0b246afa | 4547 | info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
15d1ff81 | 4548 | spin_lock(&info->lock); |
4136135b | 4549 | left = info->total_bytes - btrfs_space_info_used(info, true); |
15d1ff81 LB |
4550 | spin_unlock(&info->lock); |
4551 | ||
2ff7e61e | 4552 | num_devs = get_profile_num_devs(fs_info, type); |
39c2d7fa FM |
4553 | |
4554 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
0b246afa JM |
4555 | thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) + |
4556 | btrfs_calc_trans_metadata_size(fs_info, 1); | |
39c2d7fa | 4557 | |
0b246afa JM |
4558 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
4559 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
4560 | left, thresh, type); | |
4561 | dump_space_info(fs_info, info, 0, 0); | |
15d1ff81 LB |
4562 | } |
4563 | ||
4564 | if (left < thresh) { | |
1b86826d | 4565 | u64 flags = btrfs_system_alloc_profile(fs_info); |
15d1ff81 | 4566 | |
4fbcdf66 FM |
4567 | /* |
4568 | * Ignore failure to create system chunk. We might end up not | |
4569 | * needing it, as we might not need to COW all nodes/leafs from | |
4570 | * the paths we visit in the chunk tree (they were already COWed | |
4571 | * or created in the current transaction for example). | |
4572 | */ | |
2ff7e61e | 4573 | ret = btrfs_alloc_chunk(trans, fs_info, flags); |
4fbcdf66 FM |
4574 | } |
4575 | ||
4576 | if (!ret) { | |
0b246afa JM |
4577 | ret = btrfs_block_rsv_add(fs_info->chunk_root, |
4578 | &fs_info->chunk_block_rsv, | |
4fbcdf66 FM |
4579 | thresh, BTRFS_RESERVE_NO_FLUSH); |
4580 | if (!ret) | |
4581 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4582 | } |
4583 | } | |
4584 | ||
28b737f6 LB |
4585 | /* |
4586 | * If force is CHUNK_ALLOC_FORCE: | |
4587 | * - return 1 if it successfully allocates a chunk, | |
4588 | * - return errors including -ENOSPC otherwise. | |
4589 | * If force is NOT CHUNK_ALLOC_FORCE: | |
4590 | * - return 0 if it doesn't need to allocate a new chunk, | |
4591 | * - return 1 if it successfully allocates a chunk, | |
4592 | * - return errors including -ENOSPC otherwise. | |
4593 | */ | |
6324fbf3 | 4594 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 4595 | struct btrfs_fs_info *fs_info, u64 flags, int force) |
9ed74f2d | 4596 | { |
6324fbf3 | 4597 | struct btrfs_space_info *space_info; |
6d74119f | 4598 | int wait_for_alloc = 0; |
9ed74f2d | 4599 | int ret = 0; |
9ed74f2d | 4600 | |
c6b305a8 JB |
4601 | /* Don't re-enter if we're already allocating a chunk */ |
4602 | if (trans->allocating_chunk) | |
4603 | return -ENOSPC; | |
4604 | ||
0b246afa | 4605 | space_info = __find_space_info(fs_info, flags); |
593060d7 | 4606 | if (!space_info) { |
2be12ef7 NB |
4607 | ret = create_space_info(fs_info, flags, &space_info); |
4608 | if (ret) | |
4609 | return ret; | |
9ed74f2d JB |
4610 | } |
4611 | ||
6d74119f | 4612 | again: |
25179201 | 4613 | spin_lock(&space_info->lock); |
9e622d6b | 4614 | if (force < space_info->force_alloc) |
0e4f8f88 | 4615 | force = space_info->force_alloc; |
25179201 | 4616 | if (space_info->full) { |
2ff7e61e | 4617 | if (should_alloc_chunk(fs_info, space_info, force)) |
09fb99a6 FDBM |
4618 | ret = -ENOSPC; |
4619 | else | |
4620 | ret = 0; | |
25179201 | 4621 | spin_unlock(&space_info->lock); |
09fb99a6 | 4622 | return ret; |
9ed74f2d JB |
4623 | } |
4624 | ||
2ff7e61e | 4625 | if (!should_alloc_chunk(fs_info, space_info, force)) { |
25179201 | 4626 | spin_unlock(&space_info->lock); |
6d74119f JB |
4627 | return 0; |
4628 | } else if (space_info->chunk_alloc) { | |
4629 | wait_for_alloc = 1; | |
4630 | } else { | |
4631 | space_info->chunk_alloc = 1; | |
9ed74f2d | 4632 | } |
0e4f8f88 | 4633 | |
25179201 | 4634 | spin_unlock(&space_info->lock); |
9ed74f2d | 4635 | |
6d74119f JB |
4636 | mutex_lock(&fs_info->chunk_mutex); |
4637 | ||
4638 | /* | |
4639 | * The chunk_mutex is held throughout the entirety of a chunk | |
4640 | * allocation, so once we've acquired the chunk_mutex we know that the | |
4641 | * other guy is done and we need to recheck and see if we should | |
4642 | * allocate. | |
4643 | */ | |
4644 | if (wait_for_alloc) { | |
4645 | mutex_unlock(&fs_info->chunk_mutex); | |
4646 | wait_for_alloc = 0; | |
4647 | goto again; | |
4648 | } | |
4649 | ||
c6b305a8 JB |
4650 | trans->allocating_chunk = true; |
4651 | ||
67377734 JB |
4652 | /* |
4653 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4654 | * allocating mixed chunks instead of individual chunks. | |
4655 | */ | |
4656 | if (btrfs_mixed_space_info(space_info)) | |
4657 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4658 | ||
97e728d4 JB |
4659 | /* |
4660 | * if we're doing a data chunk, go ahead and make sure that | |
4661 | * we keep a reasonable number of metadata chunks allocated in the | |
4662 | * FS as well. | |
4663 | */ | |
9ed74f2d | 4664 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4665 | fs_info->data_chunk_allocations++; |
4666 | if (!(fs_info->data_chunk_allocations % | |
4667 | fs_info->metadata_ratio)) | |
4668 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4669 | } |
4670 | ||
15d1ff81 LB |
4671 | /* |
4672 | * Check if we have enough space in SYSTEM chunk because we may need | |
4673 | * to update devices. | |
4674 | */ | |
2ff7e61e | 4675 | check_system_chunk(trans, fs_info, flags); |
15d1ff81 | 4676 | |
2ff7e61e | 4677 | ret = btrfs_alloc_chunk(trans, fs_info, flags); |
c6b305a8 | 4678 | trans->allocating_chunk = false; |
92b8e897 | 4679 | |
9ed74f2d | 4680 | spin_lock(&space_info->lock); |
a81cb9a2 AO |
4681 | if (ret < 0 && ret != -ENOSPC) |
4682 | goto out; | |
9ed74f2d | 4683 | if (ret) |
6324fbf3 | 4684 | space_info->full = 1; |
424499db YZ |
4685 | else |
4686 | ret = 1; | |
6d74119f | 4687 | |
0e4f8f88 | 4688 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4689 | out: |
6d74119f | 4690 | space_info->chunk_alloc = 0; |
9ed74f2d | 4691 | spin_unlock(&space_info->lock); |
a25c75d5 | 4692 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4693 | /* |
4694 | * When we allocate a new chunk we reserve space in the chunk block | |
4695 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4696 | * add new nodes/leafs to it if we end up needing to do it when | |
4697 | * inserting the chunk item and updating device items as part of the | |
4698 | * second phase of chunk allocation, performed by | |
4699 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4700 | * large number of new block groups to create in our transaction | |
4701 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4702 | * in extreme cases - like having a single transaction create many new | |
4703 | * block groups when starting to write out the free space caches of all | |
4704 | * the block groups that were made dirty during the lifetime of the | |
4705 | * transaction. | |
4706 | */ | |
d9a0540a | 4707 | if (trans->can_flush_pending_bgs && |
ee22184b | 4708 | trans->chunk_bytes_reserved >= (u64)SZ_2M) { |
2ff7e61e | 4709 | btrfs_create_pending_block_groups(trans, fs_info); |
00d80e34 FM |
4710 | btrfs_trans_release_chunk_metadata(trans); |
4711 | } | |
0f9dd46c | 4712 | return ret; |
6324fbf3 | 4713 | } |
9ed74f2d | 4714 | |
c1c4919b | 4715 | static int can_overcommit(struct btrfs_fs_info *fs_info, |
a80c8dcf | 4716 | struct btrfs_space_info *space_info, u64 bytes, |
c1c4919b JM |
4717 | enum btrfs_reserve_flush_enum flush, |
4718 | bool system_chunk) | |
a80c8dcf | 4719 | { |
0b246afa | 4720 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
957780eb | 4721 | u64 profile; |
3c76cd84 | 4722 | u64 space_size; |
a80c8dcf JB |
4723 | u64 avail; |
4724 | u64 used; | |
4725 | ||
957780eb JB |
4726 | /* Don't overcommit when in mixed mode. */ |
4727 | if (space_info->flags & BTRFS_BLOCK_GROUP_DATA) | |
4728 | return 0; | |
4729 | ||
c1c4919b JM |
4730 | if (system_chunk) |
4731 | profile = btrfs_system_alloc_profile(fs_info); | |
4732 | else | |
4733 | profile = btrfs_metadata_alloc_profile(fs_info); | |
4734 | ||
4136135b | 4735 | used = btrfs_space_info_used(space_info, false); |
96f1bb57 | 4736 | |
96f1bb57 JB |
4737 | /* |
4738 | * We only want to allow over committing if we have lots of actual space | |
4739 | * free, but if we don't have enough space to handle the global reserve | |
4740 | * space then we could end up having a real enospc problem when trying | |
4741 | * to allocate a chunk or some other such important allocation. | |
4742 | */ | |
3c76cd84 MX |
4743 | spin_lock(&global_rsv->lock); |
4744 | space_size = calc_global_rsv_need_space(global_rsv); | |
4745 | spin_unlock(&global_rsv->lock); | |
4746 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
4747 | return 0; |
4748 | ||
4749 | used += space_info->bytes_may_use; | |
a80c8dcf | 4750 | |
a5ed45f8 | 4751 | avail = atomic64_read(&fs_info->free_chunk_space); |
a80c8dcf JB |
4752 | |
4753 | /* | |
4754 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
4755 | * space is actually useable. For raid56, the space info used |
4756 | * doesn't include the parity drive, so we don't have to | |
4757 | * change the math | |
a80c8dcf JB |
4758 | */ |
4759 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
4760 | BTRFS_BLOCK_GROUP_RAID1 | | |
4761 | BTRFS_BLOCK_GROUP_RAID10)) | |
4762 | avail >>= 1; | |
4763 | ||
4764 | /* | |
561c294d MX |
4765 | * If we aren't flushing all things, let us overcommit up to |
4766 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4767 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 4768 | */ |
08e007d2 | 4769 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 4770 | avail >>= 3; |
a80c8dcf | 4771 | else |
14575aef | 4772 | avail >>= 1; |
a80c8dcf | 4773 | |
14575aef | 4774 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
4775 | return 1; |
4776 | return 0; | |
4777 | } | |
4778 | ||
2ff7e61e | 4779 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info, |
6c255e67 | 4780 | unsigned long nr_pages, int nr_items) |
da633a42 | 4781 | { |
0b246afa | 4782 | struct super_block *sb = fs_info->sb; |
da633a42 | 4783 | |
925a6efb JB |
4784 | if (down_read_trylock(&sb->s_umount)) { |
4785 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4786 | up_read(&sb->s_umount); | |
4787 | } else { | |
da633a42 MX |
4788 | /* |
4789 | * We needn't worry the filesystem going from r/w to r/o though | |
4790 | * we don't acquire ->s_umount mutex, because the filesystem | |
4791 | * should guarantee the delalloc inodes list be empty after | |
4792 | * the filesystem is readonly(all dirty pages are written to | |
4793 | * the disk). | |
4794 | */ | |
0b246afa | 4795 | btrfs_start_delalloc_roots(fs_info, 0, nr_items); |
98ad69cf | 4796 | if (!current->journal_info) |
0b246afa | 4797 | btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1); |
da633a42 MX |
4798 | } |
4799 | } | |
4800 | ||
6374e57a | 4801 | static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info, |
2ff7e61e | 4802 | u64 to_reclaim) |
18cd8ea6 MX |
4803 | { |
4804 | u64 bytes; | |
6374e57a | 4805 | u64 nr; |
18cd8ea6 | 4806 | |
2ff7e61e | 4807 | bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
6374e57a | 4808 | nr = div64_u64(to_reclaim, bytes); |
18cd8ea6 MX |
4809 | if (!nr) |
4810 | nr = 1; | |
4811 | return nr; | |
4812 | } | |
4813 | ||
ee22184b | 4814 | #define EXTENT_SIZE_PER_ITEM SZ_256K |
c61a16a7 | 4815 | |
9ed74f2d | 4816 | /* |
5da9d01b | 4817 | * shrink metadata reservation for delalloc |
9ed74f2d | 4818 | */ |
c1c4919b JM |
4819 | static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim, |
4820 | u64 orig, bool wait_ordered) | |
5da9d01b | 4821 | { |
0019f10d | 4822 | struct btrfs_space_info *space_info; |
663350ac | 4823 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4824 | u64 delalloc_bytes; |
5da9d01b | 4825 | u64 max_reclaim; |
6374e57a | 4826 | u64 items; |
b1953bce | 4827 | long time_left; |
d3ee29e3 MX |
4828 | unsigned long nr_pages; |
4829 | int loops; | |
08e007d2 | 4830 | enum btrfs_reserve_flush_enum flush; |
5da9d01b | 4831 | |
c61a16a7 | 4832 | /* Calc the number of the pages we need flush for space reservation */ |
2ff7e61e | 4833 | items = calc_reclaim_items_nr(fs_info, to_reclaim); |
6374e57a | 4834 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; |
c61a16a7 | 4835 | |
663350ac | 4836 | trans = (struct btrfs_trans_handle *)current->journal_info; |
69fe2d75 | 4837 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
bf9022e0 | 4838 | |
963d678b | 4839 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4840 | &fs_info->delalloc_bytes); |
f4c738c2 | 4841 | if (delalloc_bytes == 0) { |
fdb5effd | 4842 | if (trans) |
f4c738c2 | 4843 | return; |
38c135af | 4844 | if (wait_ordered) |
0b246afa | 4845 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f4c738c2 | 4846 | return; |
fdb5effd JB |
4847 | } |
4848 | ||
d3ee29e3 | 4849 | loops = 0; |
f4c738c2 JB |
4850 | while (delalloc_bytes && loops < 3) { |
4851 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
09cbfeaf | 4852 | nr_pages = max_reclaim >> PAGE_SHIFT; |
2ff7e61e | 4853 | btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items); |
dea31f52 JB |
4854 | /* |
4855 | * We need to wait for the async pages to actually start before | |
4856 | * we do anything. | |
4857 | */ | |
0b246afa | 4858 | max_reclaim = atomic_read(&fs_info->async_delalloc_pages); |
9f3a074d MX |
4859 | if (!max_reclaim) |
4860 | goto skip_async; | |
4861 | ||
4862 | if (max_reclaim <= nr_pages) | |
4863 | max_reclaim = 0; | |
4864 | else | |
4865 | max_reclaim -= nr_pages; | |
dea31f52 | 4866 | |
0b246afa JM |
4867 | wait_event(fs_info->async_submit_wait, |
4868 | atomic_read(&fs_info->async_delalloc_pages) <= | |
9f3a074d MX |
4869 | (int)max_reclaim); |
4870 | skip_async: | |
08e007d2 MX |
4871 | if (!trans) |
4872 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
4873 | else | |
4874 | flush = BTRFS_RESERVE_NO_FLUSH; | |
0019f10d | 4875 | spin_lock(&space_info->lock); |
957780eb JB |
4876 | if (list_empty(&space_info->tickets) && |
4877 | list_empty(&space_info->priority_tickets)) { | |
4878 | spin_unlock(&space_info->lock); | |
4879 | break; | |
4880 | } | |
0019f10d | 4881 | spin_unlock(&space_info->lock); |
5da9d01b | 4882 | |
36e39c40 | 4883 | loops++; |
f104d044 | 4884 | if (wait_ordered && !trans) { |
0b246afa | 4885 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f104d044 | 4886 | } else { |
f4c738c2 | 4887 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4888 | if (time_left) |
4889 | break; | |
4890 | } | |
963d678b | 4891 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4892 | &fs_info->delalloc_bytes); |
5da9d01b | 4893 | } |
5da9d01b YZ |
4894 | } |
4895 | ||
996478ca JB |
4896 | struct reserve_ticket { |
4897 | u64 bytes; | |
4898 | int error; | |
4899 | struct list_head list; | |
4900 | wait_queue_head_t wait; | |
4901 | }; | |
4902 | ||
663350ac JB |
4903 | /** |
4904 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4905 | * @root - the root we're allocating for | |
4906 | * @bytes - the number of bytes we want to reserve | |
4907 | * @force - force the commit | |
8bb8ab2e | 4908 | * |
663350ac JB |
4909 | * This will check to make sure that committing the transaction will actually |
4910 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4911 | * will return -ENOSPC. | |
8bb8ab2e | 4912 | */ |
0c9ab349 | 4913 | static int may_commit_transaction(struct btrfs_fs_info *fs_info, |
996478ca | 4914 | struct btrfs_space_info *space_info) |
663350ac | 4915 | { |
996478ca | 4916 | struct reserve_ticket *ticket = NULL; |
0b246afa | 4917 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv; |
663350ac | 4918 | struct btrfs_trans_handle *trans; |
996478ca | 4919 | u64 bytes; |
663350ac JB |
4920 | |
4921 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4922 | if (trans) | |
4923 | return -EAGAIN; | |
4924 | ||
996478ca JB |
4925 | spin_lock(&space_info->lock); |
4926 | if (!list_empty(&space_info->priority_tickets)) | |
4927 | ticket = list_first_entry(&space_info->priority_tickets, | |
4928 | struct reserve_ticket, list); | |
4929 | else if (!list_empty(&space_info->tickets)) | |
4930 | ticket = list_first_entry(&space_info->tickets, | |
4931 | struct reserve_ticket, list); | |
4932 | bytes = (ticket) ? ticket->bytes : 0; | |
4933 | spin_unlock(&space_info->lock); | |
4934 | ||
4935 | if (!bytes) | |
4936 | return 0; | |
663350ac JB |
4937 | |
4938 | /* See if there is enough pinned space to make this reservation */ | |
b150a4f1 | 4939 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
0424c548 | 4940 | bytes) >= 0) |
663350ac | 4941 | goto commit; |
663350ac JB |
4942 | |
4943 | /* | |
4944 | * See if there is some space in the delayed insertion reservation for | |
4945 | * this reservation. | |
4946 | */ | |
4947 | if (space_info != delayed_rsv->space_info) | |
4948 | return -ENOSPC; | |
4949 | ||
4950 | spin_lock(&delayed_rsv->lock); | |
996478ca JB |
4951 | if (delayed_rsv->size > bytes) |
4952 | bytes = 0; | |
4953 | else | |
4954 | bytes -= delayed_rsv->size; | |
b150a4f1 | 4955 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
996478ca | 4956 | bytes) < 0) { |
663350ac JB |
4957 | spin_unlock(&delayed_rsv->lock); |
4958 | return -ENOSPC; | |
4959 | } | |
4960 | spin_unlock(&delayed_rsv->lock); | |
4961 | ||
4962 | commit: | |
a9b3311e | 4963 | trans = btrfs_join_transaction(fs_info->extent_root); |
663350ac JB |
4964 | if (IS_ERR(trans)) |
4965 | return -ENOSPC; | |
4966 | ||
3a45bb20 | 4967 | return btrfs_commit_transaction(trans); |
663350ac JB |
4968 | } |
4969 | ||
e38ae7a0 NB |
4970 | /* |
4971 | * Try to flush some data based on policy set by @state. This is only advisory | |
4972 | * and may fail for various reasons. The caller is supposed to examine the | |
4973 | * state of @space_info to detect the outcome. | |
4974 | */ | |
4975 | static void flush_space(struct btrfs_fs_info *fs_info, | |
96c3f433 | 4976 | struct btrfs_space_info *space_info, u64 num_bytes, |
7bdd6277 | 4977 | int state) |
96c3f433 | 4978 | { |
a9b3311e | 4979 | struct btrfs_root *root = fs_info->extent_root; |
96c3f433 JB |
4980 | struct btrfs_trans_handle *trans; |
4981 | int nr; | |
f4c738c2 | 4982 | int ret = 0; |
96c3f433 JB |
4983 | |
4984 | switch (state) { | |
96c3f433 JB |
4985 | case FLUSH_DELAYED_ITEMS_NR: |
4986 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 | 4987 | if (state == FLUSH_DELAYED_ITEMS_NR) |
2ff7e61e | 4988 | nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2; |
18cd8ea6 | 4989 | else |
96c3f433 | 4990 | nr = -1; |
18cd8ea6 | 4991 | |
96c3f433 JB |
4992 | trans = btrfs_join_transaction(root); |
4993 | if (IS_ERR(trans)) { | |
4994 | ret = PTR_ERR(trans); | |
4995 | break; | |
4996 | } | |
2ff7e61e | 4997 | ret = btrfs_run_delayed_items_nr(trans, fs_info, nr); |
3a45bb20 | 4998 | btrfs_end_transaction(trans); |
96c3f433 | 4999 | break; |
67b0fd63 JB |
5000 | case FLUSH_DELALLOC: |
5001 | case FLUSH_DELALLOC_WAIT: | |
7bdd6277 | 5002 | shrink_delalloc(fs_info, num_bytes * 2, num_bytes, |
67b0fd63 JB |
5003 | state == FLUSH_DELALLOC_WAIT); |
5004 | break; | |
ea658bad JB |
5005 | case ALLOC_CHUNK: |
5006 | trans = btrfs_join_transaction(root); | |
5007 | if (IS_ERR(trans)) { | |
5008 | ret = PTR_ERR(trans); | |
5009 | break; | |
5010 | } | |
2ff7e61e | 5011 | ret = do_chunk_alloc(trans, fs_info, |
1b86826d | 5012 | btrfs_metadata_alloc_profile(fs_info), |
ea658bad | 5013 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 5014 | btrfs_end_transaction(trans); |
eecba891 | 5015 | if (ret > 0 || ret == -ENOSPC) |
ea658bad JB |
5016 | ret = 0; |
5017 | break; | |
96c3f433 | 5018 | case COMMIT_TRANS: |
996478ca | 5019 | ret = may_commit_transaction(fs_info, space_info); |
96c3f433 JB |
5020 | break; |
5021 | default: | |
5022 | ret = -ENOSPC; | |
5023 | break; | |
5024 | } | |
5025 | ||
7bdd6277 NB |
5026 | trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state, |
5027 | ret); | |
e38ae7a0 | 5028 | return; |
96c3f433 | 5029 | } |
21c7e756 MX |
5030 | |
5031 | static inline u64 | |
c1c4919b JM |
5032 | btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info, |
5033 | struct btrfs_space_info *space_info, | |
5034 | bool system_chunk) | |
21c7e756 | 5035 | { |
957780eb | 5036 | struct reserve_ticket *ticket; |
21c7e756 MX |
5037 | u64 used; |
5038 | u64 expected; | |
957780eb | 5039 | u64 to_reclaim = 0; |
21c7e756 | 5040 | |
957780eb JB |
5041 | list_for_each_entry(ticket, &space_info->tickets, list) |
5042 | to_reclaim += ticket->bytes; | |
5043 | list_for_each_entry(ticket, &space_info->priority_tickets, list) | |
5044 | to_reclaim += ticket->bytes; | |
5045 | if (to_reclaim) | |
5046 | return to_reclaim; | |
21c7e756 | 5047 | |
e0af2484 | 5048 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); |
c1c4919b JM |
5049 | if (can_overcommit(fs_info, space_info, to_reclaim, |
5050 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
e0af2484 WX |
5051 | return 0; |
5052 | ||
0eee8a49 NB |
5053 | used = btrfs_space_info_used(space_info, true); |
5054 | ||
c1c4919b JM |
5055 | if (can_overcommit(fs_info, space_info, SZ_1M, |
5056 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
21c7e756 MX |
5057 | expected = div_factor_fine(space_info->total_bytes, 95); |
5058 | else | |
5059 | expected = div_factor_fine(space_info->total_bytes, 90); | |
5060 | ||
5061 | if (used > expected) | |
5062 | to_reclaim = used - expected; | |
5063 | else | |
5064 | to_reclaim = 0; | |
5065 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
5066 | space_info->bytes_reserved); | |
21c7e756 MX |
5067 | return to_reclaim; |
5068 | } | |
5069 | ||
c1c4919b JM |
5070 | static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info, |
5071 | struct btrfs_space_info *space_info, | |
5072 | u64 used, bool system_chunk) | |
21c7e756 | 5073 | { |
365c5313 JB |
5074 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
5075 | ||
5076 | /* If we're just plain full then async reclaim just slows us down. */ | |
baee8790 | 5077 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) |
365c5313 JB |
5078 | return 0; |
5079 | ||
c1c4919b JM |
5080 | if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5081 | system_chunk)) | |
d38b349c JB |
5082 | return 0; |
5083 | ||
0b246afa JM |
5084 | return (used >= thresh && !btrfs_fs_closing(fs_info) && |
5085 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); | |
21c7e756 MX |
5086 | } |
5087 | ||
957780eb | 5088 | static void wake_all_tickets(struct list_head *head) |
21c7e756 | 5089 | { |
957780eb | 5090 | struct reserve_ticket *ticket; |
25ce459c | 5091 | |
957780eb JB |
5092 | while (!list_empty(head)) { |
5093 | ticket = list_first_entry(head, struct reserve_ticket, list); | |
5094 | list_del_init(&ticket->list); | |
5095 | ticket->error = -ENOSPC; | |
5096 | wake_up(&ticket->wait); | |
21c7e756 | 5097 | } |
21c7e756 MX |
5098 | } |
5099 | ||
957780eb JB |
5100 | /* |
5101 | * This is for normal flushers, we can wait all goddamned day if we want to. We | |
5102 | * will loop and continuously try to flush as long as we are making progress. | |
5103 | * We count progress as clearing off tickets each time we have to loop. | |
5104 | */ | |
21c7e756 MX |
5105 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) |
5106 | { | |
5107 | struct btrfs_fs_info *fs_info; | |
5108 | struct btrfs_space_info *space_info; | |
5109 | u64 to_reclaim; | |
5110 | int flush_state; | |
957780eb | 5111 | int commit_cycles = 0; |
ce129655 | 5112 | u64 last_tickets_id; |
21c7e756 MX |
5113 | |
5114 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
5115 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
5116 | ||
957780eb | 5117 | spin_lock(&space_info->lock); |
c1c4919b JM |
5118 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5119 | false); | |
957780eb JB |
5120 | if (!to_reclaim) { |
5121 | space_info->flush = 0; | |
5122 | spin_unlock(&space_info->lock); | |
21c7e756 | 5123 | return; |
957780eb | 5124 | } |
ce129655 | 5125 | last_tickets_id = space_info->tickets_id; |
957780eb | 5126 | spin_unlock(&space_info->lock); |
21c7e756 MX |
5127 | |
5128 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
957780eb | 5129 | do { |
e38ae7a0 | 5130 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
957780eb JB |
5131 | spin_lock(&space_info->lock); |
5132 | if (list_empty(&space_info->tickets)) { | |
5133 | space_info->flush = 0; | |
5134 | spin_unlock(&space_info->lock); | |
5135 | return; | |
5136 | } | |
c1c4919b JM |
5137 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, |
5138 | space_info, | |
5139 | false); | |
ce129655 | 5140 | if (last_tickets_id == space_info->tickets_id) { |
957780eb JB |
5141 | flush_state++; |
5142 | } else { | |
ce129655 | 5143 | last_tickets_id = space_info->tickets_id; |
957780eb JB |
5144 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
5145 | if (commit_cycles) | |
5146 | commit_cycles--; | |
5147 | } | |
5148 | ||
5149 | if (flush_state > COMMIT_TRANS) { | |
5150 | commit_cycles++; | |
5151 | if (commit_cycles > 2) { | |
5152 | wake_all_tickets(&space_info->tickets); | |
5153 | space_info->flush = 0; | |
5154 | } else { | |
5155 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5156 | } | |
5157 | } | |
5158 | spin_unlock(&space_info->lock); | |
5159 | } while (flush_state <= COMMIT_TRANS); | |
5160 | } | |
5161 | ||
5162 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
5163 | { | |
5164 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
5165 | } | |
5166 | ||
5167 | static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, | |
5168 | struct btrfs_space_info *space_info, | |
5169 | struct reserve_ticket *ticket) | |
5170 | { | |
5171 | u64 to_reclaim; | |
5172 | int flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5173 | ||
5174 | spin_lock(&space_info->lock); | |
c1c4919b JM |
5175 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5176 | false); | |
957780eb JB |
5177 | if (!to_reclaim) { |
5178 | spin_unlock(&space_info->lock); | |
5179 | return; | |
5180 | } | |
5181 | spin_unlock(&space_info->lock); | |
5182 | ||
21c7e756 | 5183 | do { |
7bdd6277 | 5184 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
21c7e756 | 5185 | flush_state++; |
957780eb JB |
5186 | spin_lock(&space_info->lock); |
5187 | if (ticket->bytes == 0) { | |
5188 | spin_unlock(&space_info->lock); | |
21c7e756 | 5189 | return; |
957780eb JB |
5190 | } |
5191 | spin_unlock(&space_info->lock); | |
5192 | ||
5193 | /* | |
5194 | * Priority flushers can't wait on delalloc without | |
5195 | * deadlocking. | |
5196 | */ | |
5197 | if (flush_state == FLUSH_DELALLOC || | |
5198 | flush_state == FLUSH_DELALLOC_WAIT) | |
5199 | flush_state = ALLOC_CHUNK; | |
365c5313 | 5200 | } while (flush_state < COMMIT_TRANS); |
21c7e756 MX |
5201 | } |
5202 | ||
957780eb JB |
5203 | static int wait_reserve_ticket(struct btrfs_fs_info *fs_info, |
5204 | struct btrfs_space_info *space_info, | |
5205 | struct reserve_ticket *ticket, u64 orig_bytes) | |
5206 | ||
21c7e756 | 5207 | { |
957780eb JB |
5208 | DEFINE_WAIT(wait); |
5209 | int ret = 0; | |
5210 | ||
5211 | spin_lock(&space_info->lock); | |
5212 | while (ticket->bytes > 0 && ticket->error == 0) { | |
5213 | ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE); | |
5214 | if (ret) { | |
5215 | ret = -EINTR; | |
5216 | break; | |
5217 | } | |
5218 | spin_unlock(&space_info->lock); | |
5219 | ||
5220 | schedule(); | |
5221 | ||
5222 | finish_wait(&ticket->wait, &wait); | |
5223 | spin_lock(&space_info->lock); | |
5224 | } | |
5225 | if (!ret) | |
5226 | ret = ticket->error; | |
5227 | if (!list_empty(&ticket->list)) | |
5228 | list_del_init(&ticket->list); | |
5229 | if (ticket->bytes && ticket->bytes < orig_bytes) { | |
5230 | u64 num_bytes = orig_bytes - ticket->bytes; | |
5231 | space_info->bytes_may_use -= num_bytes; | |
5232 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5233 | space_info->flags, num_bytes, 0); | |
5234 | } | |
5235 | spin_unlock(&space_info->lock); | |
5236 | ||
5237 | return ret; | |
21c7e756 MX |
5238 | } |
5239 | ||
4a92b1b8 JB |
5240 | /** |
5241 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5242 | * @root - the root we're allocating for | |
957780eb | 5243 | * @space_info - the space info we want to allocate from |
4a92b1b8 | 5244 | * @orig_bytes - the number of bytes we want |
48fc7f7e | 5245 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 5246 | * |
01327610 | 5247 | * This will reserve orig_bytes number of bytes from the space info associated |
4a92b1b8 JB |
5248 | * with the block_rsv. If there is not enough space it will make an attempt to |
5249 | * flush out space to make room. It will do this by flushing delalloc if | |
5250 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5251 | * regain reservations will be made and this will fail if there is not enough | |
5252 | * space already. | |
8bb8ab2e | 5253 | */ |
c1c4919b | 5254 | static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info, |
957780eb JB |
5255 | struct btrfs_space_info *space_info, |
5256 | u64 orig_bytes, | |
c1c4919b JM |
5257 | enum btrfs_reserve_flush_enum flush, |
5258 | bool system_chunk) | |
9ed74f2d | 5259 | { |
957780eb | 5260 | struct reserve_ticket ticket; |
2bf64758 | 5261 | u64 used; |
8bb8ab2e | 5262 | int ret = 0; |
9ed74f2d | 5263 | |
957780eb | 5264 | ASSERT(orig_bytes); |
8ca17f0f | 5265 | ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL); |
fdb5effd | 5266 | |
8bb8ab2e | 5267 | spin_lock(&space_info->lock); |
fdb5effd | 5268 | ret = -ENOSPC; |
4136135b | 5269 | used = btrfs_space_info_used(space_info, true); |
9ed74f2d | 5270 | |
8bb8ab2e | 5271 | /* |
957780eb JB |
5272 | * If we have enough space then hooray, make our reservation and carry |
5273 | * on. If not see if we can overcommit, and if we can, hooray carry on. | |
5274 | * If not things get more complicated. | |
8bb8ab2e | 5275 | */ |
957780eb JB |
5276 | if (used + orig_bytes <= space_info->total_bytes) { |
5277 | space_info->bytes_may_use += orig_bytes; | |
0b246afa JM |
5278 | trace_btrfs_space_reservation(fs_info, "space_info", |
5279 | space_info->flags, orig_bytes, 1); | |
957780eb | 5280 | ret = 0; |
c1c4919b JM |
5281 | } else if (can_overcommit(fs_info, space_info, orig_bytes, flush, |
5282 | system_chunk)) { | |
44734ed1 | 5283 | space_info->bytes_may_use += orig_bytes; |
0b246afa JM |
5284 | trace_btrfs_space_reservation(fs_info, "space_info", |
5285 | space_info->flags, orig_bytes, 1); | |
44734ed1 | 5286 | ret = 0; |
2bf64758 JB |
5287 | } |
5288 | ||
8bb8ab2e | 5289 | /* |
957780eb JB |
5290 | * If we couldn't make a reservation then setup our reservation ticket |
5291 | * and kick the async worker if it's not already running. | |
08e007d2 | 5292 | * |
957780eb JB |
5293 | * If we are a priority flusher then we just need to add our ticket to |
5294 | * the list and we will do our own flushing further down. | |
8bb8ab2e | 5295 | */ |
72bcd99d | 5296 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
957780eb JB |
5297 | ticket.bytes = orig_bytes; |
5298 | ticket.error = 0; | |
5299 | init_waitqueue_head(&ticket.wait); | |
5300 | if (flush == BTRFS_RESERVE_FLUSH_ALL) { | |
5301 | list_add_tail(&ticket.list, &space_info->tickets); | |
5302 | if (!space_info->flush) { | |
5303 | space_info->flush = 1; | |
0b246afa | 5304 | trace_btrfs_trigger_flush(fs_info, |
f376df2b JB |
5305 | space_info->flags, |
5306 | orig_bytes, flush, | |
5307 | "enospc"); | |
957780eb | 5308 | queue_work(system_unbound_wq, |
c1c4919b | 5309 | &fs_info->async_reclaim_work); |
957780eb JB |
5310 | } |
5311 | } else { | |
5312 | list_add_tail(&ticket.list, | |
5313 | &space_info->priority_tickets); | |
5314 | } | |
21c7e756 MX |
5315 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
5316 | used += orig_bytes; | |
f6acfd50 JB |
5317 | /* |
5318 | * We will do the space reservation dance during log replay, | |
5319 | * which means we won't have fs_info->fs_root set, so don't do | |
5320 | * the async reclaim as we will panic. | |
5321 | */ | |
0b246afa | 5322 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) && |
c1c4919b JM |
5323 | need_do_async_reclaim(fs_info, space_info, |
5324 | used, system_chunk) && | |
0b246afa JM |
5325 | !work_busy(&fs_info->async_reclaim_work)) { |
5326 | trace_btrfs_trigger_flush(fs_info, space_info->flags, | |
5327 | orig_bytes, flush, "preempt"); | |
21c7e756 | 5328 | queue_work(system_unbound_wq, |
0b246afa | 5329 | &fs_info->async_reclaim_work); |
f376df2b | 5330 | } |
8bb8ab2e | 5331 | } |
f0486c68 | 5332 | spin_unlock(&space_info->lock); |
08e007d2 | 5333 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
957780eb | 5334 | return ret; |
f0486c68 | 5335 | |
957780eb | 5336 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
0b246afa | 5337 | return wait_reserve_ticket(fs_info, space_info, &ticket, |
957780eb | 5338 | orig_bytes); |
08e007d2 | 5339 | |
957780eb | 5340 | ret = 0; |
0b246afa | 5341 | priority_reclaim_metadata_space(fs_info, space_info, &ticket); |
957780eb JB |
5342 | spin_lock(&space_info->lock); |
5343 | if (ticket.bytes) { | |
5344 | if (ticket.bytes < orig_bytes) { | |
5345 | u64 num_bytes = orig_bytes - ticket.bytes; | |
5346 | space_info->bytes_may_use -= num_bytes; | |
0b246afa JM |
5347 | trace_btrfs_space_reservation(fs_info, "space_info", |
5348 | space_info->flags, | |
5349 | num_bytes, 0); | |
08e007d2 | 5350 | |
957780eb JB |
5351 | } |
5352 | list_del_init(&ticket.list); | |
5353 | ret = -ENOSPC; | |
5354 | } | |
5355 | spin_unlock(&space_info->lock); | |
5356 | ASSERT(list_empty(&ticket.list)); | |
5357 | return ret; | |
5358 | } | |
8bb8ab2e | 5359 | |
957780eb JB |
5360 | /** |
5361 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5362 | * @root - the root we're allocating for | |
5363 | * @block_rsv - the block_rsv we're allocating for | |
5364 | * @orig_bytes - the number of bytes we want | |
5365 | * @flush - whether or not we can flush to make our reservation | |
5366 | * | |
5367 | * This will reserve orgi_bytes number of bytes from the space info associated | |
5368 | * with the block_rsv. If there is not enough space it will make an attempt to | |
5369 | * flush out space to make room. It will do this by flushing delalloc if | |
5370 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5371 | * regain reservations will be made and this will fail if there is not enough | |
5372 | * space already. | |
5373 | */ | |
5374 | static int reserve_metadata_bytes(struct btrfs_root *root, | |
5375 | struct btrfs_block_rsv *block_rsv, | |
5376 | u64 orig_bytes, | |
5377 | enum btrfs_reserve_flush_enum flush) | |
5378 | { | |
0b246afa JM |
5379 | struct btrfs_fs_info *fs_info = root->fs_info; |
5380 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
957780eb | 5381 | int ret; |
c1c4919b | 5382 | bool system_chunk = (root == fs_info->chunk_root); |
957780eb | 5383 | |
c1c4919b JM |
5384 | ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info, |
5385 | orig_bytes, flush, system_chunk); | |
5d80366e JB |
5386 | if (ret == -ENOSPC && |
5387 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5d80366e JB |
5388 | if (block_rsv != global_rsv && |
5389 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5390 | ret = 0; | |
5391 | } | |
cab45e22 | 5392 | if (ret == -ENOSPC) |
0b246afa | 5393 | trace_btrfs_space_reservation(fs_info, "space_info:enospc", |
957780eb JB |
5394 | block_rsv->space_info->flags, |
5395 | orig_bytes, 1); | |
f0486c68 YZ |
5396 | return ret; |
5397 | } | |
5398 | ||
79787eaa JM |
5399 | static struct btrfs_block_rsv *get_block_rsv( |
5400 | const struct btrfs_trans_handle *trans, | |
5401 | const struct btrfs_root *root) | |
f0486c68 | 5402 | { |
0b246afa | 5403 | struct btrfs_fs_info *fs_info = root->fs_info; |
4c13d758 JB |
5404 | struct btrfs_block_rsv *block_rsv = NULL; |
5405 | ||
e9cf439f | 5406 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa JM |
5407 | (root == fs_info->csum_root && trans->adding_csums) || |
5408 | (root == fs_info->uuid_root)) | |
f7a81ea4 SB |
5409 | block_rsv = trans->block_rsv; |
5410 | ||
4c13d758 | 5411 | if (!block_rsv) |
f0486c68 YZ |
5412 | block_rsv = root->block_rsv; |
5413 | ||
5414 | if (!block_rsv) | |
0b246afa | 5415 | block_rsv = &fs_info->empty_block_rsv; |
f0486c68 YZ |
5416 | |
5417 | return block_rsv; | |
5418 | } | |
5419 | ||
5420 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
5421 | u64 num_bytes) | |
5422 | { | |
5423 | int ret = -ENOSPC; | |
5424 | spin_lock(&block_rsv->lock); | |
5425 | if (block_rsv->reserved >= num_bytes) { | |
5426 | block_rsv->reserved -= num_bytes; | |
5427 | if (block_rsv->reserved < block_rsv->size) | |
5428 | block_rsv->full = 0; | |
5429 | ret = 0; | |
5430 | } | |
5431 | spin_unlock(&block_rsv->lock); | |
5432 | return ret; | |
5433 | } | |
5434 | ||
5435 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
5436 | u64 num_bytes, int update_size) | |
5437 | { | |
5438 | spin_lock(&block_rsv->lock); | |
5439 | block_rsv->reserved += num_bytes; | |
5440 | if (update_size) | |
5441 | block_rsv->size += num_bytes; | |
5442 | else if (block_rsv->reserved >= block_rsv->size) | |
5443 | block_rsv->full = 1; | |
5444 | spin_unlock(&block_rsv->lock); | |
5445 | } | |
5446 | ||
d52be818 JB |
5447 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
5448 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5449 | int min_factor) | |
5450 | { | |
5451 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5452 | u64 min_bytes; | |
5453 | ||
5454 | if (global_rsv->space_info != dest->space_info) | |
5455 | return -ENOSPC; | |
5456 | ||
5457 | spin_lock(&global_rsv->lock); | |
5458 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5459 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5460 | spin_unlock(&global_rsv->lock); | |
5461 | return -ENOSPC; | |
5462 | } | |
5463 | global_rsv->reserved -= num_bytes; | |
5464 | if (global_rsv->reserved < global_rsv->size) | |
5465 | global_rsv->full = 0; | |
5466 | spin_unlock(&global_rsv->lock); | |
5467 | ||
5468 | block_rsv_add_bytes(dest, num_bytes, 1); | |
5469 | return 0; | |
5470 | } | |
5471 | ||
957780eb JB |
5472 | /* |
5473 | * This is for space we already have accounted in space_info->bytes_may_use, so | |
5474 | * basically when we're returning space from block_rsv's. | |
5475 | */ | |
5476 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
5477 | struct btrfs_space_info *space_info, | |
5478 | u64 num_bytes) | |
5479 | { | |
5480 | struct reserve_ticket *ticket; | |
5481 | struct list_head *head; | |
5482 | u64 used; | |
5483 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH; | |
5484 | bool check_overcommit = false; | |
5485 | ||
5486 | spin_lock(&space_info->lock); | |
5487 | head = &space_info->priority_tickets; | |
5488 | ||
5489 | /* | |
5490 | * If we are over our limit then we need to check and see if we can | |
5491 | * overcommit, and if we can't then we just need to free up our space | |
5492 | * and not satisfy any requests. | |
5493 | */ | |
0eee8a49 | 5494 | used = btrfs_space_info_used(space_info, true); |
957780eb JB |
5495 | if (used - num_bytes >= space_info->total_bytes) |
5496 | check_overcommit = true; | |
5497 | again: | |
5498 | while (!list_empty(head) && num_bytes) { | |
5499 | ticket = list_first_entry(head, struct reserve_ticket, | |
5500 | list); | |
5501 | /* | |
5502 | * We use 0 bytes because this space is already reserved, so | |
5503 | * adding the ticket space would be a double count. | |
5504 | */ | |
5505 | if (check_overcommit && | |
c1c4919b | 5506 | !can_overcommit(fs_info, space_info, 0, flush, false)) |
957780eb JB |
5507 | break; |
5508 | if (num_bytes >= ticket->bytes) { | |
5509 | list_del_init(&ticket->list); | |
5510 | num_bytes -= ticket->bytes; | |
5511 | ticket->bytes = 0; | |
ce129655 | 5512 | space_info->tickets_id++; |
957780eb JB |
5513 | wake_up(&ticket->wait); |
5514 | } else { | |
5515 | ticket->bytes -= num_bytes; | |
5516 | num_bytes = 0; | |
5517 | } | |
5518 | } | |
5519 | ||
5520 | if (num_bytes && head == &space_info->priority_tickets) { | |
5521 | head = &space_info->tickets; | |
5522 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
5523 | goto again; | |
5524 | } | |
5525 | space_info->bytes_may_use -= num_bytes; | |
5526 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5527 | space_info->flags, num_bytes, 0); | |
5528 | spin_unlock(&space_info->lock); | |
5529 | } | |
5530 | ||
5531 | /* | |
5532 | * This is for newly allocated space that isn't accounted in | |
5533 | * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent | |
5534 | * we use this helper. | |
5535 | */ | |
5536 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, | |
5537 | struct btrfs_space_info *space_info, | |
5538 | u64 num_bytes) | |
5539 | { | |
5540 | struct reserve_ticket *ticket; | |
5541 | struct list_head *head = &space_info->priority_tickets; | |
5542 | ||
5543 | again: | |
5544 | while (!list_empty(head) && num_bytes) { | |
5545 | ticket = list_first_entry(head, struct reserve_ticket, | |
5546 | list); | |
5547 | if (num_bytes >= ticket->bytes) { | |
5548 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5549 | space_info->flags, | |
5550 | ticket->bytes, 1); | |
5551 | list_del_init(&ticket->list); | |
5552 | num_bytes -= ticket->bytes; | |
5553 | space_info->bytes_may_use += ticket->bytes; | |
5554 | ticket->bytes = 0; | |
ce129655 | 5555 | space_info->tickets_id++; |
957780eb JB |
5556 | wake_up(&ticket->wait); |
5557 | } else { | |
5558 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5559 | space_info->flags, | |
5560 | num_bytes, 1); | |
5561 | space_info->bytes_may_use += num_bytes; | |
5562 | ticket->bytes -= num_bytes; | |
5563 | num_bytes = 0; | |
5564 | } | |
5565 | } | |
5566 | ||
5567 | if (num_bytes && head == &space_info->priority_tickets) { | |
5568 | head = &space_info->tickets; | |
5569 | goto again; | |
5570 | } | |
5571 | } | |
5572 | ||
69fe2d75 | 5573 | static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
8c2a3ca2 | 5574 | struct btrfs_block_rsv *block_rsv, |
62a45b60 | 5575 | struct btrfs_block_rsv *dest, u64 num_bytes) |
f0486c68 YZ |
5576 | { |
5577 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
69fe2d75 | 5578 | u64 ret; |
f0486c68 YZ |
5579 | |
5580 | spin_lock(&block_rsv->lock); | |
5581 | if (num_bytes == (u64)-1) | |
5582 | num_bytes = block_rsv->size; | |
5583 | block_rsv->size -= num_bytes; | |
5584 | if (block_rsv->reserved >= block_rsv->size) { | |
5585 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5586 | block_rsv->reserved = block_rsv->size; | |
5587 | block_rsv->full = 1; | |
5588 | } else { | |
5589 | num_bytes = 0; | |
5590 | } | |
5591 | spin_unlock(&block_rsv->lock); | |
5592 | ||
69fe2d75 | 5593 | ret = num_bytes; |
f0486c68 YZ |
5594 | if (num_bytes > 0) { |
5595 | if (dest) { | |
e9e22899 JB |
5596 | spin_lock(&dest->lock); |
5597 | if (!dest->full) { | |
5598 | u64 bytes_to_add; | |
5599 | ||
5600 | bytes_to_add = dest->size - dest->reserved; | |
5601 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5602 | dest->reserved += bytes_to_add; | |
5603 | if (dest->reserved >= dest->size) | |
5604 | dest->full = 1; | |
5605 | num_bytes -= bytes_to_add; | |
5606 | } | |
5607 | spin_unlock(&dest->lock); | |
5608 | } | |
957780eb JB |
5609 | if (num_bytes) |
5610 | space_info_add_old_bytes(fs_info, space_info, | |
5611 | num_bytes); | |
9ed74f2d | 5612 | } |
69fe2d75 | 5613 | return ret; |
f0486c68 | 5614 | } |
4e06bdd6 | 5615 | |
25d609f8 JB |
5616 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src, |
5617 | struct btrfs_block_rsv *dst, u64 num_bytes, | |
5618 | int update_size) | |
f0486c68 YZ |
5619 | { |
5620 | int ret; | |
9ed74f2d | 5621 | |
f0486c68 YZ |
5622 | ret = block_rsv_use_bytes(src, num_bytes); |
5623 | if (ret) | |
5624 | return ret; | |
9ed74f2d | 5625 | |
25d609f8 | 5626 | block_rsv_add_bytes(dst, num_bytes, update_size); |
9ed74f2d JB |
5627 | return 0; |
5628 | } | |
5629 | ||
66d8f3dd | 5630 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 5631 | { |
f0486c68 YZ |
5632 | memset(rsv, 0, sizeof(*rsv)); |
5633 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 5634 | rsv->type = type; |
f0486c68 YZ |
5635 | } |
5636 | ||
69fe2d75 JB |
5637 | void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info, |
5638 | struct btrfs_block_rsv *rsv, | |
5639 | unsigned short type) | |
5640 | { | |
5641 | btrfs_init_block_rsv(rsv, type); | |
5642 | rsv->space_info = __find_space_info(fs_info, | |
5643 | BTRFS_BLOCK_GROUP_METADATA); | |
5644 | } | |
5645 | ||
2ff7e61e | 5646 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info, |
66d8f3dd | 5647 | unsigned short type) |
f0486c68 YZ |
5648 | { |
5649 | struct btrfs_block_rsv *block_rsv; | |
9ed74f2d | 5650 | |
f0486c68 YZ |
5651 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
5652 | if (!block_rsv) | |
5653 | return NULL; | |
9ed74f2d | 5654 | |
69fe2d75 | 5655 | btrfs_init_metadata_block_rsv(fs_info, block_rsv, type); |
f0486c68 YZ |
5656 | return block_rsv; |
5657 | } | |
9ed74f2d | 5658 | |
2ff7e61e | 5659 | void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5660 | struct btrfs_block_rsv *rsv) |
5661 | { | |
2aaa6655 JB |
5662 | if (!rsv) |
5663 | return; | |
2ff7e61e | 5664 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
dabdb640 | 5665 | kfree(rsv); |
9ed74f2d JB |
5666 | } |
5667 | ||
cdfb080e CM |
5668 | void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv) |
5669 | { | |
5670 | kfree(rsv); | |
5671 | } | |
5672 | ||
08e007d2 MX |
5673 | int btrfs_block_rsv_add(struct btrfs_root *root, |
5674 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5675 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5676 | { |
f0486c68 | 5677 | int ret; |
9ed74f2d | 5678 | |
f0486c68 YZ |
5679 | if (num_bytes == 0) |
5680 | return 0; | |
8bb8ab2e | 5681 | |
61b520a9 | 5682 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
5683 | if (!ret) { |
5684 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
5685 | return 0; | |
5686 | } | |
9ed74f2d | 5687 | |
f0486c68 | 5688 | return ret; |
f0486c68 | 5689 | } |
9ed74f2d | 5690 | |
2ff7e61e | 5691 | int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
5692 | { |
5693 | u64 num_bytes = 0; | |
f0486c68 | 5694 | int ret = -ENOSPC; |
9ed74f2d | 5695 | |
f0486c68 YZ |
5696 | if (!block_rsv) |
5697 | return 0; | |
9ed74f2d | 5698 | |
f0486c68 | 5699 | spin_lock(&block_rsv->lock); |
36ba022a JB |
5700 | num_bytes = div_factor(block_rsv->size, min_factor); |
5701 | if (block_rsv->reserved >= num_bytes) | |
5702 | ret = 0; | |
5703 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 5704 | |
36ba022a JB |
5705 | return ret; |
5706 | } | |
5707 | ||
08e007d2 MX |
5708 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
5709 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5710 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
5711 | { |
5712 | u64 num_bytes = 0; | |
5713 | int ret = -ENOSPC; | |
5714 | ||
5715 | if (!block_rsv) | |
5716 | return 0; | |
5717 | ||
5718 | spin_lock(&block_rsv->lock); | |
5719 | num_bytes = min_reserved; | |
13553e52 | 5720 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 5721 | ret = 0; |
13553e52 | 5722 | else |
f0486c68 | 5723 | num_bytes -= block_rsv->reserved; |
f0486c68 | 5724 | spin_unlock(&block_rsv->lock); |
13553e52 | 5725 | |
f0486c68 YZ |
5726 | if (!ret) |
5727 | return 0; | |
5728 | ||
aa38a711 | 5729 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
5730 | if (!ret) { |
5731 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 5732 | return 0; |
6a63209f | 5733 | } |
9ed74f2d | 5734 | |
13553e52 | 5735 | return ret; |
f0486c68 YZ |
5736 | } |
5737 | ||
69fe2d75 JB |
5738 | /** |
5739 | * btrfs_inode_rsv_refill - refill the inode block rsv. | |
5740 | * @inode - the inode we are refilling. | |
5741 | * @flush - the flusing restriction. | |
5742 | * | |
5743 | * Essentially the same as btrfs_block_rsv_refill, except it uses the | |
5744 | * block_rsv->size as the minimum size. We'll either refill the missing amount | |
5745 | * or return if we already have enough space. This will also handle the resreve | |
5746 | * tracepoint for the reserved amount. | |
5747 | */ | |
5748 | int btrfs_inode_rsv_refill(struct btrfs_inode *inode, | |
5749 | enum btrfs_reserve_flush_enum flush) | |
5750 | { | |
5751 | struct btrfs_root *root = inode->root; | |
5752 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; | |
5753 | u64 num_bytes = 0; | |
5754 | int ret = -ENOSPC; | |
5755 | ||
5756 | spin_lock(&block_rsv->lock); | |
5757 | if (block_rsv->reserved < block_rsv->size) | |
5758 | num_bytes = block_rsv->size - block_rsv->reserved; | |
5759 | spin_unlock(&block_rsv->lock); | |
5760 | ||
5761 | if (num_bytes == 0) | |
5762 | return 0; | |
5763 | ||
5764 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); | |
5765 | if (!ret) { | |
5766 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
5767 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
5768 | btrfs_ino(inode), num_bytes, 1); | |
5769 | } | |
5770 | return ret; | |
5771 | } | |
5772 | ||
5773 | /** | |
5774 | * btrfs_inode_rsv_release - release any excessive reservation. | |
5775 | * @inode - the inode we need to release from. | |
5776 | * | |
5777 | * This is the same as btrfs_block_rsv_release, except that it handles the | |
5778 | * tracepoint for the reservation. | |
5779 | */ | |
5780 | void btrfs_inode_rsv_release(struct btrfs_inode *inode) | |
5781 | { | |
5782 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
5783 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5784 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; | |
5785 | u64 released = 0; | |
5786 | ||
5787 | /* | |
5788 | * Since we statically set the block_rsv->size we just want to say we | |
5789 | * are releasing 0 bytes, and then we'll just get the reservation over | |
5790 | * the size free'd. | |
5791 | */ | |
5792 | released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0); | |
5793 | if (released > 0) | |
5794 | trace_btrfs_space_reservation(fs_info, "delalloc", | |
5795 | btrfs_ino(inode), released, 0); | |
5796 | } | |
5797 | ||
2ff7e61e | 5798 | void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5799 | struct btrfs_block_rsv *block_rsv, |
5800 | u64 num_bytes) | |
5801 | { | |
0b246afa JM |
5802 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
5803 | ||
17504584 | 5804 | if (global_rsv == block_rsv || |
f0486c68 YZ |
5805 | block_rsv->space_info != global_rsv->space_info) |
5806 | global_rsv = NULL; | |
0b246afa | 5807 | block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes); |
6a63209f JB |
5808 | } |
5809 | ||
8929ecfa YZ |
5810 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5811 | { | |
5812 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5813 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5814 | u64 num_bytes; | |
6a63209f | 5815 | |
ae2e4728 JB |
5816 | /* |
5817 | * The global block rsv is based on the size of the extent tree, the | |
5818 | * checksum tree and the root tree. If the fs is empty we want to set | |
5819 | * it to a minimal amount for safety. | |
5820 | */ | |
5821 | num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) + | |
5822 | btrfs_root_used(&fs_info->csum_root->root_item) + | |
5823 | btrfs_root_used(&fs_info->tree_root->root_item); | |
5824 | num_bytes = max_t(u64, num_bytes, SZ_16M); | |
33b4d47f | 5825 | |
8929ecfa | 5826 | spin_lock(&sinfo->lock); |
1f699d38 | 5827 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5828 | |
ee22184b | 5829 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); |
4e06bdd6 | 5830 | |
fb4b10e5 | 5831 | if (block_rsv->reserved < block_rsv->size) { |
4136135b | 5832 | num_bytes = btrfs_space_info_used(sinfo, true); |
fb4b10e5 JB |
5833 | if (sinfo->total_bytes > num_bytes) { |
5834 | num_bytes = sinfo->total_bytes - num_bytes; | |
5835 | num_bytes = min(num_bytes, | |
5836 | block_rsv->size - block_rsv->reserved); | |
5837 | block_rsv->reserved += num_bytes; | |
5838 | sinfo->bytes_may_use += num_bytes; | |
5839 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5840 | sinfo->flags, num_bytes, | |
5841 | 1); | |
5842 | } | |
5843 | } else if (block_rsv->reserved > block_rsv->size) { | |
8929ecfa | 5844 | num_bytes = block_rsv->reserved - block_rsv->size; |
fb25e914 | 5845 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 5846 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5847 | sinfo->flags, num_bytes, 0); |
8929ecfa | 5848 | block_rsv->reserved = block_rsv->size; |
8929ecfa | 5849 | } |
182608c8 | 5850 | |
fb4b10e5 JB |
5851 | if (block_rsv->reserved == block_rsv->size) |
5852 | block_rsv->full = 1; | |
5853 | else | |
5854 | block_rsv->full = 0; | |
5855 | ||
8929ecfa | 5856 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5857 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5858 | } |
5859 | ||
f0486c68 | 5860 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5861 | { |
f0486c68 | 5862 | struct btrfs_space_info *space_info; |
6a63209f | 5863 | |
f0486c68 YZ |
5864 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
5865 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 5866 | |
f0486c68 | 5867 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5868 | fs_info->global_block_rsv.space_info = space_info; |
f0486c68 YZ |
5869 | fs_info->trans_block_rsv.space_info = space_info; |
5870 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5871 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 5872 | |
8929ecfa YZ |
5873 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
5874 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
5875 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
5876 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5877 | if (fs_info->quota_root) |
5878 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5879 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5880 | |
8929ecfa | 5881 | update_global_block_rsv(fs_info); |
6a63209f JB |
5882 | } |
5883 | ||
8929ecfa | 5884 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5885 | { |
8c2a3ca2 JB |
5886 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
5887 | (u64)-1); | |
8929ecfa YZ |
5888 | WARN_ON(fs_info->trans_block_rsv.size > 0); |
5889 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5890 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5891 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5892 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5893 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
5894 | } |
5895 | ||
a22285a6 | 5896 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
2ff7e61e | 5897 | struct btrfs_fs_info *fs_info) |
6a63209f | 5898 | { |
bf2681cb JB |
5899 | if (!trans->block_rsv) { |
5900 | ASSERT(!trans->bytes_reserved); | |
0e721106 | 5901 | return; |
bf2681cb | 5902 | } |
0e721106 | 5903 | |
a22285a6 YZ |
5904 | if (!trans->bytes_reserved) |
5905 | return; | |
6a63209f | 5906 | |
bf2681cb | 5907 | ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); |
0b246afa | 5908 | trace_btrfs_space_reservation(fs_info, "transaction", |
2bcc0328 | 5909 | trans->transid, trans->bytes_reserved, 0); |
2ff7e61e JM |
5910 | btrfs_block_rsv_release(fs_info, trans->block_rsv, |
5911 | trans->bytes_reserved); | |
a22285a6 YZ |
5912 | trans->bytes_reserved = 0; |
5913 | } | |
6a63209f | 5914 | |
4fbcdf66 FM |
5915 | /* |
5916 | * To be called after all the new block groups attached to the transaction | |
5917 | * handle have been created (btrfs_create_pending_block_groups()). | |
5918 | */ | |
5919 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5920 | { | |
64b63580 | 5921 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4fbcdf66 FM |
5922 | |
5923 | if (!trans->chunk_bytes_reserved) | |
5924 | return; | |
5925 | ||
5926 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5927 | ||
5928 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
5929 | trans->chunk_bytes_reserved); | |
5930 | trans->chunk_bytes_reserved = 0; | |
5931 | } | |
5932 | ||
79787eaa | 5933 | /* Can only return 0 or -ENOSPC */ |
d68fc57b | 5934 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
8ed7a2a0 | 5935 | struct btrfs_inode *inode) |
d68fc57b | 5936 | { |
8ed7a2a0 NB |
5937 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
5938 | struct btrfs_root *root = inode->root; | |
40acc3ee JB |
5939 | /* |
5940 | * We always use trans->block_rsv here as we will have reserved space | |
5941 | * for our orphan when starting the transaction, using get_block_rsv() | |
5942 | * here will sometimes make us choose the wrong block rsv as we could be | |
5943 | * doing a reloc inode for a non refcounted root. | |
5944 | */ | |
5945 | struct btrfs_block_rsv *src_rsv = trans->block_rsv; | |
d68fc57b YZ |
5946 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; |
5947 | ||
5948 | /* | |
fcb80c2a JB |
5949 | * We need to hold space in order to delete our orphan item once we've |
5950 | * added it, so this takes the reservation so we can release it later | |
5951 | * when we are truly done with the orphan item. | |
d68fc57b | 5952 | */ |
0b246afa JM |
5953 | u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
5954 | ||
8ed7a2a0 NB |
5955 | trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode), |
5956 | num_bytes, 1); | |
25d609f8 | 5957 | return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1); |
6a63209f JB |
5958 | } |
5959 | ||
703b391a | 5960 | void btrfs_orphan_release_metadata(struct btrfs_inode *inode) |
97e728d4 | 5961 | { |
703b391a NB |
5962 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
5963 | struct btrfs_root *root = inode->root; | |
0b246afa JM |
5964 | u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
5965 | ||
703b391a NB |
5966 | trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode), |
5967 | num_bytes, 0); | |
2ff7e61e | 5968 | btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes); |
d68fc57b | 5969 | } |
97e728d4 | 5970 | |
d5c12070 MX |
5971 | /* |
5972 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5973 | * root: the root of the parent directory | |
5974 | * rsv: block reservation | |
5975 | * items: the number of items that we need do reservation | |
5976 | * qgroup_reserved: used to return the reserved size in qgroup | |
5977 | * | |
5978 | * This function is used to reserve the space for snapshot/subvolume | |
5979 | * creation and deletion. Those operations are different with the | |
5980 | * common file/directory operations, they change two fs/file trees | |
5981 | * and root tree, the number of items that the qgroup reserves is | |
5982 | * different with the free space reservation. So we can not use | |
01327610 | 5983 | * the space reservation mechanism in start_transaction(). |
d5c12070 MX |
5984 | */ |
5985 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
5986 | struct btrfs_block_rsv *rsv, | |
5987 | int items, | |
ee3441b4 JM |
5988 | u64 *qgroup_reserved, |
5989 | bool use_global_rsv) | |
a22285a6 | 5990 | { |
d5c12070 MX |
5991 | u64 num_bytes; |
5992 | int ret; | |
0b246afa JM |
5993 | struct btrfs_fs_info *fs_info = root->fs_info; |
5994 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d5c12070 | 5995 | |
0b246afa | 5996 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
d5c12070 | 5997 | /* One for parent inode, two for dir entries */ |
0b246afa | 5998 | num_bytes = 3 * fs_info->nodesize; |
003d7c59 | 5999 | ret = btrfs_qgroup_reserve_meta(root, num_bytes, true); |
d5c12070 MX |
6000 | if (ret) |
6001 | return ret; | |
6002 | } else { | |
6003 | num_bytes = 0; | |
6004 | } | |
6005 | ||
6006 | *qgroup_reserved = num_bytes; | |
6007 | ||
0b246afa JM |
6008 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, items); |
6009 | rsv->space_info = __find_space_info(fs_info, | |
d5c12070 MX |
6010 | BTRFS_BLOCK_GROUP_METADATA); |
6011 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
6012 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
6013 | |
6014 | if (ret == -ENOSPC && use_global_rsv) | |
25d609f8 | 6015 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1); |
ee3441b4 | 6016 | |
7174109c QW |
6017 | if (ret && *qgroup_reserved) |
6018 | btrfs_qgroup_free_meta(root, *qgroup_reserved); | |
d5c12070 MX |
6019 | |
6020 | return ret; | |
6021 | } | |
6022 | ||
2ff7e61e | 6023 | void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info, |
7775c818 | 6024 | struct btrfs_block_rsv *rsv) |
d5c12070 | 6025 | { |
2ff7e61e | 6026 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
97e728d4 JB |
6027 | } |
6028 | ||
69fe2d75 JB |
6029 | static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info, |
6030 | struct btrfs_inode *inode) | |
9e0baf60 | 6031 | { |
69fe2d75 JB |
6032 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
6033 | u64 reserve_size = 0; | |
6034 | u64 csum_leaves; | |
6035 | unsigned outstanding_extents; | |
9e0baf60 | 6036 | |
69fe2d75 JB |
6037 | lockdep_assert_held(&inode->lock); |
6038 | outstanding_extents = inode->outstanding_extents; | |
6039 | if (outstanding_extents) | |
6040 | reserve_size = btrfs_calc_trans_metadata_size(fs_info, | |
6041 | outstanding_extents + 1); | |
6042 | csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, | |
6043 | inode->csum_bytes); | |
6044 | reserve_size += btrfs_calc_trans_metadata_size(fs_info, | |
6045 | csum_leaves); | |
9e0baf60 | 6046 | |
69fe2d75 JB |
6047 | spin_lock(&block_rsv->lock); |
6048 | block_rsv->size = reserve_size; | |
6049 | spin_unlock(&block_rsv->lock); | |
0ca1f7ce | 6050 | } |
c146afad | 6051 | |
9f3db423 | 6052 | int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 6053 | { |
9f3db423 NB |
6054 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
6055 | struct btrfs_root *root = inode->root; | |
69fe2d75 | 6056 | unsigned nr_extents; |
08e007d2 | 6057 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 6058 | int ret = 0; |
c64c2bd8 | 6059 | bool delalloc_lock = true; |
6324fbf3 | 6060 | |
c64c2bd8 JB |
6061 | /* If we are a free space inode we need to not flush since we will be in |
6062 | * the middle of a transaction commit. We also don't need the delalloc | |
6063 | * mutex since we won't race with anybody. We need this mostly to make | |
6064 | * lockdep shut its filthy mouth. | |
bac357dc JB |
6065 | * |
6066 | * If we have a transaction open (can happen if we call truncate_block | |
6067 | * from truncate), then we need FLUSH_LIMIT so we don't deadlock. | |
c64c2bd8 JB |
6068 | */ |
6069 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 6070 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 | 6071 | delalloc_lock = false; |
bac357dc JB |
6072 | } else if (current->journal_info) { |
6073 | flush = BTRFS_RESERVE_FLUSH_LIMIT; | |
c64c2bd8 | 6074 | } |
c09544e0 | 6075 | |
08e007d2 | 6076 | if (flush != BTRFS_RESERVE_NO_FLUSH && |
0b246afa | 6077 | btrfs_transaction_in_commit(fs_info)) |
0ca1f7ce | 6078 | schedule_timeout(1); |
ec44a35c | 6079 | |
c64c2bd8 | 6080 | if (delalloc_lock) |
9f3db423 | 6081 | mutex_lock(&inode->delalloc_mutex); |
c64c2bd8 | 6082 | |
0b246afa | 6083 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
69fe2d75 JB |
6084 | |
6085 | /* Add our new extents and calculate the new rsv size. */ | |
9f3db423 | 6086 | spin_lock(&inode->lock); |
69fe2d75 | 6087 | nr_extents = count_max_extents(num_bytes); |
8b62f87b | 6088 | btrfs_mod_outstanding_extents(inode, nr_extents); |
69fe2d75 JB |
6089 | inode->csum_bytes += num_bytes; |
6090 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 6091 | spin_unlock(&inode->lock); |
57a45ced | 6092 | |
0b246afa | 6093 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
7174109c | 6094 | ret = btrfs_qgroup_reserve_meta(root, |
003d7c59 | 6095 | nr_extents * fs_info->nodesize, true); |
88e081bf WS |
6096 | if (ret) |
6097 | goto out_fail; | |
6098 | } | |
c5567237 | 6099 | |
69fe2d75 | 6100 | ret = btrfs_inode_rsv_refill(inode, flush); |
88e081bf | 6101 | if (unlikely(ret)) { |
da17066c | 6102 | btrfs_qgroup_free_meta(root, |
0b246afa | 6103 | nr_extents * fs_info->nodesize); |
88e081bf | 6104 | goto out_fail; |
9e0baf60 | 6105 | } |
25179201 | 6106 | |
c64c2bd8 | 6107 | if (delalloc_lock) |
9f3db423 | 6108 | mutex_unlock(&inode->delalloc_mutex); |
0ca1f7ce | 6109 | return 0; |
88e081bf WS |
6110 | |
6111 | out_fail: | |
9f3db423 | 6112 | spin_lock(&inode->lock); |
8b62f87b JB |
6113 | nr_extents = count_max_extents(num_bytes); |
6114 | btrfs_mod_outstanding_extents(inode, -nr_extents); | |
69fe2d75 JB |
6115 | inode->csum_bytes -= num_bytes; |
6116 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 6117 | spin_unlock(&inode->lock); |
88e081bf | 6118 | |
69fe2d75 | 6119 | btrfs_inode_rsv_release(inode); |
88e081bf | 6120 | if (delalloc_lock) |
9f3db423 | 6121 | mutex_unlock(&inode->delalloc_mutex); |
88e081bf | 6122 | return ret; |
0ca1f7ce YZ |
6123 | } |
6124 | ||
7709cde3 JB |
6125 | /** |
6126 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
8b62f87b JB |
6127 | * @inode: the inode to release the reservation for. |
6128 | * @num_bytes: the number of bytes we are releasing. | |
7709cde3 JB |
6129 | * |
6130 | * This will release the metadata reservation for an inode. This can be called | |
6131 | * once we complete IO for a given set of bytes to release their metadata | |
8b62f87b | 6132 | * reservations, or on error for the same reason. |
7709cde3 | 6133 | */ |
691fa059 | 6134 | void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 6135 | { |
691fa059 | 6136 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
0ca1f7ce | 6137 | |
0b246afa | 6138 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
691fa059 | 6139 | spin_lock(&inode->lock); |
69fe2d75 JB |
6140 | inode->csum_bytes -= num_bytes; |
6141 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
691fa059 | 6142 | spin_unlock(&inode->lock); |
0ca1f7ce | 6143 | |
0b246afa | 6144 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
6145 | return; |
6146 | ||
69fe2d75 | 6147 | btrfs_inode_rsv_release(inode); |
0ca1f7ce YZ |
6148 | } |
6149 | ||
8b62f87b JB |
6150 | /** |
6151 | * btrfs_delalloc_release_extents - release our outstanding_extents | |
6152 | * @inode: the inode to balance the reservation for. | |
6153 | * @num_bytes: the number of bytes we originally reserved with | |
6154 | * | |
6155 | * When we reserve space we increase outstanding_extents for the extents we may | |
6156 | * add. Once we've set the range as delalloc or created our ordered extents we | |
6157 | * have outstanding_extents to track the real usage, so we use this to free our | |
6158 | * temporarily tracked outstanding_extents. This _must_ be used in conjunction | |
6159 | * with btrfs_delalloc_reserve_metadata. | |
6160 | */ | |
6161 | void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes) | |
6162 | { | |
6163 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); | |
6164 | unsigned num_extents; | |
8b62f87b JB |
6165 | |
6166 | spin_lock(&inode->lock); | |
6167 | num_extents = count_max_extents(num_bytes); | |
6168 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
69fe2d75 | 6169 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); |
8b62f87b JB |
6170 | spin_unlock(&inode->lock); |
6171 | ||
8b62f87b JB |
6172 | if (btrfs_is_testing(fs_info)) |
6173 | return; | |
6174 | ||
69fe2d75 | 6175 | btrfs_inode_rsv_release(inode); |
8b62f87b JB |
6176 | } |
6177 | ||
1ada3a62 | 6178 | /** |
7cf5b976 | 6179 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
6180 | * delalloc |
6181 | * @inode: inode we're writing to | |
6182 | * @start: start range we are writing to | |
6183 | * @len: how long the range we are writing to | |
364ecf36 QW |
6184 | * @reserved: mandatory parameter, record actually reserved qgroup ranges of |
6185 | * current reservation. | |
1ada3a62 | 6186 | * |
1ada3a62 QW |
6187 | * This will do the following things |
6188 | * | |
6189 | * o reserve space in data space info for num bytes | |
6190 | * and reserve precious corresponding qgroup space | |
6191 | * (Done in check_data_free_space) | |
6192 | * | |
6193 | * o reserve space for metadata space, based on the number of outstanding | |
6194 | * extents and how much csums will be needed | |
6195 | * also reserve metadata space in a per root over-reserve method. | |
6196 | * o add to the inodes->delalloc_bytes | |
6197 | * o add it to the fs_info's delalloc inodes list. | |
6198 | * (Above 3 all done in delalloc_reserve_metadata) | |
6199 | * | |
6200 | * Return 0 for success | |
6201 | * Return <0 for error(-ENOSPC or -EQUOT) | |
6202 | */ | |
364ecf36 QW |
6203 | int btrfs_delalloc_reserve_space(struct inode *inode, |
6204 | struct extent_changeset **reserved, u64 start, u64 len) | |
1ada3a62 QW |
6205 | { |
6206 | int ret; | |
6207 | ||
364ecf36 | 6208 | ret = btrfs_check_data_free_space(inode, reserved, start, len); |
1ada3a62 QW |
6209 | if (ret < 0) |
6210 | return ret; | |
9f3db423 | 6211 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len); |
1ada3a62 | 6212 | if (ret < 0) |
bc42bda2 | 6213 | btrfs_free_reserved_data_space(inode, *reserved, start, len); |
1ada3a62 QW |
6214 | return ret; |
6215 | } | |
6216 | ||
7709cde3 | 6217 | /** |
7cf5b976 | 6218 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
6219 | * @inode: inode we're releasing space for |
6220 | * @start: start position of the space already reserved | |
6221 | * @len: the len of the space already reserved | |
8b62f87b | 6222 | * @release_bytes: the len of the space we consumed or didn't use |
1ada3a62 QW |
6223 | * |
6224 | * This function will release the metadata space that was not used and will | |
6225 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
6226 | * list if there are no delalloc bytes left. | |
6227 | * Also it will handle the qgroup reserved space. | |
6228 | */ | |
bc42bda2 | 6229 | void btrfs_delalloc_release_space(struct inode *inode, |
8b62f87b JB |
6230 | struct extent_changeset *reserved, |
6231 | u64 start, u64 len) | |
1ada3a62 | 6232 | { |
691fa059 | 6233 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len); |
bc42bda2 | 6234 | btrfs_free_reserved_data_space(inode, reserved, start, len); |
6324fbf3 CM |
6235 | } |
6236 | ||
ce93ec54 | 6237 | static int update_block_group(struct btrfs_trans_handle *trans, |
6202df69 | 6238 | struct btrfs_fs_info *info, u64 bytenr, |
ce93ec54 | 6239 | u64 num_bytes, int alloc) |
9078a3e1 | 6240 | { |
0af3d00b | 6241 | struct btrfs_block_group_cache *cache = NULL; |
db94535d | 6242 | u64 total = num_bytes; |
9078a3e1 | 6243 | u64 old_val; |
db94535d | 6244 | u64 byte_in_group; |
0af3d00b | 6245 | int factor; |
3e1ad54f | 6246 | |
5d4f98a2 | 6247 | /* block accounting for super block */ |
eb73c1b7 | 6248 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 6249 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
6250 | if (alloc) |
6251 | old_val += num_bytes; | |
6252 | else | |
6253 | old_val -= num_bytes; | |
6c41761f | 6254 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 6255 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 6256 | |
d397712b | 6257 | while (total) { |
db94535d | 6258 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 6259 | if (!cache) |
79787eaa | 6260 | return -ENOENT; |
b742bb82 YZ |
6261 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
6262 | BTRFS_BLOCK_GROUP_RAID1 | | |
6263 | BTRFS_BLOCK_GROUP_RAID10)) | |
6264 | factor = 2; | |
6265 | else | |
6266 | factor = 1; | |
9d66e233 JB |
6267 | /* |
6268 | * If this block group has free space cache written out, we | |
6269 | * need to make sure to load it if we are removing space. This | |
6270 | * is because we need the unpinning stage to actually add the | |
6271 | * space back to the block group, otherwise we will leak space. | |
6272 | */ | |
6273 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 6274 | cache_block_group(cache, 1); |
0af3d00b | 6275 | |
db94535d CM |
6276 | byte_in_group = bytenr - cache->key.objectid; |
6277 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 6278 | |
25179201 | 6279 | spin_lock(&cache->space_info->lock); |
c286ac48 | 6280 | spin_lock(&cache->lock); |
0af3d00b | 6281 | |
6202df69 | 6282 | if (btrfs_test_opt(info, SPACE_CACHE) && |
0af3d00b JB |
6283 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
6284 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
6285 | ||
9078a3e1 | 6286 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 6287 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 6288 | if (alloc) { |
db94535d | 6289 | old_val += num_bytes; |
11833d66 YZ |
6290 | btrfs_set_block_group_used(&cache->item, old_val); |
6291 | cache->reserved -= num_bytes; | |
11833d66 | 6292 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
6293 | cache->space_info->bytes_used += num_bytes; |
6294 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 6295 | spin_unlock(&cache->lock); |
25179201 | 6296 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 6297 | } else { |
db94535d | 6298 | old_val -= num_bytes; |
ae0ab003 FM |
6299 | btrfs_set_block_group_used(&cache->item, old_val); |
6300 | cache->pinned += num_bytes; | |
6301 | cache->space_info->bytes_pinned += num_bytes; | |
6302 | cache->space_info->bytes_used -= num_bytes; | |
6303 | cache->space_info->disk_used -= num_bytes * factor; | |
6304 | spin_unlock(&cache->lock); | |
6305 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 6306 | |
0b246afa | 6307 | trace_btrfs_space_reservation(info, "pinned", |
c51e7bb1 JB |
6308 | cache->space_info->flags, |
6309 | num_bytes, 1); | |
d7eae340 OS |
6310 | percpu_counter_add(&cache->space_info->total_bytes_pinned, |
6311 | num_bytes); | |
ae0ab003 FM |
6312 | set_extent_dirty(info->pinned_extents, |
6313 | bytenr, bytenr + num_bytes - 1, | |
6314 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 6315 | } |
1bbc621e CM |
6316 | |
6317 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
6318 | if (list_empty(&cache->dirty_list)) { | |
6319 | list_add_tail(&cache->dirty_list, | |
6320 | &trans->transaction->dirty_bgs); | |
6321 | trans->transaction->num_dirty_bgs++; | |
6322 | btrfs_get_block_group(cache); | |
6323 | } | |
6324 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
6325 | ||
036a9348 FM |
6326 | /* |
6327 | * No longer have used bytes in this block group, queue it for | |
6328 | * deletion. We do this after adding the block group to the | |
6329 | * dirty list to avoid races between cleaner kthread and space | |
6330 | * cache writeout. | |
6331 | */ | |
6332 | if (!alloc && old_val == 0) { | |
6333 | spin_lock(&info->unused_bgs_lock); | |
6334 | if (list_empty(&cache->bg_list)) { | |
6335 | btrfs_get_block_group(cache); | |
6336 | list_add_tail(&cache->bg_list, | |
6337 | &info->unused_bgs); | |
6338 | } | |
6339 | spin_unlock(&info->unused_bgs_lock); | |
6340 | } | |
6341 | ||
fa9c0d79 | 6342 | btrfs_put_block_group(cache); |
db94535d CM |
6343 | total -= num_bytes; |
6344 | bytenr += num_bytes; | |
9078a3e1 CM |
6345 | } |
6346 | return 0; | |
6347 | } | |
6324fbf3 | 6348 | |
2ff7e61e | 6349 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 6350 | { |
0f9dd46c | 6351 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 6352 | u64 bytenr; |
0f9dd46c | 6353 | |
0b246afa JM |
6354 | spin_lock(&fs_info->block_group_cache_lock); |
6355 | bytenr = fs_info->first_logical_byte; | |
6356 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
6357 | |
6358 | if (bytenr < (u64)-1) | |
6359 | return bytenr; | |
6360 | ||
0b246afa | 6361 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 6362 | if (!cache) |
a061fc8d | 6363 | return 0; |
0f9dd46c | 6364 | |
d2fb3437 | 6365 | bytenr = cache->key.objectid; |
fa9c0d79 | 6366 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
6367 | |
6368 | return bytenr; | |
a061fc8d CM |
6369 | } |
6370 | ||
2ff7e61e | 6371 | static int pin_down_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6372 | struct btrfs_block_group_cache *cache, |
6373 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 6374 | { |
11833d66 YZ |
6375 | spin_lock(&cache->space_info->lock); |
6376 | spin_lock(&cache->lock); | |
6377 | cache->pinned += num_bytes; | |
6378 | cache->space_info->bytes_pinned += num_bytes; | |
6379 | if (reserved) { | |
6380 | cache->reserved -= num_bytes; | |
6381 | cache->space_info->bytes_reserved -= num_bytes; | |
6382 | } | |
6383 | spin_unlock(&cache->lock); | |
6384 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 6385 | |
0b246afa | 6386 | trace_btrfs_space_reservation(fs_info, "pinned", |
c51e7bb1 | 6387 | cache->space_info->flags, num_bytes, 1); |
4da8b76d | 6388 | percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes); |
0b246afa | 6389 | set_extent_dirty(fs_info->pinned_extents, bytenr, |
f0486c68 YZ |
6390 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
6391 | return 0; | |
6392 | } | |
68b38550 | 6393 | |
f0486c68 YZ |
6394 | /* |
6395 | * this function must be called within transaction | |
6396 | */ | |
2ff7e61e | 6397 | int btrfs_pin_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6398 | u64 bytenr, u64 num_bytes, int reserved) |
6399 | { | |
6400 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6401 | |
0b246afa | 6402 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
79787eaa | 6403 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 6404 | |
2ff7e61e | 6405 | pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
6406 | |
6407 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6408 | return 0; |
6409 | } | |
6410 | ||
f0486c68 | 6411 | /* |
e688b725 CM |
6412 | * this function must be called within transaction |
6413 | */ | |
2ff7e61e | 6414 | int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info, |
e688b725 CM |
6415 | u64 bytenr, u64 num_bytes) |
6416 | { | |
6417 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6418 | int ret; |
e688b725 | 6419 | |
0b246afa | 6420 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
b50c6e25 JB |
6421 | if (!cache) |
6422 | return -EINVAL; | |
e688b725 CM |
6423 | |
6424 | /* | |
6425 | * pull in the free space cache (if any) so that our pin | |
6426 | * removes the free space from the cache. We have load_only set | |
6427 | * to one because the slow code to read in the free extents does check | |
6428 | * the pinned extents. | |
6429 | */ | |
f6373bf3 | 6430 | cache_block_group(cache, 1); |
e688b725 | 6431 | |
2ff7e61e | 6432 | pin_down_extent(fs_info, cache, bytenr, num_bytes, 0); |
e688b725 CM |
6433 | |
6434 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6435 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6436 | btrfs_put_block_group(cache); |
b50c6e25 | 6437 | return ret; |
e688b725 CM |
6438 | } |
6439 | ||
2ff7e61e JM |
6440 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
6441 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
6442 | { |
6443 | int ret; | |
6444 | struct btrfs_block_group_cache *block_group; | |
6445 | struct btrfs_caching_control *caching_ctl; | |
6446 | ||
0b246afa | 6447 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
6448 | if (!block_group) |
6449 | return -EINVAL; | |
6450 | ||
6451 | cache_block_group(block_group, 0); | |
6452 | caching_ctl = get_caching_control(block_group); | |
6453 | ||
6454 | if (!caching_ctl) { | |
6455 | /* Logic error */ | |
6456 | BUG_ON(!block_group_cache_done(block_group)); | |
6457 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6458 | } else { | |
6459 | mutex_lock(&caching_ctl->mutex); | |
6460 | ||
6461 | if (start >= caching_ctl->progress) { | |
2ff7e61e | 6462 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6463 | } else if (start + num_bytes <= caching_ctl->progress) { |
6464 | ret = btrfs_remove_free_space(block_group, | |
6465 | start, num_bytes); | |
6466 | } else { | |
6467 | num_bytes = caching_ctl->progress - start; | |
6468 | ret = btrfs_remove_free_space(block_group, | |
6469 | start, num_bytes); | |
6470 | if (ret) | |
6471 | goto out_lock; | |
6472 | ||
6473 | num_bytes = (start + num_bytes) - | |
6474 | caching_ctl->progress; | |
6475 | start = caching_ctl->progress; | |
2ff7e61e | 6476 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6477 | } |
6478 | out_lock: | |
6479 | mutex_unlock(&caching_ctl->mutex); | |
6480 | put_caching_control(caching_ctl); | |
6481 | } | |
6482 | btrfs_put_block_group(block_group); | |
6483 | return ret; | |
6484 | } | |
6485 | ||
2ff7e61e | 6486 | int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info, |
8c2a1a30 JB |
6487 | struct extent_buffer *eb) |
6488 | { | |
6489 | struct btrfs_file_extent_item *item; | |
6490 | struct btrfs_key key; | |
6491 | int found_type; | |
6492 | int i; | |
6493 | ||
2ff7e61e | 6494 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
6495 | return 0; |
6496 | ||
6497 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6498 | btrfs_item_key_to_cpu(eb, &key, i); | |
6499 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6500 | continue; | |
6501 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6502 | found_type = btrfs_file_extent_type(eb, item); | |
6503 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6504 | continue; | |
6505 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6506 | continue; | |
6507 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6508 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
2ff7e61e | 6509 | __exclude_logged_extent(fs_info, key.objectid, key.offset); |
8c2a1a30 JB |
6510 | } |
6511 | ||
6512 | return 0; | |
6513 | } | |
6514 | ||
9cfa3e34 FM |
6515 | static void |
6516 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6517 | { | |
6518 | atomic_inc(&bg->reservations); | |
6519 | } | |
6520 | ||
6521 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
6522 | const u64 start) | |
6523 | { | |
6524 | struct btrfs_block_group_cache *bg; | |
6525 | ||
6526 | bg = btrfs_lookup_block_group(fs_info, start); | |
6527 | ASSERT(bg); | |
6528 | if (atomic_dec_and_test(&bg->reservations)) | |
6529 | wake_up_atomic_t(&bg->reservations); | |
6530 | btrfs_put_block_group(bg); | |
6531 | } | |
6532 | ||
6533 | static int btrfs_wait_bg_reservations_atomic_t(atomic_t *a) | |
6534 | { | |
6535 | schedule(); | |
6536 | return 0; | |
6537 | } | |
6538 | ||
6539 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6540 | { | |
6541 | struct btrfs_space_info *space_info = bg->space_info; | |
6542 | ||
6543 | ASSERT(bg->ro); | |
6544 | ||
6545 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
6546 | return; | |
6547 | ||
6548 | /* | |
6549 | * Our block group is read only but before we set it to read only, | |
6550 | * some task might have had allocated an extent from it already, but it | |
6551 | * has not yet created a respective ordered extent (and added it to a | |
6552 | * root's list of ordered extents). | |
6553 | * Therefore wait for any task currently allocating extents, since the | |
6554 | * block group's reservations counter is incremented while a read lock | |
6555 | * on the groups' semaphore is held and decremented after releasing | |
6556 | * the read access on that semaphore and creating the ordered extent. | |
6557 | */ | |
6558 | down_write(&space_info->groups_sem); | |
6559 | up_write(&space_info->groups_sem); | |
6560 | ||
6561 | wait_on_atomic_t(&bg->reservations, | |
6562 | btrfs_wait_bg_reservations_atomic_t, | |
6563 | TASK_UNINTERRUPTIBLE); | |
6564 | } | |
6565 | ||
fb25e914 | 6566 | /** |
4824f1f4 | 6567 | * btrfs_add_reserved_bytes - update the block_group and space info counters |
fb25e914 | 6568 | * @cache: The cache we are manipulating |
18513091 WX |
6569 | * @ram_bytes: The number of bytes of file content, and will be same to |
6570 | * @num_bytes except for the compress path. | |
fb25e914 | 6571 | * @num_bytes: The number of bytes in question |
e570fd27 | 6572 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 | 6573 | * |
745699ef XW |
6574 | * This is called by the allocator when it reserves space. If this is a |
6575 | * reservation and the block group has become read only we cannot make the | |
6576 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
f0486c68 | 6577 | */ |
4824f1f4 | 6578 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 6579 | u64 ram_bytes, u64 num_bytes, int delalloc) |
11833d66 | 6580 | { |
fb25e914 | 6581 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6582 | int ret = 0; |
79787eaa | 6583 | |
fb25e914 JB |
6584 | spin_lock(&space_info->lock); |
6585 | spin_lock(&cache->lock); | |
4824f1f4 WX |
6586 | if (cache->ro) { |
6587 | ret = -EAGAIN; | |
fb25e914 | 6588 | } else { |
4824f1f4 WX |
6589 | cache->reserved += num_bytes; |
6590 | space_info->bytes_reserved += num_bytes; | |
e570fd27 | 6591 | |
18513091 WX |
6592 | trace_btrfs_space_reservation(cache->fs_info, |
6593 | "space_info", space_info->flags, | |
6594 | ram_bytes, 0); | |
6595 | space_info->bytes_may_use -= ram_bytes; | |
e570fd27 | 6596 | if (delalloc) |
4824f1f4 | 6597 | cache->delalloc_bytes += num_bytes; |
324ae4df | 6598 | } |
fb25e914 JB |
6599 | spin_unlock(&cache->lock); |
6600 | spin_unlock(&space_info->lock); | |
f0486c68 | 6601 | return ret; |
324ae4df | 6602 | } |
9078a3e1 | 6603 | |
4824f1f4 WX |
6604 | /** |
6605 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
6606 | * @cache: The cache we are manipulating | |
6607 | * @num_bytes: The number of bytes in question | |
6608 | * @delalloc: The blocks are allocated for the delalloc write | |
6609 | * | |
6610 | * This is called by somebody who is freeing space that was never actually used | |
6611 | * on disk. For example if you reserve some space for a new leaf in transaction | |
6612 | * A and before transaction A commits you free that leaf, you call this with | |
6613 | * reserve set to 0 in order to clear the reservation. | |
6614 | */ | |
6615 | ||
6616 | static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, | |
6617 | u64 num_bytes, int delalloc) | |
6618 | { | |
6619 | struct btrfs_space_info *space_info = cache->space_info; | |
6620 | int ret = 0; | |
6621 | ||
6622 | spin_lock(&space_info->lock); | |
6623 | spin_lock(&cache->lock); | |
6624 | if (cache->ro) | |
6625 | space_info->bytes_readonly += num_bytes; | |
6626 | cache->reserved -= num_bytes; | |
6627 | space_info->bytes_reserved -= num_bytes; | |
6628 | ||
6629 | if (delalloc) | |
6630 | cache->delalloc_bytes -= num_bytes; | |
6631 | spin_unlock(&cache->lock); | |
6632 | spin_unlock(&space_info->lock); | |
6633 | return ret; | |
6634 | } | |
8b74c03e | 6635 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 6636 | { |
11833d66 YZ |
6637 | struct btrfs_caching_control *next; |
6638 | struct btrfs_caching_control *caching_ctl; | |
6639 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6640 | |
9e351cc8 | 6641 | down_write(&fs_info->commit_root_sem); |
25179201 | 6642 | |
11833d66 YZ |
6643 | list_for_each_entry_safe(caching_ctl, next, |
6644 | &fs_info->caching_block_groups, list) { | |
6645 | cache = caching_ctl->block_group; | |
6646 | if (block_group_cache_done(cache)) { | |
6647 | cache->last_byte_to_unpin = (u64)-1; | |
6648 | list_del_init(&caching_ctl->list); | |
6649 | put_caching_control(caching_ctl); | |
e8569813 | 6650 | } else { |
11833d66 | 6651 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6652 | } |
e8569813 | 6653 | } |
11833d66 YZ |
6654 | |
6655 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6656 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6657 | else | |
6658 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6659 | ||
9e351cc8 | 6660 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6661 | |
6662 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6663 | } |
6664 | ||
c759c4e1 JB |
6665 | /* |
6666 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6667 | * what it should be based on the mount options. | |
6668 | */ | |
6669 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
6670 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
6671 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
6672 | { |
6673 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
6674 | |
6675 | *empty_cluster = 0; | |
6676 | if (btrfs_mixed_space_info(space_info)) | |
6677 | return ret; | |
6678 | ||
c759c4e1 | 6679 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 6680 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
6681 | if (btrfs_test_opt(fs_info, SSD)) |
6682 | *empty_cluster = SZ_2M; | |
6683 | else | |
ee22184b | 6684 | *empty_cluster = SZ_64K; |
583b7231 HK |
6685 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
6686 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
6687 | *empty_cluster = SZ_2M; | |
0b246afa | 6688 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
6689 | } |
6690 | ||
6691 | return ret; | |
6692 | } | |
6693 | ||
2ff7e61e JM |
6694 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
6695 | u64 start, u64 end, | |
678886bd | 6696 | const bool return_free_space) |
ccd467d6 | 6697 | { |
11833d66 | 6698 | struct btrfs_block_group_cache *cache = NULL; |
7b398f8e JB |
6699 | struct btrfs_space_info *space_info; |
6700 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6701 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6702 | u64 len; |
c759c4e1 JB |
6703 | u64 total_unpinned = 0; |
6704 | u64 empty_cluster = 0; | |
7b398f8e | 6705 | bool readonly; |
ccd467d6 | 6706 | |
11833d66 | 6707 | while (start <= end) { |
7b398f8e | 6708 | readonly = false; |
11833d66 YZ |
6709 | if (!cache || |
6710 | start >= cache->key.objectid + cache->key.offset) { | |
6711 | if (cache) | |
6712 | btrfs_put_block_group(cache); | |
c759c4e1 | 6713 | total_unpinned = 0; |
11833d66 | 6714 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6715 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 6716 | |
2ff7e61e | 6717 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
6718 | cache->space_info, |
6719 | &empty_cluster); | |
6720 | empty_cluster <<= 1; | |
11833d66 YZ |
6721 | } |
6722 | ||
6723 | len = cache->key.objectid + cache->key.offset - start; | |
6724 | len = min(len, end + 1 - start); | |
6725 | ||
6726 | if (start < cache->last_byte_to_unpin) { | |
6727 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6728 | if (return_free_space) |
6729 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6730 | } |
6731 | ||
f0486c68 | 6732 | start += len; |
c759c4e1 | 6733 | total_unpinned += len; |
7b398f8e | 6734 | space_info = cache->space_info; |
f0486c68 | 6735 | |
c759c4e1 JB |
6736 | /* |
6737 | * If this space cluster has been marked as fragmented and we've | |
6738 | * unpinned enough in this block group to potentially allow a | |
6739 | * cluster to be created inside of it go ahead and clear the | |
6740 | * fragmented check. | |
6741 | */ | |
6742 | if (cluster && cluster->fragmented && | |
6743 | total_unpinned > empty_cluster) { | |
6744 | spin_lock(&cluster->lock); | |
6745 | cluster->fragmented = 0; | |
6746 | spin_unlock(&cluster->lock); | |
6747 | } | |
6748 | ||
7b398f8e | 6749 | spin_lock(&space_info->lock); |
11833d66 YZ |
6750 | spin_lock(&cache->lock); |
6751 | cache->pinned -= len; | |
7b398f8e | 6752 | space_info->bytes_pinned -= len; |
c51e7bb1 JB |
6753 | |
6754 | trace_btrfs_space_reservation(fs_info, "pinned", | |
6755 | space_info->flags, len, 0); | |
4f4db217 | 6756 | space_info->max_extent_size = 0; |
d288db5d | 6757 | percpu_counter_add(&space_info->total_bytes_pinned, -len); |
7b398f8e JB |
6758 | if (cache->ro) { |
6759 | space_info->bytes_readonly += len; | |
6760 | readonly = true; | |
6761 | } | |
11833d66 | 6762 | spin_unlock(&cache->lock); |
957780eb JB |
6763 | if (!readonly && return_free_space && |
6764 | global_rsv->space_info == space_info) { | |
6765 | u64 to_add = len; | |
92ac58ec | 6766 | |
7b398f8e JB |
6767 | spin_lock(&global_rsv->lock); |
6768 | if (!global_rsv->full) { | |
957780eb JB |
6769 | to_add = min(len, global_rsv->size - |
6770 | global_rsv->reserved); | |
6771 | global_rsv->reserved += to_add; | |
6772 | space_info->bytes_may_use += to_add; | |
7b398f8e JB |
6773 | if (global_rsv->reserved >= global_rsv->size) |
6774 | global_rsv->full = 1; | |
957780eb JB |
6775 | trace_btrfs_space_reservation(fs_info, |
6776 | "space_info", | |
6777 | space_info->flags, | |
6778 | to_add, 1); | |
6779 | len -= to_add; | |
7b398f8e JB |
6780 | } |
6781 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
6782 | /* Add to any tickets we may have */ |
6783 | if (len) | |
6784 | space_info_add_new_bytes(fs_info, space_info, | |
6785 | len); | |
7b398f8e JB |
6786 | } |
6787 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6788 | } |
11833d66 YZ |
6789 | |
6790 | if (cache) | |
6791 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6792 | return 0; |
6793 | } | |
6794 | ||
6795 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, | |
2ff7e61e | 6796 | struct btrfs_fs_info *fs_info) |
a28ec197 | 6797 | { |
e33e17ee JM |
6798 | struct btrfs_block_group_cache *block_group, *tmp; |
6799 | struct list_head *deleted_bgs; | |
11833d66 | 6800 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6801 | u64 start; |
6802 | u64 end; | |
a28ec197 | 6803 | int ret; |
a28ec197 | 6804 | |
11833d66 YZ |
6805 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6806 | unpin = &fs_info->freed_extents[1]; | |
6807 | else | |
6808 | unpin = &fs_info->freed_extents[0]; | |
6809 | ||
e33e17ee | 6810 | while (!trans->aborted) { |
d4b450cd | 6811 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6812 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 6813 | EXTENT_DIRTY, NULL); |
d4b450cd FM |
6814 | if (ret) { |
6815 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6816 | break; |
d4b450cd | 6817 | } |
1f3c79a2 | 6818 | |
0b246afa | 6819 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 6820 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 6821 | end + 1 - start, NULL); |
1f3c79a2 | 6822 | |
af6f8f60 | 6823 | clear_extent_dirty(unpin, start, end); |
2ff7e61e | 6824 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 6825 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
b9473439 | 6826 | cond_resched(); |
a28ec197 | 6827 | } |
817d52f8 | 6828 | |
e33e17ee JM |
6829 | /* |
6830 | * Transaction is finished. We don't need the lock anymore. We | |
6831 | * do need to clean up the block groups in case of a transaction | |
6832 | * abort. | |
6833 | */ | |
6834 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6835 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6836 | u64 trimmed = 0; | |
6837 | ||
6838 | ret = -EROFS; | |
6839 | if (!trans->aborted) | |
2ff7e61e | 6840 | ret = btrfs_discard_extent(fs_info, |
e33e17ee JM |
6841 | block_group->key.objectid, |
6842 | block_group->key.offset, | |
6843 | &trimmed); | |
6844 | ||
6845 | list_del_init(&block_group->bg_list); | |
6846 | btrfs_put_block_group_trimming(block_group); | |
6847 | btrfs_put_block_group(block_group); | |
6848 | ||
6849 | if (ret) { | |
6850 | const char *errstr = btrfs_decode_error(ret); | |
6851 | btrfs_warn(fs_info, | |
913e1535 | 6852 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
6853 | ret, errstr); |
6854 | } | |
6855 | } | |
6856 | ||
e20d96d6 CM |
6857 | return 0; |
6858 | } | |
6859 | ||
5d4f98a2 | 6860 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 6861 | struct btrfs_fs_info *info, |
c682f9b3 | 6862 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
6863 | u64 root_objectid, u64 owner_objectid, |
6864 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 6865 | struct btrfs_delayed_extent_op *extent_op) |
a28ec197 | 6866 | { |
e2fa7227 | 6867 | struct btrfs_key key; |
5d4f98a2 | 6868 | struct btrfs_path *path; |
1261ec42 | 6869 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 6870 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6871 | struct btrfs_extent_item *ei; |
6872 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6873 | int ret; |
5d4f98a2 | 6874 | int is_data; |
952fccac CM |
6875 | int extent_slot = 0; |
6876 | int found_extent = 0; | |
6877 | int num_to_del = 1; | |
5d4f98a2 YZ |
6878 | u32 item_size; |
6879 | u64 refs; | |
c682f9b3 QW |
6880 | u64 bytenr = node->bytenr; |
6881 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6882 | int last_ref = 0; |
0b246afa | 6883 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 6884 | |
5caf2a00 | 6885 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6886 | if (!path) |
6887 | return -ENOMEM; | |
5f26f772 | 6888 | |
e4058b54 | 6889 | path->reada = READA_FORWARD; |
b9473439 | 6890 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6891 | |
6892 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6893 | BUG_ON(!is_data && refs_to_drop != 1); | |
6894 | ||
3173a18f | 6895 | if (is_data) |
897ca819 | 6896 | skinny_metadata = false; |
3173a18f | 6897 | |
87bde3cd | 6898 | ret = lookup_extent_backref(trans, info, path, &iref, |
5d4f98a2 YZ |
6899 | bytenr, num_bytes, parent, |
6900 | root_objectid, owner_objectid, | |
6901 | owner_offset); | |
7bb86316 | 6902 | if (ret == 0) { |
952fccac | 6903 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6904 | while (extent_slot >= 0) { |
6905 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6906 | extent_slot); |
5d4f98a2 | 6907 | if (key.objectid != bytenr) |
952fccac | 6908 | break; |
5d4f98a2 YZ |
6909 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6910 | key.offset == num_bytes) { | |
952fccac CM |
6911 | found_extent = 1; |
6912 | break; | |
6913 | } | |
3173a18f JB |
6914 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
6915 | key.offset == owner_objectid) { | |
6916 | found_extent = 1; | |
6917 | break; | |
6918 | } | |
952fccac CM |
6919 | if (path->slots[0] - extent_slot > 5) |
6920 | break; | |
5d4f98a2 | 6921 | extent_slot--; |
952fccac | 6922 | } |
5d4f98a2 YZ |
6923 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
6924 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
6925 | if (found_extent && item_size < sizeof(*ei)) | |
6926 | found_extent = 0; | |
6927 | #endif | |
31840ae1 | 6928 | if (!found_extent) { |
5d4f98a2 | 6929 | BUG_ON(iref); |
87bde3cd JM |
6930 | ret = remove_extent_backref(trans, info, path, NULL, |
6931 | refs_to_drop, | |
fcebe456 | 6932 | is_data, &last_ref); |
005d6427 | 6933 | if (ret) { |
66642832 | 6934 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6935 | goto out; |
6936 | } | |
b3b4aa74 | 6937 | btrfs_release_path(path); |
b9473439 | 6938 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6939 | |
6940 | key.objectid = bytenr; | |
6941 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6942 | key.offset = num_bytes; | |
6943 | ||
3173a18f JB |
6944 | if (!is_data && skinny_metadata) { |
6945 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6946 | key.offset = owner_objectid; | |
6947 | } | |
6948 | ||
31840ae1 ZY |
6949 | ret = btrfs_search_slot(trans, extent_root, |
6950 | &key, path, -1, 1); | |
3173a18f JB |
6951 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
6952 | /* | |
6953 | * Couldn't find our skinny metadata item, | |
6954 | * see if we have ye olde extent item. | |
6955 | */ | |
6956 | path->slots[0]--; | |
6957 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6958 | path->slots[0]); | |
6959 | if (key.objectid == bytenr && | |
6960 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6961 | key.offset == num_bytes) | |
6962 | ret = 0; | |
6963 | } | |
6964 | ||
6965 | if (ret > 0 && skinny_metadata) { | |
6966 | skinny_metadata = false; | |
9ce49a0b | 6967 | key.objectid = bytenr; |
3173a18f JB |
6968 | key.type = BTRFS_EXTENT_ITEM_KEY; |
6969 | key.offset = num_bytes; | |
6970 | btrfs_release_path(path); | |
6971 | ret = btrfs_search_slot(trans, extent_root, | |
6972 | &key, path, -1, 1); | |
6973 | } | |
6974 | ||
f3465ca4 | 6975 | if (ret) { |
5d163e0e JM |
6976 | btrfs_err(info, |
6977 | "umm, got %d back from search, was looking for %llu", | |
6978 | ret, bytenr); | |
b783e62d | 6979 | if (ret > 0) |
a4f78750 | 6980 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 6981 | } |
005d6427 | 6982 | if (ret < 0) { |
66642832 | 6983 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6984 | goto out; |
6985 | } | |
31840ae1 ZY |
6986 | extent_slot = path->slots[0]; |
6987 | } | |
fae7f21c | 6988 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 6989 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
6990 | btrfs_err(info, |
6991 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
6992 | bytenr, parent, root_objectid, owner_objectid, |
6993 | owner_offset); | |
66642832 | 6994 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 6995 | goto out; |
79787eaa | 6996 | } else { |
66642832 | 6997 | btrfs_abort_transaction(trans, ret); |
005d6427 | 6998 | goto out; |
7bb86316 | 6999 | } |
5f39d397 CM |
7000 | |
7001 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
7002 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
7003 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
7004 | if (item_size < sizeof(*ei)) { | |
7005 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
87bde3cd JM |
7006 | ret = convert_extent_item_v0(trans, info, path, owner_objectid, |
7007 | 0); | |
005d6427 | 7008 | if (ret < 0) { |
66642832 | 7009 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7010 | goto out; |
7011 | } | |
5d4f98a2 | 7012 | |
b3b4aa74 | 7013 | btrfs_release_path(path); |
5d4f98a2 YZ |
7014 | path->leave_spinning = 1; |
7015 | ||
7016 | key.objectid = bytenr; | |
7017 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
7018 | key.offset = num_bytes; | |
7019 | ||
7020 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
7021 | -1, 1); | |
7022 | if (ret) { | |
5d163e0e JM |
7023 | btrfs_err(info, |
7024 | "umm, got %d back from search, was looking for %llu", | |
c1c9ff7c | 7025 | ret, bytenr); |
a4f78750 | 7026 | btrfs_print_leaf(path->nodes[0]); |
5d4f98a2 | 7027 | } |
005d6427 | 7028 | if (ret < 0) { |
66642832 | 7029 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7030 | goto out; |
7031 | } | |
7032 | ||
5d4f98a2 YZ |
7033 | extent_slot = path->slots[0]; |
7034 | leaf = path->nodes[0]; | |
7035 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
7036 | } | |
7037 | #endif | |
7038 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 7039 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 7040 | struct btrfs_extent_item); |
3173a18f JB |
7041 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
7042 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
7043 | struct btrfs_tree_block_info *bi; |
7044 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
7045 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
7046 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
7047 | } | |
56bec294 | 7048 | |
5d4f98a2 | 7049 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 7050 | if (refs < refs_to_drop) { |
5d163e0e JM |
7051 | btrfs_err(info, |
7052 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
7053 | refs_to_drop, refs, bytenr); | |
32b02538 | 7054 | ret = -EINVAL; |
66642832 | 7055 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
7056 | goto out; |
7057 | } | |
56bec294 | 7058 | refs -= refs_to_drop; |
5f39d397 | 7059 | |
5d4f98a2 YZ |
7060 | if (refs > 0) { |
7061 | if (extent_op) | |
7062 | __run_delayed_extent_op(extent_op, leaf, ei); | |
7063 | /* | |
7064 | * In the case of inline back ref, reference count will | |
7065 | * be updated by remove_extent_backref | |
952fccac | 7066 | */ |
5d4f98a2 YZ |
7067 | if (iref) { |
7068 | BUG_ON(!found_extent); | |
7069 | } else { | |
7070 | btrfs_set_extent_refs(leaf, ei, refs); | |
7071 | btrfs_mark_buffer_dirty(leaf); | |
7072 | } | |
7073 | if (found_extent) { | |
87bde3cd | 7074 | ret = remove_extent_backref(trans, info, path, |
5d4f98a2 | 7075 | iref, refs_to_drop, |
fcebe456 | 7076 | is_data, &last_ref); |
005d6427 | 7077 | if (ret) { |
66642832 | 7078 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7079 | goto out; |
7080 | } | |
952fccac | 7081 | } |
5d4f98a2 | 7082 | } else { |
5d4f98a2 YZ |
7083 | if (found_extent) { |
7084 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 7085 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
7086 | if (iref) { |
7087 | BUG_ON(path->slots[0] != extent_slot); | |
7088 | } else { | |
7089 | BUG_ON(path->slots[0] != extent_slot + 1); | |
7090 | path->slots[0] = extent_slot; | |
7091 | num_to_del = 2; | |
7092 | } | |
78fae27e | 7093 | } |
b9473439 | 7094 | |
fcebe456 | 7095 | last_ref = 1; |
952fccac CM |
7096 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
7097 | num_to_del); | |
005d6427 | 7098 | if (ret) { |
66642832 | 7099 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7100 | goto out; |
7101 | } | |
b3b4aa74 | 7102 | btrfs_release_path(path); |
21af804c | 7103 | |
5d4f98a2 | 7104 | if (is_data) { |
5b4aacef | 7105 | ret = btrfs_del_csums(trans, info, bytenr, num_bytes); |
005d6427 | 7106 | if (ret) { |
66642832 | 7107 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7108 | goto out; |
7109 | } | |
459931ec CM |
7110 | } |
7111 | ||
0b246afa | 7112 | ret = add_to_free_space_tree(trans, info, bytenr, num_bytes); |
1e144fb8 | 7113 | if (ret) { |
66642832 | 7114 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
7115 | goto out; |
7116 | } | |
7117 | ||
0b246afa | 7118 | ret = update_block_group(trans, info, bytenr, num_bytes, 0); |
005d6427 | 7119 | if (ret) { |
66642832 | 7120 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7121 | goto out; |
7122 | } | |
a28ec197 | 7123 | } |
fcebe456 JB |
7124 | btrfs_release_path(path); |
7125 | ||
79787eaa | 7126 | out: |
5caf2a00 | 7127 | btrfs_free_path(path); |
a28ec197 CM |
7128 | return ret; |
7129 | } | |
7130 | ||
1887be66 | 7131 | /* |
f0486c68 | 7132 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
7133 | * delayed ref for that extent as well. This searches the delayed ref tree for |
7134 | * a given extent, and if there are no other delayed refs to be processed, it | |
7135 | * removes it from the tree. | |
7136 | */ | |
7137 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 7138 | u64 bytenr) |
1887be66 CM |
7139 | { |
7140 | struct btrfs_delayed_ref_head *head; | |
7141 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 7142 | int ret = 0; |
1887be66 CM |
7143 | |
7144 | delayed_refs = &trans->transaction->delayed_refs; | |
7145 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 7146 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 7147 | if (!head) |
cf93da7b | 7148 | goto out_delayed_unlock; |
1887be66 | 7149 | |
d7df2c79 | 7150 | spin_lock(&head->lock); |
0e0adbcf | 7151 | if (!RB_EMPTY_ROOT(&head->ref_tree)) |
1887be66 CM |
7152 | goto out; |
7153 | ||
5d4f98a2 YZ |
7154 | if (head->extent_op) { |
7155 | if (!head->must_insert_reserved) | |
7156 | goto out; | |
78a6184a | 7157 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
7158 | head->extent_op = NULL; |
7159 | } | |
7160 | ||
1887be66 CM |
7161 | /* |
7162 | * waiting for the lock here would deadlock. If someone else has it | |
7163 | * locked they are already in the process of dropping it anyway | |
7164 | */ | |
7165 | if (!mutex_trylock(&head->mutex)) | |
7166 | goto out; | |
7167 | ||
7168 | /* | |
7169 | * at this point we have a head with no other entries. Go | |
7170 | * ahead and process it. | |
7171 | */ | |
c46effa6 | 7172 | rb_erase(&head->href_node, &delayed_refs->href_root); |
d278850e | 7173 | RB_CLEAR_NODE(&head->href_node); |
d7df2c79 | 7174 | atomic_dec(&delayed_refs->num_entries); |
1887be66 CM |
7175 | |
7176 | /* | |
7177 | * we don't take a ref on the node because we're removing it from the | |
7178 | * tree, so we just steal the ref the tree was holding. | |
7179 | */ | |
c3e69d58 | 7180 | delayed_refs->num_heads--; |
d7df2c79 | 7181 | if (head->processing == 0) |
c3e69d58 | 7182 | delayed_refs->num_heads_ready--; |
d7df2c79 JB |
7183 | head->processing = 0; |
7184 | spin_unlock(&head->lock); | |
1887be66 CM |
7185 | spin_unlock(&delayed_refs->lock); |
7186 | ||
f0486c68 YZ |
7187 | BUG_ON(head->extent_op); |
7188 | if (head->must_insert_reserved) | |
7189 | ret = 1; | |
7190 | ||
7191 | mutex_unlock(&head->mutex); | |
d278850e | 7192 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 7193 | return ret; |
1887be66 | 7194 | out: |
d7df2c79 | 7195 | spin_unlock(&head->lock); |
cf93da7b CM |
7196 | |
7197 | out_delayed_unlock: | |
1887be66 CM |
7198 | spin_unlock(&delayed_refs->lock); |
7199 | return 0; | |
7200 | } | |
7201 | ||
f0486c68 YZ |
7202 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
7203 | struct btrfs_root *root, | |
7204 | struct extent_buffer *buf, | |
5581a51a | 7205 | u64 parent, int last_ref) |
f0486c68 | 7206 | { |
0b246afa | 7207 | struct btrfs_fs_info *fs_info = root->fs_info; |
b150a4f1 | 7208 | int pin = 1; |
f0486c68 YZ |
7209 | int ret; |
7210 | ||
7211 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
d7eae340 OS |
7212 | int old_ref_mod, new_ref_mod; |
7213 | ||
fd708b81 JB |
7214 | btrfs_ref_tree_mod(root, buf->start, buf->len, parent, |
7215 | root->root_key.objectid, | |
7216 | btrfs_header_level(buf), 0, | |
7217 | BTRFS_DROP_DELAYED_REF); | |
7be07912 OS |
7218 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start, |
7219 | buf->len, parent, | |
0b246afa JM |
7220 | root->root_key.objectid, |
7221 | btrfs_header_level(buf), | |
7be07912 | 7222 | BTRFS_DROP_DELAYED_REF, NULL, |
d7eae340 | 7223 | &old_ref_mod, &new_ref_mod); |
79787eaa | 7224 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 7225 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
7226 | } |
7227 | ||
0a16c7d7 | 7228 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
7229 | struct btrfs_block_group_cache *cache; |
7230 | ||
f0486c68 | 7231 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 7232 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 7233 | if (!ret) |
37be25bc | 7234 | goto out; |
f0486c68 YZ |
7235 | } |
7236 | ||
4da8b76d | 7237 | pin = 0; |
0b246afa | 7238 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 7239 | |
f0486c68 | 7240 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
2ff7e61e JM |
7241 | pin_down_extent(fs_info, cache, buf->start, |
7242 | buf->len, 1); | |
6219872d | 7243 | btrfs_put_block_group(cache); |
37be25bc | 7244 | goto out; |
f0486c68 YZ |
7245 | } |
7246 | ||
7247 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
7248 | ||
7249 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 7250 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 7251 | btrfs_put_block_group(cache); |
71ff6437 | 7252 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
7253 | } |
7254 | out: | |
b150a4f1 | 7255 | if (pin) |
0b246afa | 7256 | add_pinned_bytes(fs_info, buf->len, btrfs_header_level(buf), |
b150a4f1 JB |
7257 | root->root_key.objectid); |
7258 | ||
0a16c7d7 OS |
7259 | if (last_ref) { |
7260 | /* | |
7261 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
7262 | * matter anymore. | |
7263 | */ | |
7264 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
7265 | } | |
f0486c68 YZ |
7266 | } |
7267 | ||
79787eaa | 7268 | /* Can return -ENOMEM */ |
2ff7e61e | 7269 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
84f7d8e6 | 7270 | struct btrfs_root *root, |
66d7e7f0 | 7271 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, |
b06c4bf5 | 7272 | u64 owner, u64 offset) |
925baedd | 7273 | { |
84f7d8e6 | 7274 | struct btrfs_fs_info *fs_info = root->fs_info; |
d7eae340 | 7275 | int old_ref_mod, new_ref_mod; |
925baedd CM |
7276 | int ret; |
7277 | ||
f5ee5c9a | 7278 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 7279 | return 0; |
fccb84c9 | 7280 | |
fd708b81 JB |
7281 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) |
7282 | btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, | |
7283 | root_objectid, owner, offset, | |
7284 | BTRFS_DROP_DELAYED_REF); | |
7285 | ||
56bec294 CM |
7286 | /* |
7287 | * tree log blocks never actually go into the extent allocation | |
7288 | * tree, just update pinning info and exit early. | |
56bec294 | 7289 | */ |
5d4f98a2 YZ |
7290 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
7291 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 7292 | /* unlocks the pinned mutex */ |
2ff7e61e | 7293 | btrfs_pin_extent(fs_info, bytenr, num_bytes, 1); |
d7eae340 | 7294 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 7295 | ret = 0; |
5d4f98a2 | 7296 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 | 7297 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
7be07912 OS |
7298 | num_bytes, parent, |
7299 | root_objectid, (int)owner, | |
7300 | BTRFS_DROP_DELAYED_REF, NULL, | |
d7eae340 | 7301 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 7302 | } else { |
66d7e7f0 | 7303 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
7be07912 OS |
7304 | num_bytes, parent, |
7305 | root_objectid, owner, offset, | |
7306 | 0, BTRFS_DROP_DELAYED_REF, | |
d7eae340 | 7307 | &old_ref_mod, &new_ref_mod); |
56bec294 | 7308 | } |
d7eae340 OS |
7309 | |
7310 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) | |
7311 | add_pinned_bytes(fs_info, num_bytes, owner, root_objectid); | |
7312 | ||
925baedd CM |
7313 | return ret; |
7314 | } | |
7315 | ||
817d52f8 JB |
7316 | /* |
7317 | * when we wait for progress in the block group caching, its because | |
7318 | * our allocation attempt failed at least once. So, we must sleep | |
7319 | * and let some progress happen before we try again. | |
7320 | * | |
7321 | * This function will sleep at least once waiting for new free space to | |
7322 | * show up, and then it will check the block group free space numbers | |
7323 | * for our min num_bytes. Another option is to have it go ahead | |
7324 | * and look in the rbtree for a free extent of a given size, but this | |
7325 | * is a good start. | |
36cce922 JB |
7326 | * |
7327 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
7328 | * any of the information in this block group. | |
817d52f8 | 7329 | */ |
36cce922 | 7330 | static noinline void |
817d52f8 JB |
7331 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
7332 | u64 num_bytes) | |
7333 | { | |
11833d66 | 7334 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 7335 | |
11833d66 YZ |
7336 | caching_ctl = get_caching_control(cache); |
7337 | if (!caching_ctl) | |
36cce922 | 7338 | return; |
817d52f8 | 7339 | |
11833d66 | 7340 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 7341 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
7342 | |
7343 | put_caching_control(caching_ctl); | |
11833d66 YZ |
7344 | } |
7345 | ||
7346 | static noinline int | |
7347 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
7348 | { | |
7349 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 7350 | int ret = 0; |
11833d66 YZ |
7351 | |
7352 | caching_ctl = get_caching_control(cache); | |
7353 | if (!caching_ctl) | |
36cce922 | 7354 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
7355 | |
7356 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
7357 | if (cache->cached == BTRFS_CACHE_ERROR) |
7358 | ret = -EIO; | |
11833d66 | 7359 | put_caching_control(caching_ctl); |
36cce922 | 7360 | return ret; |
817d52f8 JB |
7361 | } |
7362 | ||
31e50229 | 7363 | int __get_raid_index(u64 flags) |
b742bb82 | 7364 | { |
7738a53a | 7365 | if (flags & BTRFS_BLOCK_GROUP_RAID10) |
e6ec716f | 7366 | return BTRFS_RAID_RAID10; |
7738a53a | 7367 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
e6ec716f | 7368 | return BTRFS_RAID_RAID1; |
7738a53a | 7369 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
e6ec716f | 7370 | return BTRFS_RAID_DUP; |
7738a53a | 7371 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
e6ec716f | 7372 | return BTRFS_RAID_RAID0; |
53b381b3 | 7373 | else if (flags & BTRFS_BLOCK_GROUP_RAID5) |
e942f883 | 7374 | return BTRFS_RAID_RAID5; |
53b381b3 | 7375 | else if (flags & BTRFS_BLOCK_GROUP_RAID6) |
e942f883 | 7376 | return BTRFS_RAID_RAID6; |
7738a53a | 7377 | |
e942f883 | 7378 | return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ |
b742bb82 YZ |
7379 | } |
7380 | ||
6ab0a202 | 7381 | int get_block_group_index(struct btrfs_block_group_cache *cache) |
7738a53a | 7382 | { |
31e50229 | 7383 | return __get_raid_index(cache->flags); |
7738a53a ID |
7384 | } |
7385 | ||
6ab0a202 JM |
7386 | static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = { |
7387 | [BTRFS_RAID_RAID10] = "raid10", | |
7388 | [BTRFS_RAID_RAID1] = "raid1", | |
7389 | [BTRFS_RAID_DUP] = "dup", | |
7390 | [BTRFS_RAID_RAID0] = "raid0", | |
7391 | [BTRFS_RAID_SINGLE] = "single", | |
7392 | [BTRFS_RAID_RAID5] = "raid5", | |
7393 | [BTRFS_RAID_RAID6] = "raid6", | |
7394 | }; | |
7395 | ||
1b8e5df6 | 7396 | static const char *get_raid_name(enum btrfs_raid_types type) |
6ab0a202 JM |
7397 | { |
7398 | if (type >= BTRFS_NR_RAID_TYPES) | |
7399 | return NULL; | |
7400 | ||
7401 | return btrfs_raid_type_names[type]; | |
7402 | } | |
7403 | ||
817d52f8 | 7404 | enum btrfs_loop_type { |
285ff5af JB |
7405 | LOOP_CACHING_NOWAIT = 0, |
7406 | LOOP_CACHING_WAIT = 1, | |
7407 | LOOP_ALLOC_CHUNK = 2, | |
7408 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
7409 | }; |
7410 | ||
e570fd27 MX |
7411 | static inline void |
7412 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
7413 | int delalloc) | |
7414 | { | |
7415 | if (delalloc) | |
7416 | down_read(&cache->data_rwsem); | |
7417 | } | |
7418 | ||
7419 | static inline void | |
7420 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
7421 | int delalloc) | |
7422 | { | |
7423 | btrfs_get_block_group(cache); | |
7424 | if (delalloc) | |
7425 | down_read(&cache->data_rwsem); | |
7426 | } | |
7427 | ||
7428 | static struct btrfs_block_group_cache * | |
7429 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
7430 | struct btrfs_free_cluster *cluster, | |
7431 | int delalloc) | |
7432 | { | |
89771cc9 | 7433 | struct btrfs_block_group_cache *used_bg = NULL; |
6719afdc | 7434 | |
e570fd27 | 7435 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
7436 | while (1) { |
7437 | used_bg = cluster->block_group; | |
7438 | if (!used_bg) | |
7439 | return NULL; | |
7440 | ||
7441 | if (used_bg == block_group) | |
e570fd27 MX |
7442 | return used_bg; |
7443 | ||
6719afdc | 7444 | btrfs_get_block_group(used_bg); |
e570fd27 | 7445 | |
6719afdc GU |
7446 | if (!delalloc) |
7447 | return used_bg; | |
e570fd27 | 7448 | |
6719afdc GU |
7449 | if (down_read_trylock(&used_bg->data_rwsem)) |
7450 | return used_bg; | |
e570fd27 | 7451 | |
6719afdc | 7452 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 7453 | |
e321f8a8 LB |
7454 | /* We should only have one-level nested. */ |
7455 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 7456 | |
6719afdc GU |
7457 | spin_lock(&cluster->refill_lock); |
7458 | if (used_bg == cluster->block_group) | |
7459 | return used_bg; | |
e570fd27 | 7460 | |
6719afdc GU |
7461 | up_read(&used_bg->data_rwsem); |
7462 | btrfs_put_block_group(used_bg); | |
7463 | } | |
e570fd27 MX |
7464 | } |
7465 | ||
7466 | static inline void | |
7467 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7468 | int delalloc) | |
7469 | { | |
7470 | if (delalloc) | |
7471 | up_read(&cache->data_rwsem); | |
7472 | btrfs_put_block_group(cache); | |
7473 | } | |
7474 | ||
fec577fb CM |
7475 | /* |
7476 | * walks the btree of allocated extents and find a hole of a given size. | |
7477 | * The key ins is changed to record the hole: | |
a4820398 | 7478 | * ins->objectid == start position |
62e2749e | 7479 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7480 | * ins->offset == the size of the hole. |
fec577fb | 7481 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7482 | * |
7483 | * If there is no suitable free space, we will record the max size of | |
7484 | * the free space extent currently. | |
fec577fb | 7485 | */ |
87bde3cd | 7486 | static noinline int find_free_extent(struct btrfs_fs_info *fs_info, |
18513091 WX |
7487 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
7488 | u64 hint_byte, struct btrfs_key *ins, | |
7489 | u64 flags, int delalloc) | |
fec577fb | 7490 | { |
80eb234a | 7491 | int ret = 0; |
0b246afa | 7492 | struct btrfs_root *root = fs_info->extent_root; |
fa9c0d79 | 7493 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7494 | struct btrfs_block_group_cache *block_group = NULL; |
81c9ad23 | 7495 | u64 search_start = 0; |
a4820398 | 7496 | u64 max_extent_size = 0; |
c759c4e1 | 7497 | u64 empty_cluster = 0; |
80eb234a | 7498 | struct btrfs_space_info *space_info; |
fa9c0d79 | 7499 | int loop = 0; |
b6919a58 | 7500 | int index = __get_raid_index(flags); |
0a24325e | 7501 | bool failed_cluster_refill = false; |
1cdda9b8 | 7502 | bool failed_alloc = false; |
67377734 | 7503 | bool use_cluster = true; |
60d2adbb | 7504 | bool have_caching_bg = false; |
13a0db5a | 7505 | bool orig_have_caching_bg = false; |
a5e681d9 | 7506 | bool full_search = false; |
fec577fb | 7507 | |
0b246afa | 7508 | WARN_ON(num_bytes < fs_info->sectorsize); |
962a298f | 7509 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7510 | ins->objectid = 0; |
7511 | ins->offset = 0; | |
b1a4d965 | 7512 | |
71ff6437 | 7513 | trace_find_free_extent(fs_info, num_bytes, empty_size, flags); |
3f7de037 | 7514 | |
0b246afa | 7515 | space_info = __find_space_info(fs_info, flags); |
1b1d1f66 | 7516 | if (!space_info) { |
0b246afa | 7517 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7518 | return -ENOSPC; |
7519 | } | |
2552d17e | 7520 | |
67377734 | 7521 | /* |
4f4db217 JB |
7522 | * If our free space is heavily fragmented we may not be able to make |
7523 | * big contiguous allocations, so instead of doing the expensive search | |
7524 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7525 | * can go ahead and search for a more manageable chunk. | |
7526 | * | |
7527 | * If our max_extent_size is large enough for our allocation simply | |
7528 | * disable clustering since we will likely not be able to find enough | |
7529 | * space to create a cluster and induce latency trying. | |
67377734 | 7530 | */ |
4f4db217 JB |
7531 | if (unlikely(space_info->max_extent_size)) { |
7532 | spin_lock(&space_info->lock); | |
7533 | if (space_info->max_extent_size && | |
7534 | num_bytes > space_info->max_extent_size) { | |
7535 | ins->offset = space_info->max_extent_size; | |
7536 | spin_unlock(&space_info->lock); | |
7537 | return -ENOSPC; | |
7538 | } else if (space_info->max_extent_size) { | |
7539 | use_cluster = false; | |
7540 | } | |
7541 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7542 | } |
0f9dd46c | 7543 | |
2ff7e61e | 7544 | last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster); |
239b14b3 | 7545 | if (last_ptr) { |
fa9c0d79 CM |
7546 | spin_lock(&last_ptr->lock); |
7547 | if (last_ptr->block_group) | |
7548 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7549 | if (last_ptr->fragmented) { |
7550 | /* | |
7551 | * We still set window_start so we can keep track of the | |
7552 | * last place we found an allocation to try and save | |
7553 | * some time. | |
7554 | */ | |
7555 | hint_byte = last_ptr->window_start; | |
7556 | use_cluster = false; | |
7557 | } | |
fa9c0d79 | 7558 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7559 | } |
fa9c0d79 | 7560 | |
2ff7e61e | 7561 | search_start = max(search_start, first_logical_byte(fs_info, 0)); |
239b14b3 | 7562 | search_start = max(search_start, hint_byte); |
2552d17e | 7563 | if (search_start == hint_byte) { |
0b246afa | 7564 | block_group = btrfs_lookup_block_group(fs_info, search_start); |
817d52f8 JB |
7565 | /* |
7566 | * we don't want to use the block group if it doesn't match our | |
7567 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7568 | * |
7569 | * However if we are re-searching with an ideal block group | |
7570 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7571 | */ |
b6919a58 | 7572 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7573 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7574 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7575 | if (list_empty(&block_group->list) || |
7576 | block_group->ro) { | |
7577 | /* | |
7578 | * someone is removing this block group, | |
7579 | * we can't jump into the have_block_group | |
7580 | * target because our list pointers are not | |
7581 | * valid | |
7582 | */ | |
7583 | btrfs_put_block_group(block_group); | |
7584 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7585 | } else { |
b742bb82 | 7586 | index = get_block_group_index(block_group); |
e570fd27 | 7587 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7588 | goto have_block_group; |
ccf0e725 | 7589 | } |
2552d17e | 7590 | } else if (block_group) { |
fa9c0d79 | 7591 | btrfs_put_block_group(block_group); |
2552d17e | 7592 | } |
42e70e7a | 7593 | } |
2552d17e | 7594 | search: |
60d2adbb | 7595 | have_caching_bg = false; |
a5e681d9 JB |
7596 | if (index == 0 || index == __get_raid_index(flags)) |
7597 | full_search = true; | |
80eb234a | 7598 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
7599 | list_for_each_entry(block_group, &space_info->block_groups[index], |
7600 | list) { | |
6226cb0a | 7601 | u64 offset; |
817d52f8 | 7602 | int cached; |
8a1413a2 | 7603 | |
14443937 JM |
7604 | /* If the block group is read-only, we can skip it entirely. */ |
7605 | if (unlikely(block_group->ro)) | |
7606 | continue; | |
7607 | ||
e570fd27 | 7608 | btrfs_grab_block_group(block_group, delalloc); |
2552d17e | 7609 | search_start = block_group->key.objectid; |
42e70e7a | 7610 | |
83a50de9 CM |
7611 | /* |
7612 | * this can happen if we end up cycling through all the | |
7613 | * raid types, but we want to make sure we only allocate | |
7614 | * for the proper type. | |
7615 | */ | |
b6919a58 | 7616 | if (!block_group_bits(block_group, flags)) { |
83a50de9 CM |
7617 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
7618 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
7619 | BTRFS_BLOCK_GROUP_RAID5 | |
7620 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
7621 | BTRFS_BLOCK_GROUP_RAID10; |
7622 | ||
7623 | /* | |
7624 | * if they asked for extra copies and this block group | |
7625 | * doesn't provide them, bail. This does allow us to | |
7626 | * fill raid0 from raid1. | |
7627 | */ | |
b6919a58 | 7628 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
7629 | goto loop; |
7630 | } | |
7631 | ||
2552d17e | 7632 | have_block_group: |
291c7d2f JB |
7633 | cached = block_group_cache_done(block_group); |
7634 | if (unlikely(!cached)) { | |
a5e681d9 | 7635 | have_caching_bg = true; |
f6373bf3 | 7636 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7637 | BUG_ON(ret < 0); |
7638 | ret = 0; | |
817d52f8 JB |
7639 | } |
7640 | ||
36cce922 JB |
7641 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7642 | goto loop; | |
0f9dd46c | 7643 | |
0a24325e | 7644 | /* |
062c05c4 AO |
7645 | * Ok we want to try and use the cluster allocator, so |
7646 | * lets look there | |
0a24325e | 7647 | */ |
c759c4e1 | 7648 | if (last_ptr && use_cluster) { |
215a63d1 | 7649 | struct btrfs_block_group_cache *used_block_group; |
8de972b4 | 7650 | unsigned long aligned_cluster; |
fa9c0d79 CM |
7651 | /* |
7652 | * the refill lock keeps out other | |
7653 | * people trying to start a new cluster | |
7654 | */ | |
e570fd27 MX |
7655 | used_block_group = btrfs_lock_cluster(block_group, |
7656 | last_ptr, | |
7657 | delalloc); | |
7658 | if (!used_block_group) | |
44fb5511 | 7659 | goto refill_cluster; |
274bd4fb | 7660 | |
e570fd27 MX |
7661 | if (used_block_group != block_group && |
7662 | (used_block_group->ro || | |
7663 | !block_group_bits(used_block_group, flags))) | |
7664 | goto release_cluster; | |
44fb5511 | 7665 | |
274bd4fb | 7666 | offset = btrfs_alloc_from_cluster(used_block_group, |
a4820398 MX |
7667 | last_ptr, |
7668 | num_bytes, | |
7669 | used_block_group->key.objectid, | |
7670 | &max_extent_size); | |
fa9c0d79 CM |
7671 | if (offset) { |
7672 | /* we have a block, we're done */ | |
7673 | spin_unlock(&last_ptr->refill_lock); | |
71ff6437 | 7674 | trace_btrfs_reserve_extent_cluster(fs_info, |
89d4346a MX |
7675 | used_block_group, |
7676 | search_start, num_bytes); | |
215a63d1 | 7677 | if (used_block_group != block_group) { |
e570fd27 MX |
7678 | btrfs_release_block_group(block_group, |
7679 | delalloc); | |
215a63d1 MX |
7680 | block_group = used_block_group; |
7681 | } | |
fa9c0d79 CM |
7682 | goto checks; |
7683 | } | |
7684 | ||
274bd4fb | 7685 | WARN_ON(last_ptr->block_group != used_block_group); |
e570fd27 | 7686 | release_cluster: |
062c05c4 AO |
7687 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
7688 | * set up a new clusters, so lets just skip it | |
7689 | * and let the allocator find whatever block | |
7690 | * it can find. If we reach this point, we | |
7691 | * will have tried the cluster allocator | |
7692 | * plenty of times and not have found | |
7693 | * anything, so we are likely way too | |
7694 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
7695 | * anything. |
7696 | * | |
7697 | * However, if the cluster is taken from the | |
7698 | * current block group, release the cluster | |
7699 | * first, so that we stand a better chance of | |
7700 | * succeeding in the unclustered | |
7701 | * allocation. */ | |
7702 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
e570fd27 | 7703 | used_block_group != block_group) { |
062c05c4 | 7704 | spin_unlock(&last_ptr->refill_lock); |
e570fd27 MX |
7705 | btrfs_release_block_group(used_block_group, |
7706 | delalloc); | |
062c05c4 AO |
7707 | goto unclustered_alloc; |
7708 | } | |
7709 | ||
fa9c0d79 CM |
7710 | /* |
7711 | * this cluster didn't work out, free it and | |
7712 | * start over | |
7713 | */ | |
7714 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7715 | ||
e570fd27 MX |
7716 | if (used_block_group != block_group) |
7717 | btrfs_release_block_group(used_block_group, | |
7718 | delalloc); | |
7719 | refill_cluster: | |
a5f6f719 AO |
7720 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
7721 | spin_unlock(&last_ptr->refill_lock); | |
7722 | goto unclustered_alloc; | |
7723 | } | |
7724 | ||
8de972b4 CM |
7725 | aligned_cluster = max_t(unsigned long, |
7726 | empty_cluster + empty_size, | |
7727 | block_group->full_stripe_len); | |
7728 | ||
fa9c0d79 | 7729 | /* allocate a cluster in this block group */ |
2ff7e61e | 7730 | ret = btrfs_find_space_cluster(fs_info, block_group, |
00361589 JB |
7731 | last_ptr, search_start, |
7732 | num_bytes, | |
7733 | aligned_cluster); | |
fa9c0d79 CM |
7734 | if (ret == 0) { |
7735 | /* | |
7736 | * now pull our allocation out of this | |
7737 | * cluster | |
7738 | */ | |
7739 | offset = btrfs_alloc_from_cluster(block_group, | |
a4820398 MX |
7740 | last_ptr, |
7741 | num_bytes, | |
7742 | search_start, | |
7743 | &max_extent_size); | |
fa9c0d79 CM |
7744 | if (offset) { |
7745 | /* we found one, proceed */ | |
7746 | spin_unlock(&last_ptr->refill_lock); | |
71ff6437 | 7747 | trace_btrfs_reserve_extent_cluster(fs_info, |
3f7de037 JB |
7748 | block_group, search_start, |
7749 | num_bytes); | |
fa9c0d79 CM |
7750 | goto checks; |
7751 | } | |
0a24325e JB |
7752 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
7753 | && !failed_cluster_refill) { | |
817d52f8 JB |
7754 | spin_unlock(&last_ptr->refill_lock); |
7755 | ||
0a24325e | 7756 | failed_cluster_refill = true; |
817d52f8 JB |
7757 | wait_block_group_cache_progress(block_group, |
7758 | num_bytes + empty_cluster + empty_size); | |
7759 | goto have_block_group; | |
fa9c0d79 | 7760 | } |
817d52f8 | 7761 | |
fa9c0d79 CM |
7762 | /* |
7763 | * at this point we either didn't find a cluster | |
7764 | * or we weren't able to allocate a block from our | |
7765 | * cluster. Free the cluster we've been trying | |
7766 | * to use, and go to the next block group | |
7767 | */ | |
0a24325e | 7768 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 7769 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 7770 | goto loop; |
fa9c0d79 CM |
7771 | } |
7772 | ||
062c05c4 | 7773 | unclustered_alloc: |
c759c4e1 JB |
7774 | /* |
7775 | * We are doing an unclustered alloc, set the fragmented flag so | |
7776 | * we don't bother trying to setup a cluster again until we get | |
7777 | * more space. | |
7778 | */ | |
7779 | if (unlikely(last_ptr)) { | |
7780 | spin_lock(&last_ptr->lock); | |
7781 | last_ptr->fragmented = 1; | |
7782 | spin_unlock(&last_ptr->lock); | |
7783 | } | |
0c9b36e0 LB |
7784 | if (cached) { |
7785 | struct btrfs_free_space_ctl *ctl = | |
7786 | block_group->free_space_ctl; | |
7787 | ||
7788 | spin_lock(&ctl->tree_lock); | |
7789 | if (ctl->free_space < | |
7790 | num_bytes + empty_cluster + empty_size) { | |
7791 | if (ctl->free_space > max_extent_size) | |
7792 | max_extent_size = ctl->free_space; | |
7793 | spin_unlock(&ctl->tree_lock); | |
7794 | goto loop; | |
7795 | } | |
7796 | spin_unlock(&ctl->tree_lock); | |
a5f6f719 | 7797 | } |
a5f6f719 | 7798 | |
6226cb0a | 7799 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
a4820398 MX |
7800 | num_bytes, empty_size, |
7801 | &max_extent_size); | |
1cdda9b8 JB |
7802 | /* |
7803 | * If we didn't find a chunk, and we haven't failed on this | |
7804 | * block group before, and this block group is in the middle of | |
7805 | * caching and we are ok with waiting, then go ahead and wait | |
7806 | * for progress to be made, and set failed_alloc to true. | |
7807 | * | |
7808 | * If failed_alloc is true then we've already waited on this | |
7809 | * block group once and should move on to the next block group. | |
7810 | */ | |
7811 | if (!offset && !failed_alloc && !cached && | |
7812 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 7813 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
7814 | num_bytes + empty_size); |
7815 | failed_alloc = true; | |
817d52f8 | 7816 | goto have_block_group; |
1cdda9b8 JB |
7817 | } else if (!offset) { |
7818 | goto loop; | |
817d52f8 | 7819 | } |
fa9c0d79 | 7820 | checks: |
0b246afa | 7821 | search_start = ALIGN(offset, fs_info->stripesize); |
25179201 | 7822 | |
2552d17e JB |
7823 | /* move on to the next group */ |
7824 | if (search_start + num_bytes > | |
215a63d1 MX |
7825 | block_group->key.objectid + block_group->key.offset) { |
7826 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 7827 | goto loop; |
6226cb0a | 7828 | } |
f5a31e16 | 7829 | |
f0486c68 | 7830 | if (offset < search_start) |
215a63d1 | 7831 | btrfs_add_free_space(block_group, offset, |
f0486c68 YZ |
7832 | search_start - offset); |
7833 | BUG_ON(offset > search_start); | |
2552d17e | 7834 | |
18513091 WX |
7835 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
7836 | num_bytes, delalloc); | |
f0486c68 | 7837 | if (ret == -EAGAIN) { |
215a63d1 | 7838 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 7839 | goto loop; |
0f9dd46c | 7840 | } |
9cfa3e34 | 7841 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 7842 | |
f0486c68 | 7843 | /* we are all good, lets return */ |
2552d17e JB |
7844 | ins->objectid = search_start; |
7845 | ins->offset = num_bytes; | |
d2fb3437 | 7846 | |
71ff6437 | 7847 | trace_btrfs_reserve_extent(fs_info, block_group, |
3f7de037 | 7848 | search_start, num_bytes); |
e570fd27 | 7849 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7850 | break; |
7851 | loop: | |
0a24325e | 7852 | failed_cluster_refill = false; |
1cdda9b8 | 7853 | failed_alloc = false; |
b742bb82 | 7854 | BUG_ON(index != get_block_group_index(block_group)); |
e570fd27 | 7855 | btrfs_release_block_group(block_group, delalloc); |
14443937 | 7856 | cond_resched(); |
2552d17e JB |
7857 | } |
7858 | up_read(&space_info->groups_sem); | |
7859 | ||
13a0db5a | 7860 | if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg |
7861 | && !orig_have_caching_bg) | |
7862 | orig_have_caching_bg = true; | |
7863 | ||
60d2adbb MX |
7864 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
7865 | goto search; | |
7866 | ||
b742bb82 YZ |
7867 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
7868 | goto search; | |
7869 | ||
285ff5af | 7870 | /* |
ccf0e725 JB |
7871 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
7872 | * caching kthreads as we move along | |
817d52f8 JB |
7873 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
7874 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7875 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7876 | * again | |
fa9c0d79 | 7877 | */ |
723bda20 | 7878 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 7879 | index = 0; |
a5e681d9 JB |
7880 | if (loop == LOOP_CACHING_NOWAIT) { |
7881 | /* | |
7882 | * We want to skip the LOOP_CACHING_WAIT step if we | |
01327610 | 7883 | * don't have any uncached bgs and we've already done a |
a5e681d9 JB |
7884 | * full search through. |
7885 | */ | |
13a0db5a | 7886 | if (orig_have_caching_bg || !full_search) |
a5e681d9 JB |
7887 | loop = LOOP_CACHING_WAIT; |
7888 | else | |
7889 | loop = LOOP_ALLOC_CHUNK; | |
7890 | } else { | |
7891 | loop++; | |
7892 | } | |
7893 | ||
817d52f8 | 7894 | if (loop == LOOP_ALLOC_CHUNK) { |
00361589 | 7895 | struct btrfs_trans_handle *trans; |
f017f15f WS |
7896 | int exist = 0; |
7897 | ||
7898 | trans = current->journal_info; | |
7899 | if (trans) | |
7900 | exist = 1; | |
7901 | else | |
7902 | trans = btrfs_join_transaction(root); | |
00361589 | 7903 | |
00361589 JB |
7904 | if (IS_ERR(trans)) { |
7905 | ret = PTR_ERR(trans); | |
7906 | goto out; | |
7907 | } | |
7908 | ||
2ff7e61e | 7909 | ret = do_chunk_alloc(trans, fs_info, flags, |
ea658bad | 7910 | CHUNK_ALLOC_FORCE); |
a5e681d9 JB |
7911 | |
7912 | /* | |
7913 | * If we can't allocate a new chunk we've already looped | |
7914 | * through at least once, move on to the NO_EMPTY_SIZE | |
7915 | * case. | |
7916 | */ | |
7917 | if (ret == -ENOSPC) | |
7918 | loop = LOOP_NO_EMPTY_SIZE; | |
7919 | ||
ea658bad JB |
7920 | /* |
7921 | * Do not bail out on ENOSPC since we | |
7922 | * can do more things. | |
7923 | */ | |
00361589 | 7924 | if (ret < 0 && ret != -ENOSPC) |
66642832 | 7925 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
7926 | else |
7927 | ret = 0; | |
f017f15f | 7928 | if (!exist) |
3a45bb20 | 7929 | btrfs_end_transaction(trans); |
00361589 | 7930 | if (ret) |
ea658bad | 7931 | goto out; |
2552d17e JB |
7932 | } |
7933 | ||
723bda20 | 7934 | if (loop == LOOP_NO_EMPTY_SIZE) { |
a5e681d9 JB |
7935 | /* |
7936 | * Don't loop again if we already have no empty_size and | |
7937 | * no empty_cluster. | |
7938 | */ | |
7939 | if (empty_size == 0 && | |
7940 | empty_cluster == 0) { | |
7941 | ret = -ENOSPC; | |
7942 | goto out; | |
7943 | } | |
723bda20 JB |
7944 | empty_size = 0; |
7945 | empty_cluster = 0; | |
fa9c0d79 | 7946 | } |
723bda20 JB |
7947 | |
7948 | goto search; | |
2552d17e JB |
7949 | } else if (!ins->objectid) { |
7950 | ret = -ENOSPC; | |
d82a6f1d | 7951 | } else if (ins->objectid) { |
c759c4e1 JB |
7952 | if (!use_cluster && last_ptr) { |
7953 | spin_lock(&last_ptr->lock); | |
7954 | last_ptr->window_start = ins->objectid; | |
7955 | spin_unlock(&last_ptr->lock); | |
7956 | } | |
80eb234a | 7957 | ret = 0; |
be744175 | 7958 | } |
79787eaa | 7959 | out: |
4f4db217 JB |
7960 | if (ret == -ENOSPC) { |
7961 | spin_lock(&space_info->lock); | |
7962 | space_info->max_extent_size = max_extent_size; | |
7963 | spin_unlock(&space_info->lock); | |
a4820398 | 7964 | ins->offset = max_extent_size; |
4f4db217 | 7965 | } |
0f70abe2 | 7966 | return ret; |
fec577fb | 7967 | } |
ec44a35c | 7968 | |
ab8d0fc4 JM |
7969 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
7970 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 7971 | int dump_block_groups) |
0f9dd46c JB |
7972 | { |
7973 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 7974 | int index = 0; |
0f9dd46c | 7975 | |
9ed74f2d | 7976 | spin_lock(&info->lock); |
ab8d0fc4 JM |
7977 | btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull", |
7978 | info->flags, | |
4136135b LB |
7979 | info->total_bytes - btrfs_space_info_used(info, true), |
7980 | info->full ? "" : "not "); | |
ab8d0fc4 JM |
7981 | btrfs_info(fs_info, |
7982 | "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu", | |
7983 | info->total_bytes, info->bytes_used, info->bytes_pinned, | |
7984 | info->bytes_reserved, info->bytes_may_use, | |
7985 | info->bytes_readonly); | |
9ed74f2d JB |
7986 | spin_unlock(&info->lock); |
7987 | ||
7988 | if (!dump_block_groups) | |
7989 | return; | |
0f9dd46c | 7990 | |
80eb234a | 7991 | down_read(&info->groups_sem); |
b742bb82 YZ |
7992 | again: |
7993 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 7994 | spin_lock(&cache->lock); |
ab8d0fc4 JM |
7995 | btrfs_info(fs_info, |
7996 | "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s", | |
7997 | cache->key.objectid, cache->key.offset, | |
7998 | btrfs_block_group_used(&cache->item), cache->pinned, | |
7999 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
8000 | btrfs_dump_free_space(cache, bytes); |
8001 | spin_unlock(&cache->lock); | |
8002 | } | |
b742bb82 YZ |
8003 | if (++index < BTRFS_NR_RAID_TYPES) |
8004 | goto again; | |
80eb234a | 8005 | up_read(&info->groups_sem); |
0f9dd46c | 8006 | } |
e8569813 | 8007 | |
18513091 | 8008 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
8009 | u64 num_bytes, u64 min_alloc_size, |
8010 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 8011 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 8012 | { |
ab8d0fc4 | 8013 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 8014 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 8015 | u64 flags; |
fec577fb | 8016 | int ret; |
925baedd | 8017 | |
1b86826d | 8018 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 8019 | again: |
0b246afa | 8020 | WARN_ON(num_bytes < fs_info->sectorsize); |
87bde3cd | 8021 | ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size, |
18513091 | 8022 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 8023 | if (!ret && !is_data) { |
ab8d0fc4 | 8024 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 8025 | } else if (ret == -ENOSPC) { |
a4820398 MX |
8026 | if (!final_tried && ins->offset) { |
8027 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 8028 | num_bytes = round_down(num_bytes, |
0b246afa | 8029 | fs_info->sectorsize); |
9e622d6b | 8030 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 8031 | ram_bytes = num_bytes; |
9e622d6b MX |
8032 | if (num_bytes == min_alloc_size) |
8033 | final_tried = true; | |
8034 | goto again; | |
ab8d0fc4 | 8035 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
8036 | struct btrfs_space_info *sinfo; |
8037 | ||
ab8d0fc4 | 8038 | sinfo = __find_space_info(fs_info, flags); |
0b246afa | 8039 | btrfs_err(fs_info, |
5d163e0e JM |
8040 | "allocation failed flags %llu, wanted %llu", |
8041 | flags, num_bytes); | |
53804280 | 8042 | if (sinfo) |
ab8d0fc4 | 8043 | dump_space_info(fs_info, sinfo, num_bytes, 1); |
9e622d6b | 8044 | } |
925baedd | 8045 | } |
0f9dd46c JB |
8046 | |
8047 | return ret; | |
e6dcd2dc CM |
8048 | } |
8049 | ||
2ff7e61e | 8050 | static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
8051 | u64 start, u64 len, |
8052 | int pin, int delalloc) | |
65b51a00 | 8053 | { |
0f9dd46c | 8054 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 8055 | int ret = 0; |
0f9dd46c | 8056 | |
0b246afa | 8057 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 8058 | if (!cache) { |
0b246afa JM |
8059 | btrfs_err(fs_info, "Unable to find block group for %llu", |
8060 | start); | |
0f9dd46c JB |
8061 | return -ENOSPC; |
8062 | } | |
1f3c79a2 | 8063 | |
e688b725 | 8064 | if (pin) |
2ff7e61e | 8065 | pin_down_extent(fs_info, cache, start, len, 1); |
e688b725 | 8066 | else { |
0b246afa | 8067 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 8068 | ret = btrfs_discard_extent(fs_info, start, len, NULL); |
e688b725 | 8069 | btrfs_add_free_space(cache, start, len); |
4824f1f4 | 8070 | btrfs_free_reserved_bytes(cache, len, delalloc); |
71ff6437 | 8071 | trace_btrfs_reserved_extent_free(fs_info, start, len); |
e688b725 | 8072 | } |
31193213 | 8073 | |
fa9c0d79 | 8074 | btrfs_put_block_group(cache); |
e6dcd2dc CM |
8075 | return ret; |
8076 | } | |
8077 | ||
2ff7e61e | 8078 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 | 8079 | u64 start, u64 len, int delalloc) |
e688b725 | 8080 | { |
2ff7e61e | 8081 | return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc); |
e688b725 CM |
8082 | } |
8083 | ||
2ff7e61e | 8084 | int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info, |
e688b725 CM |
8085 | u64 start, u64 len) |
8086 | { | |
2ff7e61e | 8087 | return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0); |
e688b725 CM |
8088 | } |
8089 | ||
5d4f98a2 | 8090 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 8091 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8092 | u64 parent, u64 root_objectid, |
8093 | u64 flags, u64 owner, u64 offset, | |
8094 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
8095 | { |
8096 | int ret; | |
e6dcd2dc | 8097 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 8098 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 8099 | struct btrfs_path *path; |
5d4f98a2 YZ |
8100 | struct extent_buffer *leaf; |
8101 | int type; | |
8102 | u32 size; | |
26b8003f | 8103 | |
5d4f98a2 YZ |
8104 | if (parent > 0) |
8105 | type = BTRFS_SHARED_DATA_REF_KEY; | |
8106 | else | |
8107 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 8108 | |
5d4f98a2 | 8109 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
8110 | |
8111 | path = btrfs_alloc_path(); | |
db5b493a TI |
8112 | if (!path) |
8113 | return -ENOMEM; | |
47e4bb98 | 8114 | |
b9473439 | 8115 | path->leave_spinning = 1; |
5d4f98a2 YZ |
8116 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
8117 | ins, size); | |
79787eaa JM |
8118 | if (ret) { |
8119 | btrfs_free_path(path); | |
8120 | return ret; | |
8121 | } | |
0f9dd46c | 8122 | |
5d4f98a2 YZ |
8123 | leaf = path->nodes[0]; |
8124 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 8125 | struct btrfs_extent_item); |
5d4f98a2 YZ |
8126 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
8127 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8128 | btrfs_set_extent_flags(leaf, extent_item, | |
8129 | flags | BTRFS_EXTENT_FLAG_DATA); | |
8130 | ||
8131 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8132 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
8133 | if (parent > 0) { | |
8134 | struct btrfs_shared_data_ref *ref; | |
8135 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
8136 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8137 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
8138 | } else { | |
8139 | struct btrfs_extent_data_ref *ref; | |
8140 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
8141 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
8142 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
8143 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
8144 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
8145 | } | |
47e4bb98 CM |
8146 | |
8147 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 8148 | btrfs_free_path(path); |
f510cfec | 8149 | |
1e144fb8 OS |
8150 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
8151 | ins->offset); | |
8152 | if (ret) | |
8153 | return ret; | |
8154 | ||
6202df69 | 8155 | ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1); |
79787eaa | 8156 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8157 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8158 | ins->objectid, ins->offset); |
f5947066 CM |
8159 | BUG(); |
8160 | } | |
71ff6437 | 8161 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
8162 | return ret; |
8163 | } | |
8164 | ||
5d4f98a2 | 8165 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
2ff7e61e | 8166 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8167 | u64 parent, u64 root_objectid, |
8168 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 8169 | int level, struct btrfs_key *ins) |
e6dcd2dc CM |
8170 | { |
8171 | int ret; | |
5d4f98a2 YZ |
8172 | struct btrfs_extent_item *extent_item; |
8173 | struct btrfs_tree_block_info *block_info; | |
8174 | struct btrfs_extent_inline_ref *iref; | |
8175 | struct btrfs_path *path; | |
8176 | struct extent_buffer *leaf; | |
3173a18f | 8177 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
fcebe456 | 8178 | u64 num_bytes = ins->offset; |
0b246afa | 8179 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f JB |
8180 | |
8181 | if (!skinny_metadata) | |
8182 | size += sizeof(*block_info); | |
1c2308f8 | 8183 | |
5d4f98a2 | 8184 | path = btrfs_alloc_path(); |
857cc2fc | 8185 | if (!path) { |
2ff7e61e | 8186 | btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid, |
0b246afa | 8187 | fs_info->nodesize); |
d8926bb3 | 8188 | return -ENOMEM; |
857cc2fc | 8189 | } |
56bec294 | 8190 | |
5d4f98a2 YZ |
8191 | path->leave_spinning = 1; |
8192 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
8193 | ins, size); | |
79787eaa | 8194 | if (ret) { |
dd825259 | 8195 | btrfs_free_path(path); |
2ff7e61e | 8196 | btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid, |
0b246afa | 8197 | fs_info->nodesize); |
79787eaa JM |
8198 | return ret; |
8199 | } | |
5d4f98a2 YZ |
8200 | |
8201 | leaf = path->nodes[0]; | |
8202 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
8203 | struct btrfs_extent_item); | |
8204 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
8205 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8206 | btrfs_set_extent_flags(leaf, extent_item, | |
8207 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 8208 | |
3173a18f JB |
8209 | if (skinny_metadata) { |
8210 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
0b246afa | 8211 | num_bytes = fs_info->nodesize; |
3173a18f JB |
8212 | } else { |
8213 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
8214 | btrfs_set_tree_block_key(leaf, block_info, key); | |
8215 | btrfs_set_tree_block_level(leaf, block_info, level); | |
8216 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
8217 | } | |
5d4f98a2 | 8218 | |
5d4f98a2 YZ |
8219 | if (parent > 0) { |
8220 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
8221 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8222 | BTRFS_SHARED_BLOCK_REF_KEY); | |
8223 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8224 | } else { | |
8225 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8226 | BTRFS_TREE_BLOCK_REF_KEY); | |
8227 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
8228 | } | |
8229 | ||
8230 | btrfs_mark_buffer_dirty(leaf); | |
8231 | btrfs_free_path(path); | |
8232 | ||
1e144fb8 OS |
8233 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
8234 | num_bytes); | |
8235 | if (ret) | |
8236 | return ret; | |
8237 | ||
6202df69 JM |
8238 | ret = update_block_group(trans, fs_info, ins->objectid, |
8239 | fs_info->nodesize, 1); | |
79787eaa | 8240 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8241 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8242 | ins->objectid, ins->offset); |
5d4f98a2 YZ |
8243 | BUG(); |
8244 | } | |
0be5dc67 | 8245 | |
71ff6437 | 8246 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, |
0b246afa | 8247 | fs_info->nodesize); |
5d4f98a2 YZ |
8248 | return ret; |
8249 | } | |
8250 | ||
8251 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 8252 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
8253 | u64 offset, u64 ram_bytes, |
8254 | struct btrfs_key *ins) | |
5d4f98a2 | 8255 | { |
84f7d8e6 | 8256 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
8257 | int ret; |
8258 | ||
84f7d8e6 | 8259 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 8260 | |
fd708b81 JB |
8261 | btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0, |
8262 | root->root_key.objectid, owner, offset, | |
8263 | BTRFS_ADD_DELAYED_EXTENT); | |
8264 | ||
0b246afa | 8265 | ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid, |
84f7d8e6 JB |
8266 | ins->offset, 0, |
8267 | root->root_key.objectid, owner, | |
7be07912 OS |
8268 | offset, ram_bytes, |
8269 | BTRFS_ADD_DELAYED_EXTENT, NULL, NULL); | |
e6dcd2dc CM |
8270 | return ret; |
8271 | } | |
e02119d5 CM |
8272 | |
8273 | /* | |
8274 | * this is used by the tree logging recovery code. It records that | |
8275 | * an extent has been allocated and makes sure to clear the free | |
8276 | * space cache bits as well | |
8277 | */ | |
5d4f98a2 | 8278 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 8279 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8280 | u64 root_objectid, u64 owner, u64 offset, |
8281 | struct btrfs_key *ins) | |
e02119d5 CM |
8282 | { |
8283 | int ret; | |
8284 | struct btrfs_block_group_cache *block_group; | |
ed7a6948 | 8285 | struct btrfs_space_info *space_info; |
11833d66 | 8286 | |
8c2a1a30 JB |
8287 | /* |
8288 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 8289 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 8290 | */ |
0b246afa | 8291 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
8292 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
8293 | ins->offset); | |
b50c6e25 | 8294 | if (ret) |
8c2a1a30 | 8295 | return ret; |
11833d66 YZ |
8296 | } |
8297 | ||
0b246afa | 8298 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
8299 | if (!block_group) |
8300 | return -EINVAL; | |
8301 | ||
ed7a6948 WX |
8302 | space_info = block_group->space_info; |
8303 | spin_lock(&space_info->lock); | |
8304 | spin_lock(&block_group->lock); | |
8305 | space_info->bytes_reserved += ins->offset; | |
8306 | block_group->reserved += ins->offset; | |
8307 | spin_unlock(&block_group->lock); | |
8308 | spin_unlock(&space_info->lock); | |
8309 | ||
2ff7e61e | 8310 | ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid, |
5d4f98a2 | 8311 | 0, owner, offset, ins, 1); |
b50c6e25 | 8312 | btrfs_put_block_group(block_group); |
e02119d5 CM |
8313 | return ret; |
8314 | } | |
8315 | ||
48a3b636 ES |
8316 | static struct extent_buffer * |
8317 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
fe864576 | 8318 | u64 bytenr, int level) |
65b51a00 | 8319 | { |
0b246afa | 8320 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
8321 | struct extent_buffer *buf; |
8322 | ||
2ff7e61e | 8323 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8324 | if (IS_ERR(buf)) |
8325 | return buf; | |
8326 | ||
65b51a00 | 8327 | btrfs_set_header_generation(buf, trans->transid); |
85d4e461 | 8328 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 8329 | btrfs_tree_lock(buf); |
7c302b49 | 8330 | clean_tree_block(fs_info, buf); |
3083ee2e | 8331 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
8332 | |
8333 | btrfs_set_lock_blocking(buf); | |
4db8c528 | 8334 | set_extent_buffer_uptodate(buf); |
b4ce94de | 8335 | |
d0c803c4 | 8336 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 8337 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
8338 | /* |
8339 | * we allow two log transactions at a time, use different | |
8340 | * EXENT bit to differentiate dirty pages. | |
8341 | */ | |
656f30db | 8342 | if (buf->log_index == 0) |
8cef4e16 YZ |
8343 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
8344 | buf->start + buf->len - 1, GFP_NOFS); | |
8345 | else | |
8346 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 8347 | buf->start + buf->len - 1); |
d0c803c4 | 8348 | } else { |
656f30db | 8349 | buf->log_index = -1; |
d0c803c4 | 8350 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 8351 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 8352 | } |
64c12921 | 8353 | trans->dirty = true; |
b4ce94de | 8354 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
8355 | return buf; |
8356 | } | |
8357 | ||
f0486c68 YZ |
8358 | static struct btrfs_block_rsv * |
8359 | use_block_rsv(struct btrfs_trans_handle *trans, | |
8360 | struct btrfs_root *root, u32 blocksize) | |
8361 | { | |
0b246afa | 8362 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8363 | struct btrfs_block_rsv *block_rsv; |
0b246afa | 8364 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
f0486c68 | 8365 | int ret; |
d88033db | 8366 | bool global_updated = false; |
f0486c68 YZ |
8367 | |
8368 | block_rsv = get_block_rsv(trans, root); | |
8369 | ||
b586b323 MX |
8370 | if (unlikely(block_rsv->size == 0)) |
8371 | goto try_reserve; | |
d88033db | 8372 | again: |
f0486c68 YZ |
8373 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
8374 | if (!ret) | |
8375 | return block_rsv; | |
8376 | ||
b586b323 MX |
8377 | if (block_rsv->failfast) |
8378 | return ERR_PTR(ret); | |
8379 | ||
d88033db MX |
8380 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
8381 | global_updated = true; | |
0b246afa | 8382 | update_global_block_rsv(fs_info); |
d88033db MX |
8383 | goto again; |
8384 | } | |
8385 | ||
0b246afa | 8386 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
b586b323 MX |
8387 | static DEFINE_RATELIMIT_STATE(_rs, |
8388 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
8389 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
8390 | if (__ratelimit(&_rs)) | |
8391 | WARN(1, KERN_DEBUG | |
efe120a0 | 8392 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
8393 | } |
8394 | try_reserve: | |
8395 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
8396 | BTRFS_RESERVE_NO_FLUSH); | |
8397 | if (!ret) | |
8398 | return block_rsv; | |
8399 | /* | |
8400 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
8401 | * the global reserve if its space type is the same as the global |
8402 | * reservation. | |
b586b323 | 8403 | */ |
5881cfc9 MX |
8404 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
8405 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
8406 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
8407 | if (!ret) | |
8408 | return global_rsv; | |
8409 | } | |
8410 | return ERR_PTR(ret); | |
f0486c68 YZ |
8411 | } |
8412 | ||
8c2a3ca2 JB |
8413 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
8414 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
8415 | { |
8416 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 8417 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
8418 | } |
8419 | ||
fec577fb | 8420 | /* |
f0486c68 | 8421 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 8422 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 8423 | */ |
4d75f8a9 | 8424 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
8425 | struct btrfs_root *root, |
8426 | u64 parent, u64 root_objectid, | |
8427 | const struct btrfs_disk_key *key, | |
8428 | int level, u64 hint, | |
8429 | u64 empty_size) | |
fec577fb | 8430 | { |
0b246afa | 8431 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 8432 | struct btrfs_key ins; |
f0486c68 | 8433 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 8434 | struct extent_buffer *buf; |
67b7859e | 8435 | struct btrfs_delayed_extent_op *extent_op; |
f0486c68 YZ |
8436 | u64 flags = 0; |
8437 | int ret; | |
0b246afa JM |
8438 | u32 blocksize = fs_info->nodesize; |
8439 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 8440 | |
05653ef3 | 8441 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 8442 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 8443 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
fe864576 | 8444 | level); |
faa2dbf0 JB |
8445 | if (!IS_ERR(buf)) |
8446 | root->alloc_bytenr += blocksize; | |
8447 | return buf; | |
8448 | } | |
05653ef3 | 8449 | #endif |
fccb84c9 | 8450 | |
f0486c68 YZ |
8451 | block_rsv = use_block_rsv(trans, root, blocksize); |
8452 | if (IS_ERR(block_rsv)) | |
8453 | return ERR_CAST(block_rsv); | |
8454 | ||
18513091 | 8455 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 8456 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
8457 | if (ret) |
8458 | goto out_unuse; | |
55c69072 | 8459 | |
fe864576 | 8460 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level); |
67b7859e OS |
8461 | if (IS_ERR(buf)) { |
8462 | ret = PTR_ERR(buf); | |
8463 | goto out_free_reserved; | |
8464 | } | |
f0486c68 YZ |
8465 | |
8466 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8467 | if (parent == 0) | |
8468 | parent = ins.objectid; | |
8469 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8470 | } else | |
8471 | BUG_ON(parent > 0); | |
8472 | ||
8473 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 8474 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
8475 | if (!extent_op) { |
8476 | ret = -ENOMEM; | |
8477 | goto out_free_buf; | |
8478 | } | |
f0486c68 YZ |
8479 | if (key) |
8480 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8481 | else | |
8482 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8483 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
8484 | extent_op->update_key = skinny_metadata ? false : true; |
8485 | extent_op->update_flags = true; | |
8486 | extent_op->is_data = false; | |
b1c79e09 | 8487 | extent_op->level = level; |
f0486c68 | 8488 | |
fd708b81 JB |
8489 | btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent, |
8490 | root_objectid, level, 0, | |
8491 | BTRFS_ADD_DELAYED_EXTENT); | |
7be07912 OS |
8492 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid, |
8493 | ins.offset, parent, | |
8494 | root_objectid, level, | |
67b7859e | 8495 | BTRFS_ADD_DELAYED_EXTENT, |
7be07912 | 8496 | extent_op, NULL, NULL); |
67b7859e OS |
8497 | if (ret) |
8498 | goto out_free_delayed; | |
f0486c68 | 8499 | } |
fec577fb | 8500 | return buf; |
67b7859e OS |
8501 | |
8502 | out_free_delayed: | |
8503 | btrfs_free_delayed_extent_op(extent_op); | |
8504 | out_free_buf: | |
8505 | free_extent_buffer(buf); | |
8506 | out_free_reserved: | |
2ff7e61e | 8507 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 8508 | out_unuse: |
0b246afa | 8509 | unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 8510 | return ERR_PTR(ret); |
fec577fb | 8511 | } |
a28ec197 | 8512 | |
2c47e605 YZ |
8513 | struct walk_control { |
8514 | u64 refs[BTRFS_MAX_LEVEL]; | |
8515 | u64 flags[BTRFS_MAX_LEVEL]; | |
8516 | struct btrfs_key update_progress; | |
8517 | int stage; | |
8518 | int level; | |
8519 | int shared_level; | |
8520 | int update_ref; | |
8521 | int keep_locks; | |
1c4850e2 YZ |
8522 | int reada_slot; |
8523 | int reada_count; | |
66d7e7f0 | 8524 | int for_reloc; |
2c47e605 YZ |
8525 | }; |
8526 | ||
8527 | #define DROP_REFERENCE 1 | |
8528 | #define UPDATE_BACKREF 2 | |
8529 | ||
1c4850e2 YZ |
8530 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8531 | struct btrfs_root *root, | |
8532 | struct walk_control *wc, | |
8533 | struct btrfs_path *path) | |
6407bf6d | 8534 | { |
0b246afa | 8535 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8536 | u64 bytenr; |
8537 | u64 generation; | |
8538 | u64 refs; | |
94fcca9f | 8539 | u64 flags; |
5d4f98a2 | 8540 | u32 nritems; |
1c4850e2 YZ |
8541 | struct btrfs_key key; |
8542 | struct extent_buffer *eb; | |
6407bf6d | 8543 | int ret; |
1c4850e2 YZ |
8544 | int slot; |
8545 | int nread = 0; | |
6407bf6d | 8546 | |
1c4850e2 YZ |
8547 | if (path->slots[wc->level] < wc->reada_slot) { |
8548 | wc->reada_count = wc->reada_count * 2 / 3; | |
8549 | wc->reada_count = max(wc->reada_count, 2); | |
8550 | } else { | |
8551 | wc->reada_count = wc->reada_count * 3 / 2; | |
8552 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 8553 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 8554 | } |
7bb86316 | 8555 | |
1c4850e2 YZ |
8556 | eb = path->nodes[wc->level]; |
8557 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 8558 | |
1c4850e2 YZ |
8559 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8560 | if (nread >= wc->reada_count) | |
8561 | break; | |
bd56b302 | 8562 | |
2dd3e67b | 8563 | cond_resched(); |
1c4850e2 YZ |
8564 | bytenr = btrfs_node_blockptr(eb, slot); |
8565 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8566 | |
1c4850e2 YZ |
8567 | if (slot == path->slots[wc->level]) |
8568 | goto reada; | |
5d4f98a2 | 8569 | |
1c4850e2 YZ |
8570 | if (wc->stage == UPDATE_BACKREF && |
8571 | generation <= root->root_key.offset) | |
bd56b302 CM |
8572 | continue; |
8573 | ||
94fcca9f | 8574 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 8575 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
8576 | wc->level - 1, 1, &refs, |
8577 | &flags); | |
79787eaa JM |
8578 | /* We don't care about errors in readahead. */ |
8579 | if (ret < 0) | |
8580 | continue; | |
94fcca9f YZ |
8581 | BUG_ON(refs == 0); |
8582 | ||
1c4850e2 | 8583 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8584 | if (refs == 1) |
8585 | goto reada; | |
bd56b302 | 8586 | |
94fcca9f YZ |
8587 | if (wc->level == 1 && |
8588 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8589 | continue; | |
1c4850e2 YZ |
8590 | if (!wc->update_ref || |
8591 | generation <= root->root_key.offset) | |
8592 | continue; | |
8593 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8594 | ret = btrfs_comp_cpu_keys(&key, | |
8595 | &wc->update_progress); | |
8596 | if (ret < 0) | |
8597 | continue; | |
94fcca9f YZ |
8598 | } else { |
8599 | if (wc->level == 1 && | |
8600 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8601 | continue; | |
6407bf6d | 8602 | } |
1c4850e2 | 8603 | reada: |
2ff7e61e | 8604 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 8605 | nread++; |
20524f02 | 8606 | } |
1c4850e2 | 8607 | wc->reada_slot = slot; |
20524f02 | 8608 | } |
2c47e605 | 8609 | |
f82d02d9 | 8610 | /* |
2c016dc2 | 8611 | * helper to process tree block while walking down the tree. |
2c47e605 | 8612 | * |
2c47e605 YZ |
8613 | * when wc->stage == UPDATE_BACKREF, this function updates |
8614 | * back refs for pointers in the block. | |
8615 | * | |
8616 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8617 | */ |
2c47e605 | 8618 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8619 | struct btrfs_root *root, |
2c47e605 | 8620 | struct btrfs_path *path, |
94fcca9f | 8621 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8622 | { |
2ff7e61e | 8623 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
8624 | int level = wc->level; |
8625 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 8626 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
8627 | int ret; |
8628 | ||
2c47e605 YZ |
8629 | if (wc->stage == UPDATE_BACKREF && |
8630 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8631 | return 1; | |
f82d02d9 | 8632 | |
2c47e605 YZ |
8633 | /* |
8634 | * when reference count of tree block is 1, it won't increase | |
8635 | * again. once full backref flag is set, we never clear it. | |
8636 | */ | |
94fcca9f YZ |
8637 | if (lookup_info && |
8638 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8639 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 8640 | BUG_ON(!path->locks[level]); |
2ff7e61e | 8641 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8642 | eb->start, level, 1, |
2c47e605 YZ |
8643 | &wc->refs[level], |
8644 | &wc->flags[level]); | |
79787eaa JM |
8645 | BUG_ON(ret == -ENOMEM); |
8646 | if (ret) | |
8647 | return ret; | |
2c47e605 YZ |
8648 | BUG_ON(wc->refs[level] == 0); |
8649 | } | |
5d4f98a2 | 8650 | |
2c47e605 YZ |
8651 | if (wc->stage == DROP_REFERENCE) { |
8652 | if (wc->refs[level] > 1) | |
8653 | return 1; | |
f82d02d9 | 8654 | |
2c47e605 | 8655 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 8656 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8657 | path->locks[level] = 0; |
8658 | } | |
8659 | return 0; | |
8660 | } | |
f82d02d9 | 8661 | |
2c47e605 YZ |
8662 | /* wc->stage == UPDATE_BACKREF */ |
8663 | if (!(wc->flags[level] & flag)) { | |
8664 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 8665 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 8666 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 8667 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8668 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8669 | ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start, |
b1c79e09 JB |
8670 | eb->len, flag, |
8671 | btrfs_header_level(eb), 0); | |
79787eaa | 8672 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
8673 | wc->flags[level] |= flag; |
8674 | } | |
8675 | ||
8676 | /* | |
8677 | * the block is shared by multiple trees, so it's not good to | |
8678 | * keep the tree lock | |
8679 | */ | |
8680 | if (path->locks[level] && level > 0) { | |
bd681513 | 8681 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8682 | path->locks[level] = 0; |
8683 | } | |
8684 | return 0; | |
8685 | } | |
8686 | ||
1c4850e2 | 8687 | /* |
2c016dc2 | 8688 | * helper to process tree block pointer. |
1c4850e2 YZ |
8689 | * |
8690 | * when wc->stage == DROP_REFERENCE, this function checks | |
8691 | * reference count of the block pointed to. if the block | |
8692 | * is shared and we need update back refs for the subtree | |
8693 | * rooted at the block, this function changes wc->stage to | |
8694 | * UPDATE_BACKREF. if the block is shared and there is no | |
8695 | * need to update back, this function drops the reference | |
8696 | * to the block. | |
8697 | * | |
8698 | * NOTE: return value 1 means we should stop walking down. | |
8699 | */ | |
8700 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8701 | struct btrfs_root *root, | |
8702 | struct btrfs_path *path, | |
94fcca9f | 8703 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 8704 | { |
0b246afa | 8705 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8706 | u64 bytenr; |
8707 | u64 generation; | |
8708 | u64 parent; | |
8709 | u32 blocksize; | |
8710 | struct btrfs_key key; | |
8711 | struct extent_buffer *next; | |
8712 | int level = wc->level; | |
8713 | int reada = 0; | |
8714 | int ret = 0; | |
1152651a | 8715 | bool need_account = false; |
1c4850e2 YZ |
8716 | |
8717 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8718 | path->slots[level]); | |
8719 | /* | |
8720 | * if the lower level block was created before the snapshot | |
8721 | * was created, we know there is no need to update back refs | |
8722 | * for the subtree | |
8723 | */ | |
8724 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8725 | generation <= root->root_key.offset) { |
8726 | *lookup_info = 1; | |
1c4850e2 | 8727 | return 1; |
94fcca9f | 8728 | } |
1c4850e2 YZ |
8729 | |
8730 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
0b246afa | 8731 | blocksize = fs_info->nodesize; |
1c4850e2 | 8732 | |
0b246afa | 8733 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 8734 | if (!next) { |
2ff7e61e | 8735 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8736 | if (IS_ERR(next)) |
8737 | return PTR_ERR(next); | |
8738 | ||
b2aaaa3b JB |
8739 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8740 | level - 1); | |
1c4850e2 YZ |
8741 | reada = 1; |
8742 | } | |
8743 | btrfs_tree_lock(next); | |
8744 | btrfs_set_lock_blocking(next); | |
8745 | ||
2ff7e61e | 8746 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
8747 | &wc->refs[level - 1], |
8748 | &wc->flags[level - 1]); | |
4867268c JB |
8749 | if (ret < 0) |
8750 | goto out_unlock; | |
79787eaa | 8751 | |
c2cf52eb | 8752 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 8753 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
8754 | ret = -EIO; |
8755 | goto out_unlock; | |
c2cf52eb | 8756 | } |
94fcca9f | 8757 | *lookup_info = 0; |
1c4850e2 | 8758 | |
94fcca9f | 8759 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8760 | if (wc->refs[level - 1] > 1) { |
1152651a | 8761 | need_account = true; |
94fcca9f YZ |
8762 | if (level == 1 && |
8763 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8764 | goto skip; | |
8765 | ||
1c4850e2 YZ |
8766 | if (!wc->update_ref || |
8767 | generation <= root->root_key.offset) | |
8768 | goto skip; | |
8769 | ||
8770 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8771 | path->slots[level]); | |
8772 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8773 | if (ret < 0) | |
8774 | goto skip; | |
8775 | ||
8776 | wc->stage = UPDATE_BACKREF; | |
8777 | wc->shared_level = level - 1; | |
8778 | } | |
94fcca9f YZ |
8779 | } else { |
8780 | if (level == 1 && | |
8781 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8782 | goto skip; | |
1c4850e2 YZ |
8783 | } |
8784 | ||
b9fab919 | 8785 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8786 | btrfs_tree_unlock(next); |
8787 | free_extent_buffer(next); | |
8788 | next = NULL; | |
94fcca9f | 8789 | *lookup_info = 1; |
1c4850e2 YZ |
8790 | } |
8791 | ||
8792 | if (!next) { | |
8793 | if (reada && level == 1) | |
8794 | reada_walk_down(trans, root, wc, path); | |
2ff7e61e | 8795 | next = read_tree_block(fs_info, bytenr, generation); |
64c043de LB |
8796 | if (IS_ERR(next)) { |
8797 | return PTR_ERR(next); | |
8798 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8799 | free_extent_buffer(next); |
97d9a8a4 | 8800 | return -EIO; |
416bc658 | 8801 | } |
1c4850e2 YZ |
8802 | btrfs_tree_lock(next); |
8803 | btrfs_set_lock_blocking(next); | |
8804 | } | |
8805 | ||
8806 | level--; | |
4867268c JB |
8807 | ASSERT(level == btrfs_header_level(next)); |
8808 | if (level != btrfs_header_level(next)) { | |
8809 | btrfs_err(root->fs_info, "mismatched level"); | |
8810 | ret = -EIO; | |
8811 | goto out_unlock; | |
8812 | } | |
1c4850e2 YZ |
8813 | path->nodes[level] = next; |
8814 | path->slots[level] = 0; | |
bd681513 | 8815 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8816 | wc->level = level; |
8817 | if (wc->level == 1) | |
8818 | wc->reada_slot = 0; | |
8819 | return 0; | |
8820 | skip: | |
8821 | wc->refs[level - 1] = 0; | |
8822 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
8823 | if (wc->stage == DROP_REFERENCE) { |
8824 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8825 | parent = path->nodes[level]->start; | |
8826 | } else { | |
4867268c | 8827 | ASSERT(root->root_key.objectid == |
94fcca9f | 8828 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
8829 | if (root->root_key.objectid != |
8830 | btrfs_header_owner(path->nodes[level])) { | |
8831 | btrfs_err(root->fs_info, | |
8832 | "mismatched block owner"); | |
8833 | ret = -EIO; | |
8834 | goto out_unlock; | |
8835 | } | |
94fcca9f YZ |
8836 | parent = 0; |
8837 | } | |
1c4850e2 | 8838 | |
1152651a | 8839 | if (need_account) { |
33d1f05c QW |
8840 | ret = btrfs_qgroup_trace_subtree(trans, root, next, |
8841 | generation, level - 1); | |
1152651a | 8842 | if (ret) { |
0b246afa | 8843 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8844 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
8845 | ret); | |
1152651a MF |
8846 | } |
8847 | } | |
84f7d8e6 | 8848 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, |
2ff7e61e JM |
8849 | parent, root->root_key.objectid, |
8850 | level - 1, 0); | |
4867268c JB |
8851 | if (ret) |
8852 | goto out_unlock; | |
1c4850e2 | 8853 | } |
4867268c JB |
8854 | |
8855 | *lookup_info = 1; | |
8856 | ret = 1; | |
8857 | ||
8858 | out_unlock: | |
1c4850e2 YZ |
8859 | btrfs_tree_unlock(next); |
8860 | free_extent_buffer(next); | |
4867268c JB |
8861 | |
8862 | return ret; | |
1c4850e2 YZ |
8863 | } |
8864 | ||
2c47e605 | 8865 | /* |
2c016dc2 | 8866 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
8867 | * |
8868 | * when wc->stage == DROP_REFERENCE, this function drops | |
8869 | * reference count on the block. | |
8870 | * | |
8871 | * when wc->stage == UPDATE_BACKREF, this function changes | |
8872 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
8873 | * to UPDATE_BACKREF previously while processing the block. | |
8874 | * | |
8875 | * NOTE: return value 1 means we should stop walking up. | |
8876 | */ | |
8877 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
8878 | struct btrfs_root *root, | |
8879 | struct btrfs_path *path, | |
8880 | struct walk_control *wc) | |
8881 | { | |
0b246afa | 8882 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8883 | int ret; |
2c47e605 YZ |
8884 | int level = wc->level; |
8885 | struct extent_buffer *eb = path->nodes[level]; | |
8886 | u64 parent = 0; | |
8887 | ||
8888 | if (wc->stage == UPDATE_BACKREF) { | |
8889 | BUG_ON(wc->shared_level < level); | |
8890 | if (level < wc->shared_level) | |
8891 | goto out; | |
8892 | ||
2c47e605 YZ |
8893 | ret = find_next_key(path, level + 1, &wc->update_progress); |
8894 | if (ret > 0) | |
8895 | wc->update_ref = 0; | |
8896 | ||
8897 | wc->stage = DROP_REFERENCE; | |
8898 | wc->shared_level = -1; | |
8899 | path->slots[level] = 0; | |
8900 | ||
8901 | /* | |
8902 | * check reference count again if the block isn't locked. | |
8903 | * we should start walking down the tree again if reference | |
8904 | * count is one. | |
8905 | */ | |
8906 | if (!path->locks[level]) { | |
8907 | BUG_ON(level == 0); | |
8908 | btrfs_tree_lock(eb); | |
8909 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8910 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8911 | |
2ff7e61e | 8912 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8913 | eb->start, level, 1, |
2c47e605 YZ |
8914 | &wc->refs[level], |
8915 | &wc->flags[level]); | |
79787eaa JM |
8916 | if (ret < 0) { |
8917 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 8918 | path->locks[level] = 0; |
79787eaa JM |
8919 | return ret; |
8920 | } | |
2c47e605 YZ |
8921 | BUG_ON(wc->refs[level] == 0); |
8922 | if (wc->refs[level] == 1) { | |
bd681513 | 8923 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 8924 | path->locks[level] = 0; |
2c47e605 YZ |
8925 | return 1; |
8926 | } | |
f82d02d9 | 8927 | } |
2c47e605 | 8928 | } |
f82d02d9 | 8929 | |
2c47e605 YZ |
8930 | /* wc->stage == DROP_REFERENCE */ |
8931 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 8932 | |
2c47e605 YZ |
8933 | if (wc->refs[level] == 1) { |
8934 | if (level == 0) { | |
8935 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 8936 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 8937 | else |
e339a6b0 | 8938 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8939 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8940 | ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb); |
1152651a | 8941 | if (ret) { |
0b246afa | 8942 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8943 | "error %d accounting leaf items. Quota is out of sync, rescan required.", |
8944 | ret); | |
1152651a | 8945 | } |
2c47e605 YZ |
8946 | } |
8947 | /* make block locked assertion in clean_tree_block happy */ | |
8948 | if (!path->locks[level] && | |
8949 | btrfs_header_generation(eb) == trans->transid) { | |
8950 | btrfs_tree_lock(eb); | |
8951 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8952 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8953 | } |
7c302b49 | 8954 | clean_tree_block(fs_info, eb); |
2c47e605 YZ |
8955 | } |
8956 | ||
8957 | if (eb == root->node) { | |
8958 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8959 | parent = eb->start; | |
8960 | else | |
8961 | BUG_ON(root->root_key.objectid != | |
8962 | btrfs_header_owner(eb)); | |
8963 | } else { | |
8964 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8965 | parent = path->nodes[level + 1]->start; | |
8966 | else | |
8967 | BUG_ON(root->root_key.objectid != | |
8968 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 8969 | } |
f82d02d9 | 8970 | |
5581a51a | 8971 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
8972 | out: |
8973 | wc->refs[level] = 0; | |
8974 | wc->flags[level] = 0; | |
f0486c68 | 8975 | return 0; |
2c47e605 YZ |
8976 | } |
8977 | ||
8978 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
8979 | struct btrfs_root *root, | |
8980 | struct btrfs_path *path, | |
8981 | struct walk_control *wc) | |
8982 | { | |
2c47e605 | 8983 | int level = wc->level; |
94fcca9f | 8984 | int lookup_info = 1; |
2c47e605 YZ |
8985 | int ret; |
8986 | ||
8987 | while (level >= 0) { | |
94fcca9f | 8988 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
8989 | if (ret > 0) |
8990 | break; | |
8991 | ||
8992 | if (level == 0) | |
8993 | break; | |
8994 | ||
7a7965f8 YZ |
8995 | if (path->slots[level] >= |
8996 | btrfs_header_nritems(path->nodes[level])) | |
8997 | break; | |
8998 | ||
94fcca9f | 8999 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
9000 | if (ret > 0) { |
9001 | path->slots[level]++; | |
9002 | continue; | |
90d2c51d MX |
9003 | } else if (ret < 0) |
9004 | return ret; | |
1c4850e2 | 9005 | level = wc->level; |
f82d02d9 | 9006 | } |
f82d02d9 YZ |
9007 | return 0; |
9008 | } | |
9009 | ||
d397712b | 9010 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 9011 | struct btrfs_root *root, |
f82d02d9 | 9012 | struct btrfs_path *path, |
2c47e605 | 9013 | struct walk_control *wc, int max_level) |
20524f02 | 9014 | { |
2c47e605 | 9015 | int level = wc->level; |
20524f02 | 9016 | int ret; |
9f3a7427 | 9017 | |
2c47e605 YZ |
9018 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
9019 | while (level < max_level && path->nodes[level]) { | |
9020 | wc->level = level; | |
9021 | if (path->slots[level] + 1 < | |
9022 | btrfs_header_nritems(path->nodes[level])) { | |
9023 | path->slots[level]++; | |
20524f02 CM |
9024 | return 0; |
9025 | } else { | |
2c47e605 YZ |
9026 | ret = walk_up_proc(trans, root, path, wc); |
9027 | if (ret > 0) | |
9028 | return 0; | |
bd56b302 | 9029 | |
2c47e605 | 9030 | if (path->locks[level]) { |
bd681513 CM |
9031 | btrfs_tree_unlock_rw(path->nodes[level], |
9032 | path->locks[level]); | |
2c47e605 | 9033 | path->locks[level] = 0; |
f82d02d9 | 9034 | } |
2c47e605 YZ |
9035 | free_extent_buffer(path->nodes[level]); |
9036 | path->nodes[level] = NULL; | |
9037 | level++; | |
20524f02 CM |
9038 | } |
9039 | } | |
9040 | return 1; | |
9041 | } | |
9042 | ||
9aca1d51 | 9043 | /* |
2c47e605 YZ |
9044 | * drop a subvolume tree. |
9045 | * | |
9046 | * this function traverses the tree freeing any blocks that only | |
9047 | * referenced by the tree. | |
9048 | * | |
9049 | * when a shared tree block is found. this function decreases its | |
9050 | * reference count by one. if update_ref is true, this function | |
9051 | * also make sure backrefs for the shared block and all lower level | |
9052 | * blocks are properly updated. | |
9d1a2a3a DS |
9053 | * |
9054 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 9055 | */ |
2c536799 | 9056 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
9057 | struct btrfs_block_rsv *block_rsv, int update_ref, |
9058 | int for_reloc) | |
20524f02 | 9059 | { |
ab8d0fc4 | 9060 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 9061 | struct btrfs_path *path; |
2c47e605 | 9062 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 9063 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 9064 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
9065 | struct walk_control *wc; |
9066 | struct btrfs_key key; | |
9067 | int err = 0; | |
9068 | int ret; | |
9069 | int level; | |
d29a9f62 | 9070 | bool root_dropped = false; |
20524f02 | 9071 | |
ab8d0fc4 | 9072 | btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid); |
1152651a | 9073 | |
5caf2a00 | 9074 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
9075 | if (!path) { |
9076 | err = -ENOMEM; | |
9077 | goto out; | |
9078 | } | |
20524f02 | 9079 | |
2c47e605 | 9080 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
9081 | if (!wc) { |
9082 | btrfs_free_path(path); | |
cb1b69f4 TI |
9083 | err = -ENOMEM; |
9084 | goto out; | |
38a1a919 | 9085 | } |
2c47e605 | 9086 | |
a22285a6 | 9087 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9088 | if (IS_ERR(trans)) { |
9089 | err = PTR_ERR(trans); | |
9090 | goto out_free; | |
9091 | } | |
98d5dc13 | 9092 | |
3fd0a558 YZ |
9093 | if (block_rsv) |
9094 | trans->block_rsv = block_rsv; | |
2c47e605 | 9095 | |
9f3a7427 | 9096 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 9097 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
9098 | path->nodes[level] = btrfs_lock_root_node(root); |
9099 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 9100 | path->slots[level] = 0; |
bd681513 | 9101 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9102 | memset(&wc->update_progress, 0, |
9103 | sizeof(wc->update_progress)); | |
9f3a7427 | 9104 | } else { |
9f3a7427 | 9105 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
9106 | memcpy(&wc->update_progress, &key, |
9107 | sizeof(wc->update_progress)); | |
9108 | ||
6702ed49 | 9109 | level = root_item->drop_level; |
2c47e605 | 9110 | BUG_ON(level == 0); |
6702ed49 | 9111 | path->lowest_level = level; |
2c47e605 YZ |
9112 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
9113 | path->lowest_level = 0; | |
9114 | if (ret < 0) { | |
9115 | err = ret; | |
79787eaa | 9116 | goto out_end_trans; |
9f3a7427 | 9117 | } |
1c4850e2 | 9118 | WARN_ON(ret > 0); |
2c47e605 | 9119 | |
7d9eb12c CM |
9120 | /* |
9121 | * unlock our path, this is safe because only this | |
9122 | * function is allowed to delete this snapshot | |
9123 | */ | |
5d4f98a2 | 9124 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
9125 | |
9126 | level = btrfs_header_level(root->node); | |
9127 | while (1) { | |
9128 | btrfs_tree_lock(path->nodes[level]); | |
9129 | btrfs_set_lock_blocking(path->nodes[level]); | |
fec386ac | 9130 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9131 | |
2ff7e61e | 9132 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 9133 | path->nodes[level]->start, |
3173a18f | 9134 | level, 1, &wc->refs[level], |
2c47e605 | 9135 | &wc->flags[level]); |
79787eaa JM |
9136 | if (ret < 0) { |
9137 | err = ret; | |
9138 | goto out_end_trans; | |
9139 | } | |
2c47e605 YZ |
9140 | BUG_ON(wc->refs[level] == 0); |
9141 | ||
9142 | if (level == root_item->drop_level) | |
9143 | break; | |
9144 | ||
9145 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 9146 | path->locks[level] = 0; |
2c47e605 YZ |
9147 | WARN_ON(wc->refs[level] != 1); |
9148 | level--; | |
9149 | } | |
9f3a7427 | 9150 | } |
2c47e605 YZ |
9151 | |
9152 | wc->level = level; | |
9153 | wc->shared_level = -1; | |
9154 | wc->stage = DROP_REFERENCE; | |
9155 | wc->update_ref = update_ref; | |
9156 | wc->keep_locks = 0; | |
66d7e7f0 | 9157 | wc->for_reloc = for_reloc; |
0b246afa | 9158 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 9159 | |
d397712b | 9160 | while (1) { |
9d1a2a3a | 9161 | |
2c47e605 YZ |
9162 | ret = walk_down_tree(trans, root, path, wc); |
9163 | if (ret < 0) { | |
9164 | err = ret; | |
20524f02 | 9165 | break; |
2c47e605 | 9166 | } |
9aca1d51 | 9167 | |
2c47e605 YZ |
9168 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
9169 | if (ret < 0) { | |
9170 | err = ret; | |
20524f02 | 9171 | break; |
2c47e605 YZ |
9172 | } |
9173 | ||
9174 | if (ret > 0) { | |
9175 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
9176 | break; |
9177 | } | |
2c47e605 YZ |
9178 | |
9179 | if (wc->stage == DROP_REFERENCE) { | |
9180 | level = wc->level; | |
9181 | btrfs_node_key(path->nodes[level], | |
9182 | &root_item->drop_progress, | |
9183 | path->slots[level]); | |
9184 | root_item->drop_level = level; | |
9185 | } | |
9186 | ||
9187 | BUG_ON(wc->level == 0); | |
3a45bb20 | 9188 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 9189 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
9190 | ret = btrfs_update_root(trans, tree_root, |
9191 | &root->root_key, | |
9192 | root_item); | |
79787eaa | 9193 | if (ret) { |
66642832 | 9194 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9195 | err = ret; |
9196 | goto out_end_trans; | |
9197 | } | |
2c47e605 | 9198 | |
3a45bb20 | 9199 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 9200 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
9201 | btrfs_debug(fs_info, |
9202 | "drop snapshot early exit"); | |
3c8f2422 JB |
9203 | err = -EAGAIN; |
9204 | goto out_free; | |
9205 | } | |
9206 | ||
a22285a6 | 9207 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9208 | if (IS_ERR(trans)) { |
9209 | err = PTR_ERR(trans); | |
9210 | goto out_free; | |
9211 | } | |
3fd0a558 YZ |
9212 | if (block_rsv) |
9213 | trans->block_rsv = block_rsv; | |
c3e69d58 | 9214 | } |
20524f02 | 9215 | } |
b3b4aa74 | 9216 | btrfs_release_path(path); |
79787eaa JM |
9217 | if (err) |
9218 | goto out_end_trans; | |
2c47e605 | 9219 | |
1cd5447e | 9220 | ret = btrfs_del_root(trans, fs_info, &root->root_key); |
79787eaa | 9221 | if (ret) { |
66642832 | 9222 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9223 | goto out_end_trans; |
9224 | } | |
2c47e605 | 9225 | |
76dda93c | 9226 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
9227 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
9228 | NULL, NULL); | |
79787eaa | 9229 | if (ret < 0) { |
66642832 | 9230 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9231 | err = ret; |
9232 | goto out_end_trans; | |
9233 | } else if (ret > 0) { | |
84cd948c JB |
9234 | /* if we fail to delete the orphan item this time |
9235 | * around, it'll get picked up the next time. | |
9236 | * | |
9237 | * The most common failure here is just -ENOENT. | |
9238 | */ | |
9239 | btrfs_del_orphan_item(trans, tree_root, | |
9240 | root->root_key.objectid); | |
76dda93c YZ |
9241 | } |
9242 | } | |
9243 | ||
27cdeb70 | 9244 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 9245 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
9246 | } else { |
9247 | free_extent_buffer(root->node); | |
9248 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 9249 | btrfs_put_fs_root(root); |
76dda93c | 9250 | } |
d29a9f62 | 9251 | root_dropped = true; |
79787eaa | 9252 | out_end_trans: |
3a45bb20 | 9253 | btrfs_end_transaction_throttle(trans); |
79787eaa | 9254 | out_free: |
2c47e605 | 9255 | kfree(wc); |
5caf2a00 | 9256 | btrfs_free_path(path); |
cb1b69f4 | 9257 | out: |
d29a9f62 JB |
9258 | /* |
9259 | * So if we need to stop dropping the snapshot for whatever reason we | |
9260 | * need to make sure to add it back to the dead root list so that we | |
9261 | * keep trying to do the work later. This also cleans up roots if we | |
9262 | * don't have it in the radix (like when we recover after a power fail | |
9263 | * or unmount) so we don't leak memory. | |
9264 | */ | |
897ca819 | 9265 | if (!for_reloc && !root_dropped) |
d29a9f62 | 9266 | btrfs_add_dead_root(root); |
90515e7f | 9267 | if (err && err != -EAGAIN) |
ab8d0fc4 | 9268 | btrfs_handle_fs_error(fs_info, err, NULL); |
2c536799 | 9269 | return err; |
20524f02 | 9270 | } |
9078a3e1 | 9271 | |
2c47e605 YZ |
9272 | /* |
9273 | * drop subtree rooted at tree block 'node'. | |
9274 | * | |
9275 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 9276 | * only used by relocation code |
2c47e605 | 9277 | */ |
f82d02d9 YZ |
9278 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
9279 | struct btrfs_root *root, | |
9280 | struct extent_buffer *node, | |
9281 | struct extent_buffer *parent) | |
9282 | { | |
0b246afa | 9283 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 9284 | struct btrfs_path *path; |
2c47e605 | 9285 | struct walk_control *wc; |
f82d02d9 YZ |
9286 | int level; |
9287 | int parent_level; | |
9288 | int ret = 0; | |
9289 | int wret; | |
9290 | ||
2c47e605 YZ |
9291 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
9292 | ||
f82d02d9 | 9293 | path = btrfs_alloc_path(); |
db5b493a TI |
9294 | if (!path) |
9295 | return -ENOMEM; | |
f82d02d9 | 9296 | |
2c47e605 | 9297 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
9298 | if (!wc) { |
9299 | btrfs_free_path(path); | |
9300 | return -ENOMEM; | |
9301 | } | |
2c47e605 | 9302 | |
b9447ef8 | 9303 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
9304 | parent_level = btrfs_header_level(parent); |
9305 | extent_buffer_get(parent); | |
9306 | path->nodes[parent_level] = parent; | |
9307 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9308 | ||
b9447ef8 | 9309 | btrfs_assert_tree_locked(node); |
f82d02d9 | 9310 | level = btrfs_header_level(node); |
f82d02d9 YZ |
9311 | path->nodes[level] = node; |
9312 | path->slots[level] = 0; | |
bd681513 | 9313 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9314 | |
9315 | wc->refs[parent_level] = 1; | |
9316 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9317 | wc->level = level; | |
9318 | wc->shared_level = -1; | |
9319 | wc->stage = DROP_REFERENCE; | |
9320 | wc->update_ref = 0; | |
9321 | wc->keep_locks = 1; | |
66d7e7f0 | 9322 | wc->for_reloc = 1; |
0b246afa | 9323 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
9324 | |
9325 | while (1) { | |
2c47e605 YZ |
9326 | wret = walk_down_tree(trans, root, path, wc); |
9327 | if (wret < 0) { | |
f82d02d9 | 9328 | ret = wret; |
f82d02d9 | 9329 | break; |
2c47e605 | 9330 | } |
f82d02d9 | 9331 | |
2c47e605 | 9332 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9333 | if (wret < 0) |
9334 | ret = wret; | |
9335 | if (wret != 0) | |
9336 | break; | |
9337 | } | |
9338 | ||
2c47e605 | 9339 | kfree(wc); |
f82d02d9 YZ |
9340 | btrfs_free_path(path); |
9341 | return ret; | |
9342 | } | |
9343 | ||
6202df69 | 9344 | static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c CM |
9345 | { |
9346 | u64 num_devices; | |
fc67c450 | 9347 | u64 stripped; |
e4d8ec0f | 9348 | |
fc67c450 ID |
9349 | /* |
9350 | * if restripe for this chunk_type is on pick target profile and | |
9351 | * return, otherwise do the usual balance | |
9352 | */ | |
6202df69 | 9353 | stripped = get_restripe_target(fs_info, flags); |
fc67c450 ID |
9354 | if (stripped) |
9355 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9356 | |
6202df69 | 9357 | num_devices = fs_info->fs_devices->rw_devices; |
cd02dca5 | 9358 | |
fc67c450 | 9359 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 9360 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
9361 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
9362 | ||
ec44a35c CM |
9363 | if (num_devices == 1) { |
9364 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9365 | stripped = flags & ~stripped; | |
9366 | ||
9367 | /* turn raid0 into single device chunks */ | |
9368 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9369 | return stripped; | |
9370 | ||
9371 | /* turn mirroring into duplication */ | |
9372 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9373 | BTRFS_BLOCK_GROUP_RAID10)) | |
9374 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9375 | } else { |
9376 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9377 | if (flags & stripped) |
9378 | return flags; | |
9379 | ||
9380 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9381 | stripped = flags & ~stripped; | |
9382 | ||
9383 | /* switch duplicated blocks with raid1 */ | |
9384 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9385 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9386 | ||
e3176ca2 | 9387 | /* this is drive concat, leave it alone */ |
ec44a35c | 9388 | } |
e3176ca2 | 9389 | |
ec44a35c CM |
9390 | return flags; |
9391 | } | |
9392 | ||
868f401a | 9393 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9394 | { |
f0486c68 YZ |
9395 | struct btrfs_space_info *sinfo = cache->space_info; |
9396 | u64 num_bytes; | |
199c36ea | 9397 | u64 min_allocable_bytes; |
f0486c68 | 9398 | int ret = -ENOSPC; |
0ef3e66b | 9399 | |
199c36ea MX |
9400 | /* |
9401 | * We need some metadata space and system metadata space for | |
9402 | * allocating chunks in some corner cases until we force to set | |
9403 | * it to be readonly. | |
9404 | */ | |
9405 | if ((sinfo->flags & | |
9406 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9407 | !force) | |
ee22184b | 9408 | min_allocable_bytes = SZ_1M; |
199c36ea MX |
9409 | else |
9410 | min_allocable_bytes = 0; | |
9411 | ||
f0486c68 YZ |
9412 | spin_lock(&sinfo->lock); |
9413 | spin_lock(&cache->lock); | |
61cfea9b W |
9414 | |
9415 | if (cache->ro) { | |
868f401a | 9416 | cache->ro++; |
61cfea9b W |
9417 | ret = 0; |
9418 | goto out; | |
9419 | } | |
9420 | ||
f0486c68 YZ |
9421 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9422 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
9423 | ||
4136135b | 9424 | if (btrfs_space_info_used(sinfo, true) + num_bytes + |
37be25bc | 9425 | min_allocable_bytes <= sinfo->total_bytes) { |
f0486c68 | 9426 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9427 | cache->ro++; |
633c0aad | 9428 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9429 | ret = 0; |
9430 | } | |
61cfea9b | 9431 | out: |
f0486c68 YZ |
9432 | spin_unlock(&cache->lock); |
9433 | spin_unlock(&sinfo->lock); | |
9434 | return ret; | |
9435 | } | |
7d9eb12c | 9436 | |
5e00f193 | 9437 | int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info, |
f0486c68 | 9438 | struct btrfs_block_group_cache *cache) |
c286ac48 | 9439 | |
f0486c68 YZ |
9440 | { |
9441 | struct btrfs_trans_handle *trans; | |
9442 | u64 alloc_flags; | |
9443 | int ret; | |
7d9eb12c | 9444 | |
1bbc621e | 9445 | again: |
5e00f193 | 9446 | trans = btrfs_join_transaction(fs_info->extent_root); |
79787eaa JM |
9447 | if (IS_ERR(trans)) |
9448 | return PTR_ERR(trans); | |
5d4f98a2 | 9449 | |
1bbc621e CM |
9450 | /* |
9451 | * we're not allowed to set block groups readonly after the dirty | |
9452 | * block groups cache has started writing. If it already started, | |
9453 | * back off and let this transaction commit | |
9454 | */ | |
0b246afa | 9455 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c | 9456 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9457 | u64 transid = trans->transid; |
9458 | ||
0b246afa | 9459 | mutex_unlock(&fs_info->ro_block_group_mutex); |
3a45bb20 | 9460 | btrfs_end_transaction(trans); |
1bbc621e | 9461 | |
2ff7e61e | 9462 | ret = btrfs_wait_for_commit(fs_info, transid); |
1bbc621e CM |
9463 | if (ret) |
9464 | return ret; | |
9465 | goto again; | |
9466 | } | |
9467 | ||
153c35b6 CM |
9468 | /* |
9469 | * if we are changing raid levels, try to allocate a corresponding | |
9470 | * block group with the new raid level. | |
9471 | */ | |
0b246afa | 9472 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
153c35b6 | 9473 | if (alloc_flags != cache->flags) { |
2ff7e61e | 9474 | ret = do_chunk_alloc(trans, fs_info, alloc_flags, |
153c35b6 CM |
9475 | CHUNK_ALLOC_FORCE); |
9476 | /* | |
9477 | * ENOSPC is allowed here, we may have enough space | |
9478 | * already allocated at the new raid level to | |
9479 | * carry on | |
9480 | */ | |
9481 | if (ret == -ENOSPC) | |
9482 | ret = 0; | |
9483 | if (ret < 0) | |
9484 | goto out; | |
9485 | } | |
1bbc621e | 9486 | |
868f401a | 9487 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9488 | if (!ret) |
9489 | goto out; | |
2ff7e61e JM |
9490 | alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags); |
9491 | ret = do_chunk_alloc(trans, fs_info, alloc_flags, | |
0e4f8f88 | 9492 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9493 | if (ret < 0) |
9494 | goto out; | |
868f401a | 9495 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9496 | out: |
2f081088 | 9497 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 9498 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
34441361 | 9499 | mutex_lock(&fs_info->chunk_mutex); |
2ff7e61e | 9500 | check_system_chunk(trans, fs_info, alloc_flags); |
34441361 | 9501 | mutex_unlock(&fs_info->chunk_mutex); |
2f081088 | 9502 | } |
0b246afa | 9503 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2f081088 | 9504 | |
3a45bb20 | 9505 | btrfs_end_transaction(trans); |
f0486c68 YZ |
9506 | return ret; |
9507 | } | |
5d4f98a2 | 9508 | |
c87f08ca | 9509 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 9510 | struct btrfs_fs_info *fs_info, u64 type) |
c87f08ca | 9511 | { |
2ff7e61e JM |
9512 | u64 alloc_flags = get_alloc_profile(fs_info, type); |
9513 | ||
9514 | return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE); | |
c87f08ca CM |
9515 | } |
9516 | ||
6d07bcec MX |
9517 | /* |
9518 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9519 | * space_info. takes mirrors into account. |
6d07bcec | 9520 | */ |
633c0aad | 9521 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9522 | { |
9523 | struct btrfs_block_group_cache *block_group; | |
9524 | u64 free_bytes = 0; | |
9525 | int factor; | |
9526 | ||
01327610 | 9527 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
9528 | if (list_empty(&sinfo->ro_bgs)) |
9529 | return 0; | |
9530 | ||
9531 | spin_lock(&sinfo->lock); | |
9532 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9533 | spin_lock(&block_group->lock); |
9534 | ||
9535 | if (!block_group->ro) { | |
9536 | spin_unlock(&block_group->lock); | |
9537 | continue; | |
9538 | } | |
9539 | ||
9540 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9541 | BTRFS_BLOCK_GROUP_RAID10 | | |
9542 | BTRFS_BLOCK_GROUP_DUP)) | |
9543 | factor = 2; | |
9544 | else | |
9545 | factor = 1; | |
9546 | ||
9547 | free_bytes += (block_group->key.offset - | |
9548 | btrfs_block_group_used(&block_group->item)) * | |
9549 | factor; | |
9550 | ||
9551 | spin_unlock(&block_group->lock); | |
9552 | } | |
6d07bcec MX |
9553 | spin_unlock(&sinfo->lock); |
9554 | ||
9555 | return free_bytes; | |
9556 | } | |
9557 | ||
2ff7e61e | 9558 | void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9559 | { |
f0486c68 YZ |
9560 | struct btrfs_space_info *sinfo = cache->space_info; |
9561 | u64 num_bytes; | |
9562 | ||
9563 | BUG_ON(!cache->ro); | |
9564 | ||
9565 | spin_lock(&sinfo->lock); | |
9566 | spin_lock(&cache->lock); | |
868f401a Z |
9567 | if (!--cache->ro) { |
9568 | num_bytes = cache->key.offset - cache->reserved - | |
9569 | cache->pinned - cache->bytes_super - | |
9570 | btrfs_block_group_used(&cache->item); | |
9571 | sinfo->bytes_readonly -= num_bytes; | |
9572 | list_del_init(&cache->ro_list); | |
9573 | } | |
f0486c68 YZ |
9574 | spin_unlock(&cache->lock); |
9575 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9576 | } |
9577 | ||
ba1bf481 JB |
9578 | /* |
9579 | * checks to see if its even possible to relocate this block group. | |
9580 | * | |
9581 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9582 | * ok to go ahead and try. | |
9583 | */ | |
6bccf3ab | 9584 | int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr) |
1a40e23b | 9585 | { |
6bccf3ab | 9586 | struct btrfs_root *root = fs_info->extent_root; |
ba1bf481 JB |
9587 | struct btrfs_block_group_cache *block_group; |
9588 | struct btrfs_space_info *space_info; | |
0b246afa | 9589 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
ba1bf481 | 9590 | struct btrfs_device *device; |
6df9a95e | 9591 | struct btrfs_trans_handle *trans; |
cdcb725c | 9592 | u64 min_free; |
6719db6a JB |
9593 | u64 dev_min = 1; |
9594 | u64 dev_nr = 0; | |
4a5e98f5 | 9595 | u64 target; |
0305bc27 | 9596 | int debug; |
cdcb725c | 9597 | int index; |
ba1bf481 JB |
9598 | int full = 0; |
9599 | int ret = 0; | |
1a40e23b | 9600 | |
0b246afa | 9601 | debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG); |
0305bc27 | 9602 | |
0b246afa | 9603 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
1a40e23b | 9604 | |
ba1bf481 | 9605 | /* odd, couldn't find the block group, leave it alone */ |
0305bc27 QW |
9606 | if (!block_group) { |
9607 | if (debug) | |
0b246afa | 9608 | btrfs_warn(fs_info, |
0305bc27 QW |
9609 | "can't find block group for bytenr %llu", |
9610 | bytenr); | |
ba1bf481 | 9611 | return -1; |
0305bc27 | 9612 | } |
1a40e23b | 9613 | |
cdcb725c | 9614 | min_free = btrfs_block_group_used(&block_group->item); |
9615 | ||
ba1bf481 | 9616 | /* no bytes used, we're good */ |
cdcb725c | 9617 | if (!min_free) |
1a40e23b ZY |
9618 | goto out; |
9619 | ||
ba1bf481 JB |
9620 | space_info = block_group->space_info; |
9621 | spin_lock(&space_info->lock); | |
17d217fe | 9622 | |
ba1bf481 | 9623 | full = space_info->full; |
17d217fe | 9624 | |
ba1bf481 JB |
9625 | /* |
9626 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9627 | * relocate it unless we're able to allocate a new chunk below. |
9628 | * | |
9629 | * Otherwise, we need to make sure we have room in the space to handle | |
9630 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9631 | */ |
7ce618db | 9632 | if ((space_info->total_bytes != block_group->key.offset) && |
4136135b LB |
9633 | (btrfs_space_info_used(space_info, false) + min_free < |
9634 | space_info->total_bytes)) { | |
ba1bf481 JB |
9635 | spin_unlock(&space_info->lock); |
9636 | goto out; | |
17d217fe | 9637 | } |
ba1bf481 | 9638 | spin_unlock(&space_info->lock); |
ea8c2819 | 9639 | |
ba1bf481 JB |
9640 | /* |
9641 | * ok we don't have enough space, but maybe we have free space on our | |
9642 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9643 | * alloc devices and guess if we have enough space. if this block |
9644 | * group is going to be restriped, run checks against the target | |
9645 | * profile instead of the current one. | |
ba1bf481 JB |
9646 | */ |
9647 | ret = -1; | |
ea8c2819 | 9648 | |
cdcb725c | 9649 | /* |
9650 | * index: | |
9651 | * 0: raid10 | |
9652 | * 1: raid1 | |
9653 | * 2: dup | |
9654 | * 3: raid0 | |
9655 | * 4: single | |
9656 | */ | |
0b246afa | 9657 | target = get_restripe_target(fs_info, block_group->flags); |
4a5e98f5 | 9658 | if (target) { |
31e50229 | 9659 | index = __get_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9660 | } else { |
9661 | /* | |
9662 | * this is just a balance, so if we were marked as full | |
9663 | * we know there is no space for a new chunk | |
9664 | */ | |
0305bc27 QW |
9665 | if (full) { |
9666 | if (debug) | |
0b246afa JM |
9667 | btrfs_warn(fs_info, |
9668 | "no space to alloc new chunk for block group %llu", | |
9669 | block_group->key.objectid); | |
4a5e98f5 | 9670 | goto out; |
0305bc27 | 9671 | } |
4a5e98f5 ID |
9672 | |
9673 | index = get_block_group_index(block_group); | |
9674 | } | |
9675 | ||
e6ec716f | 9676 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9677 | dev_min = 4; |
6719db6a JB |
9678 | /* Divide by 2 */ |
9679 | min_free >>= 1; | |
e6ec716f | 9680 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9681 | dev_min = 2; |
e6ec716f | 9682 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9683 | /* Multiply by 2 */ |
9684 | min_free <<= 1; | |
e6ec716f | 9685 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9686 | dev_min = fs_devices->rw_devices; |
47c5713f | 9687 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9688 | } |
9689 | ||
6df9a95e JB |
9690 | /* We need to do this so that we can look at pending chunks */ |
9691 | trans = btrfs_join_transaction(root); | |
9692 | if (IS_ERR(trans)) { | |
9693 | ret = PTR_ERR(trans); | |
9694 | goto out; | |
9695 | } | |
9696 | ||
0b246afa | 9697 | mutex_lock(&fs_info->chunk_mutex); |
ba1bf481 | 9698 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
7bfc837d | 9699 | u64 dev_offset; |
56bec294 | 9700 | |
ba1bf481 JB |
9701 | /* |
9702 | * check to make sure we can actually find a chunk with enough | |
9703 | * space to fit our block group in. | |
9704 | */ | |
63a212ab SB |
9705 | if (device->total_bytes > device->bytes_used + min_free && |
9706 | !device->is_tgtdev_for_dev_replace) { | |
6df9a95e | 9707 | ret = find_free_dev_extent(trans, device, min_free, |
7bfc837d | 9708 | &dev_offset, NULL); |
ba1bf481 | 9709 | if (!ret) |
cdcb725c | 9710 | dev_nr++; |
9711 | ||
9712 | if (dev_nr >= dev_min) | |
73e48b27 | 9713 | break; |
cdcb725c | 9714 | |
ba1bf481 | 9715 | ret = -1; |
725c8463 | 9716 | } |
edbd8d4e | 9717 | } |
0305bc27 | 9718 | if (debug && ret == -1) |
0b246afa JM |
9719 | btrfs_warn(fs_info, |
9720 | "no space to allocate a new chunk for block group %llu", | |
9721 | block_group->key.objectid); | |
9722 | mutex_unlock(&fs_info->chunk_mutex); | |
3a45bb20 | 9723 | btrfs_end_transaction(trans); |
edbd8d4e | 9724 | out: |
ba1bf481 | 9725 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9726 | return ret; |
9727 | } | |
9728 | ||
6bccf3ab JM |
9729 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
9730 | struct btrfs_path *path, | |
9731 | struct btrfs_key *key) | |
0b86a832 | 9732 | { |
6bccf3ab | 9733 | struct btrfs_root *root = fs_info->extent_root; |
925baedd | 9734 | int ret = 0; |
0b86a832 CM |
9735 | struct btrfs_key found_key; |
9736 | struct extent_buffer *leaf; | |
9737 | int slot; | |
edbd8d4e | 9738 | |
0b86a832 CM |
9739 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9740 | if (ret < 0) | |
925baedd CM |
9741 | goto out; |
9742 | ||
d397712b | 9743 | while (1) { |
0b86a832 | 9744 | slot = path->slots[0]; |
edbd8d4e | 9745 | leaf = path->nodes[0]; |
0b86a832 CM |
9746 | if (slot >= btrfs_header_nritems(leaf)) { |
9747 | ret = btrfs_next_leaf(root, path); | |
9748 | if (ret == 0) | |
9749 | continue; | |
9750 | if (ret < 0) | |
925baedd | 9751 | goto out; |
0b86a832 | 9752 | break; |
edbd8d4e | 9753 | } |
0b86a832 | 9754 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9755 | |
0b86a832 | 9756 | if (found_key.objectid >= key->objectid && |
925baedd | 9757 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6fb37b75 LB |
9758 | struct extent_map_tree *em_tree; |
9759 | struct extent_map *em; | |
9760 | ||
9761 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
9762 | read_lock(&em_tree->lock); | |
9763 | em = lookup_extent_mapping(em_tree, found_key.objectid, | |
9764 | found_key.offset); | |
9765 | read_unlock(&em_tree->lock); | |
9766 | if (!em) { | |
0b246afa | 9767 | btrfs_err(fs_info, |
6fb37b75 LB |
9768 | "logical %llu len %llu found bg but no related chunk", |
9769 | found_key.objectid, found_key.offset); | |
9770 | ret = -ENOENT; | |
9771 | } else { | |
9772 | ret = 0; | |
9773 | } | |
187ee58c | 9774 | free_extent_map(em); |
925baedd CM |
9775 | goto out; |
9776 | } | |
0b86a832 | 9777 | path->slots[0]++; |
edbd8d4e | 9778 | } |
925baedd | 9779 | out: |
0b86a832 | 9780 | return ret; |
edbd8d4e CM |
9781 | } |
9782 | ||
0af3d00b JB |
9783 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
9784 | { | |
9785 | struct btrfs_block_group_cache *block_group; | |
9786 | u64 last = 0; | |
9787 | ||
9788 | while (1) { | |
9789 | struct inode *inode; | |
9790 | ||
9791 | block_group = btrfs_lookup_first_block_group(info, last); | |
9792 | while (block_group) { | |
9793 | spin_lock(&block_group->lock); | |
9794 | if (block_group->iref) | |
9795 | break; | |
9796 | spin_unlock(&block_group->lock); | |
2ff7e61e | 9797 | block_group = next_block_group(info, block_group); |
0af3d00b JB |
9798 | } |
9799 | if (!block_group) { | |
9800 | if (last == 0) | |
9801 | break; | |
9802 | last = 0; | |
9803 | continue; | |
9804 | } | |
9805 | ||
9806 | inode = block_group->inode; | |
9807 | block_group->iref = 0; | |
9808 | block_group->inode = NULL; | |
9809 | spin_unlock(&block_group->lock); | |
f3bca802 | 9810 | ASSERT(block_group->io_ctl.inode == NULL); |
0af3d00b JB |
9811 | iput(inode); |
9812 | last = block_group->key.objectid + block_group->key.offset; | |
9813 | btrfs_put_block_group(block_group); | |
9814 | } | |
9815 | } | |
9816 | ||
5cdd7db6 FM |
9817 | /* |
9818 | * Must be called only after stopping all workers, since we could have block | |
9819 | * group caching kthreads running, and therefore they could race with us if we | |
9820 | * freed the block groups before stopping them. | |
9821 | */ | |
1a40e23b ZY |
9822 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9823 | { | |
9824 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 9825 | struct btrfs_space_info *space_info; |
11833d66 | 9826 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
9827 | struct rb_node *n; |
9828 | ||
9e351cc8 | 9829 | down_write(&info->commit_root_sem); |
11833d66 YZ |
9830 | while (!list_empty(&info->caching_block_groups)) { |
9831 | caching_ctl = list_entry(info->caching_block_groups.next, | |
9832 | struct btrfs_caching_control, list); | |
9833 | list_del(&caching_ctl->list); | |
9834 | put_caching_control(caching_ctl); | |
9835 | } | |
9e351cc8 | 9836 | up_write(&info->commit_root_sem); |
11833d66 | 9837 | |
47ab2a6c JB |
9838 | spin_lock(&info->unused_bgs_lock); |
9839 | while (!list_empty(&info->unused_bgs)) { | |
9840 | block_group = list_first_entry(&info->unused_bgs, | |
9841 | struct btrfs_block_group_cache, | |
9842 | bg_list); | |
9843 | list_del_init(&block_group->bg_list); | |
9844 | btrfs_put_block_group(block_group); | |
9845 | } | |
9846 | spin_unlock(&info->unused_bgs_lock); | |
9847 | ||
1a40e23b ZY |
9848 | spin_lock(&info->block_group_cache_lock); |
9849 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
9850 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
9851 | cache_node); | |
1a40e23b ZY |
9852 | rb_erase(&block_group->cache_node, |
9853 | &info->block_group_cache_tree); | |
01eacb27 | 9854 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
9855 | spin_unlock(&info->block_group_cache_lock); |
9856 | ||
80eb234a | 9857 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 9858 | list_del(&block_group->list); |
80eb234a | 9859 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 9860 | |
3c14874a JB |
9861 | /* |
9862 | * We haven't cached this block group, which means we could | |
9863 | * possibly have excluded extents on this block group. | |
9864 | */ | |
36cce922 JB |
9865 | if (block_group->cached == BTRFS_CACHE_NO || |
9866 | block_group->cached == BTRFS_CACHE_ERROR) | |
2ff7e61e | 9867 | free_excluded_extents(info, block_group); |
3c14874a | 9868 | |
817d52f8 | 9869 | btrfs_remove_free_space_cache(block_group); |
5cdd7db6 | 9870 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); |
f3bca802 LB |
9871 | ASSERT(list_empty(&block_group->dirty_list)); |
9872 | ASSERT(list_empty(&block_group->io_list)); | |
9873 | ASSERT(list_empty(&block_group->bg_list)); | |
9874 | ASSERT(atomic_read(&block_group->count) == 1); | |
11dfe35a | 9875 | btrfs_put_block_group(block_group); |
d899e052 YZ |
9876 | |
9877 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
9878 | } |
9879 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
9880 | |
9881 | /* now that all the block groups are freed, go through and | |
9882 | * free all the space_info structs. This is only called during | |
9883 | * the final stages of unmount, and so we know nobody is | |
9884 | * using them. We call synchronize_rcu() once before we start, | |
9885 | * just to be on the safe side. | |
9886 | */ | |
9887 | synchronize_rcu(); | |
9888 | ||
8929ecfa YZ |
9889 | release_global_block_rsv(info); |
9890 | ||
67871254 | 9891 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
9892 | int i; |
9893 | ||
4184ea7f CM |
9894 | space_info = list_entry(info->space_info.next, |
9895 | struct btrfs_space_info, | |
9896 | list); | |
d555b6c3 JB |
9897 | |
9898 | /* | |
9899 | * Do not hide this behind enospc_debug, this is actually | |
9900 | * important and indicates a real bug if this happens. | |
9901 | */ | |
9902 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
b069e0c3 | 9903 | space_info->bytes_reserved > 0 || |
d555b6c3 | 9904 | space_info->bytes_may_use > 0)) |
ab8d0fc4 | 9905 | dump_space_info(info, space_info, 0, 0); |
4184ea7f | 9906 | list_del(&space_info->list); |
6ab0a202 JM |
9907 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
9908 | struct kobject *kobj; | |
c1895442 JM |
9909 | kobj = space_info->block_group_kobjs[i]; |
9910 | space_info->block_group_kobjs[i] = NULL; | |
9911 | if (kobj) { | |
6ab0a202 JM |
9912 | kobject_del(kobj); |
9913 | kobject_put(kobj); | |
9914 | } | |
9915 | } | |
9916 | kobject_del(&space_info->kobj); | |
9917 | kobject_put(&space_info->kobj); | |
4184ea7f | 9918 | } |
1a40e23b ZY |
9919 | return 0; |
9920 | } | |
9921 | ||
c434d21c | 9922 | static void link_block_group(struct btrfs_block_group_cache *cache) |
b742bb82 | 9923 | { |
c434d21c | 9924 | struct btrfs_space_info *space_info = cache->space_info; |
b742bb82 | 9925 | int index = get_block_group_index(cache); |
ed55b6ac | 9926 | bool first = false; |
b742bb82 YZ |
9927 | |
9928 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
9929 | if (list_empty(&space_info->block_groups[index])) |
9930 | first = true; | |
9931 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
9932 | up_write(&space_info->groups_sem); | |
9933 | ||
9934 | if (first) { | |
c1895442 | 9935 | struct raid_kobject *rkobj; |
6ab0a202 JM |
9936 | int ret; |
9937 | ||
c1895442 JM |
9938 | rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
9939 | if (!rkobj) | |
9940 | goto out_err; | |
9941 | rkobj->raid_type = index; | |
9942 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
9943 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
9944 | "%s", get_raid_name(index)); | |
6ab0a202 | 9945 | if (ret) { |
c1895442 JM |
9946 | kobject_put(&rkobj->kobj); |
9947 | goto out_err; | |
6ab0a202 | 9948 | } |
c1895442 | 9949 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 9950 | } |
c1895442 JM |
9951 | |
9952 | return; | |
9953 | out_err: | |
ab8d0fc4 JM |
9954 | btrfs_warn(cache->fs_info, |
9955 | "failed to add kobject for block cache, ignoring"); | |
b742bb82 YZ |
9956 | } |
9957 | ||
920e4a58 | 9958 | static struct btrfs_block_group_cache * |
2ff7e61e JM |
9959 | btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info, |
9960 | u64 start, u64 size) | |
920e4a58 MX |
9961 | { |
9962 | struct btrfs_block_group_cache *cache; | |
9963 | ||
9964 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
9965 | if (!cache) | |
9966 | return NULL; | |
9967 | ||
9968 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
9969 | GFP_NOFS); | |
9970 | if (!cache->free_space_ctl) { | |
9971 | kfree(cache); | |
9972 | return NULL; | |
9973 | } | |
9974 | ||
9975 | cache->key.objectid = start; | |
9976 | cache->key.offset = size; | |
9977 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
9978 | ||
0b246afa | 9979 | cache->fs_info = fs_info; |
e4ff5fb5 | 9980 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); |
1e144fb8 OS |
9981 | set_free_space_tree_thresholds(cache); |
9982 | ||
920e4a58 MX |
9983 | atomic_set(&cache->count, 1); |
9984 | spin_lock_init(&cache->lock); | |
e570fd27 | 9985 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
9986 | INIT_LIST_HEAD(&cache->list); |
9987 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 9988 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 9989 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 9990 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 9991 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 9992 | btrfs_init_free_space_ctl(cache); |
04216820 | 9993 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 9994 | mutex_init(&cache->free_space_lock); |
0966a7b1 | 9995 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); |
920e4a58 MX |
9996 | |
9997 | return cache; | |
9998 | } | |
9999 | ||
5b4aacef | 10000 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
9078a3e1 CM |
10001 | { |
10002 | struct btrfs_path *path; | |
10003 | int ret; | |
9078a3e1 | 10004 | struct btrfs_block_group_cache *cache; |
6324fbf3 | 10005 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
10006 | struct btrfs_key key; |
10007 | struct btrfs_key found_key; | |
5f39d397 | 10008 | struct extent_buffer *leaf; |
0af3d00b JB |
10009 | int need_clear = 0; |
10010 | u64 cache_gen; | |
49303381 LB |
10011 | u64 feature; |
10012 | int mixed; | |
10013 | ||
10014 | feature = btrfs_super_incompat_flags(info->super_copy); | |
10015 | mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS); | |
96b5179d | 10016 | |
9078a3e1 | 10017 | key.objectid = 0; |
0b86a832 | 10018 | key.offset = 0; |
962a298f | 10019 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
10020 | path = btrfs_alloc_path(); |
10021 | if (!path) | |
10022 | return -ENOMEM; | |
e4058b54 | 10023 | path->reada = READA_FORWARD; |
9078a3e1 | 10024 | |
0b246afa JM |
10025 | cache_gen = btrfs_super_cache_generation(info->super_copy); |
10026 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
10027 | btrfs_super_generation(info->super_copy) != cache_gen) | |
0af3d00b | 10028 | need_clear = 1; |
0b246afa | 10029 | if (btrfs_test_opt(info, CLEAR_CACHE)) |
88c2ba3b | 10030 | need_clear = 1; |
0af3d00b | 10031 | |
d397712b | 10032 | while (1) { |
6bccf3ab | 10033 | ret = find_first_block_group(info, path, &key); |
b742bb82 YZ |
10034 | if (ret > 0) |
10035 | break; | |
0b86a832 CM |
10036 | if (ret != 0) |
10037 | goto error; | |
920e4a58 | 10038 | |
5f39d397 CM |
10039 | leaf = path->nodes[0]; |
10040 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 | 10041 | |
2ff7e61e | 10042 | cache = btrfs_create_block_group_cache(info, found_key.objectid, |
920e4a58 | 10043 | found_key.offset); |
9078a3e1 | 10044 | if (!cache) { |
0b86a832 | 10045 | ret = -ENOMEM; |
f0486c68 | 10046 | goto error; |
9078a3e1 | 10047 | } |
96303081 | 10048 | |
cf7c1ef6 LB |
10049 | if (need_clear) { |
10050 | /* | |
10051 | * When we mount with old space cache, we need to | |
10052 | * set BTRFS_DC_CLEAR and set dirty flag. | |
10053 | * | |
10054 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
10055 | * truncate the old free space cache inode and | |
10056 | * setup a new one. | |
10057 | * b) Setting 'dirty flag' makes sure that we flush | |
10058 | * the new space cache info onto disk. | |
10059 | */ | |
0b246afa | 10060 | if (btrfs_test_opt(info, SPACE_CACHE)) |
ce93ec54 | 10061 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 10062 | } |
0af3d00b | 10063 | |
5f39d397 CM |
10064 | read_extent_buffer(leaf, &cache->item, |
10065 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
10066 | sizeof(cache->item)); | |
920e4a58 | 10067 | cache->flags = btrfs_block_group_flags(&cache->item); |
49303381 LB |
10068 | if (!mixed && |
10069 | ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && | |
10070 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
10071 | btrfs_err(info, | |
10072 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
10073 | cache->key.objectid); | |
10074 | ret = -EINVAL; | |
10075 | goto error; | |
10076 | } | |
0b86a832 | 10077 | |
9078a3e1 | 10078 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 10079 | btrfs_release_path(path); |
34d52cb6 | 10080 | |
3c14874a JB |
10081 | /* |
10082 | * We need to exclude the super stripes now so that the space | |
10083 | * info has super bytes accounted for, otherwise we'll think | |
10084 | * we have more space than we actually do. | |
10085 | */ | |
2ff7e61e | 10086 | ret = exclude_super_stripes(info, cache); |
835d974f JB |
10087 | if (ret) { |
10088 | /* | |
10089 | * We may have excluded something, so call this just in | |
10090 | * case. | |
10091 | */ | |
2ff7e61e | 10092 | free_excluded_extents(info, cache); |
920e4a58 | 10093 | btrfs_put_block_group(cache); |
835d974f JB |
10094 | goto error; |
10095 | } | |
3c14874a | 10096 | |
817d52f8 JB |
10097 | /* |
10098 | * check for two cases, either we are full, and therefore | |
10099 | * don't need to bother with the caching work since we won't | |
10100 | * find any space, or we are empty, and we can just add all | |
10101 | * the space in and be done with it. This saves us _alot_ of | |
10102 | * time, particularly in the full case. | |
10103 | */ | |
10104 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 10105 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10106 | cache->cached = BTRFS_CACHE_FINISHED; |
2ff7e61e | 10107 | free_excluded_extents(info, cache); |
817d52f8 | 10108 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 10109 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10110 | cache->cached = BTRFS_CACHE_FINISHED; |
0b246afa | 10111 | add_new_free_space(cache, info, |
817d52f8 JB |
10112 | found_key.objectid, |
10113 | found_key.objectid + | |
10114 | found_key.offset); | |
2ff7e61e | 10115 | free_excluded_extents(info, cache); |
817d52f8 | 10116 | } |
96b5179d | 10117 | |
0b246afa | 10118 | ret = btrfs_add_block_group_cache(info, cache); |
8c579fe7 JB |
10119 | if (ret) { |
10120 | btrfs_remove_free_space_cache(cache); | |
10121 | btrfs_put_block_group(cache); | |
10122 | goto error; | |
10123 | } | |
10124 | ||
0b246afa | 10125 | trace_btrfs_add_block_group(info, cache, 0); |
d2006e6d NB |
10126 | update_space_info(info, cache->flags, found_key.offset, |
10127 | btrfs_block_group_used(&cache->item), | |
10128 | cache->bytes_super, &space_info); | |
8c579fe7 | 10129 | |
6324fbf3 | 10130 | cache->space_info = space_info; |
1b2da372 | 10131 | |
c434d21c | 10132 | link_block_group(cache); |
0f9dd46c | 10133 | |
0b246afa | 10134 | set_avail_alloc_bits(info, cache->flags); |
2ff7e61e | 10135 | if (btrfs_chunk_readonly(info, cache->key.objectid)) { |
868f401a | 10136 | inc_block_group_ro(cache, 1); |
47ab2a6c JB |
10137 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
10138 | spin_lock(&info->unused_bgs_lock); | |
10139 | /* Should always be true but just in case. */ | |
10140 | if (list_empty(&cache->bg_list)) { | |
10141 | btrfs_get_block_group(cache); | |
10142 | list_add_tail(&cache->bg_list, | |
10143 | &info->unused_bgs); | |
10144 | } | |
10145 | spin_unlock(&info->unused_bgs_lock); | |
10146 | } | |
9078a3e1 | 10147 | } |
b742bb82 | 10148 | |
0b246afa | 10149 | list_for_each_entry_rcu(space_info, &info->space_info, list) { |
2ff7e61e | 10150 | if (!(get_alloc_profile(info, space_info->flags) & |
b742bb82 YZ |
10151 | (BTRFS_BLOCK_GROUP_RAID10 | |
10152 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
10153 | BTRFS_BLOCK_GROUP_RAID5 | |
10154 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
10155 | BTRFS_BLOCK_GROUP_DUP))) |
10156 | continue; | |
10157 | /* | |
10158 | * avoid allocating from un-mirrored block group if there are | |
10159 | * mirrored block groups. | |
10160 | */ | |
1095cc0d | 10161 | list_for_each_entry(cache, |
10162 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
10163 | list) | |
868f401a | 10164 | inc_block_group_ro(cache, 1); |
1095cc0d | 10165 | list_for_each_entry(cache, |
10166 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
10167 | list) | |
868f401a | 10168 | inc_block_group_ro(cache, 1); |
9078a3e1 | 10169 | } |
f0486c68 YZ |
10170 | |
10171 | init_global_block_rsv(info); | |
0b86a832 CM |
10172 | ret = 0; |
10173 | error: | |
9078a3e1 | 10174 | btrfs_free_path(path); |
0b86a832 | 10175 | return ret; |
9078a3e1 | 10176 | } |
6324fbf3 | 10177 | |
ea658bad | 10178 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans, |
2ff7e61e | 10179 | struct btrfs_fs_info *fs_info) |
ea658bad JB |
10180 | { |
10181 | struct btrfs_block_group_cache *block_group, *tmp; | |
0b246afa | 10182 | struct btrfs_root *extent_root = fs_info->extent_root; |
ea658bad JB |
10183 | struct btrfs_block_group_item item; |
10184 | struct btrfs_key key; | |
10185 | int ret = 0; | |
d9a0540a | 10186 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
ea658bad | 10187 | |
d9a0540a | 10188 | trans->can_flush_pending_bgs = false; |
47ab2a6c | 10189 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { |
ea658bad | 10190 | if (ret) |
c92f6be3 | 10191 | goto next; |
ea658bad JB |
10192 | |
10193 | spin_lock(&block_group->lock); | |
10194 | memcpy(&item, &block_group->item, sizeof(item)); | |
10195 | memcpy(&key, &block_group->key, sizeof(key)); | |
10196 | spin_unlock(&block_group->lock); | |
10197 | ||
10198 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
10199 | sizeof(item)); | |
10200 | if (ret) | |
66642832 | 10201 | btrfs_abort_transaction(trans, ret); |
0b246afa JM |
10202 | ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid, |
10203 | key.offset); | |
6df9a95e | 10204 | if (ret) |
66642832 | 10205 | btrfs_abort_transaction(trans, ret); |
0b246afa | 10206 | add_block_group_free_space(trans, fs_info, block_group); |
1e144fb8 | 10207 | /* already aborted the transaction if it failed. */ |
c92f6be3 FM |
10208 | next: |
10209 | list_del_init(&block_group->bg_list); | |
ea658bad | 10210 | } |
d9a0540a | 10211 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
ea658bad JB |
10212 | } |
10213 | ||
6324fbf3 | 10214 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 10215 | struct btrfs_fs_info *fs_info, u64 bytes_used, |
0174484d | 10216 | u64 type, u64 chunk_offset, u64 size) |
6324fbf3 | 10217 | { |
6324fbf3 | 10218 | struct btrfs_block_group_cache *cache; |
0b246afa | 10219 | int ret; |
6324fbf3 | 10220 | |
0b246afa | 10221 | btrfs_set_log_full_commit(fs_info, trans); |
e02119d5 | 10222 | |
2ff7e61e | 10223 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size); |
0f9dd46c JB |
10224 | if (!cache) |
10225 | return -ENOMEM; | |
34d52cb6 | 10226 | |
6324fbf3 | 10227 | btrfs_set_block_group_used(&cache->item, bytes_used); |
0174484d NB |
10228 | btrfs_set_block_group_chunk_objectid(&cache->item, |
10229 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
6324fbf3 CM |
10230 | btrfs_set_block_group_flags(&cache->item, type); |
10231 | ||
920e4a58 | 10232 | cache->flags = type; |
11833d66 | 10233 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10234 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 10235 | cache->needs_free_space = 1; |
2ff7e61e | 10236 | ret = exclude_super_stripes(fs_info, cache); |
835d974f JB |
10237 | if (ret) { |
10238 | /* | |
10239 | * We may have excluded something, so call this just in | |
10240 | * case. | |
10241 | */ | |
2ff7e61e | 10242 | free_excluded_extents(fs_info, cache); |
920e4a58 | 10243 | btrfs_put_block_group(cache); |
835d974f JB |
10244 | return ret; |
10245 | } | |
96303081 | 10246 | |
0b246afa | 10247 | add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size); |
817d52f8 | 10248 | |
2ff7e61e | 10249 | free_excluded_extents(fs_info, cache); |
11833d66 | 10250 | |
d0bd4560 | 10251 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 10252 | if (btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
10253 | u64 new_bytes_used = size - bytes_used; |
10254 | ||
10255 | bytes_used += new_bytes_used >> 1; | |
2ff7e61e | 10256 | fragment_free_space(cache); |
d0bd4560 JB |
10257 | } |
10258 | #endif | |
2e6e5183 | 10259 | /* |
2be12ef7 NB |
10260 | * Ensure the corresponding space_info object is created and |
10261 | * assigned to our block group. We want our bg to be added to the rbtree | |
10262 | * with its ->space_info set. | |
2e6e5183 | 10263 | */ |
2be12ef7 NB |
10264 | cache->space_info = __find_space_info(fs_info, cache->flags); |
10265 | if (!cache->space_info) { | |
10266 | ret = create_space_info(fs_info, cache->flags, | |
10267 | &cache->space_info); | |
10268 | if (ret) { | |
10269 | btrfs_remove_free_space_cache(cache); | |
10270 | btrfs_put_block_group(cache); | |
10271 | return ret; | |
10272 | } | |
2e6e5183 FM |
10273 | } |
10274 | ||
0b246afa | 10275 | ret = btrfs_add_block_group_cache(fs_info, cache); |
8c579fe7 JB |
10276 | if (ret) { |
10277 | btrfs_remove_free_space_cache(cache); | |
10278 | btrfs_put_block_group(cache); | |
10279 | return ret; | |
10280 | } | |
10281 | ||
2e6e5183 FM |
10282 | /* |
10283 | * Now that our block group has its ->space_info set and is inserted in | |
10284 | * the rbtree, update the space info's counters. | |
10285 | */ | |
0b246afa | 10286 | trace_btrfs_add_block_group(fs_info, cache, 1); |
d2006e6d | 10287 | update_space_info(fs_info, cache->flags, size, bytes_used, |
e40edf2d | 10288 | cache->bytes_super, &cache->space_info); |
0b246afa | 10289 | update_global_block_rsv(fs_info); |
1b2da372 | 10290 | |
c434d21c | 10291 | link_block_group(cache); |
6324fbf3 | 10292 | |
47ab2a6c | 10293 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
6324fbf3 | 10294 | |
0b246afa | 10295 | set_avail_alloc_bits(fs_info, type); |
6324fbf3 CM |
10296 | return 0; |
10297 | } | |
1a40e23b | 10298 | |
10ea00f5 ID |
10299 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
10300 | { | |
899c81ea ID |
10301 | u64 extra_flags = chunk_to_extended(flags) & |
10302 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 10303 | |
de98ced9 | 10304 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10305 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
10306 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10307 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10308 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10309 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10310 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 10311 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10312 | } |
10313 | ||
1a40e23b | 10314 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
6bccf3ab | 10315 | struct btrfs_fs_info *fs_info, u64 group_start, |
04216820 | 10316 | struct extent_map *em) |
1a40e23b | 10317 | { |
6bccf3ab | 10318 | struct btrfs_root *root = fs_info->extent_root; |
1a40e23b ZY |
10319 | struct btrfs_path *path; |
10320 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 10321 | struct btrfs_free_cluster *cluster; |
0b246afa | 10322 | struct btrfs_root *tree_root = fs_info->tree_root; |
1a40e23b | 10323 | struct btrfs_key key; |
0af3d00b | 10324 | struct inode *inode; |
c1895442 | 10325 | struct kobject *kobj = NULL; |
1a40e23b | 10326 | int ret; |
10ea00f5 | 10327 | int index; |
89a55897 | 10328 | int factor; |
4f69cb98 | 10329 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 10330 | bool remove_em; |
1a40e23b | 10331 | |
6bccf3ab | 10332 | block_group = btrfs_lookup_block_group(fs_info, group_start); |
1a40e23b | 10333 | BUG_ON(!block_group); |
c146afad | 10334 | BUG_ON(!block_group->ro); |
1a40e23b | 10335 | |
9f7c43c9 | 10336 | /* |
10337 | * Free the reserved super bytes from this block group before | |
10338 | * remove it. | |
10339 | */ | |
2ff7e61e | 10340 | free_excluded_extents(fs_info, block_group); |
fd708b81 JB |
10341 | btrfs_free_ref_tree_range(fs_info, block_group->key.objectid, |
10342 | block_group->key.offset); | |
9f7c43c9 | 10343 | |
1a40e23b | 10344 | memcpy(&key, &block_group->key, sizeof(key)); |
10ea00f5 | 10345 | index = get_block_group_index(block_group); |
89a55897 JB |
10346 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
10347 | BTRFS_BLOCK_GROUP_RAID1 | | |
10348 | BTRFS_BLOCK_GROUP_RAID10)) | |
10349 | factor = 2; | |
10350 | else | |
10351 | factor = 1; | |
1a40e23b | 10352 | |
44fb5511 | 10353 | /* make sure this block group isn't part of an allocation cluster */ |
0b246afa | 10354 | cluster = &fs_info->data_alloc_cluster; |
44fb5511 CM |
10355 | spin_lock(&cluster->refill_lock); |
10356 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10357 | spin_unlock(&cluster->refill_lock); | |
10358 | ||
10359 | /* | |
10360 | * make sure this block group isn't part of a metadata | |
10361 | * allocation cluster | |
10362 | */ | |
0b246afa | 10363 | cluster = &fs_info->meta_alloc_cluster; |
44fb5511 CM |
10364 | spin_lock(&cluster->refill_lock); |
10365 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10366 | spin_unlock(&cluster->refill_lock); | |
10367 | ||
1a40e23b | 10368 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10369 | if (!path) { |
10370 | ret = -ENOMEM; | |
10371 | goto out; | |
10372 | } | |
1a40e23b | 10373 | |
1bbc621e CM |
10374 | /* |
10375 | * get the inode first so any iput calls done for the io_list | |
10376 | * aren't the final iput (no unlinks allowed now) | |
10377 | */ | |
77ab86bf | 10378 | inode = lookup_free_space_inode(fs_info, block_group, path); |
1bbc621e CM |
10379 | |
10380 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10381 | /* | |
10382 | * make sure our free spache cache IO is done before remove the | |
10383 | * free space inode | |
10384 | */ | |
10385 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10386 | if (!list_empty(&block_group->io_list)) { | |
10387 | list_del_init(&block_group->io_list); | |
10388 | ||
10389 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10390 | ||
10391 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
afdb5718 | 10392 | btrfs_wait_cache_io(trans, block_group, path); |
1bbc621e CM |
10393 | btrfs_put_block_group(block_group); |
10394 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10395 | } | |
10396 | ||
10397 | if (!list_empty(&block_group->dirty_list)) { | |
10398 | list_del_init(&block_group->dirty_list); | |
10399 | btrfs_put_block_group(block_group); | |
10400 | } | |
10401 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10402 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10403 | ||
0af3d00b | 10404 | if (!IS_ERR(inode)) { |
73f2e545 | 10405 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
79787eaa JM |
10406 | if (ret) { |
10407 | btrfs_add_delayed_iput(inode); | |
10408 | goto out; | |
10409 | } | |
0af3d00b JB |
10410 | clear_nlink(inode); |
10411 | /* One for the block groups ref */ | |
10412 | spin_lock(&block_group->lock); | |
10413 | if (block_group->iref) { | |
10414 | block_group->iref = 0; | |
10415 | block_group->inode = NULL; | |
10416 | spin_unlock(&block_group->lock); | |
10417 | iput(inode); | |
10418 | } else { | |
10419 | spin_unlock(&block_group->lock); | |
10420 | } | |
10421 | /* One for our lookup ref */ | |
455757c3 | 10422 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10423 | } |
10424 | ||
10425 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10426 | key.offset = block_group->key.objectid; | |
10427 | key.type = 0; | |
10428 | ||
10429 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10430 | if (ret < 0) | |
10431 | goto out; | |
10432 | if (ret > 0) | |
b3b4aa74 | 10433 | btrfs_release_path(path); |
0af3d00b JB |
10434 | if (ret == 0) { |
10435 | ret = btrfs_del_item(trans, tree_root, path); | |
10436 | if (ret) | |
10437 | goto out; | |
b3b4aa74 | 10438 | btrfs_release_path(path); |
0af3d00b JB |
10439 | } |
10440 | ||
0b246afa | 10441 | spin_lock(&fs_info->block_group_cache_lock); |
1a40e23b | 10442 | rb_erase(&block_group->cache_node, |
0b246afa | 10443 | &fs_info->block_group_cache_tree); |
292cbd51 | 10444 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd | 10445 | |
0b246afa JM |
10446 | if (fs_info->first_logical_byte == block_group->key.objectid) |
10447 | fs_info->first_logical_byte = (u64)-1; | |
10448 | spin_unlock(&fs_info->block_group_cache_lock); | |
817d52f8 | 10449 | |
80eb234a | 10450 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10451 | /* |
10452 | * we must use list_del_init so people can check to see if they | |
10453 | * are still on the list after taking the semaphore | |
10454 | */ | |
10455 | list_del_init(&block_group->list); | |
6ab0a202 | 10456 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10457 | kobj = block_group->space_info->block_group_kobjs[index]; |
10458 | block_group->space_info->block_group_kobjs[index] = NULL; | |
0b246afa | 10459 | clear_avail_alloc_bits(fs_info, block_group->flags); |
6ab0a202 | 10460 | } |
80eb234a | 10461 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
10462 | if (kobj) { |
10463 | kobject_del(kobj); | |
10464 | kobject_put(kobj); | |
10465 | } | |
1a40e23b | 10466 | |
4f69cb98 FM |
10467 | if (block_group->has_caching_ctl) |
10468 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10469 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10470 | wait_block_group_cache_done(block_group); |
4f69cb98 | 10471 | if (block_group->has_caching_ctl) { |
0b246afa | 10472 | down_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10473 | if (!caching_ctl) { |
10474 | struct btrfs_caching_control *ctl; | |
10475 | ||
10476 | list_for_each_entry(ctl, | |
0b246afa | 10477 | &fs_info->caching_block_groups, list) |
4f69cb98 FM |
10478 | if (ctl->block_group == block_group) { |
10479 | caching_ctl = ctl; | |
1e4f4714 | 10480 | refcount_inc(&caching_ctl->count); |
4f69cb98 FM |
10481 | break; |
10482 | } | |
10483 | } | |
10484 | if (caching_ctl) | |
10485 | list_del_init(&caching_ctl->list); | |
0b246afa | 10486 | up_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10487 | if (caching_ctl) { |
10488 | /* Once for the caching bgs list and once for us. */ | |
10489 | put_caching_control(caching_ctl); | |
10490 | put_caching_control(caching_ctl); | |
10491 | } | |
10492 | } | |
817d52f8 | 10493 | |
ce93ec54 JB |
10494 | spin_lock(&trans->transaction->dirty_bgs_lock); |
10495 | if (!list_empty(&block_group->dirty_list)) { | |
1bbc621e CM |
10496 | WARN_ON(1); |
10497 | } | |
10498 | if (!list_empty(&block_group->io_list)) { | |
10499 | WARN_ON(1); | |
ce93ec54 JB |
10500 | } |
10501 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
817d52f8 JB |
10502 | btrfs_remove_free_space_cache(block_group); |
10503 | ||
c146afad | 10504 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10505 | list_del_init(&block_group->ro_list); |
18d018ad | 10506 | |
0b246afa | 10507 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
18d018ad ZL |
10508 | WARN_ON(block_group->space_info->total_bytes |
10509 | < block_group->key.offset); | |
10510 | WARN_ON(block_group->space_info->bytes_readonly | |
10511 | < block_group->key.offset); | |
10512 | WARN_ON(block_group->space_info->disk_total | |
10513 | < block_group->key.offset * factor); | |
10514 | } | |
c146afad YZ |
10515 | block_group->space_info->total_bytes -= block_group->key.offset; |
10516 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10517 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10518 | |
c146afad | 10519 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10520 | |
0af3d00b JB |
10521 | memcpy(&key, &block_group->key, sizeof(key)); |
10522 | ||
34441361 | 10523 | mutex_lock(&fs_info->chunk_mutex); |
495e64f4 FM |
10524 | if (!list_empty(&em->list)) { |
10525 | /* We're in the transaction->pending_chunks list. */ | |
10526 | free_extent_map(em); | |
10527 | } | |
04216820 FM |
10528 | spin_lock(&block_group->lock); |
10529 | block_group->removed = 1; | |
10530 | /* | |
10531 | * At this point trimming can't start on this block group, because we | |
10532 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10533 | * so no one can't find it anymore and even if someone already got this | |
10534 | * block group before we removed it from the rbtree, they have already | |
10535 | * incremented block_group->trimming - if they didn't, they won't find | |
10536 | * any free space entries because we already removed them all when we | |
10537 | * called btrfs_remove_free_space_cache(). | |
10538 | * | |
10539 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10540 | * to prevent the same logical address range and physical device space | |
10541 | * ranges from being reused for a new block group. This is because our | |
10542 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10543 | * completely transactionless, so while it is trimming a range the | |
10544 | * currently running transaction might finish and a new one start, | |
10545 | * allowing for new block groups to be created that can reuse the same | |
10546 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10547 | * |
10548 | * There may also be an implicit trim operation if the file system | |
10549 | * is mounted with -odiscard. The same protections must remain | |
10550 | * in place until the extents have been discarded completely when | |
10551 | * the transaction commit has completed. | |
04216820 FM |
10552 | */ |
10553 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
10554 | /* | |
10555 | * Make sure a trimmer task always sees the em in the pinned_chunks list | |
10556 | * if it sees block_group->removed == 1 (needs to lock block_group->lock | |
10557 | * before checking block_group->removed). | |
10558 | */ | |
10559 | if (!remove_em) { | |
10560 | /* | |
10561 | * Our em might be in trans->transaction->pending_chunks which | |
10562 | * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks), | |
10563 | * and so is the fs_info->pinned_chunks list. | |
10564 | * | |
10565 | * So at this point we must be holding the chunk_mutex to avoid | |
10566 | * any races with chunk allocation (more specifically at | |
10567 | * volumes.c:contains_pending_extent()), to ensure it always | |
10568 | * sees the em, either in the pending_chunks list or in the | |
10569 | * pinned_chunks list. | |
10570 | */ | |
0b246afa | 10571 | list_move_tail(&em->list, &fs_info->pinned_chunks); |
04216820 FM |
10572 | } |
10573 | spin_unlock(&block_group->lock); | |
04216820 FM |
10574 | |
10575 | if (remove_em) { | |
10576 | struct extent_map_tree *em_tree; | |
10577 | ||
0b246afa | 10578 | em_tree = &fs_info->mapping_tree.map_tree; |
04216820 | 10579 | write_lock(&em_tree->lock); |
8dbcd10f FM |
10580 | /* |
10581 | * The em might be in the pending_chunks list, so make sure the | |
10582 | * chunk mutex is locked, since remove_extent_mapping() will | |
10583 | * delete us from that list. | |
10584 | */ | |
04216820 FM |
10585 | remove_extent_mapping(em_tree, em); |
10586 | write_unlock(&em_tree->lock); | |
10587 | /* once for the tree */ | |
10588 | free_extent_map(em); | |
10589 | } | |
10590 | ||
34441361 | 10591 | mutex_unlock(&fs_info->chunk_mutex); |
8dbcd10f | 10592 | |
0b246afa | 10593 | ret = remove_block_group_free_space(trans, fs_info, block_group); |
1e144fb8 OS |
10594 | if (ret) |
10595 | goto out; | |
10596 | ||
fa9c0d79 CM |
10597 | btrfs_put_block_group(block_group); |
10598 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10599 | |
10600 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10601 | if (ret > 0) | |
10602 | ret = -EIO; | |
10603 | if (ret < 0) | |
10604 | goto out; | |
10605 | ||
10606 | ret = btrfs_del_item(trans, root, path); | |
10607 | out: | |
10608 | btrfs_free_path(path); | |
10609 | return ret; | |
10610 | } | |
acce952b | 10611 | |
8eab77ff | 10612 | struct btrfs_trans_handle * |
7fd01182 FM |
10613 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10614 | const u64 chunk_offset) | |
8eab77ff | 10615 | { |
7fd01182 FM |
10616 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; |
10617 | struct extent_map *em; | |
10618 | struct map_lookup *map; | |
10619 | unsigned int num_items; | |
10620 | ||
10621 | read_lock(&em_tree->lock); | |
10622 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10623 | read_unlock(&em_tree->lock); | |
10624 | ASSERT(em && em->start == chunk_offset); | |
10625 | ||
8eab77ff | 10626 | /* |
7fd01182 FM |
10627 | * We need to reserve 3 + N units from the metadata space info in order |
10628 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10629 | * btrfs_remove_block_group()), which are used for: | |
10630 | * | |
8eab77ff FM |
10631 | * 1 unit for adding the free space inode's orphan (located in the tree |
10632 | * of tree roots). | |
7fd01182 FM |
10633 | * 1 unit for deleting the block group item (located in the extent |
10634 | * tree). | |
10635 | * 1 unit for deleting the free space item (located in tree of tree | |
10636 | * roots). | |
10637 | * N units for deleting N device extent items corresponding to each | |
10638 | * stripe (located in the device tree). | |
10639 | * | |
10640 | * In order to remove a block group we also need to reserve units in the | |
10641 | * system space info in order to update the chunk tree (update one or | |
10642 | * more device items and remove one chunk item), but this is done at | |
10643 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10644 | */ |
95617d69 | 10645 | map = em->map_lookup; |
7fd01182 FM |
10646 | num_items = 3 + map->num_stripes; |
10647 | free_extent_map(em); | |
10648 | ||
8eab77ff | 10649 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10650 | num_items, 1); |
8eab77ff FM |
10651 | } |
10652 | ||
47ab2a6c JB |
10653 | /* |
10654 | * Process the unused_bgs list and remove any that don't have any allocated | |
10655 | * space inside of them. | |
10656 | */ | |
10657 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10658 | { | |
10659 | struct btrfs_block_group_cache *block_group; | |
10660 | struct btrfs_space_info *space_info; | |
47ab2a6c JB |
10661 | struct btrfs_trans_handle *trans; |
10662 | int ret = 0; | |
10663 | ||
afcdd129 | 10664 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) |
47ab2a6c JB |
10665 | return; |
10666 | ||
10667 | spin_lock(&fs_info->unused_bgs_lock); | |
10668 | while (!list_empty(&fs_info->unused_bgs)) { | |
10669 | u64 start, end; | |
e33e17ee | 10670 | int trimming; |
47ab2a6c JB |
10671 | |
10672 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10673 | struct btrfs_block_group_cache, | |
10674 | bg_list); | |
47ab2a6c | 10675 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10676 | |
10677 | space_info = block_group->space_info; | |
10678 | ||
47ab2a6c JB |
10679 | if (ret || btrfs_mixed_space_info(space_info)) { |
10680 | btrfs_put_block_group(block_group); | |
10681 | continue; | |
10682 | } | |
10683 | spin_unlock(&fs_info->unused_bgs_lock); | |
10684 | ||
d5f2e33b | 10685 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10686 | |
47ab2a6c JB |
10687 | /* Don't want to race with allocators so take the groups_sem */ |
10688 | down_write(&space_info->groups_sem); | |
10689 | spin_lock(&block_group->lock); | |
10690 | if (block_group->reserved || | |
10691 | btrfs_block_group_used(&block_group->item) || | |
19c4d2f9 | 10692 | block_group->ro || |
aefbe9a6 | 10693 | list_is_singular(&block_group->list)) { |
47ab2a6c JB |
10694 | /* |
10695 | * We want to bail if we made new allocations or have | |
10696 | * outstanding allocations in this block group. We do | |
10697 | * the ro check in case balance is currently acting on | |
10698 | * this block group. | |
10699 | */ | |
10700 | spin_unlock(&block_group->lock); | |
10701 | up_write(&space_info->groups_sem); | |
10702 | goto next; | |
10703 | } | |
10704 | spin_unlock(&block_group->lock); | |
10705 | ||
10706 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 10707 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
10708 | up_write(&space_info->groups_sem); |
10709 | if (ret < 0) { | |
10710 | ret = 0; | |
10711 | goto next; | |
10712 | } | |
10713 | ||
10714 | /* | |
10715 | * Want to do this before we do anything else so we can recover | |
10716 | * properly if we fail to join the transaction. | |
10717 | */ | |
7fd01182 FM |
10718 | trans = btrfs_start_trans_remove_block_group(fs_info, |
10719 | block_group->key.objectid); | |
47ab2a6c | 10720 | if (IS_ERR(trans)) { |
2ff7e61e | 10721 | btrfs_dec_block_group_ro(block_group); |
47ab2a6c JB |
10722 | ret = PTR_ERR(trans); |
10723 | goto next; | |
10724 | } | |
10725 | ||
10726 | /* | |
10727 | * We could have pending pinned extents for this block group, | |
10728 | * just delete them, we don't care about them anymore. | |
10729 | */ | |
10730 | start = block_group->key.objectid; | |
10731 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
10732 | /* |
10733 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
10734 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
10735 | * another task might be running finish_extent_commit() for the | |
10736 | * previous transaction N - 1, and have seen a range belonging | |
10737 | * to the block group in freed_extents[] before we were able to | |
10738 | * clear the whole block group range from freed_extents[]. This | |
10739 | * means that task can lookup for the block group after we | |
10740 | * unpinned it from freed_extents[] and removed it, leading to | |
10741 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
10742 | */ | |
10743 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 10744 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
91166212 | 10745 | EXTENT_DIRTY); |
758eb51e | 10746 | if (ret) { |
d4b450cd | 10747 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10748 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10749 | goto end_trans; |
10750 | } | |
10751 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
91166212 | 10752 | EXTENT_DIRTY); |
758eb51e | 10753 | if (ret) { |
d4b450cd | 10754 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10755 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10756 | goto end_trans; |
10757 | } | |
d4b450cd | 10758 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
10759 | |
10760 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
10761 | spin_lock(&space_info->lock); |
10762 | spin_lock(&block_group->lock); | |
10763 | ||
10764 | space_info->bytes_pinned -= block_group->pinned; | |
10765 | space_info->bytes_readonly += block_group->pinned; | |
10766 | percpu_counter_add(&space_info->total_bytes_pinned, | |
10767 | -block_group->pinned); | |
47ab2a6c JB |
10768 | block_group->pinned = 0; |
10769 | ||
c30666d4 ZL |
10770 | spin_unlock(&block_group->lock); |
10771 | spin_unlock(&space_info->lock); | |
10772 | ||
e33e17ee | 10773 | /* DISCARD can flip during remount */ |
0b246afa | 10774 | trimming = btrfs_test_opt(fs_info, DISCARD); |
e33e17ee JM |
10775 | |
10776 | /* Implicit trim during transaction commit. */ | |
10777 | if (trimming) | |
10778 | btrfs_get_block_group_trimming(block_group); | |
10779 | ||
47ab2a6c JB |
10780 | /* |
10781 | * Btrfs_remove_chunk will abort the transaction if things go | |
10782 | * horribly wrong. | |
10783 | */ | |
5b4aacef | 10784 | ret = btrfs_remove_chunk(trans, fs_info, |
47ab2a6c | 10785 | block_group->key.objectid); |
e33e17ee JM |
10786 | |
10787 | if (ret) { | |
10788 | if (trimming) | |
10789 | btrfs_put_block_group_trimming(block_group); | |
10790 | goto end_trans; | |
10791 | } | |
10792 | ||
10793 | /* | |
10794 | * If we're not mounted with -odiscard, we can just forget | |
10795 | * about this block group. Otherwise we'll need to wait | |
10796 | * until transaction commit to do the actual discard. | |
10797 | */ | |
10798 | if (trimming) { | |
348a0013 FM |
10799 | spin_lock(&fs_info->unused_bgs_lock); |
10800 | /* | |
10801 | * A concurrent scrub might have added us to the list | |
10802 | * fs_info->unused_bgs, so use a list_move operation | |
10803 | * to add the block group to the deleted_bgs list. | |
10804 | */ | |
e33e17ee JM |
10805 | list_move(&block_group->bg_list, |
10806 | &trans->transaction->deleted_bgs); | |
348a0013 | 10807 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
10808 | btrfs_get_block_group(block_group); |
10809 | } | |
758eb51e | 10810 | end_trans: |
3a45bb20 | 10811 | btrfs_end_transaction(trans); |
47ab2a6c | 10812 | next: |
d5f2e33b | 10813 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
10814 | btrfs_put_block_group(block_group); |
10815 | spin_lock(&fs_info->unused_bgs_lock); | |
10816 | } | |
10817 | spin_unlock(&fs_info->unused_bgs_lock); | |
10818 | } | |
10819 | ||
c59021f8 | 10820 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
10821 | { | |
10822 | struct btrfs_space_info *space_info; | |
1aba86d6 | 10823 | struct btrfs_super_block *disk_super; |
10824 | u64 features; | |
10825 | u64 flags; | |
10826 | int mixed = 0; | |
c59021f8 | 10827 | int ret; |
10828 | ||
6c41761f | 10829 | disk_super = fs_info->super_copy; |
1aba86d6 | 10830 | if (!btrfs_super_root(disk_super)) |
0dc924c5 | 10831 | return -EINVAL; |
c59021f8 | 10832 | |
1aba86d6 | 10833 | features = btrfs_super_incompat_flags(disk_super); |
10834 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
10835 | mixed = 1; | |
c59021f8 | 10836 | |
1aba86d6 | 10837 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
2be12ef7 | 10838 | ret = create_space_info(fs_info, flags, &space_info); |
c59021f8 | 10839 | if (ret) |
1aba86d6 | 10840 | goto out; |
c59021f8 | 10841 | |
1aba86d6 | 10842 | if (mixed) { |
10843 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
2be12ef7 | 10844 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 10845 | } else { |
10846 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
2be12ef7 | 10847 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 10848 | if (ret) |
10849 | goto out; | |
10850 | ||
10851 | flags = BTRFS_BLOCK_GROUP_DATA; | |
2be12ef7 | 10852 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 10853 | } |
10854 | out: | |
c59021f8 | 10855 | return ret; |
10856 | } | |
10857 | ||
2ff7e61e JM |
10858 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
10859 | u64 start, u64 end) | |
acce952b | 10860 | { |
2ff7e61e | 10861 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 10862 | } |
10863 | ||
499f377f JM |
10864 | /* |
10865 | * It used to be that old block groups would be left around forever. | |
10866 | * Iterating over them would be enough to trim unused space. Since we | |
10867 | * now automatically remove them, we also need to iterate over unallocated | |
10868 | * space. | |
10869 | * | |
10870 | * We don't want a transaction for this since the discard may take a | |
10871 | * substantial amount of time. We don't require that a transaction be | |
10872 | * running, but we do need to take a running transaction into account | |
10873 | * to ensure that we're not discarding chunks that were released in | |
10874 | * the current transaction. | |
10875 | * | |
10876 | * Holding the chunks lock will prevent other threads from allocating | |
10877 | * or releasing chunks, but it won't prevent a running transaction | |
10878 | * from committing and releasing the memory that the pending chunks | |
10879 | * list head uses. For that, we need to take a reference to the | |
10880 | * transaction. | |
10881 | */ | |
10882 | static int btrfs_trim_free_extents(struct btrfs_device *device, | |
10883 | u64 minlen, u64 *trimmed) | |
10884 | { | |
10885 | u64 start = 0, len = 0; | |
10886 | int ret; | |
10887 | ||
10888 | *trimmed = 0; | |
10889 | ||
10890 | /* Not writeable = nothing to do. */ | |
10891 | if (!device->writeable) | |
10892 | return 0; | |
10893 | ||
10894 | /* No free space = nothing to do. */ | |
10895 | if (device->total_bytes <= device->bytes_used) | |
10896 | return 0; | |
10897 | ||
10898 | ret = 0; | |
10899 | ||
10900 | while (1) { | |
fb456252 | 10901 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
10902 | struct btrfs_transaction *trans; |
10903 | u64 bytes; | |
10904 | ||
10905 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
10906 | if (ret) | |
10907 | return ret; | |
10908 | ||
10909 | down_read(&fs_info->commit_root_sem); | |
10910 | ||
10911 | spin_lock(&fs_info->trans_lock); | |
10912 | trans = fs_info->running_transaction; | |
10913 | if (trans) | |
9b64f57d | 10914 | refcount_inc(&trans->use_count); |
499f377f JM |
10915 | spin_unlock(&fs_info->trans_lock); |
10916 | ||
10917 | ret = find_free_dev_extent_start(trans, device, minlen, start, | |
10918 | &start, &len); | |
10919 | if (trans) | |
10920 | btrfs_put_transaction(trans); | |
10921 | ||
10922 | if (ret) { | |
10923 | up_read(&fs_info->commit_root_sem); | |
10924 | mutex_unlock(&fs_info->chunk_mutex); | |
10925 | if (ret == -ENOSPC) | |
10926 | ret = 0; | |
10927 | break; | |
10928 | } | |
10929 | ||
10930 | ret = btrfs_issue_discard(device->bdev, start, len, &bytes); | |
10931 | up_read(&fs_info->commit_root_sem); | |
10932 | mutex_unlock(&fs_info->chunk_mutex); | |
10933 | ||
10934 | if (ret) | |
10935 | break; | |
10936 | ||
10937 | start += len; | |
10938 | *trimmed += bytes; | |
10939 | ||
10940 | if (fatal_signal_pending(current)) { | |
10941 | ret = -ERESTARTSYS; | |
10942 | break; | |
10943 | } | |
10944 | ||
10945 | cond_resched(); | |
10946 | } | |
10947 | ||
10948 | return ret; | |
10949 | } | |
10950 | ||
2ff7e61e | 10951 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 10952 | { |
f7039b1d | 10953 | struct btrfs_block_group_cache *cache = NULL; |
499f377f JM |
10954 | struct btrfs_device *device; |
10955 | struct list_head *devices; | |
f7039b1d LD |
10956 | u64 group_trimmed; |
10957 | u64 start; | |
10958 | u64 end; | |
10959 | u64 trimmed = 0; | |
2cac13e4 | 10960 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
10961 | int ret = 0; |
10962 | ||
2cac13e4 LB |
10963 | /* |
10964 | * try to trim all FS space, our block group may start from non-zero. | |
10965 | */ | |
10966 | if (range->len == total_bytes) | |
10967 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
10968 | else | |
10969 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
10970 | |
10971 | while (cache) { | |
10972 | if (cache->key.objectid >= (range->start + range->len)) { | |
10973 | btrfs_put_block_group(cache); | |
10974 | break; | |
10975 | } | |
10976 | ||
10977 | start = max(range->start, cache->key.objectid); | |
10978 | end = min(range->start + range->len, | |
10979 | cache->key.objectid + cache->key.offset); | |
10980 | ||
10981 | if (end - start >= range->minlen) { | |
10982 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 10983 | ret = cache_block_group(cache, 0); |
1be41b78 JB |
10984 | if (ret) { |
10985 | btrfs_put_block_group(cache); | |
10986 | break; | |
10987 | } | |
10988 | ret = wait_block_group_cache_done(cache); | |
10989 | if (ret) { | |
10990 | btrfs_put_block_group(cache); | |
10991 | break; | |
10992 | } | |
f7039b1d LD |
10993 | } |
10994 | ret = btrfs_trim_block_group(cache, | |
10995 | &group_trimmed, | |
10996 | start, | |
10997 | end, | |
10998 | range->minlen); | |
10999 | ||
11000 | trimmed += group_trimmed; | |
11001 | if (ret) { | |
11002 | btrfs_put_block_group(cache); | |
11003 | break; | |
11004 | } | |
11005 | } | |
11006 | ||
2ff7e61e | 11007 | cache = next_block_group(fs_info, cache); |
f7039b1d LD |
11008 | } |
11009 | ||
0b246afa JM |
11010 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
11011 | devices = &fs_info->fs_devices->alloc_list; | |
499f377f JM |
11012 | list_for_each_entry(device, devices, dev_alloc_list) { |
11013 | ret = btrfs_trim_free_extents(device, range->minlen, | |
11014 | &group_trimmed); | |
11015 | if (ret) | |
11016 | break; | |
11017 | ||
11018 | trimmed += group_trimmed; | |
11019 | } | |
0b246afa | 11020 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 11021 | |
f7039b1d LD |
11022 | range->len = trimmed; |
11023 | return ret; | |
11024 | } | |
8257b2dc MX |
11025 | |
11026 | /* | |
ea14b57f | 11027 | * btrfs_{start,end}_write_no_snapshotting() are similar to |
9ea24bbe FM |
11028 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing |
11029 | * data into the page cache through nocow before the subvolume is snapshoted, | |
11030 | * but flush the data into disk after the snapshot creation, or to prevent | |
ea14b57f | 11031 | * operations while snapshotting is ongoing and that cause the snapshot to be |
9ea24bbe | 11032 | * inconsistent (writes followed by expanding truncates for example). |
8257b2dc | 11033 | */ |
ea14b57f | 11034 | void btrfs_end_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc MX |
11035 | { |
11036 | percpu_counter_dec(&root->subv_writers->counter); | |
11037 | /* | |
a83342aa | 11038 | * Make sure counter is updated before we wake up waiters. |
8257b2dc MX |
11039 | */ |
11040 | smp_mb(); | |
11041 | if (waitqueue_active(&root->subv_writers->wait)) | |
11042 | wake_up(&root->subv_writers->wait); | |
11043 | } | |
11044 | ||
ea14b57f | 11045 | int btrfs_start_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc | 11046 | { |
ea14b57f | 11047 | if (atomic_read(&root->will_be_snapshotted)) |
8257b2dc MX |
11048 | return 0; |
11049 | ||
11050 | percpu_counter_inc(&root->subv_writers->counter); | |
11051 | /* | |
11052 | * Make sure counter is updated before we check for snapshot creation. | |
11053 | */ | |
11054 | smp_mb(); | |
ea14b57f DS |
11055 | if (atomic_read(&root->will_be_snapshotted)) { |
11056 | btrfs_end_write_no_snapshotting(root); | |
8257b2dc MX |
11057 | return 0; |
11058 | } | |
11059 | return 1; | |
11060 | } | |
0bc19f90 | 11061 | |
ea14b57f | 11062 | static int wait_snapshotting_atomic_t(atomic_t *a) |
0bc19f90 ZL |
11063 | { |
11064 | schedule(); | |
11065 | return 0; | |
11066 | } | |
11067 | ||
11068 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) | |
11069 | { | |
11070 | while (true) { | |
11071 | int ret; | |
11072 | ||
ea14b57f | 11073 | ret = btrfs_start_write_no_snapshotting(root); |
0bc19f90 ZL |
11074 | if (ret) |
11075 | break; | |
ea14b57f DS |
11076 | wait_on_atomic_t(&root->will_be_snapshotted, |
11077 | wait_snapshotting_atomic_t, | |
0bc19f90 ZL |
11078 | TASK_UNINTERRUPTIBLE); |
11079 | } | |
11080 | } |