Commit | Line | Data |
---|---|---|
6cbd5570 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
ec6b910f | 18 | #include <linux/sched.h> |
f361bf4a | 19 | #include <linux/sched/signal.h> |
edbd8d4e | 20 | #include <linux/pagemap.h> |
ec44a35c | 21 | #include <linux/writeback.h> |
21af804c | 22 | #include <linux/blkdev.h> |
b7a9f29f | 23 | #include <linux/sort.h> |
4184ea7f | 24 | #include <linux/rcupdate.h> |
817d52f8 | 25 | #include <linux/kthread.h> |
5a0e3ad6 | 26 | #include <linux/slab.h> |
dff51cd1 | 27 | #include <linux/ratelimit.h> |
b150a4f1 | 28 | #include <linux/percpu_counter.h> |
69fe2d75 | 29 | #include <linux/lockdep.h> |
9678c543 | 30 | #include <linux/crc32c.h> |
995946dd | 31 | #include "tree-log.h" |
fec577fb CM |
32 | #include "disk-io.h" |
33 | #include "print-tree.h" | |
0b86a832 | 34 | #include "volumes.h" |
53b381b3 | 35 | #include "raid56.h" |
925baedd | 36 | #include "locking.h" |
fa9c0d79 | 37 | #include "free-space-cache.h" |
1e144fb8 | 38 | #include "free-space-tree.h" |
3fed40cc | 39 | #include "math.h" |
6ab0a202 | 40 | #include "sysfs.h" |
fcebe456 | 41 | #include "qgroup.h" |
fd708b81 | 42 | #include "ref-verify.h" |
fec577fb | 43 | |
709c0486 AJ |
44 | #undef SCRAMBLE_DELAYED_REFS |
45 | ||
9e622d6b MX |
46 | /* |
47 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
48 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
49 | * if we really need one. | |
50 | * | |
0e4f8f88 CM |
51 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
52 | * if we have very few chunks already allocated. This is | |
53 | * used as part of the clustering code to help make sure | |
54 | * we have a good pool of storage to cluster in, without | |
55 | * filling the FS with empty chunks | |
56 | * | |
9e622d6b MX |
57 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
58 | * | |
0e4f8f88 CM |
59 | */ |
60 | enum { | |
61 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
62 | CHUNK_ALLOC_LIMITED = 1, |
63 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
64 | }; |
65 | ||
5d4f98a2 | 66 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 67 | struct btrfs_fs_info *fs_info, |
c682f9b3 | 68 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
69 | u64 root_objectid, u64 owner_objectid, |
70 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 71 | struct btrfs_delayed_extent_op *extra_op); |
5d4f98a2 YZ |
72 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
73 | struct extent_buffer *leaf, | |
74 | struct btrfs_extent_item *ei); | |
75 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
2ff7e61e | 76 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
77 | u64 parent, u64 root_objectid, |
78 | u64 flags, u64 owner, u64 offset, | |
79 | struct btrfs_key *ins, int ref_mod); | |
80 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
2ff7e61e | 81 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
82 | u64 parent, u64 root_objectid, |
83 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 84 | int level, struct btrfs_key *ins); |
6a63209f | 85 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 86 | struct btrfs_fs_info *fs_info, u64 flags, |
698d0082 | 87 | int force); |
11833d66 YZ |
88 | static int find_next_key(struct btrfs_path *path, int level, |
89 | struct btrfs_key *key); | |
ab8d0fc4 JM |
90 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
91 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 92 | int dump_block_groups); |
5d80366e JB |
93 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
94 | u64 num_bytes); | |
957780eb JB |
95 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, |
96 | struct btrfs_space_info *space_info, | |
97 | u64 num_bytes); | |
98 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
99 | struct btrfs_space_info *space_info, | |
100 | u64 num_bytes); | |
6a63209f | 101 | |
817d52f8 JB |
102 | static noinline int |
103 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
104 | { | |
105 | smp_mb(); | |
36cce922 JB |
106 | return cache->cached == BTRFS_CACHE_FINISHED || |
107 | cache->cached == BTRFS_CACHE_ERROR; | |
817d52f8 JB |
108 | } |
109 | ||
0f9dd46c JB |
110 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
111 | { | |
112 | return (cache->flags & bits) == bits; | |
113 | } | |
114 | ||
758f2dfc | 115 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
116 | { |
117 | atomic_inc(&cache->count); | |
118 | } | |
119 | ||
120 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
121 | { | |
f0486c68 YZ |
122 | if (atomic_dec_and_test(&cache->count)) { |
123 | WARN_ON(cache->pinned > 0); | |
124 | WARN_ON(cache->reserved > 0); | |
0966a7b1 QW |
125 | |
126 | /* | |
127 | * If not empty, someone is still holding mutex of | |
128 | * full_stripe_lock, which can only be released by caller. | |
129 | * And it will definitely cause use-after-free when caller | |
130 | * tries to release full stripe lock. | |
131 | * | |
132 | * No better way to resolve, but only to warn. | |
133 | */ | |
134 | WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); | |
34d52cb6 | 135 | kfree(cache->free_space_ctl); |
11dfe35a | 136 | kfree(cache); |
f0486c68 | 137 | } |
11dfe35a JB |
138 | } |
139 | ||
0f9dd46c JB |
140 | /* |
141 | * this adds the block group to the fs_info rb tree for the block group | |
142 | * cache | |
143 | */ | |
b2950863 | 144 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
145 | struct btrfs_block_group_cache *block_group) |
146 | { | |
147 | struct rb_node **p; | |
148 | struct rb_node *parent = NULL; | |
149 | struct btrfs_block_group_cache *cache; | |
150 | ||
151 | spin_lock(&info->block_group_cache_lock); | |
152 | p = &info->block_group_cache_tree.rb_node; | |
153 | ||
154 | while (*p) { | |
155 | parent = *p; | |
156 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
157 | cache_node); | |
158 | if (block_group->key.objectid < cache->key.objectid) { | |
159 | p = &(*p)->rb_left; | |
160 | } else if (block_group->key.objectid > cache->key.objectid) { | |
161 | p = &(*p)->rb_right; | |
162 | } else { | |
163 | spin_unlock(&info->block_group_cache_lock); | |
164 | return -EEXIST; | |
165 | } | |
166 | } | |
167 | ||
168 | rb_link_node(&block_group->cache_node, parent, p); | |
169 | rb_insert_color(&block_group->cache_node, | |
170 | &info->block_group_cache_tree); | |
a1897fdd LB |
171 | |
172 | if (info->first_logical_byte > block_group->key.objectid) | |
173 | info->first_logical_byte = block_group->key.objectid; | |
174 | ||
0f9dd46c JB |
175 | spin_unlock(&info->block_group_cache_lock); |
176 | ||
177 | return 0; | |
178 | } | |
179 | ||
180 | /* | |
181 | * This will return the block group at or after bytenr if contains is 0, else | |
182 | * it will return the block group that contains the bytenr | |
183 | */ | |
184 | static struct btrfs_block_group_cache * | |
185 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
186 | int contains) | |
187 | { | |
188 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
189 | struct rb_node *n; | |
190 | u64 end, start; | |
191 | ||
192 | spin_lock(&info->block_group_cache_lock); | |
193 | n = info->block_group_cache_tree.rb_node; | |
194 | ||
195 | while (n) { | |
196 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
197 | cache_node); | |
198 | end = cache->key.objectid + cache->key.offset - 1; | |
199 | start = cache->key.objectid; | |
200 | ||
201 | if (bytenr < start) { | |
202 | if (!contains && (!ret || start < ret->key.objectid)) | |
203 | ret = cache; | |
204 | n = n->rb_left; | |
205 | } else if (bytenr > start) { | |
206 | if (contains && bytenr <= end) { | |
207 | ret = cache; | |
208 | break; | |
209 | } | |
210 | n = n->rb_right; | |
211 | } else { | |
212 | ret = cache; | |
213 | break; | |
214 | } | |
215 | } | |
a1897fdd | 216 | if (ret) { |
11dfe35a | 217 | btrfs_get_block_group(ret); |
a1897fdd LB |
218 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
219 | info->first_logical_byte = ret->key.objectid; | |
220 | } | |
0f9dd46c JB |
221 | spin_unlock(&info->block_group_cache_lock); |
222 | ||
223 | return ret; | |
224 | } | |
225 | ||
2ff7e61e | 226 | static int add_excluded_extent(struct btrfs_fs_info *fs_info, |
11833d66 | 227 | u64 start, u64 num_bytes) |
817d52f8 | 228 | { |
11833d66 | 229 | u64 end = start + num_bytes - 1; |
0b246afa | 230 | set_extent_bits(&fs_info->freed_extents[0], |
ceeb0ae7 | 231 | start, end, EXTENT_UPTODATE); |
0b246afa | 232 | set_extent_bits(&fs_info->freed_extents[1], |
ceeb0ae7 | 233 | start, end, EXTENT_UPTODATE); |
11833d66 YZ |
234 | return 0; |
235 | } | |
817d52f8 | 236 | |
2ff7e61e | 237 | static void free_excluded_extents(struct btrfs_fs_info *fs_info, |
11833d66 YZ |
238 | struct btrfs_block_group_cache *cache) |
239 | { | |
240 | u64 start, end; | |
817d52f8 | 241 | |
11833d66 YZ |
242 | start = cache->key.objectid; |
243 | end = start + cache->key.offset - 1; | |
244 | ||
0b246afa | 245 | clear_extent_bits(&fs_info->freed_extents[0], |
91166212 | 246 | start, end, EXTENT_UPTODATE); |
0b246afa | 247 | clear_extent_bits(&fs_info->freed_extents[1], |
91166212 | 248 | start, end, EXTENT_UPTODATE); |
817d52f8 JB |
249 | } |
250 | ||
2ff7e61e | 251 | static int exclude_super_stripes(struct btrfs_fs_info *fs_info, |
11833d66 | 252 | struct btrfs_block_group_cache *cache) |
817d52f8 | 253 | { |
817d52f8 JB |
254 | u64 bytenr; |
255 | u64 *logical; | |
256 | int stripe_len; | |
257 | int i, nr, ret; | |
258 | ||
06b2331f YZ |
259 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
260 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
261 | cache->bytes_super += stripe_len; | |
2ff7e61e | 262 | ret = add_excluded_extent(fs_info, cache->key.objectid, |
06b2331f | 263 | stripe_len); |
835d974f JB |
264 | if (ret) |
265 | return ret; | |
06b2331f YZ |
266 | } |
267 | ||
817d52f8 JB |
268 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
269 | bytenr = btrfs_sb_offset(i); | |
0b246afa | 270 | ret = btrfs_rmap_block(fs_info, cache->key.objectid, |
ab8d0fc4 | 271 | bytenr, 0, &logical, &nr, &stripe_len); |
835d974f JB |
272 | if (ret) |
273 | return ret; | |
11833d66 | 274 | |
817d52f8 | 275 | while (nr--) { |
51bf5f0b JB |
276 | u64 start, len; |
277 | ||
278 | if (logical[nr] > cache->key.objectid + | |
279 | cache->key.offset) | |
280 | continue; | |
281 | ||
282 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
283 | continue; | |
284 | ||
285 | start = logical[nr]; | |
286 | if (start < cache->key.objectid) { | |
287 | start = cache->key.objectid; | |
288 | len = (logical[nr] + stripe_len) - start; | |
289 | } else { | |
290 | len = min_t(u64, stripe_len, | |
291 | cache->key.objectid + | |
292 | cache->key.offset - start); | |
293 | } | |
294 | ||
295 | cache->bytes_super += len; | |
2ff7e61e | 296 | ret = add_excluded_extent(fs_info, start, len); |
835d974f JB |
297 | if (ret) { |
298 | kfree(logical); | |
299 | return ret; | |
300 | } | |
817d52f8 | 301 | } |
11833d66 | 302 | |
817d52f8 JB |
303 | kfree(logical); |
304 | } | |
817d52f8 JB |
305 | return 0; |
306 | } | |
307 | ||
11833d66 YZ |
308 | static struct btrfs_caching_control * |
309 | get_caching_control(struct btrfs_block_group_cache *cache) | |
310 | { | |
311 | struct btrfs_caching_control *ctl; | |
312 | ||
313 | spin_lock(&cache->lock); | |
dde5abee JB |
314 | if (!cache->caching_ctl) { |
315 | spin_unlock(&cache->lock); | |
11833d66 YZ |
316 | return NULL; |
317 | } | |
318 | ||
319 | ctl = cache->caching_ctl; | |
1e4f4714 | 320 | refcount_inc(&ctl->count); |
11833d66 YZ |
321 | spin_unlock(&cache->lock); |
322 | return ctl; | |
323 | } | |
324 | ||
325 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
326 | { | |
1e4f4714 | 327 | if (refcount_dec_and_test(&ctl->count)) |
11833d66 YZ |
328 | kfree(ctl); |
329 | } | |
330 | ||
d0bd4560 | 331 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 332 | static void fragment_free_space(struct btrfs_block_group_cache *block_group) |
d0bd4560 | 333 | { |
2ff7e61e | 334 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
d0bd4560 JB |
335 | u64 start = block_group->key.objectid; |
336 | u64 len = block_group->key.offset; | |
337 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? | |
0b246afa | 338 | fs_info->nodesize : fs_info->sectorsize; |
d0bd4560 JB |
339 | u64 step = chunk << 1; |
340 | ||
341 | while (len > chunk) { | |
342 | btrfs_remove_free_space(block_group, start, chunk); | |
343 | start += step; | |
344 | if (len < step) | |
345 | len = 0; | |
346 | else | |
347 | len -= step; | |
348 | } | |
349 | } | |
350 | #endif | |
351 | ||
0f9dd46c JB |
352 | /* |
353 | * this is only called by cache_block_group, since we could have freed extents | |
354 | * we need to check the pinned_extents for any extents that can't be used yet | |
355 | * since their free space will be released as soon as the transaction commits. | |
356 | */ | |
a5ed9182 OS |
357 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
358 | struct btrfs_fs_info *info, u64 start, u64 end) | |
0f9dd46c | 359 | { |
817d52f8 | 360 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
361 | int ret; |
362 | ||
363 | while (start < end) { | |
11833d66 | 364 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 365 | &extent_start, &extent_end, |
e6138876 JB |
366 | EXTENT_DIRTY | EXTENT_UPTODATE, |
367 | NULL); | |
0f9dd46c JB |
368 | if (ret) |
369 | break; | |
370 | ||
06b2331f | 371 | if (extent_start <= start) { |
0f9dd46c JB |
372 | start = extent_end + 1; |
373 | } else if (extent_start > start && extent_start < end) { | |
374 | size = extent_start - start; | |
817d52f8 | 375 | total_added += size; |
ea6a478e JB |
376 | ret = btrfs_add_free_space(block_group, start, |
377 | size); | |
79787eaa | 378 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
379 | start = extent_end + 1; |
380 | } else { | |
381 | break; | |
382 | } | |
383 | } | |
384 | ||
385 | if (start < end) { | |
386 | size = end - start; | |
817d52f8 | 387 | total_added += size; |
ea6a478e | 388 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 389 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
390 | } |
391 | ||
817d52f8 | 392 | return total_added; |
0f9dd46c JB |
393 | } |
394 | ||
73fa48b6 | 395 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) |
e37c9e69 | 396 | { |
0b246afa JM |
397 | struct btrfs_block_group_cache *block_group = caching_ctl->block_group; |
398 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
399 | struct btrfs_root *extent_root = fs_info->extent_root; | |
e37c9e69 | 400 | struct btrfs_path *path; |
5f39d397 | 401 | struct extent_buffer *leaf; |
11833d66 | 402 | struct btrfs_key key; |
817d52f8 | 403 | u64 total_found = 0; |
11833d66 YZ |
404 | u64 last = 0; |
405 | u32 nritems; | |
73fa48b6 | 406 | int ret; |
d0bd4560 | 407 | bool wakeup = true; |
f510cfec | 408 | |
e37c9e69 CM |
409 | path = btrfs_alloc_path(); |
410 | if (!path) | |
73fa48b6 | 411 | return -ENOMEM; |
7d7d6068 | 412 | |
817d52f8 | 413 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 414 | |
d0bd4560 JB |
415 | #ifdef CONFIG_BTRFS_DEBUG |
416 | /* | |
417 | * If we're fragmenting we don't want to make anybody think we can | |
418 | * allocate from this block group until we've had a chance to fragment | |
419 | * the free space. | |
420 | */ | |
2ff7e61e | 421 | if (btrfs_should_fragment_free_space(block_group)) |
d0bd4560 JB |
422 | wakeup = false; |
423 | #endif | |
5cd57b2c | 424 | /* |
817d52f8 JB |
425 | * We don't want to deadlock with somebody trying to allocate a new |
426 | * extent for the extent root while also trying to search the extent | |
427 | * root to add free space. So we skip locking and search the commit | |
428 | * root, since its read-only | |
5cd57b2c CM |
429 | */ |
430 | path->skip_locking = 1; | |
817d52f8 | 431 | path->search_commit_root = 1; |
e4058b54 | 432 | path->reada = READA_FORWARD; |
817d52f8 | 433 | |
e4404d6e | 434 | key.objectid = last; |
e37c9e69 | 435 | key.offset = 0; |
11833d66 | 436 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 437 | |
52ee28d2 | 438 | next: |
11833d66 | 439 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 440 | if (ret < 0) |
73fa48b6 | 441 | goto out; |
a512bbf8 | 442 | |
11833d66 YZ |
443 | leaf = path->nodes[0]; |
444 | nritems = btrfs_header_nritems(leaf); | |
445 | ||
d397712b | 446 | while (1) { |
7841cb28 | 447 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 448 | last = (u64)-1; |
817d52f8 | 449 | break; |
f25784b3 | 450 | } |
817d52f8 | 451 | |
11833d66 YZ |
452 | if (path->slots[0] < nritems) { |
453 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
454 | } else { | |
455 | ret = find_next_key(path, 0, &key); | |
456 | if (ret) | |
e37c9e69 | 457 | break; |
817d52f8 | 458 | |
c9ea7b24 | 459 | if (need_resched() || |
9e351cc8 | 460 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
d0bd4560 JB |
461 | if (wakeup) |
462 | caching_ctl->progress = last; | |
ff5714cc | 463 | btrfs_release_path(path); |
9e351cc8 | 464 | up_read(&fs_info->commit_root_sem); |
589d8ade | 465 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 466 | cond_resched(); |
73fa48b6 OS |
467 | mutex_lock(&caching_ctl->mutex); |
468 | down_read(&fs_info->commit_root_sem); | |
469 | goto next; | |
589d8ade | 470 | } |
0a3896d0 JB |
471 | |
472 | ret = btrfs_next_leaf(extent_root, path); | |
473 | if (ret < 0) | |
73fa48b6 | 474 | goto out; |
0a3896d0 JB |
475 | if (ret) |
476 | break; | |
589d8ade JB |
477 | leaf = path->nodes[0]; |
478 | nritems = btrfs_header_nritems(leaf); | |
479 | continue; | |
11833d66 | 480 | } |
817d52f8 | 481 | |
52ee28d2 LB |
482 | if (key.objectid < last) { |
483 | key.objectid = last; | |
484 | key.offset = 0; | |
485 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
486 | ||
d0bd4560 JB |
487 | if (wakeup) |
488 | caching_ctl->progress = last; | |
52ee28d2 LB |
489 | btrfs_release_path(path); |
490 | goto next; | |
491 | } | |
492 | ||
11833d66 YZ |
493 | if (key.objectid < block_group->key.objectid) { |
494 | path->slots[0]++; | |
817d52f8 | 495 | continue; |
e37c9e69 | 496 | } |
0f9dd46c | 497 | |
e37c9e69 | 498 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 499 | block_group->key.offset) |
e37c9e69 | 500 | break; |
7d7d6068 | 501 | |
3173a18f JB |
502 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
503 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
817d52f8 JB |
504 | total_found += add_new_free_space(block_group, |
505 | fs_info, last, | |
506 | key.objectid); | |
3173a18f JB |
507 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
508 | last = key.objectid + | |
da17066c | 509 | fs_info->nodesize; |
3173a18f JB |
510 | else |
511 | last = key.objectid + key.offset; | |
817d52f8 | 512 | |
73fa48b6 | 513 | if (total_found > CACHING_CTL_WAKE_UP) { |
11833d66 | 514 | total_found = 0; |
d0bd4560 JB |
515 | if (wakeup) |
516 | wake_up(&caching_ctl->wait); | |
11833d66 | 517 | } |
817d52f8 | 518 | } |
e37c9e69 CM |
519 | path->slots[0]++; |
520 | } | |
817d52f8 | 521 | ret = 0; |
e37c9e69 | 522 | |
817d52f8 JB |
523 | total_found += add_new_free_space(block_group, fs_info, last, |
524 | block_group->key.objectid + | |
525 | block_group->key.offset); | |
11833d66 | 526 | caching_ctl->progress = (u64)-1; |
817d52f8 | 527 | |
73fa48b6 OS |
528 | out: |
529 | btrfs_free_path(path); | |
530 | return ret; | |
531 | } | |
532 | ||
533 | static noinline void caching_thread(struct btrfs_work *work) | |
534 | { | |
535 | struct btrfs_block_group_cache *block_group; | |
536 | struct btrfs_fs_info *fs_info; | |
537 | struct btrfs_caching_control *caching_ctl; | |
538 | int ret; | |
539 | ||
540 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
541 | block_group = caching_ctl->block_group; | |
542 | fs_info = block_group->fs_info; | |
543 | ||
544 | mutex_lock(&caching_ctl->mutex); | |
545 | down_read(&fs_info->commit_root_sem); | |
546 | ||
1e144fb8 OS |
547 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
548 | ret = load_free_space_tree(caching_ctl); | |
549 | else | |
550 | ret = load_extent_tree_free(caching_ctl); | |
73fa48b6 | 551 | |
817d52f8 | 552 | spin_lock(&block_group->lock); |
11833d66 | 553 | block_group->caching_ctl = NULL; |
73fa48b6 | 554 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; |
817d52f8 | 555 | spin_unlock(&block_group->lock); |
0f9dd46c | 556 | |
d0bd4560 | 557 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 558 | if (btrfs_should_fragment_free_space(block_group)) { |
d0bd4560 JB |
559 | u64 bytes_used; |
560 | ||
561 | spin_lock(&block_group->space_info->lock); | |
562 | spin_lock(&block_group->lock); | |
563 | bytes_used = block_group->key.offset - | |
564 | btrfs_block_group_used(&block_group->item); | |
565 | block_group->space_info->bytes_used += bytes_used >> 1; | |
566 | spin_unlock(&block_group->lock); | |
567 | spin_unlock(&block_group->space_info->lock); | |
2ff7e61e | 568 | fragment_free_space(block_group); |
d0bd4560 JB |
569 | } |
570 | #endif | |
571 | ||
572 | caching_ctl->progress = (u64)-1; | |
11833d66 | 573 | |
9e351cc8 | 574 | up_read(&fs_info->commit_root_sem); |
2ff7e61e | 575 | free_excluded_extents(fs_info, block_group); |
11833d66 | 576 | mutex_unlock(&caching_ctl->mutex); |
73fa48b6 | 577 | |
11833d66 YZ |
578 | wake_up(&caching_ctl->wait); |
579 | ||
580 | put_caching_control(caching_ctl); | |
11dfe35a | 581 | btrfs_put_block_group(block_group); |
817d52f8 JB |
582 | } |
583 | ||
9d66e233 | 584 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 585 | int load_cache_only) |
817d52f8 | 586 | { |
291c7d2f | 587 | DEFINE_WAIT(wait); |
11833d66 YZ |
588 | struct btrfs_fs_info *fs_info = cache->fs_info; |
589 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
590 | int ret = 0; |
591 | ||
291c7d2f | 592 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
593 | if (!caching_ctl) |
594 | return -ENOMEM; | |
291c7d2f JB |
595 | |
596 | INIT_LIST_HEAD(&caching_ctl->list); | |
597 | mutex_init(&caching_ctl->mutex); | |
598 | init_waitqueue_head(&caching_ctl->wait); | |
599 | caching_ctl->block_group = cache; | |
600 | caching_ctl->progress = cache->key.objectid; | |
1e4f4714 | 601 | refcount_set(&caching_ctl->count, 1); |
9e0af237 LB |
602 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, |
603 | caching_thread, NULL, NULL); | |
291c7d2f JB |
604 | |
605 | spin_lock(&cache->lock); | |
606 | /* | |
607 | * This should be a rare occasion, but this could happen I think in the | |
608 | * case where one thread starts to load the space cache info, and then | |
609 | * some other thread starts a transaction commit which tries to do an | |
610 | * allocation while the other thread is still loading the space cache | |
611 | * info. The previous loop should have kept us from choosing this block | |
612 | * group, but if we've moved to the state where we will wait on caching | |
613 | * block groups we need to first check if we're doing a fast load here, | |
614 | * so we can wait for it to finish, otherwise we could end up allocating | |
615 | * from a block group who's cache gets evicted for one reason or | |
616 | * another. | |
617 | */ | |
618 | while (cache->cached == BTRFS_CACHE_FAST) { | |
619 | struct btrfs_caching_control *ctl; | |
620 | ||
621 | ctl = cache->caching_ctl; | |
1e4f4714 | 622 | refcount_inc(&ctl->count); |
291c7d2f JB |
623 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); |
624 | spin_unlock(&cache->lock); | |
625 | ||
626 | schedule(); | |
627 | ||
628 | finish_wait(&ctl->wait, &wait); | |
629 | put_caching_control(ctl); | |
630 | spin_lock(&cache->lock); | |
631 | } | |
632 | ||
633 | if (cache->cached != BTRFS_CACHE_NO) { | |
634 | spin_unlock(&cache->lock); | |
635 | kfree(caching_ctl); | |
11833d66 | 636 | return 0; |
291c7d2f JB |
637 | } |
638 | WARN_ON(cache->caching_ctl); | |
639 | cache->caching_ctl = caching_ctl; | |
640 | cache->cached = BTRFS_CACHE_FAST; | |
641 | spin_unlock(&cache->lock); | |
11833d66 | 642 | |
d8953d69 | 643 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cb83b7b8 | 644 | mutex_lock(&caching_ctl->mutex); |
9d66e233 JB |
645 | ret = load_free_space_cache(fs_info, cache); |
646 | ||
647 | spin_lock(&cache->lock); | |
648 | if (ret == 1) { | |
291c7d2f | 649 | cache->caching_ctl = NULL; |
9d66e233 JB |
650 | cache->cached = BTRFS_CACHE_FINISHED; |
651 | cache->last_byte_to_unpin = (u64)-1; | |
cb83b7b8 | 652 | caching_ctl->progress = (u64)-1; |
9d66e233 | 653 | } else { |
291c7d2f JB |
654 | if (load_cache_only) { |
655 | cache->caching_ctl = NULL; | |
656 | cache->cached = BTRFS_CACHE_NO; | |
657 | } else { | |
658 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 659 | cache->has_caching_ctl = 1; |
291c7d2f | 660 | } |
9d66e233 JB |
661 | } |
662 | spin_unlock(&cache->lock); | |
d0bd4560 JB |
663 | #ifdef CONFIG_BTRFS_DEBUG |
664 | if (ret == 1 && | |
2ff7e61e | 665 | btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
666 | u64 bytes_used; |
667 | ||
668 | spin_lock(&cache->space_info->lock); | |
669 | spin_lock(&cache->lock); | |
670 | bytes_used = cache->key.offset - | |
671 | btrfs_block_group_used(&cache->item); | |
672 | cache->space_info->bytes_used += bytes_used >> 1; | |
673 | spin_unlock(&cache->lock); | |
674 | spin_unlock(&cache->space_info->lock); | |
2ff7e61e | 675 | fragment_free_space(cache); |
d0bd4560 JB |
676 | } |
677 | #endif | |
cb83b7b8 JB |
678 | mutex_unlock(&caching_ctl->mutex); |
679 | ||
291c7d2f | 680 | wake_up(&caching_ctl->wait); |
3c14874a | 681 | if (ret == 1) { |
291c7d2f | 682 | put_caching_control(caching_ctl); |
2ff7e61e | 683 | free_excluded_extents(fs_info, cache); |
9d66e233 | 684 | return 0; |
3c14874a | 685 | } |
291c7d2f JB |
686 | } else { |
687 | /* | |
1e144fb8 OS |
688 | * We're either using the free space tree or no caching at all. |
689 | * Set cached to the appropriate value and wakeup any waiters. | |
291c7d2f JB |
690 | */ |
691 | spin_lock(&cache->lock); | |
692 | if (load_cache_only) { | |
693 | cache->caching_ctl = NULL; | |
694 | cache->cached = BTRFS_CACHE_NO; | |
695 | } else { | |
696 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 697 | cache->has_caching_ctl = 1; |
291c7d2f JB |
698 | } |
699 | spin_unlock(&cache->lock); | |
700 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
701 | } |
702 | ||
291c7d2f JB |
703 | if (load_cache_only) { |
704 | put_caching_control(caching_ctl); | |
11833d66 | 705 | return 0; |
817d52f8 | 706 | } |
817d52f8 | 707 | |
9e351cc8 | 708 | down_write(&fs_info->commit_root_sem); |
1e4f4714 | 709 | refcount_inc(&caching_ctl->count); |
11833d66 | 710 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
9e351cc8 | 711 | up_write(&fs_info->commit_root_sem); |
11833d66 | 712 | |
11dfe35a | 713 | btrfs_get_block_group(cache); |
11833d66 | 714 | |
e66f0bb1 | 715 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 716 | |
ef8bbdfe | 717 | return ret; |
e37c9e69 CM |
718 | } |
719 | ||
0f9dd46c JB |
720 | /* |
721 | * return the block group that starts at or after bytenr | |
722 | */ | |
d397712b CM |
723 | static struct btrfs_block_group_cache * |
724 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 725 | { |
e2c89907 | 726 | return block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b CM |
727 | } |
728 | ||
0f9dd46c | 729 | /* |
9f55684c | 730 | * return the block group that contains the given bytenr |
0f9dd46c | 731 | */ |
d397712b CM |
732 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
733 | struct btrfs_fs_info *info, | |
734 | u64 bytenr) | |
be744175 | 735 | { |
e2c89907 | 736 | return block_group_cache_tree_search(info, bytenr, 1); |
be744175 | 737 | } |
0b86a832 | 738 | |
0f9dd46c JB |
739 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
740 | u64 flags) | |
6324fbf3 | 741 | { |
0f9dd46c | 742 | struct list_head *head = &info->space_info; |
0f9dd46c | 743 | struct btrfs_space_info *found; |
4184ea7f | 744 | |
52ba6929 | 745 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 746 | |
4184ea7f CM |
747 | rcu_read_lock(); |
748 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 749 | if (found->flags & flags) { |
4184ea7f | 750 | rcu_read_unlock(); |
0f9dd46c | 751 | return found; |
4184ea7f | 752 | } |
0f9dd46c | 753 | } |
4184ea7f | 754 | rcu_read_unlock(); |
0f9dd46c | 755 | return NULL; |
6324fbf3 CM |
756 | } |
757 | ||
0d9f824d OS |
758 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes, |
759 | u64 owner, u64 root_objectid) | |
760 | { | |
761 | struct btrfs_space_info *space_info; | |
762 | u64 flags; | |
763 | ||
764 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
765 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) | |
766 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
767 | else | |
768 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
769 | } else { | |
770 | flags = BTRFS_BLOCK_GROUP_DATA; | |
771 | } | |
772 | ||
773 | space_info = __find_space_info(fs_info, flags); | |
55e8196a | 774 | ASSERT(space_info); |
0d9f824d OS |
775 | percpu_counter_add(&space_info->total_bytes_pinned, num_bytes); |
776 | } | |
777 | ||
4184ea7f CM |
778 | /* |
779 | * after adding space to the filesystem, we need to clear the full flags | |
780 | * on all the space infos. | |
781 | */ | |
782 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
783 | { | |
784 | struct list_head *head = &info->space_info; | |
785 | struct btrfs_space_info *found; | |
786 | ||
787 | rcu_read_lock(); | |
788 | list_for_each_entry_rcu(found, head, list) | |
789 | found->full = 0; | |
790 | rcu_read_unlock(); | |
791 | } | |
792 | ||
1a4ed8fd | 793 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 794 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
795 | { |
796 | int ret; | |
797 | struct btrfs_key key; | |
31840ae1 | 798 | struct btrfs_path *path; |
e02119d5 | 799 | |
31840ae1 | 800 | path = btrfs_alloc_path(); |
d8926bb3 MF |
801 | if (!path) |
802 | return -ENOMEM; | |
803 | ||
e02119d5 CM |
804 | key.objectid = start; |
805 | key.offset = len; | |
3173a18f | 806 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 807 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 808 | btrfs_free_path(path); |
7bb86316 CM |
809 | return ret; |
810 | } | |
811 | ||
a22285a6 | 812 | /* |
3173a18f | 813 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
814 | * |
815 | * the head node for delayed ref is used to store the sum of all the | |
816 | * reference count modifications queued up in the rbtree. the head | |
817 | * node may also store the extent flags to set. This way you can check | |
818 | * to see what the reference count and extent flags would be if all of | |
819 | * the delayed refs are not processed. | |
820 | */ | |
821 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 822 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 823 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
824 | { |
825 | struct btrfs_delayed_ref_head *head; | |
826 | struct btrfs_delayed_ref_root *delayed_refs; | |
827 | struct btrfs_path *path; | |
828 | struct btrfs_extent_item *ei; | |
829 | struct extent_buffer *leaf; | |
830 | struct btrfs_key key; | |
831 | u32 item_size; | |
832 | u64 num_refs; | |
833 | u64 extent_flags; | |
834 | int ret; | |
835 | ||
3173a18f JB |
836 | /* |
837 | * If we don't have skinny metadata, don't bother doing anything | |
838 | * different | |
839 | */ | |
0b246afa JM |
840 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
841 | offset = fs_info->nodesize; | |
3173a18f JB |
842 | metadata = 0; |
843 | } | |
844 | ||
a22285a6 YZ |
845 | path = btrfs_alloc_path(); |
846 | if (!path) | |
847 | return -ENOMEM; | |
848 | ||
a22285a6 YZ |
849 | if (!trans) { |
850 | path->skip_locking = 1; | |
851 | path->search_commit_root = 1; | |
852 | } | |
639eefc8 FDBM |
853 | |
854 | search_again: | |
855 | key.objectid = bytenr; | |
856 | key.offset = offset; | |
857 | if (metadata) | |
858 | key.type = BTRFS_METADATA_ITEM_KEY; | |
859 | else | |
860 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
861 | ||
0b246afa | 862 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
863 | if (ret < 0) |
864 | goto out_free; | |
865 | ||
3173a18f | 866 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
867 | if (path->slots[0]) { |
868 | path->slots[0]--; | |
869 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
870 | path->slots[0]); | |
871 | if (key.objectid == bytenr && | |
872 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 873 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
874 | ret = 0; |
875 | } | |
3173a18f JB |
876 | } |
877 | ||
a22285a6 YZ |
878 | if (ret == 0) { |
879 | leaf = path->nodes[0]; | |
880 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
881 | if (item_size >= sizeof(*ei)) { | |
882 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
883 | struct btrfs_extent_item); | |
884 | num_refs = btrfs_extent_refs(leaf, ei); | |
885 | extent_flags = btrfs_extent_flags(leaf, ei); | |
886 | } else { | |
887 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
888 | struct btrfs_extent_item_v0 *ei0; | |
889 | BUG_ON(item_size != sizeof(*ei0)); | |
890 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
891 | struct btrfs_extent_item_v0); | |
892 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
893 | /* FIXME: this isn't correct for data */ | |
894 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
895 | #else | |
896 | BUG(); | |
897 | #endif | |
898 | } | |
899 | BUG_ON(num_refs == 0); | |
900 | } else { | |
901 | num_refs = 0; | |
902 | extent_flags = 0; | |
903 | ret = 0; | |
904 | } | |
905 | ||
906 | if (!trans) | |
907 | goto out; | |
908 | ||
909 | delayed_refs = &trans->transaction->delayed_refs; | |
910 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 911 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
912 | if (head) { |
913 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 914 | refcount_inc(&head->refs); |
a22285a6 YZ |
915 | spin_unlock(&delayed_refs->lock); |
916 | ||
b3b4aa74 | 917 | btrfs_release_path(path); |
a22285a6 | 918 | |
8cc33e5c DS |
919 | /* |
920 | * Mutex was contended, block until it's released and try | |
921 | * again | |
922 | */ | |
a22285a6 YZ |
923 | mutex_lock(&head->mutex); |
924 | mutex_unlock(&head->mutex); | |
d278850e | 925 | btrfs_put_delayed_ref_head(head); |
639eefc8 | 926 | goto search_again; |
a22285a6 | 927 | } |
d7df2c79 | 928 | spin_lock(&head->lock); |
a22285a6 YZ |
929 | if (head->extent_op && head->extent_op->update_flags) |
930 | extent_flags |= head->extent_op->flags_to_set; | |
931 | else | |
932 | BUG_ON(num_refs == 0); | |
933 | ||
d278850e | 934 | num_refs += head->ref_mod; |
d7df2c79 | 935 | spin_unlock(&head->lock); |
a22285a6 YZ |
936 | mutex_unlock(&head->mutex); |
937 | } | |
938 | spin_unlock(&delayed_refs->lock); | |
939 | out: | |
940 | WARN_ON(num_refs == 0); | |
941 | if (refs) | |
942 | *refs = num_refs; | |
943 | if (flags) | |
944 | *flags = extent_flags; | |
945 | out_free: | |
946 | btrfs_free_path(path); | |
947 | return ret; | |
948 | } | |
949 | ||
d8d5f3e1 CM |
950 | /* |
951 | * Back reference rules. Back refs have three main goals: | |
952 | * | |
953 | * 1) differentiate between all holders of references to an extent so that | |
954 | * when a reference is dropped we can make sure it was a valid reference | |
955 | * before freeing the extent. | |
956 | * | |
957 | * 2) Provide enough information to quickly find the holders of an extent | |
958 | * if we notice a given block is corrupted or bad. | |
959 | * | |
960 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
961 | * maintenance. This is actually the same as #2, but with a slightly | |
962 | * different use case. | |
963 | * | |
5d4f98a2 YZ |
964 | * There are two kinds of back refs. The implicit back refs is optimized |
965 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
966 | * back refs of this kind provide information about the block's owner tree | |
967 | * and the pointer's key. These information allow us to find the block by | |
968 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
969 | * referenced by their owner trees. The location of tree block is recorded | |
970 | * in the back refs. Actually the full back refs is generic, and can be | |
971 | * used in all cases the implicit back refs is used. The major shortcoming | |
972 | * of the full back refs is its overhead. Every time a tree block gets | |
973 | * COWed, we have to update back refs entry for all pointers in it. | |
974 | * | |
975 | * For a newly allocated tree block, we use implicit back refs for | |
976 | * pointers in it. This means most tree related operations only involve | |
977 | * implicit back refs. For a tree block created in old transaction, the | |
978 | * only way to drop a reference to it is COW it. So we can detect the | |
979 | * event that tree block loses its owner tree's reference and do the | |
980 | * back refs conversion. | |
981 | * | |
01327610 | 982 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
983 | * |
984 | * The reference count of the block is one and the tree is the block's | |
985 | * owner tree. Nothing to do in this case. | |
986 | * | |
987 | * The reference count of the block is one and the tree is not the | |
988 | * block's owner tree. In this case, full back refs is used for pointers | |
989 | * in the block. Remove these full back refs, add implicit back refs for | |
990 | * every pointers in the new block. | |
991 | * | |
992 | * The reference count of the block is greater than one and the tree is | |
993 | * the block's owner tree. In this case, implicit back refs is used for | |
994 | * pointers in the block. Add full back refs for every pointers in the | |
995 | * block, increase lower level extents' reference counts. The original | |
996 | * implicit back refs are entailed to the new block. | |
997 | * | |
998 | * The reference count of the block is greater than one and the tree is | |
999 | * not the block's owner tree. Add implicit back refs for every pointer in | |
1000 | * the new block, increase lower level extents' reference count. | |
1001 | * | |
1002 | * Back Reference Key composing: | |
1003 | * | |
1004 | * The key objectid corresponds to the first byte in the extent, | |
1005 | * The key type is used to differentiate between types of back refs. | |
1006 | * There are different meanings of the key offset for different types | |
1007 | * of back refs. | |
1008 | * | |
d8d5f3e1 CM |
1009 | * File extents can be referenced by: |
1010 | * | |
1011 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 1012 | * - different files inside a single subvolume |
d8d5f3e1 CM |
1013 | * - different offsets inside a file (bookend extents in file.c) |
1014 | * | |
5d4f98a2 | 1015 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
1016 | * |
1017 | * - Objectid of the subvolume root | |
d8d5f3e1 | 1018 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
1019 | * - original offset in the file |
1020 | * - how many bookend extents | |
d8d5f3e1 | 1021 | * |
5d4f98a2 YZ |
1022 | * The key offset for the implicit back refs is hash of the first |
1023 | * three fields. | |
d8d5f3e1 | 1024 | * |
5d4f98a2 | 1025 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 1026 | * |
5d4f98a2 | 1027 | * - number of pointers in the tree leaf |
d8d5f3e1 | 1028 | * |
5d4f98a2 YZ |
1029 | * The key offset for the implicit back refs is the first byte of |
1030 | * the tree leaf | |
d8d5f3e1 | 1031 | * |
5d4f98a2 YZ |
1032 | * When a file extent is allocated, The implicit back refs is used. |
1033 | * the fields are filled in: | |
d8d5f3e1 | 1034 | * |
5d4f98a2 | 1035 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 1036 | * |
5d4f98a2 YZ |
1037 | * When a file extent is removed file truncation, we find the |
1038 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 1039 | * |
5d4f98a2 | 1040 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 1041 | * |
5d4f98a2 | 1042 | * Btree extents can be referenced by: |
d8d5f3e1 | 1043 | * |
5d4f98a2 | 1044 | * - Different subvolumes |
d8d5f3e1 | 1045 | * |
5d4f98a2 YZ |
1046 | * Both the implicit back refs and the full back refs for tree blocks |
1047 | * only consist of key. The key offset for the implicit back refs is | |
1048 | * objectid of block's owner tree. The key offset for the full back refs | |
1049 | * is the first byte of parent block. | |
d8d5f3e1 | 1050 | * |
5d4f98a2 YZ |
1051 | * When implicit back refs is used, information about the lowest key and |
1052 | * level of the tree block are required. These information are stored in | |
1053 | * tree block info structure. | |
d8d5f3e1 | 1054 | */ |
31840ae1 | 1055 | |
5d4f98a2 YZ |
1056 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
1057 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
87bde3cd | 1058 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1059 | struct btrfs_path *path, |
1060 | u64 owner, u32 extra_size) | |
7bb86316 | 1061 | { |
87bde3cd | 1062 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1063 | struct btrfs_extent_item *item; |
1064 | struct btrfs_extent_item_v0 *ei0; | |
1065 | struct btrfs_extent_ref_v0 *ref0; | |
1066 | struct btrfs_tree_block_info *bi; | |
1067 | struct extent_buffer *leaf; | |
7bb86316 | 1068 | struct btrfs_key key; |
5d4f98a2 YZ |
1069 | struct btrfs_key found_key; |
1070 | u32 new_size = sizeof(*item); | |
1071 | u64 refs; | |
1072 | int ret; | |
1073 | ||
1074 | leaf = path->nodes[0]; | |
1075 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
1076 | ||
1077 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1078 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
1079 | struct btrfs_extent_item_v0); | |
1080 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
1081 | ||
1082 | if (owner == (u64)-1) { | |
1083 | while (1) { | |
1084 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1085 | ret = btrfs_next_leaf(root, path); | |
1086 | if (ret < 0) | |
1087 | return ret; | |
79787eaa | 1088 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
1089 | leaf = path->nodes[0]; |
1090 | } | |
1091 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1092 | path->slots[0]); | |
1093 | BUG_ON(key.objectid != found_key.objectid); | |
1094 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
1095 | path->slots[0]++; | |
1096 | continue; | |
1097 | } | |
1098 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1099 | struct btrfs_extent_ref_v0); | |
1100 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
1101 | break; | |
1102 | } | |
1103 | } | |
b3b4aa74 | 1104 | btrfs_release_path(path); |
5d4f98a2 YZ |
1105 | |
1106 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1107 | new_size += sizeof(*bi); | |
1108 | ||
1109 | new_size -= sizeof(*ei0); | |
1110 | ret = btrfs_search_slot(trans, root, &key, path, | |
1111 | new_size + extra_size, 1); | |
1112 | if (ret < 0) | |
1113 | return ret; | |
79787eaa | 1114 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 1115 | |
87bde3cd | 1116 | btrfs_extend_item(fs_info, path, new_size); |
5d4f98a2 YZ |
1117 | |
1118 | leaf = path->nodes[0]; | |
1119 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1120 | btrfs_set_extent_refs(leaf, item, refs); | |
1121 | /* FIXME: get real generation */ | |
1122 | btrfs_set_extent_generation(leaf, item, 0); | |
1123 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1124 | btrfs_set_extent_flags(leaf, item, | |
1125 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1126 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1127 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1128 | /* FIXME: get first key of the block */ | |
b159fa28 | 1129 | memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi)); |
5d4f98a2 YZ |
1130 | btrfs_set_tree_block_level(leaf, bi, (int)owner); |
1131 | } else { | |
1132 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1133 | } | |
1134 | btrfs_mark_buffer_dirty(leaf); | |
1135 | return 0; | |
1136 | } | |
1137 | #endif | |
1138 | ||
167ce953 LB |
1139 | /* |
1140 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
1141 | * is_data == BTRFS_REF_TYPE_DATA, data type is requried, | |
1142 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. | |
1143 | */ | |
1144 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
1145 | struct btrfs_extent_inline_ref *iref, | |
1146 | enum btrfs_inline_ref_type is_data) | |
1147 | { | |
1148 | int type = btrfs_extent_inline_ref_type(eb, iref); | |
64ecdb64 | 1149 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 LB |
1150 | |
1151 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
1152 | type == BTRFS_SHARED_BLOCK_REF_KEY || | |
1153 | type == BTRFS_SHARED_DATA_REF_KEY || | |
1154 | type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1155 | if (is_data == BTRFS_REF_TYPE_BLOCK) { | |
64ecdb64 | 1156 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
167ce953 | 1157 | return type; |
64ecdb64 LB |
1158 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
1159 | ASSERT(eb->fs_info); | |
1160 | /* | |
1161 | * Every shared one has parent tree | |
1162 | * block, which must be aligned to | |
1163 | * nodesize. | |
1164 | */ | |
1165 | if (offset && | |
1166 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1167 | return type; | |
1168 | } | |
167ce953 | 1169 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 1170 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 1171 | return type; |
64ecdb64 LB |
1172 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
1173 | ASSERT(eb->fs_info); | |
1174 | /* | |
1175 | * Every shared one has parent tree | |
1176 | * block, which must be aligned to | |
1177 | * nodesize. | |
1178 | */ | |
1179 | if (offset && | |
1180 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1181 | return type; | |
1182 | } | |
167ce953 LB |
1183 | } else { |
1184 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
1185 | return type; | |
1186 | } | |
1187 | } | |
1188 | ||
1189 | btrfs_print_leaf((struct extent_buffer *)eb); | |
1190 | btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d", | |
1191 | eb->start, type); | |
1192 | WARN_ON(1); | |
1193 | ||
1194 | return BTRFS_REF_TYPE_INVALID; | |
1195 | } | |
1196 | ||
5d4f98a2 YZ |
1197 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
1198 | { | |
1199 | u32 high_crc = ~(u32)0; | |
1200 | u32 low_crc = ~(u32)0; | |
1201 | __le64 lenum; | |
1202 | ||
1203 | lenum = cpu_to_le64(root_objectid); | |
9678c543 | 1204 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1205 | lenum = cpu_to_le64(owner); |
9678c543 | 1206 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1207 | lenum = cpu_to_le64(offset); |
9678c543 | 1208 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1209 | |
1210 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1211 | } | |
1212 | ||
1213 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1214 | struct btrfs_extent_data_ref *ref) | |
1215 | { | |
1216 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1217 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1218 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1219 | } | |
1220 | ||
1221 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1222 | struct btrfs_extent_data_ref *ref, | |
1223 | u64 root_objectid, u64 owner, u64 offset) | |
1224 | { | |
1225 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1226 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1227 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1228 | return 0; | |
1229 | return 1; | |
1230 | } | |
1231 | ||
1232 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1233 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1234 | struct btrfs_path *path, |
1235 | u64 bytenr, u64 parent, | |
1236 | u64 root_objectid, | |
1237 | u64 owner, u64 offset) | |
1238 | { | |
87bde3cd | 1239 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1240 | struct btrfs_key key; |
1241 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1242 | struct extent_buffer *leaf; |
5d4f98a2 | 1243 | u32 nritems; |
74493f7a | 1244 | int ret; |
5d4f98a2 YZ |
1245 | int recow; |
1246 | int err = -ENOENT; | |
74493f7a | 1247 | |
31840ae1 | 1248 | key.objectid = bytenr; |
5d4f98a2 YZ |
1249 | if (parent) { |
1250 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1251 | key.offset = parent; | |
1252 | } else { | |
1253 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1254 | key.offset = hash_extent_data_ref(root_objectid, | |
1255 | owner, offset); | |
1256 | } | |
1257 | again: | |
1258 | recow = 0; | |
1259 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1260 | if (ret < 0) { | |
1261 | err = ret; | |
1262 | goto fail; | |
1263 | } | |
31840ae1 | 1264 | |
5d4f98a2 YZ |
1265 | if (parent) { |
1266 | if (!ret) | |
1267 | return 0; | |
1268 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1269 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1270 | btrfs_release_path(path); |
5d4f98a2 YZ |
1271 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1272 | if (ret < 0) { | |
1273 | err = ret; | |
1274 | goto fail; | |
1275 | } | |
1276 | if (!ret) | |
1277 | return 0; | |
1278 | #endif | |
1279 | goto fail; | |
31840ae1 ZY |
1280 | } |
1281 | ||
1282 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1283 | nritems = btrfs_header_nritems(leaf); |
1284 | while (1) { | |
1285 | if (path->slots[0] >= nritems) { | |
1286 | ret = btrfs_next_leaf(root, path); | |
1287 | if (ret < 0) | |
1288 | err = ret; | |
1289 | if (ret) | |
1290 | goto fail; | |
1291 | ||
1292 | leaf = path->nodes[0]; | |
1293 | nritems = btrfs_header_nritems(leaf); | |
1294 | recow = 1; | |
1295 | } | |
1296 | ||
1297 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1298 | if (key.objectid != bytenr || | |
1299 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1300 | goto fail; | |
1301 | ||
1302 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1303 | struct btrfs_extent_data_ref); | |
1304 | ||
1305 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1306 | owner, offset)) { | |
1307 | if (recow) { | |
b3b4aa74 | 1308 | btrfs_release_path(path); |
5d4f98a2 YZ |
1309 | goto again; |
1310 | } | |
1311 | err = 0; | |
1312 | break; | |
1313 | } | |
1314 | path->slots[0]++; | |
31840ae1 | 1315 | } |
5d4f98a2 YZ |
1316 | fail: |
1317 | return err; | |
31840ae1 ZY |
1318 | } |
1319 | ||
5d4f98a2 | 1320 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1321 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1322 | struct btrfs_path *path, |
1323 | u64 bytenr, u64 parent, | |
1324 | u64 root_objectid, u64 owner, | |
1325 | u64 offset, int refs_to_add) | |
31840ae1 | 1326 | { |
87bde3cd | 1327 | struct btrfs_root *root = fs_info->extent_root; |
31840ae1 ZY |
1328 | struct btrfs_key key; |
1329 | struct extent_buffer *leaf; | |
5d4f98a2 | 1330 | u32 size; |
31840ae1 ZY |
1331 | u32 num_refs; |
1332 | int ret; | |
74493f7a | 1333 | |
74493f7a | 1334 | key.objectid = bytenr; |
5d4f98a2 YZ |
1335 | if (parent) { |
1336 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1337 | key.offset = parent; | |
1338 | size = sizeof(struct btrfs_shared_data_ref); | |
1339 | } else { | |
1340 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1341 | key.offset = hash_extent_data_ref(root_objectid, | |
1342 | owner, offset); | |
1343 | size = sizeof(struct btrfs_extent_data_ref); | |
1344 | } | |
74493f7a | 1345 | |
5d4f98a2 YZ |
1346 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1347 | if (ret && ret != -EEXIST) | |
1348 | goto fail; | |
1349 | ||
1350 | leaf = path->nodes[0]; | |
1351 | if (parent) { | |
1352 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1353 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1354 | struct btrfs_shared_data_ref); |
1355 | if (ret == 0) { | |
1356 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1357 | } else { | |
1358 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1359 | num_refs += refs_to_add; | |
1360 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1361 | } |
5d4f98a2 YZ |
1362 | } else { |
1363 | struct btrfs_extent_data_ref *ref; | |
1364 | while (ret == -EEXIST) { | |
1365 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1366 | struct btrfs_extent_data_ref); | |
1367 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1368 | owner, offset)) | |
1369 | break; | |
b3b4aa74 | 1370 | btrfs_release_path(path); |
5d4f98a2 YZ |
1371 | key.offset++; |
1372 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1373 | size); | |
1374 | if (ret && ret != -EEXIST) | |
1375 | goto fail; | |
31840ae1 | 1376 | |
5d4f98a2 YZ |
1377 | leaf = path->nodes[0]; |
1378 | } | |
1379 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1380 | struct btrfs_extent_data_ref); | |
1381 | if (ret == 0) { | |
1382 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1383 | root_objectid); | |
1384 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1385 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1386 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1387 | } else { | |
1388 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1389 | num_refs += refs_to_add; | |
1390 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1391 | } |
31840ae1 | 1392 | } |
5d4f98a2 YZ |
1393 | btrfs_mark_buffer_dirty(leaf); |
1394 | ret = 0; | |
1395 | fail: | |
b3b4aa74 | 1396 | btrfs_release_path(path); |
7bb86316 | 1397 | return ret; |
74493f7a CM |
1398 | } |
1399 | ||
5d4f98a2 | 1400 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1401 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 1402 | struct btrfs_path *path, |
fcebe456 | 1403 | int refs_to_drop, int *last_ref) |
31840ae1 | 1404 | { |
5d4f98a2 YZ |
1405 | struct btrfs_key key; |
1406 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1407 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1408 | struct extent_buffer *leaf; |
5d4f98a2 | 1409 | u32 num_refs = 0; |
31840ae1 ZY |
1410 | int ret = 0; |
1411 | ||
1412 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1413 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1414 | ||
1415 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1416 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1417 | struct btrfs_extent_data_ref); | |
1418 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1419 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1420 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1421 | struct btrfs_shared_data_ref); | |
1422 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1423 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1424 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1425 | struct btrfs_extent_ref_v0 *ref0; | |
1426 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1427 | struct btrfs_extent_ref_v0); | |
1428 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1429 | #endif | |
1430 | } else { | |
1431 | BUG(); | |
1432 | } | |
1433 | ||
56bec294 CM |
1434 | BUG_ON(num_refs < refs_to_drop); |
1435 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1436 | |
31840ae1 | 1437 | if (num_refs == 0) { |
87bde3cd | 1438 | ret = btrfs_del_item(trans, fs_info->extent_root, path); |
fcebe456 | 1439 | *last_ref = 1; |
31840ae1 | 1440 | } else { |
5d4f98a2 YZ |
1441 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1442 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1443 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1444 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1445 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1446 | else { | |
1447 | struct btrfs_extent_ref_v0 *ref0; | |
1448 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1449 | struct btrfs_extent_ref_v0); | |
1450 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1451 | } | |
1452 | #endif | |
31840ae1 ZY |
1453 | btrfs_mark_buffer_dirty(leaf); |
1454 | } | |
31840ae1 ZY |
1455 | return ret; |
1456 | } | |
1457 | ||
9ed0dea0 | 1458 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1459 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1460 | { |
5d4f98a2 YZ |
1461 | struct btrfs_key key; |
1462 | struct extent_buffer *leaf; | |
1463 | struct btrfs_extent_data_ref *ref1; | |
1464 | struct btrfs_shared_data_ref *ref2; | |
1465 | u32 num_refs = 0; | |
3de28d57 | 1466 | int type; |
5d4f98a2 YZ |
1467 | |
1468 | leaf = path->nodes[0]; | |
1469 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1470 | if (iref) { | |
3de28d57 LB |
1471 | /* |
1472 | * If type is invalid, we should have bailed out earlier than | |
1473 | * this call. | |
1474 | */ | |
1475 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
1476 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
1477 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
1478 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
1479 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1480 | } else { | |
1481 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1482 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1483 | } | |
1484 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1485 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1486 | struct btrfs_extent_data_ref); | |
1487 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1488 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1489 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1490 | struct btrfs_shared_data_ref); | |
1491 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1492 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1493 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1494 | struct btrfs_extent_ref_v0 *ref0; | |
1495 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1496 | struct btrfs_extent_ref_v0); | |
1497 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1498 | #endif |
5d4f98a2 YZ |
1499 | } else { |
1500 | WARN_ON(1); | |
1501 | } | |
1502 | return num_refs; | |
1503 | } | |
15916de8 | 1504 | |
5d4f98a2 | 1505 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1506 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1507 | struct btrfs_path *path, |
1508 | u64 bytenr, u64 parent, | |
1509 | u64 root_objectid) | |
1f3c79a2 | 1510 | { |
87bde3cd | 1511 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 | 1512 | struct btrfs_key key; |
1f3c79a2 | 1513 | int ret; |
1f3c79a2 | 1514 | |
5d4f98a2 YZ |
1515 | key.objectid = bytenr; |
1516 | if (parent) { | |
1517 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1518 | key.offset = parent; | |
1519 | } else { | |
1520 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1521 | key.offset = root_objectid; | |
1f3c79a2 LH |
1522 | } |
1523 | ||
5d4f98a2 YZ |
1524 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1525 | if (ret > 0) | |
1526 | ret = -ENOENT; | |
1527 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1528 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1529 | btrfs_release_path(path); |
5d4f98a2 YZ |
1530 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1531 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1532 | if (ret > 0) | |
1533 | ret = -ENOENT; | |
1534 | } | |
1f3c79a2 | 1535 | #endif |
5d4f98a2 | 1536 | return ret; |
1f3c79a2 LH |
1537 | } |
1538 | ||
5d4f98a2 | 1539 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1540 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1541 | struct btrfs_path *path, |
1542 | u64 bytenr, u64 parent, | |
1543 | u64 root_objectid) | |
31840ae1 | 1544 | { |
5d4f98a2 | 1545 | struct btrfs_key key; |
31840ae1 | 1546 | int ret; |
31840ae1 | 1547 | |
5d4f98a2 YZ |
1548 | key.objectid = bytenr; |
1549 | if (parent) { | |
1550 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1551 | key.offset = parent; | |
1552 | } else { | |
1553 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1554 | key.offset = root_objectid; | |
1555 | } | |
1556 | ||
87bde3cd JM |
1557 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, |
1558 | path, &key, 0); | |
b3b4aa74 | 1559 | btrfs_release_path(path); |
31840ae1 ZY |
1560 | return ret; |
1561 | } | |
1562 | ||
5d4f98a2 | 1563 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1564 | { |
5d4f98a2 YZ |
1565 | int type; |
1566 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1567 | if (parent > 0) | |
1568 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1569 | else | |
1570 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1571 | } else { | |
1572 | if (parent > 0) | |
1573 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1574 | else | |
1575 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1576 | } | |
1577 | return type; | |
31840ae1 | 1578 | } |
56bec294 | 1579 | |
2c47e605 YZ |
1580 | static int find_next_key(struct btrfs_path *path, int level, |
1581 | struct btrfs_key *key) | |
56bec294 | 1582 | |
02217ed2 | 1583 | { |
2c47e605 | 1584 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1585 | if (!path->nodes[level]) |
1586 | break; | |
5d4f98a2 YZ |
1587 | if (path->slots[level] + 1 >= |
1588 | btrfs_header_nritems(path->nodes[level])) | |
1589 | continue; | |
1590 | if (level == 0) | |
1591 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1592 | path->slots[level] + 1); | |
1593 | else | |
1594 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1595 | path->slots[level] + 1); | |
1596 | return 0; | |
1597 | } | |
1598 | return 1; | |
1599 | } | |
037e6390 | 1600 | |
5d4f98a2 YZ |
1601 | /* |
1602 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1603 | * to the address of inline back ref, and 0 is returned. | |
1604 | * | |
1605 | * if back ref isn't found, *ref_ret is set to the address where it | |
1606 | * should be inserted, and -ENOENT is returned. | |
1607 | * | |
1608 | * if insert is true and there are too many inline back refs, the path | |
1609 | * points to the extent item, and -EAGAIN is returned. | |
1610 | * | |
1611 | * NOTE: inline back refs are ordered in the same way that back ref | |
1612 | * items in the tree are ordered. | |
1613 | */ | |
1614 | static noinline_for_stack | |
1615 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1616 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1617 | struct btrfs_path *path, |
1618 | struct btrfs_extent_inline_ref **ref_ret, | |
1619 | u64 bytenr, u64 num_bytes, | |
1620 | u64 parent, u64 root_objectid, | |
1621 | u64 owner, u64 offset, int insert) | |
1622 | { | |
87bde3cd | 1623 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1624 | struct btrfs_key key; |
1625 | struct extent_buffer *leaf; | |
1626 | struct btrfs_extent_item *ei; | |
1627 | struct btrfs_extent_inline_ref *iref; | |
1628 | u64 flags; | |
1629 | u64 item_size; | |
1630 | unsigned long ptr; | |
1631 | unsigned long end; | |
1632 | int extra_size; | |
1633 | int type; | |
1634 | int want; | |
1635 | int ret; | |
1636 | int err = 0; | |
0b246afa | 1637 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 1638 | int needed; |
26b8003f | 1639 | |
db94535d | 1640 | key.objectid = bytenr; |
31840ae1 | 1641 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1642 | key.offset = num_bytes; |
31840ae1 | 1643 | |
5d4f98a2 YZ |
1644 | want = extent_ref_type(parent, owner); |
1645 | if (insert) { | |
1646 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1647 | path->keep_locks = 1; |
5d4f98a2 YZ |
1648 | } else |
1649 | extra_size = -1; | |
3173a18f JB |
1650 | |
1651 | /* | |
1652 | * Owner is our parent level, so we can just add one to get the level | |
1653 | * for the block we are interested in. | |
1654 | */ | |
1655 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1656 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1657 | key.offset = owner; | |
1658 | } | |
1659 | ||
1660 | again: | |
5d4f98a2 | 1661 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1662 | if (ret < 0) { |
5d4f98a2 YZ |
1663 | err = ret; |
1664 | goto out; | |
1665 | } | |
3173a18f JB |
1666 | |
1667 | /* | |
1668 | * We may be a newly converted file system which still has the old fat | |
1669 | * extent entries for metadata, so try and see if we have one of those. | |
1670 | */ | |
1671 | if (ret > 0 && skinny_metadata) { | |
1672 | skinny_metadata = false; | |
1673 | if (path->slots[0]) { | |
1674 | path->slots[0]--; | |
1675 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1676 | path->slots[0]); | |
1677 | if (key.objectid == bytenr && | |
1678 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1679 | key.offset == num_bytes) | |
1680 | ret = 0; | |
1681 | } | |
1682 | if (ret) { | |
9ce49a0b | 1683 | key.objectid = bytenr; |
3173a18f JB |
1684 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1685 | key.offset = num_bytes; | |
1686 | btrfs_release_path(path); | |
1687 | goto again; | |
1688 | } | |
1689 | } | |
1690 | ||
79787eaa JM |
1691 | if (ret && !insert) { |
1692 | err = -ENOENT; | |
1693 | goto out; | |
fae7f21c | 1694 | } else if (WARN_ON(ret)) { |
492104c8 | 1695 | err = -EIO; |
492104c8 | 1696 | goto out; |
79787eaa | 1697 | } |
5d4f98a2 YZ |
1698 | |
1699 | leaf = path->nodes[0]; | |
1700 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1701 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1702 | if (item_size < sizeof(*ei)) { | |
1703 | if (!insert) { | |
1704 | err = -ENOENT; | |
1705 | goto out; | |
1706 | } | |
87bde3cd | 1707 | ret = convert_extent_item_v0(trans, fs_info, path, owner, |
5d4f98a2 YZ |
1708 | extra_size); |
1709 | if (ret < 0) { | |
1710 | err = ret; | |
1711 | goto out; | |
1712 | } | |
1713 | leaf = path->nodes[0]; | |
1714 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1715 | } | |
1716 | #endif | |
1717 | BUG_ON(item_size < sizeof(*ei)); | |
1718 | ||
5d4f98a2 YZ |
1719 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1720 | flags = btrfs_extent_flags(leaf, ei); | |
1721 | ||
1722 | ptr = (unsigned long)(ei + 1); | |
1723 | end = (unsigned long)ei + item_size; | |
1724 | ||
3173a18f | 1725 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1726 | ptr += sizeof(struct btrfs_tree_block_info); |
1727 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1728 | } |
1729 | ||
3de28d57 LB |
1730 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
1731 | needed = BTRFS_REF_TYPE_DATA; | |
1732 | else | |
1733 | needed = BTRFS_REF_TYPE_BLOCK; | |
1734 | ||
5d4f98a2 YZ |
1735 | err = -ENOENT; |
1736 | while (1) { | |
1737 | if (ptr >= end) { | |
1738 | WARN_ON(ptr > end); | |
1739 | break; | |
1740 | } | |
1741 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
1742 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
1743 | if (type == BTRFS_REF_TYPE_INVALID) { | |
1744 | err = -EINVAL; | |
1745 | goto out; | |
1746 | } | |
1747 | ||
5d4f98a2 YZ |
1748 | if (want < type) |
1749 | break; | |
1750 | if (want > type) { | |
1751 | ptr += btrfs_extent_inline_ref_size(type); | |
1752 | continue; | |
1753 | } | |
1754 | ||
1755 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1756 | struct btrfs_extent_data_ref *dref; | |
1757 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1758 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1759 | owner, offset)) { | |
1760 | err = 0; | |
1761 | break; | |
1762 | } | |
1763 | if (hash_extent_data_ref_item(leaf, dref) < | |
1764 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1765 | break; | |
1766 | } else { | |
1767 | u64 ref_offset; | |
1768 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1769 | if (parent > 0) { | |
1770 | if (parent == ref_offset) { | |
1771 | err = 0; | |
1772 | break; | |
1773 | } | |
1774 | if (ref_offset < parent) | |
1775 | break; | |
1776 | } else { | |
1777 | if (root_objectid == ref_offset) { | |
1778 | err = 0; | |
1779 | break; | |
1780 | } | |
1781 | if (ref_offset < root_objectid) | |
1782 | break; | |
1783 | } | |
1784 | } | |
1785 | ptr += btrfs_extent_inline_ref_size(type); | |
1786 | } | |
1787 | if (err == -ENOENT && insert) { | |
1788 | if (item_size + extra_size >= | |
1789 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1790 | err = -EAGAIN; | |
1791 | goto out; | |
1792 | } | |
1793 | /* | |
1794 | * To add new inline back ref, we have to make sure | |
1795 | * there is no corresponding back ref item. | |
1796 | * For simplicity, we just do not add new inline back | |
1797 | * ref if there is any kind of item for this block | |
1798 | */ | |
2c47e605 YZ |
1799 | if (find_next_key(path, 0, &key) == 0 && |
1800 | key.objectid == bytenr && | |
85d4198e | 1801 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1802 | err = -EAGAIN; |
1803 | goto out; | |
1804 | } | |
1805 | } | |
1806 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1807 | out: | |
85d4198e | 1808 | if (insert) { |
5d4f98a2 YZ |
1809 | path->keep_locks = 0; |
1810 | btrfs_unlock_up_safe(path, 1); | |
1811 | } | |
1812 | return err; | |
1813 | } | |
1814 | ||
1815 | /* | |
1816 | * helper to add new inline back ref | |
1817 | */ | |
1818 | static noinline_for_stack | |
87bde3cd | 1819 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1820 | struct btrfs_path *path, |
1821 | struct btrfs_extent_inline_ref *iref, | |
1822 | u64 parent, u64 root_objectid, | |
1823 | u64 owner, u64 offset, int refs_to_add, | |
1824 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1825 | { |
1826 | struct extent_buffer *leaf; | |
1827 | struct btrfs_extent_item *ei; | |
1828 | unsigned long ptr; | |
1829 | unsigned long end; | |
1830 | unsigned long item_offset; | |
1831 | u64 refs; | |
1832 | int size; | |
1833 | int type; | |
5d4f98a2 YZ |
1834 | |
1835 | leaf = path->nodes[0]; | |
1836 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1837 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1838 | ||
1839 | type = extent_ref_type(parent, owner); | |
1840 | size = btrfs_extent_inline_ref_size(type); | |
1841 | ||
87bde3cd | 1842 | btrfs_extend_item(fs_info, path, size); |
5d4f98a2 YZ |
1843 | |
1844 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1845 | refs = btrfs_extent_refs(leaf, ei); | |
1846 | refs += refs_to_add; | |
1847 | btrfs_set_extent_refs(leaf, ei, refs); | |
1848 | if (extent_op) | |
1849 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1850 | ||
1851 | ptr = (unsigned long)ei + item_offset; | |
1852 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1853 | if (ptr < end - size) | |
1854 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1855 | end - size - ptr); | |
1856 | ||
1857 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1858 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1859 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1860 | struct btrfs_extent_data_ref *dref; | |
1861 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1862 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1863 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1864 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1865 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1866 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1867 | struct btrfs_shared_data_ref *sref; | |
1868 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1869 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1870 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1871 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1872 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1873 | } else { | |
1874 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1875 | } | |
1876 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1877 | } |
1878 | ||
1879 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1880 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1881 | struct btrfs_path *path, |
1882 | struct btrfs_extent_inline_ref **ref_ret, | |
1883 | u64 bytenr, u64 num_bytes, u64 parent, | |
1884 | u64 root_objectid, u64 owner, u64 offset) | |
1885 | { | |
1886 | int ret; | |
1887 | ||
87bde3cd | 1888 | ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret, |
5d4f98a2 YZ |
1889 | bytenr, num_bytes, parent, |
1890 | root_objectid, owner, offset, 0); | |
1891 | if (ret != -ENOENT) | |
54aa1f4d | 1892 | return ret; |
5d4f98a2 | 1893 | |
b3b4aa74 | 1894 | btrfs_release_path(path); |
5d4f98a2 YZ |
1895 | *ref_ret = NULL; |
1896 | ||
1897 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
87bde3cd JM |
1898 | ret = lookup_tree_block_ref(trans, fs_info, path, bytenr, |
1899 | parent, root_objectid); | |
5d4f98a2 | 1900 | } else { |
87bde3cd JM |
1901 | ret = lookup_extent_data_ref(trans, fs_info, path, bytenr, |
1902 | parent, root_objectid, owner, | |
1903 | offset); | |
b9473439 | 1904 | } |
5d4f98a2 YZ |
1905 | return ret; |
1906 | } | |
31840ae1 | 1907 | |
5d4f98a2 YZ |
1908 | /* |
1909 | * helper to update/remove inline back ref | |
1910 | */ | |
1911 | static noinline_for_stack | |
87bde3cd | 1912 | void update_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1913 | struct btrfs_path *path, |
1914 | struct btrfs_extent_inline_ref *iref, | |
1915 | int refs_to_mod, | |
fcebe456 JB |
1916 | struct btrfs_delayed_extent_op *extent_op, |
1917 | int *last_ref) | |
5d4f98a2 YZ |
1918 | { |
1919 | struct extent_buffer *leaf; | |
1920 | struct btrfs_extent_item *ei; | |
1921 | struct btrfs_extent_data_ref *dref = NULL; | |
1922 | struct btrfs_shared_data_ref *sref = NULL; | |
1923 | unsigned long ptr; | |
1924 | unsigned long end; | |
1925 | u32 item_size; | |
1926 | int size; | |
1927 | int type; | |
5d4f98a2 YZ |
1928 | u64 refs; |
1929 | ||
1930 | leaf = path->nodes[0]; | |
1931 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1932 | refs = btrfs_extent_refs(leaf, ei); | |
1933 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1934 | refs += refs_to_mod; | |
1935 | btrfs_set_extent_refs(leaf, ei, refs); | |
1936 | if (extent_op) | |
1937 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1938 | ||
3de28d57 LB |
1939 | /* |
1940 | * If type is invalid, we should have bailed out after | |
1941 | * lookup_inline_extent_backref(). | |
1942 | */ | |
1943 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
1944 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
5d4f98a2 YZ |
1945 | |
1946 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1947 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1948 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1949 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1950 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1951 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1952 | } else { | |
1953 | refs = 1; | |
1954 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1955 | } |
31840ae1 | 1956 | |
5d4f98a2 YZ |
1957 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1958 | refs += refs_to_mod; | |
1959 | ||
1960 | if (refs > 0) { | |
1961 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1962 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1963 | else | |
1964 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1965 | } else { | |
fcebe456 | 1966 | *last_ref = 1; |
5d4f98a2 YZ |
1967 | size = btrfs_extent_inline_ref_size(type); |
1968 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1969 | ptr = (unsigned long)iref; | |
1970 | end = (unsigned long)ei + item_size; | |
1971 | if (ptr + size < end) | |
1972 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1973 | end - ptr - size); | |
1974 | item_size -= size; | |
87bde3cd | 1975 | btrfs_truncate_item(fs_info, path, item_size, 1); |
5d4f98a2 YZ |
1976 | } |
1977 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1978 | } |
1979 | ||
1980 | static noinline_for_stack | |
1981 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1982 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1983 | struct btrfs_path *path, |
1984 | u64 bytenr, u64 num_bytes, u64 parent, | |
1985 | u64 root_objectid, u64 owner, | |
1986 | u64 offset, int refs_to_add, | |
1987 | struct btrfs_delayed_extent_op *extent_op) | |
1988 | { | |
1989 | struct btrfs_extent_inline_ref *iref; | |
1990 | int ret; | |
1991 | ||
87bde3cd | 1992 | ret = lookup_inline_extent_backref(trans, fs_info, path, &iref, |
5d4f98a2 YZ |
1993 | bytenr, num_bytes, parent, |
1994 | root_objectid, owner, offset, 1); | |
1995 | if (ret == 0) { | |
1996 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
87bde3cd | 1997 | update_inline_extent_backref(fs_info, path, iref, |
fcebe456 | 1998 | refs_to_add, extent_op, NULL); |
5d4f98a2 | 1999 | } else if (ret == -ENOENT) { |
87bde3cd | 2000 | setup_inline_extent_backref(fs_info, path, iref, parent, |
143bede5 JM |
2001 | root_objectid, owner, offset, |
2002 | refs_to_add, extent_op); | |
2003 | ret = 0; | |
771ed689 | 2004 | } |
5d4f98a2 YZ |
2005 | return ret; |
2006 | } | |
31840ae1 | 2007 | |
5d4f98a2 | 2008 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
87bde3cd | 2009 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2010 | struct btrfs_path *path, |
2011 | u64 bytenr, u64 parent, u64 root_objectid, | |
2012 | u64 owner, u64 offset, int refs_to_add) | |
2013 | { | |
2014 | int ret; | |
2015 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
2016 | BUG_ON(refs_to_add != 1); | |
87bde3cd | 2017 | ret = insert_tree_block_ref(trans, fs_info, path, bytenr, |
5d4f98a2 YZ |
2018 | parent, root_objectid); |
2019 | } else { | |
87bde3cd | 2020 | ret = insert_extent_data_ref(trans, fs_info, path, bytenr, |
5d4f98a2 YZ |
2021 | parent, root_objectid, |
2022 | owner, offset, refs_to_add); | |
2023 | } | |
2024 | return ret; | |
2025 | } | |
56bec294 | 2026 | |
5d4f98a2 | 2027 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
87bde3cd | 2028 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2029 | struct btrfs_path *path, |
2030 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 2031 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 2032 | { |
143bede5 | 2033 | int ret = 0; |
b9473439 | 2034 | |
5d4f98a2 YZ |
2035 | BUG_ON(!is_data && refs_to_drop != 1); |
2036 | if (iref) { | |
87bde3cd | 2037 | update_inline_extent_backref(fs_info, path, iref, |
fcebe456 | 2038 | -refs_to_drop, NULL, last_ref); |
5d4f98a2 | 2039 | } else if (is_data) { |
87bde3cd | 2040 | ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop, |
fcebe456 | 2041 | last_ref); |
5d4f98a2 | 2042 | } else { |
fcebe456 | 2043 | *last_ref = 1; |
87bde3cd | 2044 | ret = btrfs_del_item(trans, fs_info->extent_root, path); |
5d4f98a2 YZ |
2045 | } |
2046 | return ret; | |
2047 | } | |
2048 | ||
86557861 | 2049 | #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len)) |
d04c6b88 JM |
2050 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
2051 | u64 *discarded_bytes) | |
5d4f98a2 | 2052 | { |
86557861 JM |
2053 | int j, ret = 0; |
2054 | u64 bytes_left, end; | |
4d89d377 | 2055 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 2056 | |
4d89d377 JM |
2057 | if (WARN_ON(start != aligned_start)) { |
2058 | len -= aligned_start - start; | |
2059 | len = round_down(len, 1 << 9); | |
2060 | start = aligned_start; | |
2061 | } | |
d04c6b88 | 2062 | |
4d89d377 | 2063 | *discarded_bytes = 0; |
86557861 JM |
2064 | |
2065 | if (!len) | |
2066 | return 0; | |
2067 | ||
2068 | end = start + len; | |
2069 | bytes_left = len; | |
2070 | ||
2071 | /* Skip any superblocks on this device. */ | |
2072 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
2073 | u64 sb_start = btrfs_sb_offset(j); | |
2074 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
2075 | u64 size = sb_start - start; | |
2076 | ||
2077 | if (!in_range(sb_start, start, bytes_left) && | |
2078 | !in_range(sb_end, start, bytes_left) && | |
2079 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
2080 | continue; | |
2081 | ||
2082 | /* | |
2083 | * Superblock spans beginning of range. Adjust start and | |
2084 | * try again. | |
2085 | */ | |
2086 | if (sb_start <= start) { | |
2087 | start += sb_end - start; | |
2088 | if (start > end) { | |
2089 | bytes_left = 0; | |
2090 | break; | |
2091 | } | |
2092 | bytes_left = end - start; | |
2093 | continue; | |
2094 | } | |
2095 | ||
2096 | if (size) { | |
2097 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
2098 | GFP_NOFS, 0); | |
2099 | if (!ret) | |
2100 | *discarded_bytes += size; | |
2101 | else if (ret != -EOPNOTSUPP) | |
2102 | return ret; | |
2103 | } | |
2104 | ||
2105 | start = sb_end; | |
2106 | if (start > end) { | |
2107 | bytes_left = 0; | |
2108 | break; | |
2109 | } | |
2110 | bytes_left = end - start; | |
2111 | } | |
2112 | ||
2113 | if (bytes_left) { | |
2114 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
2115 | GFP_NOFS, 0); |
2116 | if (!ret) | |
86557861 | 2117 | *discarded_bytes += bytes_left; |
4d89d377 | 2118 | } |
d04c6b88 | 2119 | return ret; |
5d4f98a2 | 2120 | } |
5d4f98a2 | 2121 | |
2ff7e61e | 2122 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 2123 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 2124 | { |
5d4f98a2 | 2125 | int ret; |
5378e607 | 2126 | u64 discarded_bytes = 0; |
a1d3c478 | 2127 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 2128 | |
e244a0ae | 2129 | |
2999241d FM |
2130 | /* |
2131 | * Avoid races with device replace and make sure our bbio has devices | |
2132 | * associated to its stripes that don't go away while we are discarding. | |
2133 | */ | |
0b246afa | 2134 | btrfs_bio_counter_inc_blocked(fs_info); |
5d4f98a2 | 2135 | /* Tell the block device(s) that the sectors can be discarded */ |
0b246afa JM |
2136 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes, |
2137 | &bbio, 0); | |
79787eaa | 2138 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 2139 | if (!ret) { |
a1d3c478 | 2140 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
2141 | int i; |
2142 | ||
5d4f98a2 | 2143 | |
a1d3c478 | 2144 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 2145 | u64 bytes; |
38b5f68e AJ |
2146 | struct request_queue *req_q; |
2147 | ||
627e0873 FM |
2148 | if (!stripe->dev->bdev) { |
2149 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); | |
2150 | continue; | |
2151 | } | |
38b5f68e AJ |
2152 | req_q = bdev_get_queue(stripe->dev->bdev); |
2153 | if (!blk_queue_discard(req_q)) | |
d5e2003c JB |
2154 | continue; |
2155 | ||
5378e607 LD |
2156 | ret = btrfs_issue_discard(stripe->dev->bdev, |
2157 | stripe->physical, | |
d04c6b88 JM |
2158 | stripe->length, |
2159 | &bytes); | |
5378e607 | 2160 | if (!ret) |
d04c6b88 | 2161 | discarded_bytes += bytes; |
5378e607 | 2162 | else if (ret != -EOPNOTSUPP) |
79787eaa | 2163 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
2164 | |
2165 | /* | |
2166 | * Just in case we get back EOPNOTSUPP for some reason, | |
2167 | * just ignore the return value so we don't screw up | |
2168 | * people calling discard_extent. | |
2169 | */ | |
2170 | ret = 0; | |
5d4f98a2 | 2171 | } |
6e9606d2 | 2172 | btrfs_put_bbio(bbio); |
5d4f98a2 | 2173 | } |
0b246afa | 2174 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
2175 | |
2176 | if (actual_bytes) | |
2177 | *actual_bytes = discarded_bytes; | |
2178 | ||
5d4f98a2 | 2179 | |
53b381b3 DW |
2180 | if (ret == -EOPNOTSUPP) |
2181 | ret = 0; | |
5d4f98a2 | 2182 | return ret; |
5d4f98a2 YZ |
2183 | } |
2184 | ||
79787eaa | 2185 | /* Can return -ENOMEM */ |
5d4f98a2 | 2186 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
84f7d8e6 | 2187 | struct btrfs_root *root, |
5d4f98a2 | 2188 | u64 bytenr, u64 num_bytes, u64 parent, |
b06c4bf5 | 2189 | u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 | 2190 | { |
84f7d8e6 | 2191 | struct btrfs_fs_info *fs_info = root->fs_info; |
d7eae340 | 2192 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 2193 | int ret; |
66d7e7f0 | 2194 | |
5d4f98a2 YZ |
2195 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
2196 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
2197 | ||
fd708b81 JB |
2198 | btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid, |
2199 | owner, offset, BTRFS_ADD_DELAYED_REF); | |
2200 | ||
5d4f98a2 | 2201 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 | 2202 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
7be07912 OS |
2203 | num_bytes, parent, |
2204 | root_objectid, (int)owner, | |
2205 | BTRFS_ADD_DELAYED_REF, NULL, | |
d7eae340 | 2206 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 2207 | } else { |
66d7e7f0 | 2208 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
7be07912 OS |
2209 | num_bytes, parent, |
2210 | root_objectid, owner, offset, | |
d7eae340 OS |
2211 | 0, BTRFS_ADD_DELAYED_REF, |
2212 | &old_ref_mod, &new_ref_mod); | |
5d4f98a2 | 2213 | } |
d7eae340 OS |
2214 | |
2215 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) | |
2216 | add_pinned_bytes(fs_info, -num_bytes, owner, root_objectid); | |
2217 | ||
5d4f98a2 YZ |
2218 | return ret; |
2219 | } | |
2220 | ||
2221 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2222 | struct btrfs_fs_info *fs_info, |
c682f9b3 | 2223 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2224 | u64 parent, u64 root_objectid, |
2225 | u64 owner, u64 offset, int refs_to_add, | |
2226 | struct btrfs_delayed_extent_op *extent_op) | |
2227 | { | |
2228 | struct btrfs_path *path; | |
2229 | struct extent_buffer *leaf; | |
2230 | struct btrfs_extent_item *item; | |
fcebe456 | 2231 | struct btrfs_key key; |
c682f9b3 QW |
2232 | u64 bytenr = node->bytenr; |
2233 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2234 | u64 refs; |
2235 | int ret; | |
5d4f98a2 YZ |
2236 | |
2237 | path = btrfs_alloc_path(); | |
2238 | if (!path) | |
2239 | return -ENOMEM; | |
2240 | ||
e4058b54 | 2241 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2242 | path->leave_spinning = 1; |
2243 | /* this will setup the path even if it fails to insert the back ref */ | |
87bde3cd JM |
2244 | ret = insert_inline_extent_backref(trans, fs_info, path, bytenr, |
2245 | num_bytes, parent, root_objectid, | |
2246 | owner, offset, | |
5d4f98a2 | 2247 | refs_to_add, extent_op); |
0ed4792a | 2248 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2249 | goto out; |
fcebe456 JB |
2250 | |
2251 | /* | |
2252 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2253 | * inline extent ref, so just update the reference count and add a | |
2254 | * normal backref. | |
2255 | */ | |
5d4f98a2 | 2256 | leaf = path->nodes[0]; |
fcebe456 | 2257 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2258 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2259 | refs = btrfs_extent_refs(leaf, item); | |
2260 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2261 | if (extent_op) | |
2262 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2263 | |
5d4f98a2 | 2264 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2265 | btrfs_release_path(path); |
56bec294 | 2266 | |
e4058b54 | 2267 | path->reada = READA_FORWARD; |
b9473439 | 2268 | path->leave_spinning = 1; |
56bec294 | 2269 | /* now insert the actual backref */ |
87bde3cd JM |
2270 | ret = insert_extent_backref(trans, fs_info, path, bytenr, parent, |
2271 | root_objectid, owner, offset, refs_to_add); | |
79787eaa | 2272 | if (ret) |
66642832 | 2273 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 2274 | out: |
56bec294 | 2275 | btrfs_free_path(path); |
30d133fc | 2276 | return ret; |
56bec294 CM |
2277 | } |
2278 | ||
5d4f98a2 | 2279 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2280 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2281 | struct btrfs_delayed_ref_node *node, |
2282 | struct btrfs_delayed_extent_op *extent_op, | |
2283 | int insert_reserved) | |
56bec294 | 2284 | { |
5d4f98a2 YZ |
2285 | int ret = 0; |
2286 | struct btrfs_delayed_data_ref *ref; | |
2287 | struct btrfs_key ins; | |
2288 | u64 parent = 0; | |
2289 | u64 ref_root = 0; | |
2290 | u64 flags = 0; | |
2291 | ||
2292 | ins.objectid = node->bytenr; | |
2293 | ins.offset = node->num_bytes; | |
2294 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2295 | ||
2296 | ref = btrfs_delayed_node_to_data_ref(node); | |
0b246afa | 2297 | trace_run_delayed_data_ref(fs_info, node, ref, node->action); |
599c75ec | 2298 | |
5d4f98a2 YZ |
2299 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2300 | parent = ref->parent; | |
fcebe456 | 2301 | ref_root = ref->root; |
5d4f98a2 YZ |
2302 | |
2303 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2304 | if (extent_op) |
5d4f98a2 | 2305 | flags |= extent_op->flags_to_set; |
2ff7e61e | 2306 | ret = alloc_reserved_file_extent(trans, fs_info, |
5d4f98a2 YZ |
2307 | parent, ref_root, flags, |
2308 | ref->objectid, ref->offset, | |
2309 | &ins, node->ref_mod); | |
5d4f98a2 | 2310 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2ff7e61e | 2311 | ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent, |
5d4f98a2 YZ |
2312 | ref_root, ref->objectid, |
2313 | ref->offset, node->ref_mod, | |
c682f9b3 | 2314 | extent_op); |
5d4f98a2 | 2315 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
2ff7e61e | 2316 | ret = __btrfs_free_extent(trans, fs_info, node, parent, |
5d4f98a2 YZ |
2317 | ref_root, ref->objectid, |
2318 | ref->offset, node->ref_mod, | |
c682f9b3 | 2319 | extent_op); |
5d4f98a2 YZ |
2320 | } else { |
2321 | BUG(); | |
2322 | } | |
2323 | return ret; | |
2324 | } | |
2325 | ||
2326 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2327 | struct extent_buffer *leaf, | |
2328 | struct btrfs_extent_item *ei) | |
2329 | { | |
2330 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2331 | if (extent_op->update_flags) { | |
2332 | flags |= extent_op->flags_to_set; | |
2333 | btrfs_set_extent_flags(leaf, ei, flags); | |
2334 | } | |
2335 | ||
2336 | if (extent_op->update_key) { | |
2337 | struct btrfs_tree_block_info *bi; | |
2338 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2339 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2340 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2341 | } | |
2342 | } | |
2343 | ||
2344 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2345 | struct btrfs_fs_info *fs_info, |
d278850e | 2346 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
2347 | struct btrfs_delayed_extent_op *extent_op) |
2348 | { | |
2349 | struct btrfs_key key; | |
2350 | struct btrfs_path *path; | |
2351 | struct btrfs_extent_item *ei; | |
2352 | struct extent_buffer *leaf; | |
2353 | u32 item_size; | |
56bec294 | 2354 | int ret; |
5d4f98a2 | 2355 | int err = 0; |
b1c79e09 | 2356 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2357 | |
79787eaa JM |
2358 | if (trans->aborted) |
2359 | return 0; | |
2360 | ||
0b246afa | 2361 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
2362 | metadata = 0; |
2363 | ||
5d4f98a2 YZ |
2364 | path = btrfs_alloc_path(); |
2365 | if (!path) | |
2366 | return -ENOMEM; | |
2367 | ||
d278850e | 2368 | key.objectid = head->bytenr; |
5d4f98a2 | 2369 | |
3173a18f | 2370 | if (metadata) { |
3173a18f | 2371 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2372 | key.offset = extent_op->level; |
3173a18f JB |
2373 | } else { |
2374 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 2375 | key.offset = head->num_bytes; |
3173a18f JB |
2376 | } |
2377 | ||
2378 | again: | |
e4058b54 | 2379 | path->reada = READA_FORWARD; |
5d4f98a2 | 2380 | path->leave_spinning = 1; |
0b246afa | 2381 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
2382 | if (ret < 0) { |
2383 | err = ret; | |
2384 | goto out; | |
2385 | } | |
2386 | if (ret > 0) { | |
3173a18f | 2387 | if (metadata) { |
55994887 FDBM |
2388 | if (path->slots[0] > 0) { |
2389 | path->slots[0]--; | |
2390 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2391 | path->slots[0]); | |
d278850e | 2392 | if (key.objectid == head->bytenr && |
55994887 | 2393 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 2394 | key.offset == head->num_bytes) |
55994887 FDBM |
2395 | ret = 0; |
2396 | } | |
2397 | if (ret > 0) { | |
2398 | btrfs_release_path(path); | |
2399 | metadata = 0; | |
3173a18f | 2400 | |
d278850e JB |
2401 | key.objectid = head->bytenr; |
2402 | key.offset = head->num_bytes; | |
55994887 FDBM |
2403 | key.type = BTRFS_EXTENT_ITEM_KEY; |
2404 | goto again; | |
2405 | } | |
2406 | } else { | |
2407 | err = -EIO; | |
2408 | goto out; | |
3173a18f | 2409 | } |
5d4f98a2 YZ |
2410 | } |
2411 | ||
2412 | leaf = path->nodes[0]; | |
2413 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2414 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2415 | if (item_size < sizeof(*ei)) { | |
87bde3cd | 2416 | ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0); |
5d4f98a2 YZ |
2417 | if (ret < 0) { |
2418 | err = ret; | |
2419 | goto out; | |
2420 | } | |
2421 | leaf = path->nodes[0]; | |
2422 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2423 | } | |
2424 | #endif | |
2425 | BUG_ON(item_size < sizeof(*ei)); | |
2426 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2427 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2428 | |
5d4f98a2 YZ |
2429 | btrfs_mark_buffer_dirty(leaf); |
2430 | out: | |
2431 | btrfs_free_path(path); | |
2432 | return err; | |
56bec294 CM |
2433 | } |
2434 | ||
5d4f98a2 | 2435 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2436 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2437 | struct btrfs_delayed_ref_node *node, |
2438 | struct btrfs_delayed_extent_op *extent_op, | |
2439 | int insert_reserved) | |
56bec294 CM |
2440 | { |
2441 | int ret = 0; | |
5d4f98a2 YZ |
2442 | struct btrfs_delayed_tree_ref *ref; |
2443 | struct btrfs_key ins; | |
2444 | u64 parent = 0; | |
2445 | u64 ref_root = 0; | |
0b246afa | 2446 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
56bec294 | 2447 | |
5d4f98a2 | 2448 | ref = btrfs_delayed_node_to_tree_ref(node); |
0b246afa | 2449 | trace_run_delayed_tree_ref(fs_info, node, ref, node->action); |
599c75ec | 2450 | |
5d4f98a2 YZ |
2451 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2452 | parent = ref->parent; | |
fcebe456 | 2453 | ref_root = ref->root; |
5d4f98a2 | 2454 | |
3173a18f JB |
2455 | ins.objectid = node->bytenr; |
2456 | if (skinny_metadata) { | |
2457 | ins.offset = ref->level; | |
2458 | ins.type = BTRFS_METADATA_ITEM_KEY; | |
2459 | } else { | |
2460 | ins.offset = node->num_bytes; | |
2461 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2462 | } | |
2463 | ||
02794222 | 2464 | if (node->ref_mod != 1) { |
2ff7e61e | 2465 | btrfs_err(fs_info, |
02794222 LB |
2466 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
2467 | node->bytenr, node->ref_mod, node->action, ref_root, | |
2468 | parent); | |
2469 | return -EIO; | |
2470 | } | |
5d4f98a2 | 2471 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 2472 | BUG_ON(!extent_op || !extent_op->update_flags); |
2ff7e61e | 2473 | ret = alloc_reserved_tree_block(trans, fs_info, |
5d4f98a2 YZ |
2474 | parent, ref_root, |
2475 | extent_op->flags_to_set, | |
2476 | &extent_op->key, | |
b06c4bf5 | 2477 | ref->level, &ins); |
5d4f98a2 | 2478 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2ff7e61e | 2479 | ret = __btrfs_inc_extent_ref(trans, fs_info, node, |
c682f9b3 QW |
2480 | parent, ref_root, |
2481 | ref->level, 0, 1, | |
fcebe456 | 2482 | extent_op); |
5d4f98a2 | 2483 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
2ff7e61e | 2484 | ret = __btrfs_free_extent(trans, fs_info, node, |
c682f9b3 QW |
2485 | parent, ref_root, |
2486 | ref->level, 0, 1, extent_op); | |
5d4f98a2 YZ |
2487 | } else { |
2488 | BUG(); | |
2489 | } | |
56bec294 CM |
2490 | return ret; |
2491 | } | |
2492 | ||
2493 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 2494 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2495 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2496 | struct btrfs_delayed_ref_node *node, |
2497 | struct btrfs_delayed_extent_op *extent_op, | |
2498 | int insert_reserved) | |
56bec294 | 2499 | { |
79787eaa JM |
2500 | int ret = 0; |
2501 | ||
857cc2fc JB |
2502 | if (trans->aborted) { |
2503 | if (insert_reserved) | |
2ff7e61e | 2504 | btrfs_pin_extent(fs_info, node->bytenr, |
857cc2fc | 2505 | node->num_bytes, 1); |
79787eaa | 2506 | return 0; |
857cc2fc | 2507 | } |
79787eaa | 2508 | |
5d4f98a2 YZ |
2509 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2510 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2ff7e61e | 2511 | ret = run_delayed_tree_ref(trans, fs_info, node, extent_op, |
5d4f98a2 YZ |
2512 | insert_reserved); |
2513 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2514 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2ff7e61e | 2515 | ret = run_delayed_data_ref(trans, fs_info, node, extent_op, |
5d4f98a2 YZ |
2516 | insert_reserved); |
2517 | else | |
2518 | BUG(); | |
2519 | return ret; | |
56bec294 CM |
2520 | } |
2521 | ||
c6fc2454 | 2522 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2523 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2524 | { | |
cffc3374 FM |
2525 | struct btrfs_delayed_ref_node *ref; |
2526 | ||
0e0adbcf | 2527 | if (RB_EMPTY_ROOT(&head->ref_tree)) |
c6fc2454 | 2528 | return NULL; |
d7df2c79 | 2529 | |
cffc3374 FM |
2530 | /* |
2531 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2532 | * This is to prevent a ref count from going down to zero, which deletes | |
2533 | * the extent item from the extent tree, when there still are references | |
2534 | * to add, which would fail because they would not find the extent item. | |
2535 | */ | |
1d57ee94 WX |
2536 | if (!list_empty(&head->ref_add_list)) |
2537 | return list_first_entry(&head->ref_add_list, | |
2538 | struct btrfs_delayed_ref_node, add_list); | |
2539 | ||
0e0adbcf JB |
2540 | ref = rb_entry(rb_first(&head->ref_tree), |
2541 | struct btrfs_delayed_ref_node, ref_node); | |
1d57ee94 WX |
2542 | ASSERT(list_empty(&ref->add_list)); |
2543 | return ref; | |
56bec294 CM |
2544 | } |
2545 | ||
2eadaa22 JB |
2546 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
2547 | struct btrfs_delayed_ref_head *head) | |
2548 | { | |
2549 | spin_lock(&delayed_refs->lock); | |
2550 | head->processing = 0; | |
2551 | delayed_refs->num_heads_ready++; | |
2552 | spin_unlock(&delayed_refs->lock); | |
2553 | btrfs_delayed_ref_unlock(head); | |
2554 | } | |
2555 | ||
b00e6250 JB |
2556 | static int cleanup_extent_op(struct btrfs_trans_handle *trans, |
2557 | struct btrfs_fs_info *fs_info, | |
2558 | struct btrfs_delayed_ref_head *head) | |
2559 | { | |
2560 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
2561 | int ret; | |
2562 | ||
2563 | if (!extent_op) | |
2564 | return 0; | |
2565 | head->extent_op = NULL; | |
2566 | if (head->must_insert_reserved) { | |
2567 | btrfs_free_delayed_extent_op(extent_op); | |
2568 | return 0; | |
2569 | } | |
2570 | spin_unlock(&head->lock); | |
d278850e | 2571 | ret = run_delayed_extent_op(trans, fs_info, head, extent_op); |
b00e6250 JB |
2572 | btrfs_free_delayed_extent_op(extent_op); |
2573 | return ret ? ret : 1; | |
2574 | } | |
2575 | ||
194ab0bc JB |
2576 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
2577 | struct btrfs_fs_info *fs_info, | |
2578 | struct btrfs_delayed_ref_head *head) | |
2579 | { | |
2580 | struct btrfs_delayed_ref_root *delayed_refs; | |
2581 | int ret; | |
2582 | ||
2583 | delayed_refs = &trans->transaction->delayed_refs; | |
2584 | ||
2585 | ret = cleanup_extent_op(trans, fs_info, head); | |
2586 | if (ret < 0) { | |
2587 | unselect_delayed_ref_head(delayed_refs, head); | |
2588 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
2589 | return ret; | |
2590 | } else if (ret) { | |
2591 | return ret; | |
2592 | } | |
2593 | ||
2594 | /* | |
2595 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
2596 | * and then re-check to make sure nobody got added. | |
2597 | */ | |
2598 | spin_unlock(&head->lock); | |
2599 | spin_lock(&delayed_refs->lock); | |
2600 | spin_lock(&head->lock); | |
0e0adbcf | 2601 | if (!RB_EMPTY_ROOT(&head->ref_tree) || head->extent_op) { |
194ab0bc JB |
2602 | spin_unlock(&head->lock); |
2603 | spin_unlock(&delayed_refs->lock); | |
2604 | return 1; | |
2605 | } | |
194ab0bc JB |
2606 | delayed_refs->num_heads--; |
2607 | rb_erase(&head->href_node, &delayed_refs->href_root); | |
d278850e | 2608 | RB_CLEAR_NODE(&head->href_node); |
194ab0bc | 2609 | spin_unlock(&delayed_refs->lock); |
c1103f7a JB |
2610 | spin_unlock(&head->lock); |
2611 | atomic_dec(&delayed_refs->num_entries); | |
2612 | ||
d278850e | 2613 | trace_run_delayed_ref_head(fs_info, head, 0); |
c1103f7a JB |
2614 | |
2615 | if (head->total_ref_mod < 0) { | |
2616 | struct btrfs_block_group_cache *cache; | |
2617 | ||
d278850e | 2618 | cache = btrfs_lookup_block_group(fs_info, head->bytenr); |
c1103f7a JB |
2619 | ASSERT(cache); |
2620 | percpu_counter_add(&cache->space_info->total_bytes_pinned, | |
d278850e | 2621 | -head->num_bytes); |
c1103f7a JB |
2622 | btrfs_put_block_group(cache); |
2623 | ||
2624 | if (head->is_data) { | |
2625 | spin_lock(&delayed_refs->lock); | |
d278850e | 2626 | delayed_refs->pending_csums -= head->num_bytes; |
c1103f7a JB |
2627 | spin_unlock(&delayed_refs->lock); |
2628 | } | |
2629 | } | |
2630 | ||
2631 | if (head->must_insert_reserved) { | |
d278850e JB |
2632 | btrfs_pin_extent(fs_info, head->bytenr, |
2633 | head->num_bytes, 1); | |
c1103f7a | 2634 | if (head->is_data) { |
d278850e JB |
2635 | ret = btrfs_del_csums(trans, fs_info, head->bytenr, |
2636 | head->num_bytes); | |
c1103f7a JB |
2637 | } |
2638 | } | |
2639 | ||
2640 | /* Also free its reserved qgroup space */ | |
2641 | btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root, | |
2642 | head->qgroup_reserved); | |
2643 | btrfs_delayed_ref_unlock(head); | |
d278850e | 2644 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
2645 | return 0; |
2646 | } | |
2647 | ||
79787eaa JM |
2648 | /* |
2649 | * Returns 0 on success or if called with an already aborted transaction. | |
2650 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2651 | */ | |
d7df2c79 | 2652 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
d7df2c79 | 2653 | unsigned long nr) |
56bec294 | 2654 | { |
0a1e458a | 2655 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 CM |
2656 | struct btrfs_delayed_ref_root *delayed_refs; |
2657 | struct btrfs_delayed_ref_node *ref; | |
2658 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2659 | struct btrfs_delayed_extent_op *extent_op; |
0a2b2a84 | 2660 | ktime_t start = ktime_get(); |
56bec294 | 2661 | int ret; |
d7df2c79 | 2662 | unsigned long count = 0; |
0a2b2a84 | 2663 | unsigned long actual_count = 0; |
56bec294 | 2664 | int must_insert_reserved = 0; |
56bec294 CM |
2665 | |
2666 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2667 | while (1) { |
2668 | if (!locked_ref) { | |
d7df2c79 | 2669 | if (count >= nr) |
56bec294 | 2670 | break; |
56bec294 | 2671 | |
d7df2c79 JB |
2672 | spin_lock(&delayed_refs->lock); |
2673 | locked_ref = btrfs_select_ref_head(trans); | |
2674 | if (!locked_ref) { | |
2675 | spin_unlock(&delayed_refs->lock); | |
2676 | break; | |
2677 | } | |
c3e69d58 CM |
2678 | |
2679 | /* grab the lock that says we are going to process | |
2680 | * all the refs for this head */ | |
2681 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
d7df2c79 | 2682 | spin_unlock(&delayed_refs->lock); |
c3e69d58 CM |
2683 | /* |
2684 | * we may have dropped the spin lock to get the head | |
2685 | * mutex lock, and that might have given someone else | |
2686 | * time to free the head. If that's true, it has been | |
2687 | * removed from our list and we can move on. | |
2688 | */ | |
2689 | if (ret == -EAGAIN) { | |
2690 | locked_ref = NULL; | |
2691 | count++; | |
2692 | continue; | |
56bec294 CM |
2693 | } |
2694 | } | |
a28ec197 | 2695 | |
2c3cf7d5 FM |
2696 | /* |
2697 | * We need to try and merge add/drops of the same ref since we | |
2698 | * can run into issues with relocate dropping the implicit ref | |
2699 | * and then it being added back again before the drop can | |
2700 | * finish. If we merged anything we need to re-loop so we can | |
2701 | * get a good ref. | |
2702 | * Or we can get node references of the same type that weren't | |
2703 | * merged when created due to bumps in the tree mod seq, and | |
2704 | * we need to merge them to prevent adding an inline extent | |
2705 | * backref before dropping it (triggering a BUG_ON at | |
2706 | * insert_inline_extent_backref()). | |
2707 | */ | |
d7df2c79 | 2708 | spin_lock(&locked_ref->lock); |
2c3cf7d5 FM |
2709 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, |
2710 | locked_ref); | |
ae1e206b | 2711 | |
d1270cd9 AJ |
2712 | /* |
2713 | * locked_ref is the head node, so we have to go one | |
2714 | * node back for any delayed ref updates | |
2715 | */ | |
2716 | ref = select_delayed_ref(locked_ref); | |
2717 | ||
2718 | if (ref && ref->seq && | |
097b8a7c | 2719 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
d7df2c79 | 2720 | spin_unlock(&locked_ref->lock); |
2eadaa22 | 2721 | unselect_delayed_ref_head(delayed_refs, locked_ref); |
d7df2c79 | 2722 | locked_ref = NULL; |
d1270cd9 | 2723 | cond_resched(); |
27a377db | 2724 | count++; |
d1270cd9 AJ |
2725 | continue; |
2726 | } | |
2727 | ||
c1103f7a JB |
2728 | /* |
2729 | * We're done processing refs in this ref_head, clean everything | |
2730 | * up and move on to the next ref_head. | |
2731 | */ | |
56bec294 | 2732 | if (!ref) { |
194ab0bc JB |
2733 | ret = cleanup_ref_head(trans, fs_info, locked_ref); |
2734 | if (ret > 0 ) { | |
b00e6250 JB |
2735 | /* We dropped our lock, we need to loop. */ |
2736 | ret = 0; | |
d7df2c79 | 2737 | continue; |
194ab0bc JB |
2738 | } else if (ret) { |
2739 | return ret; | |
5d4f98a2 | 2740 | } |
c1103f7a JB |
2741 | locked_ref = NULL; |
2742 | count++; | |
2743 | continue; | |
2744 | } | |
02217ed2 | 2745 | |
c1103f7a JB |
2746 | actual_count++; |
2747 | ref->in_tree = 0; | |
0e0adbcf JB |
2748 | rb_erase(&ref->ref_node, &locked_ref->ref_tree); |
2749 | RB_CLEAR_NODE(&ref->ref_node); | |
c1103f7a JB |
2750 | if (!list_empty(&ref->add_list)) |
2751 | list_del(&ref->add_list); | |
2752 | /* | |
2753 | * When we play the delayed ref, also correct the ref_mod on | |
2754 | * head | |
2755 | */ | |
2756 | switch (ref->action) { | |
2757 | case BTRFS_ADD_DELAYED_REF: | |
2758 | case BTRFS_ADD_DELAYED_EXTENT: | |
d278850e | 2759 | locked_ref->ref_mod -= ref->ref_mod; |
c1103f7a JB |
2760 | break; |
2761 | case BTRFS_DROP_DELAYED_REF: | |
d278850e | 2762 | locked_ref->ref_mod += ref->ref_mod; |
c1103f7a JB |
2763 | break; |
2764 | default: | |
2765 | WARN_ON(1); | |
22cd2e7d | 2766 | } |
1ce7a5ec JB |
2767 | atomic_dec(&delayed_refs->num_entries); |
2768 | ||
b00e6250 JB |
2769 | /* |
2770 | * Record the must-insert_reserved flag before we drop the spin | |
2771 | * lock. | |
2772 | */ | |
2773 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2774 | locked_ref->must_insert_reserved = 0; | |
2775 | ||
2776 | extent_op = locked_ref->extent_op; | |
2777 | locked_ref->extent_op = NULL; | |
d7df2c79 | 2778 | spin_unlock(&locked_ref->lock); |
925baedd | 2779 | |
2ff7e61e | 2780 | ret = run_one_delayed_ref(trans, fs_info, ref, extent_op, |
56bec294 | 2781 | must_insert_reserved); |
eb099670 | 2782 | |
78a6184a | 2783 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2784 | if (ret) { |
2eadaa22 | 2785 | unselect_delayed_ref_head(delayed_refs, locked_ref); |
093486c4 | 2786 | btrfs_put_delayed_ref(ref); |
5d163e0e JM |
2787 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", |
2788 | ret); | |
79787eaa JM |
2789 | return ret; |
2790 | } | |
2791 | ||
093486c4 MX |
2792 | btrfs_put_delayed_ref(ref); |
2793 | count++; | |
c3e69d58 | 2794 | cond_resched(); |
c3e69d58 | 2795 | } |
0a2b2a84 JB |
2796 | |
2797 | /* | |
2798 | * We don't want to include ref heads since we can have empty ref heads | |
2799 | * and those will drastically skew our runtime down since we just do | |
2800 | * accounting, no actual extent tree updates. | |
2801 | */ | |
2802 | if (actual_count > 0) { | |
2803 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2804 | u64 avg; | |
2805 | ||
2806 | /* | |
2807 | * We weigh the current average higher than our current runtime | |
2808 | * to avoid large swings in the average. | |
2809 | */ | |
2810 | spin_lock(&delayed_refs->lock); | |
2811 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2812 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2813 | spin_unlock(&delayed_refs->lock); |
2814 | } | |
d7df2c79 | 2815 | return 0; |
c3e69d58 CM |
2816 | } |
2817 | ||
709c0486 AJ |
2818 | #ifdef SCRAMBLE_DELAYED_REFS |
2819 | /* | |
2820 | * Normally delayed refs get processed in ascending bytenr order. This | |
2821 | * correlates in most cases to the order added. To expose dependencies on this | |
2822 | * order, we start to process the tree in the middle instead of the beginning | |
2823 | */ | |
2824 | static u64 find_middle(struct rb_root *root) | |
2825 | { | |
2826 | struct rb_node *n = root->rb_node; | |
2827 | struct btrfs_delayed_ref_node *entry; | |
2828 | int alt = 1; | |
2829 | u64 middle; | |
2830 | u64 first = 0, last = 0; | |
2831 | ||
2832 | n = rb_first(root); | |
2833 | if (n) { | |
2834 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2835 | first = entry->bytenr; | |
2836 | } | |
2837 | n = rb_last(root); | |
2838 | if (n) { | |
2839 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2840 | last = entry->bytenr; | |
2841 | } | |
2842 | n = root->rb_node; | |
2843 | ||
2844 | while (n) { | |
2845 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2846 | WARN_ON(!entry->in_tree); | |
2847 | ||
2848 | middle = entry->bytenr; | |
2849 | ||
2850 | if (alt) | |
2851 | n = n->rb_left; | |
2852 | else | |
2853 | n = n->rb_right; | |
2854 | ||
2855 | alt = 1 - alt; | |
2856 | } | |
2857 | return middle; | |
2858 | } | |
2859 | #endif | |
2860 | ||
2ff7e61e | 2861 | static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads) |
1be41b78 JB |
2862 | { |
2863 | u64 num_bytes; | |
2864 | ||
2865 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2866 | sizeof(struct btrfs_extent_inline_ref)); | |
0b246afa | 2867 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1be41b78 JB |
2868 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
2869 | ||
2870 | /* | |
2871 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2872 | * closer to what we're really going to want to use. |
1be41b78 | 2873 | */ |
0b246afa | 2874 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info)); |
1be41b78 JB |
2875 | } |
2876 | ||
1262133b JB |
2877 | /* |
2878 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2879 | * would require to store the csums for that many bytes. | |
2880 | */ | |
2ff7e61e | 2881 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2882 | { |
2883 | u64 csum_size; | |
2884 | u64 num_csums_per_leaf; | |
2885 | u64 num_csums; | |
2886 | ||
0b246afa | 2887 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2888 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2889 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2890 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2891 | num_csums += num_csums_per_leaf - 1; |
2892 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2893 | return num_csums; | |
2894 | } | |
2895 | ||
0a2b2a84 | 2896 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2897 | struct btrfs_fs_info *fs_info) |
1be41b78 JB |
2898 | { |
2899 | struct btrfs_block_rsv *global_rsv; | |
2900 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
1262133b | 2901 | u64 csum_bytes = trans->transaction->delayed_refs.pending_csums; |
165c8b02 | 2902 | unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs; |
cb723e49 | 2903 | u64 num_bytes, num_dirty_bgs_bytes; |
1be41b78 JB |
2904 | int ret = 0; |
2905 | ||
0b246afa | 2906 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
2ff7e61e | 2907 | num_heads = heads_to_leaves(fs_info, num_heads); |
1be41b78 | 2908 | if (num_heads > 1) |
0b246afa | 2909 | num_bytes += (num_heads - 1) * fs_info->nodesize; |
1be41b78 | 2910 | num_bytes <<= 1; |
2ff7e61e JM |
2911 | num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) * |
2912 | fs_info->nodesize; | |
0b246afa | 2913 | num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info, |
cb723e49 | 2914 | num_dirty_bgs); |
0b246afa | 2915 | global_rsv = &fs_info->global_block_rsv; |
1be41b78 JB |
2916 | |
2917 | /* | |
2918 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2919 | * wiggle room since running delayed refs can create more delayed refs. | |
2920 | */ | |
cb723e49 JB |
2921 | if (global_rsv->space_info->full) { |
2922 | num_dirty_bgs_bytes <<= 1; | |
1be41b78 | 2923 | num_bytes <<= 1; |
cb723e49 | 2924 | } |
1be41b78 JB |
2925 | |
2926 | spin_lock(&global_rsv->lock); | |
cb723e49 | 2927 | if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes) |
1be41b78 JB |
2928 | ret = 1; |
2929 | spin_unlock(&global_rsv->lock); | |
2930 | return ret; | |
2931 | } | |
2932 | ||
0a2b2a84 | 2933 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2934 | struct btrfs_fs_info *fs_info) |
0a2b2a84 | 2935 | { |
0a2b2a84 JB |
2936 | u64 num_entries = |
2937 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2938 | u64 avg_runtime; | |
a79b7d4b | 2939 | u64 val; |
0a2b2a84 JB |
2940 | |
2941 | smp_mb(); | |
2942 | avg_runtime = fs_info->avg_delayed_ref_runtime; | |
a79b7d4b | 2943 | val = num_entries * avg_runtime; |
dc1a90c6 | 2944 | if (val >= NSEC_PER_SEC) |
0a2b2a84 | 2945 | return 1; |
a79b7d4b CM |
2946 | if (val >= NSEC_PER_SEC / 2) |
2947 | return 2; | |
0a2b2a84 | 2948 | |
2ff7e61e | 2949 | return btrfs_check_space_for_delayed_refs(trans, fs_info); |
0a2b2a84 JB |
2950 | } |
2951 | ||
a79b7d4b CM |
2952 | struct async_delayed_refs { |
2953 | struct btrfs_root *root; | |
31b9655f | 2954 | u64 transid; |
a79b7d4b CM |
2955 | int count; |
2956 | int error; | |
2957 | int sync; | |
2958 | struct completion wait; | |
2959 | struct btrfs_work work; | |
2960 | }; | |
2961 | ||
2ff7e61e JM |
2962 | static inline struct async_delayed_refs * |
2963 | to_async_delayed_refs(struct btrfs_work *work) | |
2964 | { | |
2965 | return container_of(work, struct async_delayed_refs, work); | |
2966 | } | |
2967 | ||
a79b7d4b CM |
2968 | static void delayed_ref_async_start(struct btrfs_work *work) |
2969 | { | |
2ff7e61e | 2970 | struct async_delayed_refs *async = to_async_delayed_refs(work); |
a79b7d4b | 2971 | struct btrfs_trans_handle *trans; |
2ff7e61e | 2972 | struct btrfs_fs_info *fs_info = async->root->fs_info; |
a79b7d4b CM |
2973 | int ret; |
2974 | ||
0f873eca | 2975 | /* if the commit is already started, we don't need to wait here */ |
2ff7e61e | 2976 | if (btrfs_transaction_blocked(fs_info)) |
31b9655f | 2977 | goto done; |
31b9655f | 2978 | |
0f873eca CM |
2979 | trans = btrfs_join_transaction(async->root); |
2980 | if (IS_ERR(trans)) { | |
2981 | async->error = PTR_ERR(trans); | |
a79b7d4b CM |
2982 | goto done; |
2983 | } | |
2984 | ||
2985 | /* | |
01327610 | 2986 | * trans->sync means that when we call end_transaction, we won't |
a79b7d4b CM |
2987 | * wait on delayed refs |
2988 | */ | |
2989 | trans->sync = true; | |
0f873eca CM |
2990 | |
2991 | /* Don't bother flushing if we got into a different transaction */ | |
2992 | if (trans->transid > async->transid) | |
2993 | goto end; | |
2994 | ||
c79a70b1 | 2995 | ret = btrfs_run_delayed_refs(trans, async->count); |
a79b7d4b CM |
2996 | if (ret) |
2997 | async->error = ret; | |
0f873eca | 2998 | end: |
3a45bb20 | 2999 | ret = btrfs_end_transaction(trans); |
a79b7d4b CM |
3000 | if (ret && !async->error) |
3001 | async->error = ret; | |
3002 | done: | |
3003 | if (async->sync) | |
3004 | complete(&async->wait); | |
3005 | else | |
3006 | kfree(async); | |
3007 | } | |
3008 | ||
2ff7e61e | 3009 | int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info, |
31b9655f | 3010 | unsigned long count, u64 transid, int wait) |
a79b7d4b CM |
3011 | { |
3012 | struct async_delayed_refs *async; | |
3013 | int ret; | |
3014 | ||
3015 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
3016 | if (!async) | |
3017 | return -ENOMEM; | |
3018 | ||
0b246afa | 3019 | async->root = fs_info->tree_root; |
a79b7d4b CM |
3020 | async->count = count; |
3021 | async->error = 0; | |
31b9655f | 3022 | async->transid = transid; |
a79b7d4b CM |
3023 | if (wait) |
3024 | async->sync = 1; | |
3025 | else | |
3026 | async->sync = 0; | |
3027 | init_completion(&async->wait); | |
3028 | ||
9e0af237 LB |
3029 | btrfs_init_work(&async->work, btrfs_extent_refs_helper, |
3030 | delayed_ref_async_start, NULL, NULL); | |
a79b7d4b | 3031 | |
0b246afa | 3032 | btrfs_queue_work(fs_info->extent_workers, &async->work); |
a79b7d4b CM |
3033 | |
3034 | if (wait) { | |
3035 | wait_for_completion(&async->wait); | |
3036 | ret = async->error; | |
3037 | kfree(async); | |
3038 | return ret; | |
3039 | } | |
3040 | return 0; | |
3041 | } | |
3042 | ||
c3e69d58 CM |
3043 | /* |
3044 | * this starts processing the delayed reference count updates and | |
3045 | * extent insertions we have queued up so far. count can be | |
3046 | * 0, which means to process everything in the tree at the start | |
3047 | * of the run (but not newly added entries), or it can be some target | |
3048 | * number you'd like to process. | |
79787eaa JM |
3049 | * |
3050 | * Returns 0 on success or if called with an aborted transaction | |
3051 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
3052 | */ |
3053 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
c79a70b1 | 3054 | unsigned long count) |
c3e69d58 | 3055 | { |
c79a70b1 | 3056 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 CM |
3057 | struct rb_node *node; |
3058 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 3059 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
3060 | int ret; |
3061 | int run_all = count == (unsigned long)-1; | |
d9a0540a | 3062 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
c3e69d58 | 3063 | |
79787eaa JM |
3064 | /* We'll clean this up in btrfs_cleanup_transaction */ |
3065 | if (trans->aborted) | |
3066 | return 0; | |
3067 | ||
0b246afa | 3068 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
3069 | return 0; |
3070 | ||
c3e69d58 | 3071 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 3072 | if (count == 0) |
d7df2c79 | 3073 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 3074 | |
c3e69d58 | 3075 | again: |
709c0486 AJ |
3076 | #ifdef SCRAMBLE_DELAYED_REFS |
3077 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
3078 | #endif | |
d9a0540a | 3079 | trans->can_flush_pending_bgs = false; |
0a1e458a | 3080 | ret = __btrfs_run_delayed_refs(trans, count); |
d7df2c79 | 3081 | if (ret < 0) { |
66642832 | 3082 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 3083 | return ret; |
eb099670 | 3084 | } |
c3e69d58 | 3085 | |
56bec294 | 3086 | if (run_all) { |
d7df2c79 | 3087 | if (!list_empty(&trans->new_bgs)) |
6c686b35 | 3088 | btrfs_create_pending_block_groups(trans); |
ea658bad | 3089 | |
d7df2c79 | 3090 | spin_lock(&delayed_refs->lock); |
c46effa6 | 3091 | node = rb_first(&delayed_refs->href_root); |
d7df2c79 JB |
3092 | if (!node) { |
3093 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 3094 | goto out; |
d7df2c79 | 3095 | } |
d278850e JB |
3096 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
3097 | href_node); | |
3098 | refcount_inc(&head->refs); | |
3099 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 3100 | |
d278850e JB |
3101 | /* Mutex was contended, block until it's released and retry. */ |
3102 | mutex_lock(&head->mutex); | |
3103 | mutex_unlock(&head->mutex); | |
56bec294 | 3104 | |
d278850e | 3105 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 3106 | cond_resched(); |
56bec294 | 3107 | goto again; |
5f39d397 | 3108 | } |
54aa1f4d | 3109 | out: |
d9a0540a | 3110 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
a28ec197 CM |
3111 | return 0; |
3112 | } | |
3113 | ||
5d4f98a2 | 3114 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2ff7e61e | 3115 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 3116 | u64 bytenr, u64 num_bytes, u64 flags, |
b1c79e09 | 3117 | int level, int is_data) |
5d4f98a2 YZ |
3118 | { |
3119 | struct btrfs_delayed_extent_op *extent_op; | |
3120 | int ret; | |
3121 | ||
78a6184a | 3122 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
3123 | if (!extent_op) |
3124 | return -ENOMEM; | |
3125 | ||
3126 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
3127 | extent_op->update_flags = true; |
3128 | extent_op->update_key = false; | |
3129 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 3130 | extent_op->level = level; |
5d4f98a2 | 3131 | |
0b246afa | 3132 | ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr, |
66d7e7f0 | 3133 | num_bytes, extent_op); |
5d4f98a2 | 3134 | if (ret) |
78a6184a | 3135 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
3136 | return ret; |
3137 | } | |
3138 | ||
e4c3b2dc | 3139 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3140 | struct btrfs_path *path, |
3141 | u64 objectid, u64 offset, u64 bytenr) | |
3142 | { | |
3143 | struct btrfs_delayed_ref_head *head; | |
3144 | struct btrfs_delayed_ref_node *ref; | |
3145 | struct btrfs_delayed_data_ref *data_ref; | |
3146 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 3147 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 3148 | struct rb_node *node; |
5d4f98a2 YZ |
3149 | int ret = 0; |
3150 | ||
e4c3b2dc LB |
3151 | cur_trans = root->fs_info->running_transaction; |
3152 | if (!cur_trans) | |
3153 | return 0; | |
3154 | ||
3155 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 3156 | spin_lock(&delayed_refs->lock); |
f72ad18e | 3157 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
3158 | if (!head) { |
3159 | spin_unlock(&delayed_refs->lock); | |
3160 | return 0; | |
3161 | } | |
5d4f98a2 YZ |
3162 | |
3163 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 3164 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
3165 | spin_unlock(&delayed_refs->lock); |
3166 | ||
b3b4aa74 | 3167 | btrfs_release_path(path); |
5d4f98a2 | 3168 | |
8cc33e5c DS |
3169 | /* |
3170 | * Mutex was contended, block until it's released and let | |
3171 | * caller try again | |
3172 | */ | |
5d4f98a2 YZ |
3173 | mutex_lock(&head->mutex); |
3174 | mutex_unlock(&head->mutex); | |
d278850e | 3175 | btrfs_put_delayed_ref_head(head); |
5d4f98a2 YZ |
3176 | return -EAGAIN; |
3177 | } | |
d7df2c79 | 3178 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 3179 | |
d7df2c79 | 3180 | spin_lock(&head->lock); |
0e0adbcf JB |
3181 | /* |
3182 | * XXX: We should replace this with a proper search function in the | |
3183 | * future. | |
3184 | */ | |
3185 | for (node = rb_first(&head->ref_tree); node; node = rb_next(node)) { | |
3186 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); | |
d7df2c79 JB |
3187 | /* If it's a shared ref we know a cross reference exists */ |
3188 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3189 | ret = 1; | |
3190 | break; | |
3191 | } | |
5d4f98a2 | 3192 | |
d7df2c79 | 3193 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 3194 | |
d7df2c79 JB |
3195 | /* |
3196 | * If our ref doesn't match the one we're currently looking at | |
3197 | * then we have a cross reference. | |
3198 | */ | |
3199 | if (data_ref->root != root->root_key.objectid || | |
3200 | data_ref->objectid != objectid || | |
3201 | data_ref->offset != offset) { | |
3202 | ret = 1; | |
3203 | break; | |
3204 | } | |
5d4f98a2 | 3205 | } |
d7df2c79 | 3206 | spin_unlock(&head->lock); |
5d4f98a2 | 3207 | mutex_unlock(&head->mutex); |
5d4f98a2 YZ |
3208 | return ret; |
3209 | } | |
3210 | ||
e4c3b2dc | 3211 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3212 | struct btrfs_path *path, |
3213 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d | 3214 | { |
0b246afa JM |
3215 | struct btrfs_fs_info *fs_info = root->fs_info; |
3216 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 3217 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3218 | struct btrfs_extent_data_ref *ref; |
3219 | struct btrfs_extent_inline_ref *iref; | |
3220 | struct btrfs_extent_item *ei; | |
f321e491 | 3221 | struct btrfs_key key; |
5d4f98a2 | 3222 | u32 item_size; |
3de28d57 | 3223 | int type; |
be20aa9d | 3224 | int ret; |
925baedd | 3225 | |
be20aa9d | 3226 | key.objectid = bytenr; |
31840ae1 | 3227 | key.offset = (u64)-1; |
f321e491 | 3228 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 3229 | |
be20aa9d CM |
3230 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
3231 | if (ret < 0) | |
3232 | goto out; | |
79787eaa | 3233 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
3234 | |
3235 | ret = -ENOENT; | |
3236 | if (path->slots[0] == 0) | |
31840ae1 | 3237 | goto out; |
be20aa9d | 3238 | |
31840ae1 | 3239 | path->slots[0]--; |
f321e491 | 3240 | leaf = path->nodes[0]; |
5d4f98a2 | 3241 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3242 | |
5d4f98a2 | 3243 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3244 | goto out; |
f321e491 | 3245 | |
5d4f98a2 YZ |
3246 | ret = 1; |
3247 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3248 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3249 | if (item_size < sizeof(*ei)) { | |
3250 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
3251 | goto out; | |
3252 | } | |
3253 | #endif | |
3254 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 3255 | |
5d4f98a2 YZ |
3256 | if (item_size != sizeof(*ei) + |
3257 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3258 | goto out; | |
be20aa9d | 3259 | |
5d4f98a2 YZ |
3260 | if (btrfs_extent_generation(leaf, ei) <= |
3261 | btrfs_root_last_snapshot(&root->root_item)) | |
3262 | goto out; | |
3263 | ||
3264 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 LB |
3265 | |
3266 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
3267 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
3268 | goto out; |
3269 | ||
3270 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3271 | if (btrfs_extent_refs(leaf, ei) != | |
3272 | btrfs_extent_data_ref_count(leaf, ref) || | |
3273 | btrfs_extent_data_ref_root(leaf, ref) != | |
3274 | root->root_key.objectid || | |
3275 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3276 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3277 | goto out; | |
3278 | ||
3279 | ret = 0; | |
3280 | out: | |
3281 | return ret; | |
3282 | } | |
3283 | ||
e4c3b2dc LB |
3284 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
3285 | u64 bytenr) | |
5d4f98a2 YZ |
3286 | { |
3287 | struct btrfs_path *path; | |
3288 | int ret; | |
3289 | int ret2; | |
3290 | ||
3291 | path = btrfs_alloc_path(); | |
3292 | if (!path) | |
3293 | return -ENOENT; | |
3294 | ||
3295 | do { | |
e4c3b2dc | 3296 | ret = check_committed_ref(root, path, objectid, |
5d4f98a2 YZ |
3297 | offset, bytenr); |
3298 | if (ret && ret != -ENOENT) | |
f321e491 | 3299 | goto out; |
80ff3856 | 3300 | |
e4c3b2dc | 3301 | ret2 = check_delayed_ref(root, path, objectid, |
5d4f98a2 YZ |
3302 | offset, bytenr); |
3303 | } while (ret2 == -EAGAIN); | |
3304 | ||
3305 | if (ret2 && ret2 != -ENOENT) { | |
3306 | ret = ret2; | |
3307 | goto out; | |
f321e491 | 3308 | } |
5d4f98a2 YZ |
3309 | |
3310 | if (ret != -ENOENT || ret2 != -ENOENT) | |
3311 | ret = 0; | |
be20aa9d | 3312 | out: |
80ff3856 | 3313 | btrfs_free_path(path); |
f0486c68 YZ |
3314 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3315 | WARN_ON(ret > 0); | |
f321e491 | 3316 | return ret; |
be20aa9d | 3317 | } |
c5739bba | 3318 | |
5d4f98a2 | 3319 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3320 | struct btrfs_root *root, |
5d4f98a2 | 3321 | struct extent_buffer *buf, |
e339a6b0 | 3322 | int full_backref, int inc) |
31840ae1 | 3323 | { |
0b246afa | 3324 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 3325 | u64 bytenr; |
5d4f98a2 YZ |
3326 | u64 num_bytes; |
3327 | u64 parent; | |
31840ae1 | 3328 | u64 ref_root; |
31840ae1 | 3329 | u32 nritems; |
31840ae1 ZY |
3330 | struct btrfs_key key; |
3331 | struct btrfs_file_extent_item *fi; | |
3332 | int i; | |
3333 | int level; | |
3334 | int ret = 0; | |
2ff7e61e | 3335 | int (*process_func)(struct btrfs_trans_handle *, |
84f7d8e6 | 3336 | struct btrfs_root *, |
b06c4bf5 | 3337 | u64, u64, u64, u64, u64, u64); |
31840ae1 | 3338 | |
fccb84c9 | 3339 | |
0b246afa | 3340 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3341 | return 0; |
fccb84c9 | 3342 | |
31840ae1 | 3343 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3344 | nritems = btrfs_header_nritems(buf); |
3345 | level = btrfs_header_level(buf); | |
3346 | ||
27cdeb70 | 3347 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3348 | return 0; |
31840ae1 | 3349 | |
5d4f98a2 YZ |
3350 | if (inc) |
3351 | process_func = btrfs_inc_extent_ref; | |
3352 | else | |
3353 | process_func = btrfs_free_extent; | |
31840ae1 | 3354 | |
5d4f98a2 YZ |
3355 | if (full_backref) |
3356 | parent = buf->start; | |
3357 | else | |
3358 | parent = 0; | |
3359 | ||
3360 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3361 | if (level == 0) { |
5d4f98a2 | 3362 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3363 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3364 | continue; |
5d4f98a2 | 3365 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3366 | struct btrfs_file_extent_item); |
3367 | if (btrfs_file_extent_type(buf, fi) == | |
3368 | BTRFS_FILE_EXTENT_INLINE) | |
3369 | continue; | |
3370 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3371 | if (bytenr == 0) | |
3372 | continue; | |
5d4f98a2 YZ |
3373 | |
3374 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3375 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
84f7d8e6 | 3376 | ret = process_func(trans, root, bytenr, num_bytes, |
5d4f98a2 | 3377 | parent, ref_root, key.objectid, |
b06c4bf5 | 3378 | key.offset); |
31840ae1 ZY |
3379 | if (ret) |
3380 | goto fail; | |
3381 | } else { | |
5d4f98a2 | 3382 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 3383 | num_bytes = fs_info->nodesize; |
84f7d8e6 | 3384 | ret = process_func(trans, root, bytenr, num_bytes, |
b06c4bf5 | 3385 | parent, ref_root, level - 1, 0); |
31840ae1 ZY |
3386 | if (ret) |
3387 | goto fail; | |
3388 | } | |
3389 | } | |
3390 | return 0; | |
3391 | fail: | |
5d4f98a2 YZ |
3392 | return ret; |
3393 | } | |
3394 | ||
3395 | int btrfs_inc_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, 1); |
5d4f98a2 YZ |
3399 | } |
3400 | ||
3401 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3402 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3403 | { |
e339a6b0 | 3404 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3405 | } |
3406 | ||
9078a3e1 | 3407 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 3408 | struct btrfs_fs_info *fs_info, |
9078a3e1 CM |
3409 | struct btrfs_path *path, |
3410 | struct btrfs_block_group_cache *cache) | |
3411 | { | |
3412 | int ret; | |
0b246afa | 3413 | struct btrfs_root *extent_root = fs_info->extent_root; |
5f39d397 CM |
3414 | unsigned long bi; |
3415 | struct extent_buffer *leaf; | |
9078a3e1 | 3416 | |
9078a3e1 | 3417 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3418 | if (ret) { |
3419 | if (ret > 0) | |
3420 | ret = -ENOENT; | |
54aa1f4d | 3421 | goto fail; |
df95e7f0 | 3422 | } |
5f39d397 CM |
3423 | |
3424 | leaf = path->nodes[0]; | |
3425 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3426 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3427 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3428 | fail: |
24b89d08 | 3429 | btrfs_release_path(path); |
df95e7f0 | 3430 | return ret; |
9078a3e1 CM |
3431 | |
3432 | } | |
3433 | ||
4a8c9a62 | 3434 | static struct btrfs_block_group_cache * |
2ff7e61e | 3435 | next_block_group(struct btrfs_fs_info *fs_info, |
4a8c9a62 YZ |
3436 | struct btrfs_block_group_cache *cache) |
3437 | { | |
3438 | struct rb_node *node; | |
292cbd51 | 3439 | |
0b246afa | 3440 | spin_lock(&fs_info->block_group_cache_lock); |
292cbd51 FM |
3441 | |
3442 | /* If our block group was removed, we need a full search. */ | |
3443 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3444 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3445 | ||
0b246afa | 3446 | spin_unlock(&fs_info->block_group_cache_lock); |
292cbd51 | 3447 | btrfs_put_block_group(cache); |
0b246afa | 3448 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; |
292cbd51 | 3449 | } |
4a8c9a62 YZ |
3450 | node = rb_next(&cache->cache_node); |
3451 | btrfs_put_block_group(cache); | |
3452 | if (node) { | |
3453 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3454 | cache_node); | |
11dfe35a | 3455 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3456 | } else |
3457 | cache = NULL; | |
0b246afa | 3458 | spin_unlock(&fs_info->block_group_cache_lock); |
4a8c9a62 YZ |
3459 | return cache; |
3460 | } | |
3461 | ||
0af3d00b JB |
3462 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3463 | struct btrfs_trans_handle *trans, | |
3464 | struct btrfs_path *path) | |
3465 | { | |
0b246afa JM |
3466 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
3467 | struct btrfs_root *root = fs_info->tree_root; | |
0af3d00b | 3468 | struct inode *inode = NULL; |
364ecf36 | 3469 | struct extent_changeset *data_reserved = NULL; |
0af3d00b | 3470 | u64 alloc_hint = 0; |
2b20982e | 3471 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3472 | u64 num_pages = 0; |
0af3d00b JB |
3473 | int retries = 0; |
3474 | int ret = 0; | |
3475 | ||
3476 | /* | |
3477 | * If this block group is smaller than 100 megs don't bother caching the | |
3478 | * block group. | |
3479 | */ | |
ee22184b | 3480 | if (block_group->key.offset < (100 * SZ_1M)) { |
0af3d00b JB |
3481 | spin_lock(&block_group->lock); |
3482 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3483 | spin_unlock(&block_group->lock); | |
3484 | return 0; | |
3485 | } | |
3486 | ||
0c0ef4bc JB |
3487 | if (trans->aborted) |
3488 | return 0; | |
0af3d00b | 3489 | again: |
77ab86bf | 3490 | inode = lookup_free_space_inode(fs_info, block_group, path); |
0af3d00b JB |
3491 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { |
3492 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3493 | btrfs_release_path(path); |
0af3d00b JB |
3494 | goto out; |
3495 | } | |
3496 | ||
3497 | if (IS_ERR(inode)) { | |
3498 | BUG_ON(retries); | |
3499 | retries++; | |
3500 | ||
3501 | if (block_group->ro) | |
3502 | goto out_free; | |
3503 | ||
77ab86bf JM |
3504 | ret = create_free_space_inode(fs_info, trans, block_group, |
3505 | path); | |
0af3d00b JB |
3506 | if (ret) |
3507 | goto out_free; | |
3508 | goto again; | |
3509 | } | |
3510 | ||
3511 | /* | |
3512 | * We want to set the generation to 0, that way if anything goes wrong | |
3513 | * from here on out we know not to trust this cache when we load up next | |
3514 | * time. | |
3515 | */ | |
3516 | BTRFS_I(inode)->generation = 0; | |
3517 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3518 | if (ret) { |
3519 | /* | |
3520 | * So theoretically we could recover from this, simply set the | |
3521 | * super cache generation to 0 so we know to invalidate the | |
3522 | * cache, but then we'd have to keep track of the block groups | |
3523 | * that fail this way so we know we _have_ to reset this cache | |
3524 | * before the next commit or risk reading stale cache. So to | |
3525 | * limit our exposure to horrible edge cases lets just abort the | |
3526 | * transaction, this only happens in really bad situations | |
3527 | * anyway. | |
3528 | */ | |
66642832 | 3529 | btrfs_abort_transaction(trans, ret); |
0c0ef4bc JB |
3530 | goto out_put; |
3531 | } | |
0af3d00b JB |
3532 | WARN_ON(ret); |
3533 | ||
8e138e0d JB |
3534 | /* We've already setup this transaction, go ahead and exit */ |
3535 | if (block_group->cache_generation == trans->transid && | |
3536 | i_size_read(inode)) { | |
3537 | dcs = BTRFS_DC_SETUP; | |
3538 | goto out_put; | |
3539 | } | |
3540 | ||
0af3d00b | 3541 | if (i_size_read(inode) > 0) { |
2ff7e61e | 3542 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
0b246afa | 3543 | &fs_info->global_block_rsv); |
7b61cd92 MX |
3544 | if (ret) |
3545 | goto out_put; | |
3546 | ||
77ab86bf | 3547 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); |
0af3d00b JB |
3548 | if (ret) |
3549 | goto out_put; | |
3550 | } | |
3551 | ||
3552 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3553 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
0b246afa | 3554 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cf7c1ef6 LB |
3555 | /* |
3556 | * don't bother trying to write stuff out _if_ | |
3557 | * a) we're not cached, | |
1a79c1f2 LB |
3558 | * b) we're with nospace_cache mount option, |
3559 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
cf7c1ef6 | 3560 | */ |
2b20982e | 3561 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3562 | spin_unlock(&block_group->lock); |
3563 | goto out_put; | |
3564 | } | |
3565 | spin_unlock(&block_group->lock); | |
3566 | ||
2968b1f4 JB |
3567 | /* |
3568 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3569 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3570 | */ | |
3571 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3572 | ret = -ENOSPC; | |
3573 | goto out_put; | |
3574 | } | |
3575 | ||
6fc823b1 JB |
3576 | /* |
3577 | * Try to preallocate enough space based on how big the block group is. | |
3578 | * Keep in mind this has to include any pinned space which could end up | |
3579 | * taking up quite a bit since it's not folded into the other space | |
3580 | * cache. | |
3581 | */ | |
ee22184b | 3582 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
0af3d00b JB |
3583 | if (!num_pages) |
3584 | num_pages = 1; | |
3585 | ||
0af3d00b | 3586 | num_pages *= 16; |
09cbfeaf | 3587 | num_pages *= PAGE_SIZE; |
0af3d00b | 3588 | |
364ecf36 | 3589 | ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages); |
0af3d00b JB |
3590 | if (ret) |
3591 | goto out_put; | |
3592 | ||
3593 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3594 | num_pages, num_pages, | |
3595 | &alloc_hint); | |
2968b1f4 JB |
3596 | /* |
3597 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3598 | * of metadata or split extents when writing the cache out, which means | |
3599 | * we can enospc if we are heavily fragmented in addition to just normal | |
3600 | * out of space conditions. So if we hit this just skip setting up any | |
3601 | * other block groups for this transaction, maybe we'll unpin enough | |
3602 | * space the next time around. | |
3603 | */ | |
2b20982e JB |
3604 | if (!ret) |
3605 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3606 | else if (ret == -ENOSPC) |
3607 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
c09544e0 | 3608 | |
0af3d00b JB |
3609 | out_put: |
3610 | iput(inode); | |
3611 | out_free: | |
b3b4aa74 | 3612 | btrfs_release_path(path); |
0af3d00b JB |
3613 | out: |
3614 | spin_lock(&block_group->lock); | |
e65cbb94 | 3615 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3616 | block_group->cache_generation = trans->transid; |
2b20982e | 3617 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3618 | spin_unlock(&block_group->lock); |
3619 | ||
364ecf36 | 3620 | extent_changeset_free(data_reserved); |
0af3d00b JB |
3621 | return ret; |
3622 | } | |
3623 | ||
dcdf7f6d | 3624 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans, |
2ff7e61e | 3625 | struct btrfs_fs_info *fs_info) |
dcdf7f6d JB |
3626 | { |
3627 | struct btrfs_block_group_cache *cache, *tmp; | |
3628 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3629 | struct btrfs_path *path; | |
3630 | ||
3631 | if (list_empty(&cur_trans->dirty_bgs) || | |
0b246afa | 3632 | !btrfs_test_opt(fs_info, SPACE_CACHE)) |
dcdf7f6d JB |
3633 | return 0; |
3634 | ||
3635 | path = btrfs_alloc_path(); | |
3636 | if (!path) | |
3637 | return -ENOMEM; | |
3638 | ||
3639 | /* Could add new block groups, use _safe just in case */ | |
3640 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3641 | dirty_list) { | |
3642 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3643 | cache_save_setup(cache, trans, path); | |
3644 | } | |
3645 | ||
3646 | btrfs_free_path(path); | |
3647 | return 0; | |
3648 | } | |
3649 | ||
1bbc621e CM |
3650 | /* |
3651 | * transaction commit does final block group cache writeback during a | |
3652 | * critical section where nothing is allowed to change the FS. This is | |
3653 | * required in order for the cache to actually match the block group, | |
3654 | * but can introduce a lot of latency into the commit. | |
3655 | * | |
3656 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3657 | * cache IO. There's a chance we'll have to redo some of it if the | |
3658 | * block group changes again during the commit, but it greatly reduces | |
3659 | * the commit latency by getting rid of the easy block groups while | |
3660 | * we're still allowing others to join the commit. | |
3661 | */ | |
21217054 | 3662 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) |
9078a3e1 | 3663 | { |
21217054 | 3664 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4a8c9a62 | 3665 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3666 | struct btrfs_transaction *cur_trans = trans->transaction; |
3667 | int ret = 0; | |
c9dc4c65 | 3668 | int should_put; |
1bbc621e CM |
3669 | struct btrfs_path *path = NULL; |
3670 | LIST_HEAD(dirty); | |
3671 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3672 | int num_started = 0; |
1bbc621e CM |
3673 | int loops = 0; |
3674 | ||
3675 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3676 | if (list_empty(&cur_trans->dirty_bgs)) { |
3677 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3678 | return 0; | |
1bbc621e | 3679 | } |
b58d1a9e | 3680 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3681 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3682 | |
1bbc621e | 3683 | again: |
1bbc621e CM |
3684 | /* |
3685 | * make sure all the block groups on our dirty list actually | |
3686 | * exist | |
3687 | */ | |
6c686b35 | 3688 | btrfs_create_pending_block_groups(trans); |
1bbc621e CM |
3689 | |
3690 | if (!path) { | |
3691 | path = btrfs_alloc_path(); | |
3692 | if (!path) | |
3693 | return -ENOMEM; | |
3694 | } | |
3695 | ||
b58d1a9e FM |
3696 | /* |
3697 | * cache_write_mutex is here only to save us from balance or automatic | |
3698 | * removal of empty block groups deleting this block group while we are | |
3699 | * writing out the cache | |
3700 | */ | |
3701 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e CM |
3702 | while (!list_empty(&dirty)) { |
3703 | cache = list_first_entry(&dirty, | |
3704 | struct btrfs_block_group_cache, | |
3705 | dirty_list); | |
1bbc621e CM |
3706 | /* |
3707 | * this can happen if something re-dirties a block | |
3708 | * group that is already under IO. Just wait for it to | |
3709 | * finish and then do it all again | |
3710 | */ | |
3711 | if (!list_empty(&cache->io_list)) { | |
3712 | list_del_init(&cache->io_list); | |
afdb5718 | 3713 | btrfs_wait_cache_io(trans, cache, path); |
1bbc621e CM |
3714 | btrfs_put_block_group(cache); |
3715 | } | |
3716 | ||
3717 | ||
3718 | /* | |
3719 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3720 | * if it should update the cache_state. Don't delete | |
3721 | * until after we wait. | |
3722 | * | |
3723 | * Since we're not running in the commit critical section | |
3724 | * we need the dirty_bgs_lock to protect from update_block_group | |
3725 | */ | |
3726 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3727 | list_del_init(&cache->dirty_list); | |
3728 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3729 | ||
3730 | should_put = 1; | |
3731 | ||
3732 | cache_save_setup(cache, trans, path); | |
3733 | ||
3734 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3735 | cache->io_ctl.inode = NULL; | |
0b246afa | 3736 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3737 | cache, path); |
1bbc621e CM |
3738 | if (ret == 0 && cache->io_ctl.inode) { |
3739 | num_started++; | |
3740 | should_put = 0; | |
3741 | ||
3742 | /* | |
45ae2c18 NB |
3743 | * The cache_write_mutex is protecting the |
3744 | * io_list, also refer to the definition of | |
3745 | * btrfs_transaction::io_bgs for more details | |
1bbc621e CM |
3746 | */ |
3747 | list_add_tail(&cache->io_list, io); | |
3748 | } else { | |
3749 | /* | |
3750 | * if we failed to write the cache, the | |
3751 | * generation will be bad and life goes on | |
3752 | */ | |
3753 | ret = 0; | |
3754 | } | |
3755 | } | |
ff1f8250 | 3756 | if (!ret) { |
2ff7e61e JM |
3757 | ret = write_one_cache_group(trans, fs_info, |
3758 | path, cache); | |
ff1f8250 FM |
3759 | /* |
3760 | * Our block group might still be attached to the list | |
3761 | * of new block groups in the transaction handle of some | |
3762 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3763 | * means its block group item isn't yet in the extent | |
3764 | * tree. If this happens ignore the error, as we will | |
3765 | * try again later in the critical section of the | |
3766 | * transaction commit. | |
3767 | */ | |
3768 | if (ret == -ENOENT) { | |
3769 | ret = 0; | |
3770 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3771 | if (list_empty(&cache->dirty_list)) { | |
3772 | list_add_tail(&cache->dirty_list, | |
3773 | &cur_trans->dirty_bgs); | |
3774 | btrfs_get_block_group(cache); | |
3775 | } | |
3776 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3777 | } else if (ret) { | |
66642832 | 3778 | btrfs_abort_transaction(trans, ret); |
ff1f8250 FM |
3779 | } |
3780 | } | |
1bbc621e CM |
3781 | |
3782 | /* if its not on the io list, we need to put the block group */ | |
3783 | if (should_put) | |
3784 | btrfs_put_block_group(cache); | |
3785 | ||
3786 | if (ret) | |
3787 | break; | |
b58d1a9e FM |
3788 | |
3789 | /* | |
3790 | * Avoid blocking other tasks for too long. It might even save | |
3791 | * us from writing caches for block groups that are going to be | |
3792 | * removed. | |
3793 | */ | |
3794 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3795 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3796 | } |
b58d1a9e | 3797 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3798 | |
3799 | /* | |
3800 | * go through delayed refs for all the stuff we've just kicked off | |
3801 | * and then loop back (just once) | |
3802 | */ | |
c79a70b1 | 3803 | ret = btrfs_run_delayed_refs(trans, 0); |
1bbc621e CM |
3804 | if (!ret && loops == 0) { |
3805 | loops++; | |
3806 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3807 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3808 | /* |
3809 | * dirty_bgs_lock protects us from concurrent block group | |
3810 | * deletes too (not just cache_write_mutex). | |
3811 | */ | |
3812 | if (!list_empty(&dirty)) { | |
3813 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3814 | goto again; | |
3815 | } | |
1bbc621e | 3816 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c79a1751 | 3817 | } else if (ret < 0) { |
2ff7e61e | 3818 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
1bbc621e CM |
3819 | } |
3820 | ||
3821 | btrfs_free_path(path); | |
3822 | return ret; | |
3823 | } | |
3824 | ||
3825 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3826 | struct btrfs_fs_info *fs_info) |
1bbc621e CM |
3827 | { |
3828 | struct btrfs_block_group_cache *cache; | |
3829 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3830 | int ret = 0; | |
3831 | int should_put; | |
3832 | struct btrfs_path *path; | |
3833 | struct list_head *io = &cur_trans->io_bgs; | |
3834 | int num_started = 0; | |
9078a3e1 CM |
3835 | |
3836 | path = btrfs_alloc_path(); | |
3837 | if (!path) | |
3838 | return -ENOMEM; | |
3839 | ||
ce93ec54 | 3840 | /* |
e44081ef FM |
3841 | * Even though we are in the critical section of the transaction commit, |
3842 | * we can still have concurrent tasks adding elements to this | |
3843 | * transaction's list of dirty block groups. These tasks correspond to | |
3844 | * endio free space workers started when writeback finishes for a | |
3845 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3846 | * allocate new block groups as a result of COWing nodes of the root | |
3847 | * tree when updating the free space inode. The writeback for the space | |
3848 | * caches is triggered by an earlier call to | |
3849 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3850 | * loop. | |
3851 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3852 | * delayed refs to make sure we have the best chance at doing this all |
3853 | * in one shot. | |
3854 | */ | |
e44081ef | 3855 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3856 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3857 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3858 | struct btrfs_block_group_cache, | |
3859 | dirty_list); | |
c9dc4c65 CM |
3860 | |
3861 | /* | |
3862 | * this can happen if cache_save_setup re-dirties a block | |
3863 | * group that is already under IO. Just wait for it to | |
3864 | * finish and then do it all again | |
3865 | */ | |
3866 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3867 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3868 | list_del_init(&cache->io_list); |
afdb5718 | 3869 | btrfs_wait_cache_io(trans, cache, path); |
c9dc4c65 | 3870 | btrfs_put_block_group(cache); |
e44081ef | 3871 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3872 | } |
3873 | ||
1bbc621e CM |
3874 | /* |
3875 | * don't remove from the dirty list until after we've waited | |
3876 | * on any pending IO | |
3877 | */ | |
ce93ec54 | 3878 | list_del_init(&cache->dirty_list); |
e44081ef | 3879 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3880 | should_put = 1; |
3881 | ||
1bbc621e | 3882 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3883 | |
ce93ec54 | 3884 | if (!ret) |
c79a70b1 | 3885 | ret = btrfs_run_delayed_refs(trans, |
2ff7e61e | 3886 | (unsigned long) -1); |
c9dc4c65 CM |
3887 | |
3888 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3889 | cache->io_ctl.inode = NULL; | |
0b246afa | 3890 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3891 | cache, path); |
c9dc4c65 CM |
3892 | if (ret == 0 && cache->io_ctl.inode) { |
3893 | num_started++; | |
3894 | should_put = 0; | |
1bbc621e | 3895 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3896 | } else { |
3897 | /* | |
3898 | * if we failed to write the cache, the | |
3899 | * generation will be bad and life goes on | |
3900 | */ | |
3901 | ret = 0; | |
3902 | } | |
3903 | } | |
ff1f8250 | 3904 | if (!ret) { |
2ff7e61e JM |
3905 | ret = write_one_cache_group(trans, fs_info, |
3906 | path, cache); | |
2bc0bb5f FM |
3907 | /* |
3908 | * One of the free space endio workers might have | |
3909 | * created a new block group while updating a free space | |
3910 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3911 | * and hasn't released its transaction handle yet, in | |
3912 | * which case the new block group is still attached to | |
3913 | * its transaction handle and its creation has not | |
3914 | * finished yet (no block group item in the extent tree | |
3915 | * yet, etc). If this is the case, wait for all free | |
3916 | * space endio workers to finish and retry. This is a | |
3917 | * a very rare case so no need for a more efficient and | |
3918 | * complex approach. | |
3919 | */ | |
3920 | if (ret == -ENOENT) { | |
3921 | wait_event(cur_trans->writer_wait, | |
3922 | atomic_read(&cur_trans->num_writers) == 1); | |
2ff7e61e JM |
3923 | ret = write_one_cache_group(trans, fs_info, |
3924 | path, cache); | |
2bc0bb5f | 3925 | } |
ff1f8250 | 3926 | if (ret) |
66642832 | 3927 | btrfs_abort_transaction(trans, ret); |
ff1f8250 | 3928 | } |
c9dc4c65 CM |
3929 | |
3930 | /* if its not on the io list, we need to put the block group */ | |
3931 | if (should_put) | |
3932 | btrfs_put_block_group(cache); | |
e44081ef | 3933 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3934 | } |
e44081ef | 3935 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3936 | |
45ae2c18 NB |
3937 | /* |
3938 | * Refer to the definition of io_bgs member for details why it's safe | |
3939 | * to use it without any locking | |
3940 | */ | |
1bbc621e CM |
3941 | while (!list_empty(io)) { |
3942 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3943 | io_list); |
3944 | list_del_init(&cache->io_list); | |
afdb5718 | 3945 | btrfs_wait_cache_io(trans, cache, path); |
0cb59c99 JB |
3946 | btrfs_put_block_group(cache); |
3947 | } | |
3948 | ||
9078a3e1 | 3949 | btrfs_free_path(path); |
ce93ec54 | 3950 | return ret; |
9078a3e1 CM |
3951 | } |
3952 | ||
2ff7e61e | 3953 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 YZ |
3954 | { |
3955 | struct btrfs_block_group_cache *block_group; | |
3956 | int readonly = 0; | |
3957 | ||
0b246afa | 3958 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
3959 | if (!block_group || block_group->ro) |
3960 | readonly = 1; | |
3961 | if (block_group) | |
fa9c0d79 | 3962 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3963 | return readonly; |
3964 | } | |
3965 | ||
f78c436c FM |
3966 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
3967 | { | |
3968 | struct btrfs_block_group_cache *bg; | |
3969 | bool ret = true; | |
3970 | ||
3971 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3972 | if (!bg) | |
3973 | return false; | |
3974 | ||
3975 | spin_lock(&bg->lock); | |
3976 | if (bg->ro) | |
3977 | ret = false; | |
3978 | else | |
3979 | atomic_inc(&bg->nocow_writers); | |
3980 | spin_unlock(&bg->lock); | |
3981 | ||
3982 | /* no put on block group, done by btrfs_dec_nocow_writers */ | |
3983 | if (!ret) | |
3984 | btrfs_put_block_group(bg); | |
3985 | ||
3986 | return ret; | |
3987 | ||
3988 | } | |
3989 | ||
3990 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
3991 | { | |
3992 | struct btrfs_block_group_cache *bg; | |
3993 | ||
3994 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3995 | ASSERT(bg); | |
3996 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
3997 | wake_up_atomic_t(&bg->nocow_writers); | |
3998 | /* | |
3999 | * Once for our lookup and once for the lookup done by a previous call | |
4000 | * to btrfs_inc_nocow_writers() | |
4001 | */ | |
4002 | btrfs_put_block_group(bg); | |
4003 | btrfs_put_block_group(bg); | |
4004 | } | |
4005 | ||
f78c436c FM |
4006 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) |
4007 | { | |
5e4def20 | 4008 | wait_on_atomic_t(&bg->nocow_writers, atomic_t_wait, |
f78c436c FM |
4009 | TASK_UNINTERRUPTIBLE); |
4010 | } | |
4011 | ||
6ab0a202 JM |
4012 | static const char *alloc_name(u64 flags) |
4013 | { | |
4014 | switch (flags) { | |
4015 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
4016 | return "mixed"; | |
4017 | case BTRFS_BLOCK_GROUP_METADATA: | |
4018 | return "metadata"; | |
4019 | case BTRFS_BLOCK_GROUP_DATA: | |
4020 | return "data"; | |
4021 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
4022 | return "system"; | |
4023 | default: | |
4024 | WARN_ON(1); | |
4025 | return "invalid-combination"; | |
4026 | }; | |
4027 | } | |
4028 | ||
2be12ef7 NB |
4029 | static int create_space_info(struct btrfs_fs_info *info, u64 flags, |
4030 | struct btrfs_space_info **new) | |
4031 | { | |
4032 | ||
4033 | struct btrfs_space_info *space_info; | |
4034 | int i; | |
4035 | int ret; | |
4036 | ||
4037 | space_info = kzalloc(sizeof(*space_info), GFP_NOFS); | |
4038 | if (!space_info) | |
4039 | return -ENOMEM; | |
4040 | ||
4041 | ret = percpu_counter_init(&space_info->total_bytes_pinned, 0, | |
4042 | GFP_KERNEL); | |
4043 | if (ret) { | |
4044 | kfree(space_info); | |
4045 | return ret; | |
4046 | } | |
4047 | ||
4048 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
4049 | INIT_LIST_HEAD(&space_info->block_groups[i]); | |
4050 | init_rwsem(&space_info->groups_sem); | |
4051 | spin_lock_init(&space_info->lock); | |
4052 | space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; | |
4053 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
4054 | init_waitqueue_head(&space_info->wait); | |
4055 | INIT_LIST_HEAD(&space_info->ro_bgs); | |
4056 | INIT_LIST_HEAD(&space_info->tickets); | |
4057 | INIT_LIST_HEAD(&space_info->priority_tickets); | |
4058 | ||
4059 | ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype, | |
4060 | info->space_info_kobj, "%s", | |
4061 | alloc_name(space_info->flags)); | |
4062 | if (ret) { | |
4063 | percpu_counter_destroy(&space_info->total_bytes_pinned); | |
4064 | kfree(space_info); | |
4065 | return ret; | |
4066 | } | |
4067 | ||
4068 | *new = space_info; | |
4069 | list_add_rcu(&space_info->list, &info->space_info); | |
4070 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
4071 | info->data_sinfo = space_info; | |
4072 | ||
4073 | return ret; | |
4074 | } | |
4075 | ||
d2006e6d | 4076 | static void update_space_info(struct btrfs_fs_info *info, u64 flags, |
593060d7 | 4077 | u64 total_bytes, u64 bytes_used, |
e40edf2d | 4078 | u64 bytes_readonly, |
593060d7 CM |
4079 | struct btrfs_space_info **space_info) |
4080 | { | |
4081 | struct btrfs_space_info *found; | |
b742bb82 YZ |
4082 | int factor; |
4083 | ||
4084 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
4085 | BTRFS_BLOCK_GROUP_RAID10)) | |
4086 | factor = 2; | |
4087 | else | |
4088 | factor = 1; | |
593060d7 CM |
4089 | |
4090 | found = __find_space_info(info, flags); | |
d2006e6d NB |
4091 | ASSERT(found); |
4092 | spin_lock(&found->lock); | |
4093 | found->total_bytes += total_bytes; | |
4094 | found->disk_total += total_bytes * factor; | |
4095 | found->bytes_used += bytes_used; | |
4096 | found->disk_used += bytes_used * factor; | |
4097 | found->bytes_readonly += bytes_readonly; | |
4098 | if (total_bytes > 0) | |
4099 | found->full = 0; | |
4100 | space_info_add_new_bytes(info, found, total_bytes - | |
4101 | bytes_used - bytes_readonly); | |
4102 | spin_unlock(&found->lock); | |
4103 | *space_info = found; | |
593060d7 CM |
4104 | } |
4105 | ||
8790d502 CM |
4106 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
4107 | { | |
899c81ea ID |
4108 | u64 extra_flags = chunk_to_extended(flags) & |
4109 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 4110 | |
de98ced9 | 4111 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
4112 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
4113 | fs_info->avail_data_alloc_bits |= extra_flags; | |
4114 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
4115 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
4116 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
4117 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 4118 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 4119 | } |
593060d7 | 4120 | |
fc67c450 ID |
4121 | /* |
4122 | * returns target flags in extended format or 0 if restripe for this | |
4123 | * chunk_type is not in progress | |
c6664b42 ID |
4124 | * |
4125 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
4126 | */ |
4127 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
4128 | { | |
4129 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4130 | u64 target = 0; | |
4131 | ||
fc67c450 ID |
4132 | if (!bctl) |
4133 | return 0; | |
4134 | ||
4135 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
4136 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4137 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
4138 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
4139 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4140 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
4141 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
4142 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4143 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
4144 | } | |
4145 | ||
4146 | return target; | |
4147 | } | |
4148 | ||
a46d11a8 ID |
4149 | /* |
4150 | * @flags: available profiles in extended format (see ctree.h) | |
4151 | * | |
e4d8ec0f ID |
4152 | * Returns reduced profile in chunk format. If profile changing is in |
4153 | * progress (either running or paused) picks the target profile (if it's | |
4154 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 4155 | */ |
2ff7e61e | 4156 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c | 4157 | { |
0b246afa | 4158 | u64 num_devices = fs_info->fs_devices->rw_devices; |
fc67c450 | 4159 | u64 target; |
9c170b26 ZL |
4160 | u64 raid_type; |
4161 | u64 allowed = 0; | |
a061fc8d | 4162 | |
fc67c450 ID |
4163 | /* |
4164 | * see if restripe for this chunk_type is in progress, if so | |
4165 | * try to reduce to the target profile | |
4166 | */ | |
0b246afa JM |
4167 | spin_lock(&fs_info->balance_lock); |
4168 | target = get_restripe_target(fs_info, flags); | |
fc67c450 ID |
4169 | if (target) { |
4170 | /* pick target profile only if it's already available */ | |
4171 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
0b246afa | 4172 | spin_unlock(&fs_info->balance_lock); |
fc67c450 | 4173 | return extended_to_chunk(target); |
e4d8ec0f ID |
4174 | } |
4175 | } | |
0b246afa | 4176 | spin_unlock(&fs_info->balance_lock); |
e4d8ec0f | 4177 | |
53b381b3 | 4178 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
4179 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
4180 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
4181 | allowed |= btrfs_raid_group[raid_type]; | |
4182 | } | |
4183 | allowed &= flags; | |
4184 | ||
4185 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
4186 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
4187 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
4188 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
4189 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
4190 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
4191 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
4192 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
4193 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
4194 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
4195 | ||
4196 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
4197 | ||
4198 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
4199 | } |
4200 | ||
2ff7e61e | 4201 | static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
6a63209f | 4202 | { |
de98ced9 | 4203 | unsigned seq; |
f8213bdc | 4204 | u64 flags; |
de98ced9 MX |
4205 | |
4206 | do { | |
f8213bdc | 4207 | flags = orig_flags; |
0b246afa | 4208 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 MX |
4209 | |
4210 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
0b246afa | 4211 | flags |= fs_info->avail_data_alloc_bits; |
de98ced9 | 4212 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
0b246afa | 4213 | flags |= fs_info->avail_system_alloc_bits; |
de98ced9 | 4214 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
0b246afa JM |
4215 | flags |= fs_info->avail_metadata_alloc_bits; |
4216 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
6fef8df1 | 4217 | |
2ff7e61e | 4218 | return btrfs_reduce_alloc_profile(fs_info, flags); |
6a63209f JB |
4219 | } |
4220 | ||
1b86826d | 4221 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 4222 | { |
0b246afa | 4223 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 4224 | u64 flags; |
53b381b3 | 4225 | u64 ret; |
9ed74f2d | 4226 | |
b742bb82 YZ |
4227 | if (data) |
4228 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 4229 | else if (root == fs_info->chunk_root) |
b742bb82 | 4230 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 4231 | else |
b742bb82 | 4232 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 4233 | |
2ff7e61e | 4234 | ret = get_alloc_profile(fs_info, flags); |
53b381b3 | 4235 | return ret; |
6a63209f | 4236 | } |
9ed74f2d | 4237 | |
1b86826d JM |
4238 | u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info) |
4239 | { | |
4240 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA); | |
4241 | } | |
4242 | ||
4243 | u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info) | |
4244 | { | |
4245 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4246 | } | |
4247 | ||
4248 | u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info) | |
4249 | { | |
4250 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
4251 | } | |
4252 | ||
4136135b LB |
4253 | static u64 btrfs_space_info_used(struct btrfs_space_info *s_info, |
4254 | bool may_use_included) | |
4255 | { | |
4256 | ASSERT(s_info); | |
4257 | return s_info->bytes_used + s_info->bytes_reserved + | |
4258 | s_info->bytes_pinned + s_info->bytes_readonly + | |
4259 | (may_use_included ? s_info->bytes_may_use : 0); | |
4260 | } | |
4261 | ||
04f4f916 | 4262 | int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes) |
6a63209f | 4263 | { |
04f4f916 | 4264 | struct btrfs_root *root = inode->root; |
b4d7c3c9 | 4265 | struct btrfs_fs_info *fs_info = root->fs_info; |
1174cade | 4266 | struct btrfs_space_info *data_sinfo = fs_info->data_sinfo; |
ab6e2410 | 4267 | u64 used; |
94b947b2 | 4268 | int ret = 0; |
c99f1b0c ZL |
4269 | int need_commit = 2; |
4270 | int have_pinned_space; | |
6a63209f | 4271 | |
6a63209f | 4272 | /* make sure bytes are sectorsize aligned */ |
0b246afa | 4273 | bytes = ALIGN(bytes, fs_info->sectorsize); |
6a63209f | 4274 | |
9dced186 | 4275 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 4276 | need_commit = 0; |
9dced186 | 4277 | ASSERT(current->journal_info); |
0af3d00b JB |
4278 | } |
4279 | ||
6a63209f JB |
4280 | again: |
4281 | /* make sure we have enough space to handle the data first */ | |
4282 | spin_lock(&data_sinfo->lock); | |
4136135b | 4283 | used = btrfs_space_info_used(data_sinfo, true); |
ab6e2410 JB |
4284 | |
4285 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 4286 | struct btrfs_trans_handle *trans; |
9ed74f2d | 4287 | |
6a63209f JB |
4288 | /* |
4289 | * if we don't have enough free bytes in this space then we need | |
4290 | * to alloc a new chunk. | |
4291 | */ | |
b9fd47cd | 4292 | if (!data_sinfo->full) { |
6a63209f | 4293 | u64 alloc_target; |
9ed74f2d | 4294 | |
0e4f8f88 | 4295 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 4296 | spin_unlock(&data_sinfo->lock); |
1174cade | 4297 | |
1b86826d | 4298 | alloc_target = btrfs_data_alloc_profile(fs_info); |
9dced186 MX |
4299 | /* |
4300 | * It is ugly that we don't call nolock join | |
4301 | * transaction for the free space inode case here. | |
4302 | * But it is safe because we only do the data space | |
4303 | * reservation for the free space cache in the | |
4304 | * transaction context, the common join transaction | |
4305 | * just increase the counter of the current transaction | |
4306 | * handler, doesn't try to acquire the trans_lock of | |
4307 | * the fs. | |
4308 | */ | |
7a7eaa40 | 4309 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4310 | if (IS_ERR(trans)) |
4311 | return PTR_ERR(trans); | |
9ed74f2d | 4312 | |
2ff7e61e | 4313 | ret = do_chunk_alloc(trans, fs_info, alloc_target, |
0e4f8f88 | 4314 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 4315 | btrfs_end_transaction(trans); |
d52a5b5f MX |
4316 | if (ret < 0) { |
4317 | if (ret != -ENOSPC) | |
4318 | return ret; | |
c99f1b0c ZL |
4319 | else { |
4320 | have_pinned_space = 1; | |
d52a5b5f | 4321 | goto commit_trans; |
c99f1b0c | 4322 | } |
d52a5b5f | 4323 | } |
9ed74f2d | 4324 | |
6a63209f JB |
4325 | goto again; |
4326 | } | |
f2bb8f5c JB |
4327 | |
4328 | /* | |
b150a4f1 | 4329 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
4330 | * allocation, and no removed chunk in current transaction, |
4331 | * don't bother committing the transaction. | |
f2bb8f5c | 4332 | */ |
c99f1b0c ZL |
4333 | have_pinned_space = percpu_counter_compare( |
4334 | &data_sinfo->total_bytes_pinned, | |
4335 | used + bytes - data_sinfo->total_bytes); | |
6a63209f | 4336 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4337 | |
4e06bdd6 | 4338 | /* commit the current transaction and try again */ |
d52a5b5f | 4339 | commit_trans: |
92e2f7e3 | 4340 | if (need_commit) { |
c99f1b0c | 4341 | need_commit--; |
b150a4f1 | 4342 | |
e1746e83 ZL |
4343 | if (need_commit > 0) { |
4344 | btrfs_start_delalloc_roots(fs_info, 0, -1); | |
6374e57a | 4345 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, |
0b246afa | 4346 | (u64)-1); |
e1746e83 | 4347 | } |
9a4e7276 | 4348 | |
7a7eaa40 | 4349 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4350 | if (IS_ERR(trans)) |
4351 | return PTR_ERR(trans); | |
c99f1b0c | 4352 | if (have_pinned_space >= 0 || |
3204d33c JB |
4353 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4354 | &trans->transaction->flags) || | |
c99f1b0c | 4355 | need_commit > 0) { |
3a45bb20 | 4356 | ret = btrfs_commit_transaction(trans); |
94b947b2 ZL |
4357 | if (ret) |
4358 | return ret; | |
d7c15171 | 4359 | /* |
c2d6cb16 FM |
4360 | * The cleaner kthread might still be doing iput |
4361 | * operations. Wait for it to finish so that | |
4362 | * more space is released. | |
d7c15171 | 4363 | */ |
0b246afa JM |
4364 | mutex_lock(&fs_info->cleaner_delayed_iput_mutex); |
4365 | mutex_unlock(&fs_info->cleaner_delayed_iput_mutex); | |
94b947b2 ZL |
4366 | goto again; |
4367 | } else { | |
3a45bb20 | 4368 | btrfs_end_transaction(trans); |
94b947b2 | 4369 | } |
4e06bdd6 | 4370 | } |
9ed74f2d | 4371 | |
0b246afa | 4372 | trace_btrfs_space_reservation(fs_info, |
cab45e22 JM |
4373 | "space_info:enospc", |
4374 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4375 | return -ENOSPC; |
4376 | } | |
4377 | data_sinfo->bytes_may_use += bytes; | |
0b246afa | 4378 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4379 | data_sinfo->flags, bytes, 1); |
6a63209f | 4380 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4381 | |
237c0e9f | 4382 | return ret; |
9ed74f2d | 4383 | } |
6a63209f | 4384 | |
364ecf36 QW |
4385 | int btrfs_check_data_free_space(struct inode *inode, |
4386 | struct extent_changeset **reserved, u64 start, u64 len) | |
4ceff079 | 4387 | { |
0b246afa | 4388 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4389 | int ret; |
4390 | ||
4391 | /* align the range */ | |
0b246afa JM |
4392 | len = round_up(start + len, fs_info->sectorsize) - |
4393 | round_down(start, fs_info->sectorsize); | |
4394 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4395 | |
04f4f916 | 4396 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len); |
4ceff079 QW |
4397 | if (ret < 0) |
4398 | return ret; | |
4399 | ||
1e5ec2e7 | 4400 | /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */ |
364ecf36 | 4401 | ret = btrfs_qgroup_reserve_data(inode, reserved, start, len); |
7bc329c1 | 4402 | if (ret < 0) |
1e5ec2e7 | 4403 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
364ecf36 QW |
4404 | else |
4405 | ret = 0; | |
4ceff079 QW |
4406 | return ret; |
4407 | } | |
4408 | ||
4ceff079 QW |
4409 | /* |
4410 | * Called if we need to clear a data reservation for this inode | |
4411 | * Normally in a error case. | |
4412 | * | |
51773bec QW |
4413 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4414 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4415 | * Like clear_bit_hook(). | |
4ceff079 | 4416 | */ |
51773bec QW |
4417 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4418 | u64 len) | |
4ceff079 | 4419 | { |
0b246afa | 4420 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4421 | struct btrfs_space_info *data_sinfo; |
4422 | ||
4423 | /* Make sure the range is aligned to sectorsize */ | |
0b246afa JM |
4424 | len = round_up(start + len, fs_info->sectorsize) - |
4425 | round_down(start, fs_info->sectorsize); | |
4426 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4427 | |
0b246afa | 4428 | data_sinfo = fs_info->data_sinfo; |
4ceff079 QW |
4429 | spin_lock(&data_sinfo->lock); |
4430 | if (WARN_ON(data_sinfo->bytes_may_use < len)) | |
4431 | data_sinfo->bytes_may_use = 0; | |
4432 | else | |
4433 | data_sinfo->bytes_may_use -= len; | |
0b246afa | 4434 | trace_btrfs_space_reservation(fs_info, "space_info", |
4ceff079 QW |
4435 | data_sinfo->flags, len, 0); |
4436 | spin_unlock(&data_sinfo->lock); | |
4437 | } | |
4438 | ||
51773bec QW |
4439 | /* |
4440 | * Called if we need to clear a data reservation for this inode | |
4441 | * Normally in a error case. | |
4442 | * | |
01327610 | 4443 | * This one will handle the per-inode data rsv map for accurate reserved |
51773bec QW |
4444 | * space framework. |
4445 | */ | |
bc42bda2 QW |
4446 | void btrfs_free_reserved_data_space(struct inode *inode, |
4447 | struct extent_changeset *reserved, u64 start, u64 len) | |
51773bec | 4448 | { |
0c476a5d JM |
4449 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4450 | ||
4451 | /* Make sure the range is aligned to sectorsize */ | |
da17066c JM |
4452 | len = round_up(start + len, root->fs_info->sectorsize) - |
4453 | round_down(start, root->fs_info->sectorsize); | |
4454 | start = round_down(start, root->fs_info->sectorsize); | |
0c476a5d | 4455 | |
51773bec | 4456 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
bc42bda2 | 4457 | btrfs_qgroup_free_data(inode, reserved, start, len); |
51773bec QW |
4458 | } |
4459 | ||
97e728d4 | 4460 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4461 | { |
97e728d4 JB |
4462 | struct list_head *head = &info->space_info; |
4463 | struct btrfs_space_info *found; | |
e3ccfa98 | 4464 | |
97e728d4 JB |
4465 | rcu_read_lock(); |
4466 | list_for_each_entry_rcu(found, head, list) { | |
4467 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4468 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4469 | } |
97e728d4 | 4470 | rcu_read_unlock(); |
e3ccfa98 JB |
4471 | } |
4472 | ||
3c76cd84 MX |
4473 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
4474 | { | |
4475 | return (global->size << 1); | |
4476 | } | |
4477 | ||
2ff7e61e | 4478 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, |
698d0082 | 4479 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4480 | { |
0b246afa | 4481 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
8d8aafee | 4482 | u64 bytes_used = btrfs_space_info_used(sinfo, false); |
e5bc2458 | 4483 | u64 thresh; |
e3ccfa98 | 4484 | |
0e4f8f88 CM |
4485 | if (force == CHUNK_ALLOC_FORCE) |
4486 | return 1; | |
4487 | ||
fb25e914 JB |
4488 | /* |
4489 | * We need to take into account the global rsv because for all intents | |
4490 | * and purposes it's used space. Don't worry about locking the | |
4491 | * global_rsv, it doesn't change except when the transaction commits. | |
4492 | */ | |
54338b5c | 4493 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
8d8aafee | 4494 | bytes_used += calc_global_rsv_need_space(global_rsv); |
fb25e914 | 4495 | |
0e4f8f88 CM |
4496 | /* |
4497 | * in limited mode, we want to have some free space up to | |
4498 | * about 1% of the FS size. | |
4499 | */ | |
4500 | if (force == CHUNK_ALLOC_LIMITED) { | |
0b246afa | 4501 | thresh = btrfs_super_total_bytes(fs_info->super_copy); |
ee22184b | 4502 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
0e4f8f88 | 4503 | |
8d8aafee | 4504 | if (sinfo->total_bytes - bytes_used < thresh) |
0e4f8f88 CM |
4505 | return 1; |
4506 | } | |
0e4f8f88 | 4507 | |
8d8aafee | 4508 | if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) |
14ed0ca6 | 4509 | return 0; |
424499db | 4510 | return 1; |
32c00aff JB |
4511 | } |
4512 | ||
2ff7e61e | 4513 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4514 | { |
4515 | u64 num_dev; | |
4516 | ||
53b381b3 DW |
4517 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
4518 | BTRFS_BLOCK_GROUP_RAID0 | | |
4519 | BTRFS_BLOCK_GROUP_RAID5 | | |
4520 | BTRFS_BLOCK_GROUP_RAID6)) | |
0b246afa | 4521 | num_dev = fs_info->fs_devices->rw_devices; |
15d1ff81 LB |
4522 | else if (type & BTRFS_BLOCK_GROUP_RAID1) |
4523 | num_dev = 2; | |
4524 | else | |
4525 | num_dev = 1; /* DUP or single */ | |
4526 | ||
39c2d7fa | 4527 | return num_dev; |
15d1ff81 LB |
4528 | } |
4529 | ||
39c2d7fa FM |
4530 | /* |
4531 | * If @is_allocation is true, reserve space in the system space info necessary | |
4532 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4533 | * removing a chunk. | |
4534 | */ | |
4535 | void check_system_chunk(struct btrfs_trans_handle *trans, | |
2ff7e61e | 4536 | struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4537 | { |
4538 | struct btrfs_space_info *info; | |
4539 | u64 left; | |
4540 | u64 thresh; | |
4fbcdf66 | 4541 | int ret = 0; |
39c2d7fa | 4542 | u64 num_devs; |
4fbcdf66 FM |
4543 | |
4544 | /* | |
4545 | * Needed because we can end up allocating a system chunk and for an | |
4546 | * atomic and race free space reservation in the chunk block reserve. | |
4547 | */ | |
a32bf9a3 | 4548 | lockdep_assert_held(&fs_info->chunk_mutex); |
15d1ff81 | 4549 | |
0b246afa | 4550 | info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
15d1ff81 | 4551 | spin_lock(&info->lock); |
4136135b | 4552 | left = info->total_bytes - btrfs_space_info_used(info, true); |
15d1ff81 LB |
4553 | spin_unlock(&info->lock); |
4554 | ||
2ff7e61e | 4555 | num_devs = get_profile_num_devs(fs_info, type); |
39c2d7fa FM |
4556 | |
4557 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
0b246afa JM |
4558 | thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) + |
4559 | btrfs_calc_trans_metadata_size(fs_info, 1); | |
39c2d7fa | 4560 | |
0b246afa JM |
4561 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
4562 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
4563 | left, thresh, type); | |
4564 | dump_space_info(fs_info, info, 0, 0); | |
15d1ff81 LB |
4565 | } |
4566 | ||
4567 | if (left < thresh) { | |
1b86826d | 4568 | u64 flags = btrfs_system_alloc_profile(fs_info); |
15d1ff81 | 4569 | |
4fbcdf66 FM |
4570 | /* |
4571 | * Ignore failure to create system chunk. We might end up not | |
4572 | * needing it, as we might not need to COW all nodes/leafs from | |
4573 | * the paths we visit in the chunk tree (they were already COWed | |
4574 | * or created in the current transaction for example). | |
4575 | */ | |
2ff7e61e | 4576 | ret = btrfs_alloc_chunk(trans, fs_info, flags); |
4fbcdf66 FM |
4577 | } |
4578 | ||
4579 | if (!ret) { | |
0b246afa JM |
4580 | ret = btrfs_block_rsv_add(fs_info->chunk_root, |
4581 | &fs_info->chunk_block_rsv, | |
4fbcdf66 FM |
4582 | thresh, BTRFS_RESERVE_NO_FLUSH); |
4583 | if (!ret) | |
4584 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4585 | } |
4586 | } | |
4587 | ||
28b737f6 LB |
4588 | /* |
4589 | * If force is CHUNK_ALLOC_FORCE: | |
4590 | * - return 1 if it successfully allocates a chunk, | |
4591 | * - return errors including -ENOSPC otherwise. | |
4592 | * If force is NOT CHUNK_ALLOC_FORCE: | |
4593 | * - return 0 if it doesn't need to allocate a new chunk, | |
4594 | * - return 1 if it successfully allocates a chunk, | |
4595 | * - return errors including -ENOSPC otherwise. | |
4596 | */ | |
6324fbf3 | 4597 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 4598 | struct btrfs_fs_info *fs_info, u64 flags, int force) |
9ed74f2d | 4599 | { |
6324fbf3 | 4600 | struct btrfs_space_info *space_info; |
6d74119f | 4601 | int wait_for_alloc = 0; |
9ed74f2d | 4602 | int ret = 0; |
9ed74f2d | 4603 | |
c6b305a8 JB |
4604 | /* Don't re-enter if we're already allocating a chunk */ |
4605 | if (trans->allocating_chunk) | |
4606 | return -ENOSPC; | |
4607 | ||
0b246afa | 4608 | space_info = __find_space_info(fs_info, flags); |
dc2d3005 | 4609 | ASSERT(space_info); |
9ed74f2d | 4610 | |
6d74119f | 4611 | again: |
25179201 | 4612 | spin_lock(&space_info->lock); |
9e622d6b | 4613 | if (force < space_info->force_alloc) |
0e4f8f88 | 4614 | force = space_info->force_alloc; |
25179201 | 4615 | if (space_info->full) { |
2ff7e61e | 4616 | if (should_alloc_chunk(fs_info, space_info, force)) |
09fb99a6 FDBM |
4617 | ret = -ENOSPC; |
4618 | else | |
4619 | ret = 0; | |
25179201 | 4620 | spin_unlock(&space_info->lock); |
09fb99a6 | 4621 | return ret; |
9ed74f2d JB |
4622 | } |
4623 | ||
2ff7e61e | 4624 | if (!should_alloc_chunk(fs_info, space_info, force)) { |
25179201 | 4625 | spin_unlock(&space_info->lock); |
6d74119f JB |
4626 | return 0; |
4627 | } else if (space_info->chunk_alloc) { | |
4628 | wait_for_alloc = 1; | |
4629 | } else { | |
4630 | space_info->chunk_alloc = 1; | |
9ed74f2d | 4631 | } |
0e4f8f88 | 4632 | |
25179201 | 4633 | spin_unlock(&space_info->lock); |
9ed74f2d | 4634 | |
6d74119f JB |
4635 | mutex_lock(&fs_info->chunk_mutex); |
4636 | ||
4637 | /* | |
4638 | * The chunk_mutex is held throughout the entirety of a chunk | |
4639 | * allocation, so once we've acquired the chunk_mutex we know that the | |
4640 | * other guy is done and we need to recheck and see if we should | |
4641 | * allocate. | |
4642 | */ | |
4643 | if (wait_for_alloc) { | |
4644 | mutex_unlock(&fs_info->chunk_mutex); | |
4645 | wait_for_alloc = 0; | |
1e1c50a9 | 4646 | cond_resched(); |
6d74119f JB |
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) { |
6c686b35 | 4709 | btrfs_create_pending_block_groups(trans); |
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; | |
5da9d01b | 4830 | |
c61a16a7 | 4831 | /* Calc the number of the pages we need flush for space reservation */ |
2ff7e61e | 4832 | items = calc_reclaim_items_nr(fs_info, to_reclaim); |
6374e57a | 4833 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; |
c61a16a7 | 4834 | |
663350ac | 4835 | trans = (struct btrfs_trans_handle *)current->journal_info; |
69fe2d75 | 4836 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
bf9022e0 | 4837 | |
963d678b | 4838 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4839 | &fs_info->delalloc_bytes); |
f4c738c2 | 4840 | if (delalloc_bytes == 0) { |
fdb5effd | 4841 | if (trans) |
f4c738c2 | 4842 | return; |
38c135af | 4843 | if (wait_ordered) |
0b246afa | 4844 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f4c738c2 | 4845 | return; |
fdb5effd JB |
4846 | } |
4847 | ||
d3ee29e3 | 4848 | loops = 0; |
f4c738c2 JB |
4849 | while (delalloc_bytes && loops < 3) { |
4850 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
09cbfeaf | 4851 | nr_pages = max_reclaim >> PAGE_SHIFT; |
2ff7e61e | 4852 | btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items); |
dea31f52 JB |
4853 | /* |
4854 | * We need to wait for the async pages to actually start before | |
4855 | * we do anything. | |
4856 | */ | |
0b246afa | 4857 | max_reclaim = atomic_read(&fs_info->async_delalloc_pages); |
9f3a074d MX |
4858 | if (!max_reclaim) |
4859 | goto skip_async; | |
4860 | ||
4861 | if (max_reclaim <= nr_pages) | |
4862 | max_reclaim = 0; | |
4863 | else | |
4864 | max_reclaim -= nr_pages; | |
dea31f52 | 4865 | |
0b246afa JM |
4866 | wait_event(fs_info->async_submit_wait, |
4867 | atomic_read(&fs_info->async_delalloc_pages) <= | |
9f3a074d MX |
4868 | (int)max_reclaim); |
4869 | skip_async: | |
0019f10d | 4870 | spin_lock(&space_info->lock); |
957780eb JB |
4871 | if (list_empty(&space_info->tickets) && |
4872 | list_empty(&space_info->priority_tickets)) { | |
4873 | spin_unlock(&space_info->lock); | |
4874 | break; | |
4875 | } | |
0019f10d | 4876 | spin_unlock(&space_info->lock); |
5da9d01b | 4877 | |
36e39c40 | 4878 | loops++; |
f104d044 | 4879 | if (wait_ordered && !trans) { |
0b246afa | 4880 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f104d044 | 4881 | } else { |
f4c738c2 | 4882 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4883 | if (time_left) |
4884 | break; | |
4885 | } | |
963d678b | 4886 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4887 | &fs_info->delalloc_bytes); |
5da9d01b | 4888 | } |
5da9d01b YZ |
4889 | } |
4890 | ||
996478ca JB |
4891 | struct reserve_ticket { |
4892 | u64 bytes; | |
4893 | int error; | |
4894 | struct list_head list; | |
4895 | wait_queue_head_t wait; | |
4896 | }; | |
4897 | ||
663350ac JB |
4898 | /** |
4899 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4900 | * @root - the root we're allocating for | |
4901 | * @bytes - the number of bytes we want to reserve | |
4902 | * @force - force the commit | |
8bb8ab2e | 4903 | * |
663350ac JB |
4904 | * This will check to make sure that committing the transaction will actually |
4905 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4906 | * will return -ENOSPC. | |
8bb8ab2e | 4907 | */ |
0c9ab349 | 4908 | static int may_commit_transaction(struct btrfs_fs_info *fs_info, |
996478ca | 4909 | struct btrfs_space_info *space_info) |
663350ac | 4910 | { |
996478ca | 4911 | struct reserve_ticket *ticket = NULL; |
0b246afa | 4912 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv; |
663350ac | 4913 | struct btrfs_trans_handle *trans; |
996478ca | 4914 | u64 bytes; |
663350ac JB |
4915 | |
4916 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4917 | if (trans) | |
4918 | return -EAGAIN; | |
4919 | ||
996478ca JB |
4920 | spin_lock(&space_info->lock); |
4921 | if (!list_empty(&space_info->priority_tickets)) | |
4922 | ticket = list_first_entry(&space_info->priority_tickets, | |
4923 | struct reserve_ticket, list); | |
4924 | else if (!list_empty(&space_info->tickets)) | |
4925 | ticket = list_first_entry(&space_info->tickets, | |
4926 | struct reserve_ticket, list); | |
4927 | bytes = (ticket) ? ticket->bytes : 0; | |
4928 | spin_unlock(&space_info->lock); | |
4929 | ||
4930 | if (!bytes) | |
4931 | return 0; | |
663350ac JB |
4932 | |
4933 | /* See if there is enough pinned space to make this reservation */ | |
b150a4f1 | 4934 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
0424c548 | 4935 | bytes) >= 0) |
663350ac | 4936 | goto commit; |
663350ac JB |
4937 | |
4938 | /* | |
4939 | * See if there is some space in the delayed insertion reservation for | |
4940 | * this reservation. | |
4941 | */ | |
4942 | if (space_info != delayed_rsv->space_info) | |
4943 | return -ENOSPC; | |
4944 | ||
4945 | spin_lock(&delayed_rsv->lock); | |
996478ca JB |
4946 | if (delayed_rsv->size > bytes) |
4947 | bytes = 0; | |
4948 | else | |
4949 | bytes -= delayed_rsv->size; | |
057aac3e NB |
4950 | spin_unlock(&delayed_rsv->lock); |
4951 | ||
b150a4f1 | 4952 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
996478ca | 4953 | bytes) < 0) { |
663350ac JB |
4954 | return -ENOSPC; |
4955 | } | |
663350ac JB |
4956 | |
4957 | commit: | |
a9b3311e | 4958 | trans = btrfs_join_transaction(fs_info->extent_root); |
663350ac JB |
4959 | if (IS_ERR(trans)) |
4960 | return -ENOSPC; | |
4961 | ||
3a45bb20 | 4962 | return btrfs_commit_transaction(trans); |
663350ac JB |
4963 | } |
4964 | ||
e38ae7a0 NB |
4965 | /* |
4966 | * Try to flush some data based on policy set by @state. This is only advisory | |
4967 | * and may fail for various reasons. The caller is supposed to examine the | |
4968 | * state of @space_info to detect the outcome. | |
4969 | */ | |
4970 | static void flush_space(struct btrfs_fs_info *fs_info, | |
96c3f433 | 4971 | struct btrfs_space_info *space_info, u64 num_bytes, |
7bdd6277 | 4972 | int state) |
96c3f433 | 4973 | { |
a9b3311e | 4974 | struct btrfs_root *root = fs_info->extent_root; |
96c3f433 JB |
4975 | struct btrfs_trans_handle *trans; |
4976 | int nr; | |
f4c738c2 | 4977 | int ret = 0; |
96c3f433 JB |
4978 | |
4979 | switch (state) { | |
96c3f433 JB |
4980 | case FLUSH_DELAYED_ITEMS_NR: |
4981 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 | 4982 | if (state == FLUSH_DELAYED_ITEMS_NR) |
2ff7e61e | 4983 | nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2; |
18cd8ea6 | 4984 | else |
96c3f433 | 4985 | nr = -1; |
18cd8ea6 | 4986 | |
96c3f433 JB |
4987 | trans = btrfs_join_transaction(root); |
4988 | if (IS_ERR(trans)) { | |
4989 | ret = PTR_ERR(trans); | |
4990 | break; | |
4991 | } | |
e5c304e6 | 4992 | ret = btrfs_run_delayed_items_nr(trans, nr); |
3a45bb20 | 4993 | btrfs_end_transaction(trans); |
96c3f433 | 4994 | break; |
67b0fd63 JB |
4995 | case FLUSH_DELALLOC: |
4996 | case FLUSH_DELALLOC_WAIT: | |
7bdd6277 | 4997 | shrink_delalloc(fs_info, num_bytes * 2, num_bytes, |
67b0fd63 JB |
4998 | state == FLUSH_DELALLOC_WAIT); |
4999 | break; | |
ea658bad JB |
5000 | case ALLOC_CHUNK: |
5001 | trans = btrfs_join_transaction(root); | |
5002 | if (IS_ERR(trans)) { | |
5003 | ret = PTR_ERR(trans); | |
5004 | break; | |
5005 | } | |
2ff7e61e | 5006 | ret = do_chunk_alloc(trans, fs_info, |
1b86826d | 5007 | btrfs_metadata_alloc_profile(fs_info), |
ea658bad | 5008 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 5009 | btrfs_end_transaction(trans); |
eecba891 | 5010 | if (ret > 0 || ret == -ENOSPC) |
ea658bad JB |
5011 | ret = 0; |
5012 | break; | |
96c3f433 | 5013 | case COMMIT_TRANS: |
996478ca | 5014 | ret = may_commit_transaction(fs_info, space_info); |
96c3f433 JB |
5015 | break; |
5016 | default: | |
5017 | ret = -ENOSPC; | |
5018 | break; | |
5019 | } | |
5020 | ||
7bdd6277 NB |
5021 | trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state, |
5022 | ret); | |
e38ae7a0 | 5023 | return; |
96c3f433 | 5024 | } |
21c7e756 MX |
5025 | |
5026 | static inline u64 | |
c1c4919b JM |
5027 | btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info, |
5028 | struct btrfs_space_info *space_info, | |
5029 | bool system_chunk) | |
21c7e756 | 5030 | { |
957780eb | 5031 | struct reserve_ticket *ticket; |
21c7e756 MX |
5032 | u64 used; |
5033 | u64 expected; | |
957780eb | 5034 | u64 to_reclaim = 0; |
21c7e756 | 5035 | |
957780eb JB |
5036 | list_for_each_entry(ticket, &space_info->tickets, list) |
5037 | to_reclaim += ticket->bytes; | |
5038 | list_for_each_entry(ticket, &space_info->priority_tickets, list) | |
5039 | to_reclaim += ticket->bytes; | |
5040 | if (to_reclaim) | |
5041 | return to_reclaim; | |
21c7e756 | 5042 | |
e0af2484 | 5043 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); |
c1c4919b JM |
5044 | if (can_overcommit(fs_info, space_info, to_reclaim, |
5045 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
e0af2484 WX |
5046 | return 0; |
5047 | ||
0eee8a49 NB |
5048 | used = btrfs_space_info_used(space_info, true); |
5049 | ||
c1c4919b JM |
5050 | if (can_overcommit(fs_info, space_info, SZ_1M, |
5051 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
21c7e756 MX |
5052 | expected = div_factor_fine(space_info->total_bytes, 95); |
5053 | else | |
5054 | expected = div_factor_fine(space_info->total_bytes, 90); | |
5055 | ||
5056 | if (used > expected) | |
5057 | to_reclaim = used - expected; | |
5058 | else | |
5059 | to_reclaim = 0; | |
5060 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
5061 | space_info->bytes_reserved); | |
21c7e756 MX |
5062 | return to_reclaim; |
5063 | } | |
5064 | ||
c1c4919b JM |
5065 | static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info, |
5066 | struct btrfs_space_info *space_info, | |
5067 | u64 used, bool system_chunk) | |
21c7e756 | 5068 | { |
365c5313 JB |
5069 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
5070 | ||
5071 | /* If we're just plain full then async reclaim just slows us down. */ | |
baee8790 | 5072 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) |
365c5313 JB |
5073 | return 0; |
5074 | ||
c1c4919b JM |
5075 | if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5076 | system_chunk)) | |
d38b349c JB |
5077 | return 0; |
5078 | ||
0b246afa JM |
5079 | return (used >= thresh && !btrfs_fs_closing(fs_info) && |
5080 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); | |
21c7e756 MX |
5081 | } |
5082 | ||
957780eb | 5083 | static void wake_all_tickets(struct list_head *head) |
21c7e756 | 5084 | { |
957780eb | 5085 | struct reserve_ticket *ticket; |
25ce459c | 5086 | |
957780eb JB |
5087 | while (!list_empty(head)) { |
5088 | ticket = list_first_entry(head, struct reserve_ticket, list); | |
5089 | list_del_init(&ticket->list); | |
5090 | ticket->error = -ENOSPC; | |
5091 | wake_up(&ticket->wait); | |
21c7e756 | 5092 | } |
21c7e756 MX |
5093 | } |
5094 | ||
957780eb JB |
5095 | /* |
5096 | * This is for normal flushers, we can wait all goddamned day if we want to. We | |
5097 | * will loop and continuously try to flush as long as we are making progress. | |
5098 | * We count progress as clearing off tickets each time we have to loop. | |
5099 | */ | |
21c7e756 MX |
5100 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) |
5101 | { | |
5102 | struct btrfs_fs_info *fs_info; | |
5103 | struct btrfs_space_info *space_info; | |
5104 | u64 to_reclaim; | |
5105 | int flush_state; | |
957780eb | 5106 | int commit_cycles = 0; |
ce129655 | 5107 | u64 last_tickets_id; |
21c7e756 MX |
5108 | |
5109 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
5110 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
5111 | ||
957780eb | 5112 | spin_lock(&space_info->lock); |
c1c4919b JM |
5113 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5114 | false); | |
957780eb JB |
5115 | if (!to_reclaim) { |
5116 | space_info->flush = 0; | |
5117 | spin_unlock(&space_info->lock); | |
21c7e756 | 5118 | return; |
957780eb | 5119 | } |
ce129655 | 5120 | last_tickets_id = space_info->tickets_id; |
957780eb | 5121 | spin_unlock(&space_info->lock); |
21c7e756 MX |
5122 | |
5123 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
957780eb | 5124 | do { |
e38ae7a0 | 5125 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
957780eb JB |
5126 | spin_lock(&space_info->lock); |
5127 | if (list_empty(&space_info->tickets)) { | |
5128 | space_info->flush = 0; | |
5129 | spin_unlock(&space_info->lock); | |
5130 | return; | |
5131 | } | |
c1c4919b JM |
5132 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, |
5133 | space_info, | |
5134 | false); | |
ce129655 | 5135 | if (last_tickets_id == space_info->tickets_id) { |
957780eb JB |
5136 | flush_state++; |
5137 | } else { | |
ce129655 | 5138 | last_tickets_id = space_info->tickets_id; |
957780eb JB |
5139 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
5140 | if (commit_cycles) | |
5141 | commit_cycles--; | |
5142 | } | |
5143 | ||
5144 | if (flush_state > COMMIT_TRANS) { | |
5145 | commit_cycles++; | |
5146 | if (commit_cycles > 2) { | |
5147 | wake_all_tickets(&space_info->tickets); | |
5148 | space_info->flush = 0; | |
5149 | } else { | |
5150 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5151 | } | |
5152 | } | |
5153 | spin_unlock(&space_info->lock); | |
5154 | } while (flush_state <= COMMIT_TRANS); | |
5155 | } | |
5156 | ||
5157 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
5158 | { | |
5159 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
5160 | } | |
5161 | ||
5162 | static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, | |
5163 | struct btrfs_space_info *space_info, | |
5164 | struct reserve_ticket *ticket) | |
5165 | { | |
5166 | u64 to_reclaim; | |
5167 | int flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5168 | ||
5169 | spin_lock(&space_info->lock); | |
c1c4919b JM |
5170 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5171 | false); | |
957780eb JB |
5172 | if (!to_reclaim) { |
5173 | spin_unlock(&space_info->lock); | |
5174 | return; | |
5175 | } | |
5176 | spin_unlock(&space_info->lock); | |
5177 | ||
21c7e756 | 5178 | do { |
7bdd6277 | 5179 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
21c7e756 | 5180 | flush_state++; |
957780eb JB |
5181 | spin_lock(&space_info->lock); |
5182 | if (ticket->bytes == 0) { | |
5183 | spin_unlock(&space_info->lock); | |
21c7e756 | 5184 | return; |
957780eb JB |
5185 | } |
5186 | spin_unlock(&space_info->lock); | |
5187 | ||
5188 | /* | |
5189 | * Priority flushers can't wait on delalloc without | |
5190 | * deadlocking. | |
5191 | */ | |
5192 | if (flush_state == FLUSH_DELALLOC || | |
5193 | flush_state == FLUSH_DELALLOC_WAIT) | |
5194 | flush_state = ALLOC_CHUNK; | |
365c5313 | 5195 | } while (flush_state < COMMIT_TRANS); |
21c7e756 MX |
5196 | } |
5197 | ||
957780eb JB |
5198 | static int wait_reserve_ticket(struct btrfs_fs_info *fs_info, |
5199 | struct btrfs_space_info *space_info, | |
5200 | struct reserve_ticket *ticket, u64 orig_bytes) | |
5201 | ||
21c7e756 | 5202 | { |
957780eb JB |
5203 | DEFINE_WAIT(wait); |
5204 | int ret = 0; | |
5205 | ||
5206 | spin_lock(&space_info->lock); | |
5207 | while (ticket->bytes > 0 && ticket->error == 0) { | |
5208 | ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE); | |
5209 | if (ret) { | |
5210 | ret = -EINTR; | |
5211 | break; | |
5212 | } | |
5213 | spin_unlock(&space_info->lock); | |
5214 | ||
5215 | schedule(); | |
5216 | ||
5217 | finish_wait(&ticket->wait, &wait); | |
5218 | spin_lock(&space_info->lock); | |
5219 | } | |
5220 | if (!ret) | |
5221 | ret = ticket->error; | |
5222 | if (!list_empty(&ticket->list)) | |
5223 | list_del_init(&ticket->list); | |
5224 | if (ticket->bytes && ticket->bytes < orig_bytes) { | |
5225 | u64 num_bytes = orig_bytes - ticket->bytes; | |
5226 | space_info->bytes_may_use -= num_bytes; | |
5227 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5228 | space_info->flags, num_bytes, 0); | |
5229 | } | |
5230 | spin_unlock(&space_info->lock); | |
5231 | ||
5232 | return ret; | |
21c7e756 MX |
5233 | } |
5234 | ||
4a92b1b8 JB |
5235 | /** |
5236 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5237 | * @root - the root we're allocating for | |
957780eb | 5238 | * @space_info - the space info we want to allocate from |
4a92b1b8 | 5239 | * @orig_bytes - the number of bytes we want |
48fc7f7e | 5240 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 5241 | * |
01327610 | 5242 | * This will reserve orig_bytes number of bytes from the space info associated |
4a92b1b8 JB |
5243 | * with the block_rsv. If there is not enough space it will make an attempt to |
5244 | * flush out space to make room. It will do this by flushing delalloc if | |
5245 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5246 | * regain reservations will be made and this will fail if there is not enough | |
5247 | * space already. | |
8bb8ab2e | 5248 | */ |
c1c4919b | 5249 | static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info, |
957780eb JB |
5250 | struct btrfs_space_info *space_info, |
5251 | u64 orig_bytes, | |
c1c4919b JM |
5252 | enum btrfs_reserve_flush_enum flush, |
5253 | bool system_chunk) | |
9ed74f2d | 5254 | { |
957780eb | 5255 | struct reserve_ticket ticket; |
2bf64758 | 5256 | u64 used; |
8bb8ab2e | 5257 | int ret = 0; |
9ed74f2d | 5258 | |
957780eb | 5259 | ASSERT(orig_bytes); |
8ca17f0f | 5260 | ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL); |
fdb5effd | 5261 | |
8bb8ab2e | 5262 | spin_lock(&space_info->lock); |
fdb5effd | 5263 | ret = -ENOSPC; |
4136135b | 5264 | used = btrfs_space_info_used(space_info, true); |
9ed74f2d | 5265 | |
8bb8ab2e | 5266 | /* |
957780eb JB |
5267 | * If we have enough space then hooray, make our reservation and carry |
5268 | * on. If not see if we can overcommit, and if we can, hooray carry on. | |
5269 | * If not things get more complicated. | |
8bb8ab2e | 5270 | */ |
957780eb JB |
5271 | if (used + orig_bytes <= space_info->total_bytes) { |
5272 | space_info->bytes_may_use += orig_bytes; | |
0b246afa JM |
5273 | trace_btrfs_space_reservation(fs_info, "space_info", |
5274 | space_info->flags, orig_bytes, 1); | |
957780eb | 5275 | ret = 0; |
c1c4919b JM |
5276 | } else if (can_overcommit(fs_info, space_info, orig_bytes, flush, |
5277 | system_chunk)) { | |
44734ed1 | 5278 | space_info->bytes_may_use += orig_bytes; |
0b246afa JM |
5279 | trace_btrfs_space_reservation(fs_info, "space_info", |
5280 | space_info->flags, orig_bytes, 1); | |
44734ed1 | 5281 | ret = 0; |
2bf64758 JB |
5282 | } |
5283 | ||
8bb8ab2e | 5284 | /* |
957780eb JB |
5285 | * If we couldn't make a reservation then setup our reservation ticket |
5286 | * and kick the async worker if it's not already running. | |
08e007d2 | 5287 | * |
957780eb JB |
5288 | * If we are a priority flusher then we just need to add our ticket to |
5289 | * the list and we will do our own flushing further down. | |
8bb8ab2e | 5290 | */ |
72bcd99d | 5291 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
957780eb JB |
5292 | ticket.bytes = orig_bytes; |
5293 | ticket.error = 0; | |
5294 | init_waitqueue_head(&ticket.wait); | |
5295 | if (flush == BTRFS_RESERVE_FLUSH_ALL) { | |
5296 | list_add_tail(&ticket.list, &space_info->tickets); | |
5297 | if (!space_info->flush) { | |
5298 | space_info->flush = 1; | |
0b246afa | 5299 | trace_btrfs_trigger_flush(fs_info, |
f376df2b JB |
5300 | space_info->flags, |
5301 | orig_bytes, flush, | |
5302 | "enospc"); | |
957780eb | 5303 | queue_work(system_unbound_wq, |
c1c4919b | 5304 | &fs_info->async_reclaim_work); |
957780eb JB |
5305 | } |
5306 | } else { | |
5307 | list_add_tail(&ticket.list, | |
5308 | &space_info->priority_tickets); | |
5309 | } | |
21c7e756 MX |
5310 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
5311 | used += orig_bytes; | |
f6acfd50 JB |
5312 | /* |
5313 | * We will do the space reservation dance during log replay, | |
5314 | * which means we won't have fs_info->fs_root set, so don't do | |
5315 | * the async reclaim as we will panic. | |
5316 | */ | |
0b246afa | 5317 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) && |
c1c4919b JM |
5318 | need_do_async_reclaim(fs_info, space_info, |
5319 | used, system_chunk) && | |
0b246afa JM |
5320 | !work_busy(&fs_info->async_reclaim_work)) { |
5321 | trace_btrfs_trigger_flush(fs_info, space_info->flags, | |
5322 | orig_bytes, flush, "preempt"); | |
21c7e756 | 5323 | queue_work(system_unbound_wq, |
0b246afa | 5324 | &fs_info->async_reclaim_work); |
f376df2b | 5325 | } |
8bb8ab2e | 5326 | } |
f0486c68 | 5327 | spin_unlock(&space_info->lock); |
08e007d2 | 5328 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
957780eb | 5329 | return ret; |
f0486c68 | 5330 | |
957780eb | 5331 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
0b246afa | 5332 | return wait_reserve_ticket(fs_info, space_info, &ticket, |
957780eb | 5333 | orig_bytes); |
08e007d2 | 5334 | |
957780eb | 5335 | ret = 0; |
0b246afa | 5336 | priority_reclaim_metadata_space(fs_info, space_info, &ticket); |
957780eb JB |
5337 | spin_lock(&space_info->lock); |
5338 | if (ticket.bytes) { | |
5339 | if (ticket.bytes < orig_bytes) { | |
5340 | u64 num_bytes = orig_bytes - ticket.bytes; | |
5341 | space_info->bytes_may_use -= num_bytes; | |
0b246afa JM |
5342 | trace_btrfs_space_reservation(fs_info, "space_info", |
5343 | space_info->flags, | |
5344 | num_bytes, 0); | |
08e007d2 | 5345 | |
957780eb JB |
5346 | } |
5347 | list_del_init(&ticket.list); | |
5348 | ret = -ENOSPC; | |
5349 | } | |
5350 | spin_unlock(&space_info->lock); | |
5351 | ASSERT(list_empty(&ticket.list)); | |
5352 | return ret; | |
5353 | } | |
8bb8ab2e | 5354 | |
957780eb JB |
5355 | /** |
5356 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5357 | * @root - the root we're allocating for | |
5358 | * @block_rsv - the block_rsv we're allocating for | |
5359 | * @orig_bytes - the number of bytes we want | |
5360 | * @flush - whether or not we can flush to make our reservation | |
5361 | * | |
5362 | * This will reserve orgi_bytes number of bytes from the space info associated | |
5363 | * with the block_rsv. If there is not enough space it will make an attempt to | |
5364 | * flush out space to make room. It will do this by flushing delalloc if | |
5365 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5366 | * regain reservations will be made and this will fail if there is not enough | |
5367 | * space already. | |
5368 | */ | |
5369 | static int reserve_metadata_bytes(struct btrfs_root *root, | |
5370 | struct btrfs_block_rsv *block_rsv, | |
5371 | u64 orig_bytes, | |
5372 | enum btrfs_reserve_flush_enum flush) | |
5373 | { | |
0b246afa JM |
5374 | struct btrfs_fs_info *fs_info = root->fs_info; |
5375 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
957780eb | 5376 | int ret; |
c1c4919b | 5377 | bool system_chunk = (root == fs_info->chunk_root); |
957780eb | 5378 | |
c1c4919b JM |
5379 | ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info, |
5380 | orig_bytes, flush, system_chunk); | |
5d80366e JB |
5381 | if (ret == -ENOSPC && |
5382 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5d80366e JB |
5383 | if (block_rsv != global_rsv && |
5384 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5385 | ret = 0; | |
5386 | } | |
9a3daff3 | 5387 | if (ret == -ENOSPC) { |
0b246afa | 5388 | trace_btrfs_space_reservation(fs_info, "space_info:enospc", |
957780eb JB |
5389 | block_rsv->space_info->flags, |
5390 | orig_bytes, 1); | |
9a3daff3 NB |
5391 | |
5392 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) | |
5393 | dump_space_info(fs_info, block_rsv->space_info, | |
5394 | orig_bytes, 0); | |
5395 | } | |
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 | */ | |
3f2dd7a0 QW |
5748 | static int btrfs_inode_rsv_refill(struct btrfs_inode *inode, |
5749 | enum btrfs_reserve_flush_enum flush) | |
69fe2d75 JB |
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 | ||
43b18595 QW |
5764 | ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes, true); |
5765 | if (ret) | |
5766 | return ret; | |
69fe2d75 JB |
5767 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
5768 | if (!ret) { | |
5769 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
5770 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
5771 | btrfs_ino(inode), num_bytes, 1); | |
5772 | } | |
5773 | return ret; | |
5774 | } | |
5775 | ||
5776 | /** | |
5777 | * btrfs_inode_rsv_release - release any excessive reservation. | |
5778 | * @inode - the inode we need to release from. | |
43b18595 QW |
5779 | * @qgroup_free - free or convert qgroup meta. |
5780 | * Unlike normal operation, qgroup meta reservation needs to know if we are | |
5781 | * freeing qgroup reservation or just converting it into per-trans. Normally | |
5782 | * @qgroup_free is true for error handling, and false for normal release. | |
69fe2d75 JB |
5783 | * |
5784 | * This is the same as btrfs_block_rsv_release, except that it handles the | |
5785 | * tracepoint for the reservation. | |
5786 | */ | |
43b18595 | 5787 | static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free) |
69fe2d75 JB |
5788 | { |
5789 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
5790 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5791 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; | |
5792 | u64 released = 0; | |
5793 | ||
5794 | /* | |
5795 | * Since we statically set the block_rsv->size we just want to say we | |
5796 | * are releasing 0 bytes, and then we'll just get the reservation over | |
5797 | * the size free'd. | |
5798 | */ | |
5799 | released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0); | |
5800 | if (released > 0) | |
5801 | trace_btrfs_space_reservation(fs_info, "delalloc", | |
5802 | btrfs_ino(inode), released, 0); | |
43b18595 QW |
5803 | if (qgroup_free) |
5804 | btrfs_qgroup_free_meta_prealloc(inode->root, released); | |
5805 | else | |
5806 | btrfs_qgroup_convert_reserved_meta(inode->root, released); | |
69fe2d75 JB |
5807 | } |
5808 | ||
2ff7e61e | 5809 | void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5810 | struct btrfs_block_rsv *block_rsv, |
5811 | u64 num_bytes) | |
5812 | { | |
0b246afa JM |
5813 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
5814 | ||
17504584 | 5815 | if (global_rsv == block_rsv || |
f0486c68 YZ |
5816 | block_rsv->space_info != global_rsv->space_info) |
5817 | global_rsv = NULL; | |
0b246afa | 5818 | block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes); |
6a63209f JB |
5819 | } |
5820 | ||
8929ecfa YZ |
5821 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5822 | { | |
5823 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5824 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5825 | u64 num_bytes; | |
6a63209f | 5826 | |
ae2e4728 JB |
5827 | /* |
5828 | * The global block rsv is based on the size of the extent tree, the | |
5829 | * checksum tree and the root tree. If the fs is empty we want to set | |
5830 | * it to a minimal amount for safety. | |
5831 | */ | |
5832 | num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) + | |
5833 | btrfs_root_used(&fs_info->csum_root->root_item) + | |
5834 | btrfs_root_used(&fs_info->tree_root->root_item); | |
5835 | num_bytes = max_t(u64, num_bytes, SZ_16M); | |
33b4d47f | 5836 | |
8929ecfa | 5837 | spin_lock(&sinfo->lock); |
1f699d38 | 5838 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5839 | |
ee22184b | 5840 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); |
4e06bdd6 | 5841 | |
fb4b10e5 | 5842 | if (block_rsv->reserved < block_rsv->size) { |
4136135b | 5843 | num_bytes = btrfs_space_info_used(sinfo, true); |
fb4b10e5 JB |
5844 | if (sinfo->total_bytes > num_bytes) { |
5845 | num_bytes = sinfo->total_bytes - num_bytes; | |
5846 | num_bytes = min(num_bytes, | |
5847 | block_rsv->size - block_rsv->reserved); | |
5848 | block_rsv->reserved += num_bytes; | |
5849 | sinfo->bytes_may_use += num_bytes; | |
5850 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5851 | sinfo->flags, num_bytes, | |
5852 | 1); | |
5853 | } | |
5854 | } else if (block_rsv->reserved > block_rsv->size) { | |
8929ecfa | 5855 | num_bytes = block_rsv->reserved - block_rsv->size; |
fb25e914 | 5856 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 5857 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5858 | sinfo->flags, num_bytes, 0); |
8929ecfa | 5859 | block_rsv->reserved = block_rsv->size; |
8929ecfa | 5860 | } |
182608c8 | 5861 | |
fb4b10e5 JB |
5862 | if (block_rsv->reserved == block_rsv->size) |
5863 | block_rsv->full = 1; | |
5864 | else | |
5865 | block_rsv->full = 0; | |
5866 | ||
8929ecfa | 5867 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5868 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5869 | } |
5870 | ||
f0486c68 | 5871 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5872 | { |
f0486c68 | 5873 | struct btrfs_space_info *space_info; |
6a63209f | 5874 | |
f0486c68 YZ |
5875 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
5876 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 5877 | |
f0486c68 | 5878 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5879 | fs_info->global_block_rsv.space_info = space_info; |
f0486c68 YZ |
5880 | fs_info->trans_block_rsv.space_info = space_info; |
5881 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5882 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 5883 | |
8929ecfa YZ |
5884 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
5885 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
5886 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
5887 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5888 | if (fs_info->quota_root) |
5889 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5890 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5891 | |
8929ecfa | 5892 | update_global_block_rsv(fs_info); |
6a63209f JB |
5893 | } |
5894 | ||
8929ecfa | 5895 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5896 | { |
8c2a3ca2 JB |
5897 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
5898 | (u64)-1); | |
8929ecfa YZ |
5899 | WARN_ON(fs_info->trans_block_rsv.size > 0); |
5900 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5901 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5902 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5903 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5904 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
5905 | } |
5906 | ||
6a63209f | 5907 | |
4fbcdf66 FM |
5908 | /* |
5909 | * To be called after all the new block groups attached to the transaction | |
5910 | * handle have been created (btrfs_create_pending_block_groups()). | |
5911 | */ | |
5912 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5913 | { | |
64b63580 | 5914 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4fbcdf66 FM |
5915 | |
5916 | if (!trans->chunk_bytes_reserved) | |
5917 | return; | |
5918 | ||
5919 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5920 | ||
5921 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
5922 | trans->chunk_bytes_reserved); | |
5923 | trans->chunk_bytes_reserved = 0; | |
5924 | } | |
5925 | ||
79787eaa | 5926 | /* Can only return 0 or -ENOSPC */ |
d68fc57b | 5927 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
8ed7a2a0 | 5928 | struct btrfs_inode *inode) |
d68fc57b | 5929 | { |
8ed7a2a0 NB |
5930 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
5931 | struct btrfs_root *root = inode->root; | |
40acc3ee JB |
5932 | /* |
5933 | * We always use trans->block_rsv here as we will have reserved space | |
5934 | * for our orphan when starting the transaction, using get_block_rsv() | |
5935 | * here will sometimes make us choose the wrong block rsv as we could be | |
5936 | * doing a reloc inode for a non refcounted root. | |
5937 | */ | |
5938 | struct btrfs_block_rsv *src_rsv = trans->block_rsv; | |
d68fc57b YZ |
5939 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; |
5940 | ||
5941 | /* | |
fcb80c2a JB |
5942 | * We need to hold space in order to delete our orphan item once we've |
5943 | * added it, so this takes the reservation so we can release it later | |
5944 | * when we are truly done with the orphan item. | |
d68fc57b | 5945 | */ |
0b246afa JM |
5946 | u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
5947 | ||
9678c543 | 5948 | trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode), |
8ed7a2a0 | 5949 | num_bytes, 1); |
25d609f8 | 5950 | return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1); |
6a63209f JB |
5951 | } |
5952 | ||
703b391a | 5953 | void btrfs_orphan_release_metadata(struct btrfs_inode *inode) |
97e728d4 | 5954 | { |
703b391a NB |
5955 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
5956 | struct btrfs_root *root = inode->root; | |
0b246afa JM |
5957 | u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
5958 | ||
703b391a NB |
5959 | trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode), |
5960 | num_bytes, 0); | |
2ff7e61e | 5961 | btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes); |
d68fc57b | 5962 | } |
97e728d4 | 5963 | |
d5c12070 MX |
5964 | /* |
5965 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5966 | * root: the root of the parent directory | |
5967 | * rsv: block reservation | |
5968 | * items: the number of items that we need do reservation | |
5969 | * qgroup_reserved: used to return the reserved size in qgroup | |
5970 | * | |
5971 | * This function is used to reserve the space for snapshot/subvolume | |
5972 | * creation and deletion. Those operations are different with the | |
5973 | * common file/directory operations, they change two fs/file trees | |
5974 | * and root tree, the number of items that the qgroup reserves is | |
5975 | * different with the free space reservation. So we can not use | |
01327610 | 5976 | * the space reservation mechanism in start_transaction(). |
d5c12070 MX |
5977 | */ |
5978 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
5979 | struct btrfs_block_rsv *rsv, | |
5980 | int items, | |
ee3441b4 JM |
5981 | u64 *qgroup_reserved, |
5982 | bool use_global_rsv) | |
a22285a6 | 5983 | { |
d5c12070 MX |
5984 | u64 num_bytes; |
5985 | int ret; | |
0b246afa JM |
5986 | struct btrfs_fs_info *fs_info = root->fs_info; |
5987 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d5c12070 | 5988 | |
0b246afa | 5989 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
d5c12070 | 5990 | /* One for parent inode, two for dir entries */ |
0b246afa | 5991 | num_bytes = 3 * fs_info->nodesize; |
733e03a0 | 5992 | ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes, true); |
d5c12070 MX |
5993 | if (ret) |
5994 | return ret; | |
5995 | } else { | |
5996 | num_bytes = 0; | |
5997 | } | |
5998 | ||
5999 | *qgroup_reserved = num_bytes; | |
6000 | ||
0b246afa JM |
6001 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, items); |
6002 | rsv->space_info = __find_space_info(fs_info, | |
d5c12070 MX |
6003 | BTRFS_BLOCK_GROUP_METADATA); |
6004 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
6005 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
6006 | |
6007 | if (ret == -ENOSPC && use_global_rsv) | |
25d609f8 | 6008 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1); |
ee3441b4 | 6009 | |
7174109c | 6010 | if (ret && *qgroup_reserved) |
733e03a0 | 6011 | btrfs_qgroup_free_meta_prealloc(root, *qgroup_reserved); |
d5c12070 MX |
6012 | |
6013 | return ret; | |
6014 | } | |
6015 | ||
2ff7e61e | 6016 | void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info, |
7775c818 | 6017 | struct btrfs_block_rsv *rsv) |
d5c12070 | 6018 | { |
2ff7e61e | 6019 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
97e728d4 JB |
6020 | } |
6021 | ||
69fe2d75 JB |
6022 | static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info, |
6023 | struct btrfs_inode *inode) | |
9e0baf60 | 6024 | { |
69fe2d75 JB |
6025 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
6026 | u64 reserve_size = 0; | |
6027 | u64 csum_leaves; | |
6028 | unsigned outstanding_extents; | |
9e0baf60 | 6029 | |
69fe2d75 JB |
6030 | lockdep_assert_held(&inode->lock); |
6031 | outstanding_extents = inode->outstanding_extents; | |
6032 | if (outstanding_extents) | |
6033 | reserve_size = btrfs_calc_trans_metadata_size(fs_info, | |
6034 | outstanding_extents + 1); | |
6035 | csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, | |
6036 | inode->csum_bytes); | |
6037 | reserve_size += btrfs_calc_trans_metadata_size(fs_info, | |
6038 | csum_leaves); | |
9e0baf60 | 6039 | |
69fe2d75 JB |
6040 | spin_lock(&block_rsv->lock); |
6041 | block_rsv->size = reserve_size; | |
6042 | spin_unlock(&block_rsv->lock); | |
0ca1f7ce | 6043 | } |
c146afad | 6044 | |
9f3db423 | 6045 | int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 6046 | { |
9f3db423 | 6047 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
69fe2d75 | 6048 | unsigned nr_extents; |
08e007d2 | 6049 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 6050 | int ret = 0; |
c64c2bd8 | 6051 | bool delalloc_lock = true; |
6324fbf3 | 6052 | |
c64c2bd8 JB |
6053 | /* If we are a free space inode we need to not flush since we will be in |
6054 | * the middle of a transaction commit. We also don't need the delalloc | |
6055 | * mutex since we won't race with anybody. We need this mostly to make | |
6056 | * lockdep shut its filthy mouth. | |
bac357dc JB |
6057 | * |
6058 | * If we have a transaction open (can happen if we call truncate_block | |
6059 | * from truncate), then we need FLUSH_LIMIT so we don't deadlock. | |
c64c2bd8 JB |
6060 | */ |
6061 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 6062 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 | 6063 | delalloc_lock = false; |
da07d4ab NB |
6064 | } else { |
6065 | if (current->journal_info) | |
6066 | flush = BTRFS_RESERVE_FLUSH_LIMIT; | |
c09544e0 | 6067 | |
da07d4ab NB |
6068 | if (btrfs_transaction_in_commit(fs_info)) |
6069 | schedule_timeout(1); | |
6070 | } | |
ec44a35c | 6071 | |
c64c2bd8 | 6072 | if (delalloc_lock) |
9f3db423 | 6073 | mutex_lock(&inode->delalloc_mutex); |
c64c2bd8 | 6074 | |
0b246afa | 6075 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
69fe2d75 JB |
6076 | |
6077 | /* Add our new extents and calculate the new rsv size. */ | |
9f3db423 | 6078 | spin_lock(&inode->lock); |
69fe2d75 | 6079 | nr_extents = count_max_extents(num_bytes); |
8b62f87b | 6080 | btrfs_mod_outstanding_extents(inode, nr_extents); |
69fe2d75 JB |
6081 | inode->csum_bytes += num_bytes; |
6082 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 6083 | spin_unlock(&inode->lock); |
57a45ced | 6084 | |
69fe2d75 | 6085 | ret = btrfs_inode_rsv_refill(inode, flush); |
43b18595 | 6086 | if (unlikely(ret)) |
88e081bf | 6087 | goto out_fail; |
25179201 | 6088 | |
c64c2bd8 | 6089 | if (delalloc_lock) |
9f3db423 | 6090 | mutex_unlock(&inode->delalloc_mutex); |
0ca1f7ce | 6091 | return 0; |
88e081bf WS |
6092 | |
6093 | out_fail: | |
9f3db423 | 6094 | spin_lock(&inode->lock); |
8b62f87b JB |
6095 | nr_extents = count_max_extents(num_bytes); |
6096 | btrfs_mod_outstanding_extents(inode, -nr_extents); | |
69fe2d75 JB |
6097 | inode->csum_bytes -= num_bytes; |
6098 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 6099 | spin_unlock(&inode->lock); |
88e081bf | 6100 | |
43b18595 | 6101 | btrfs_inode_rsv_release(inode, true); |
88e081bf | 6102 | if (delalloc_lock) |
9f3db423 | 6103 | mutex_unlock(&inode->delalloc_mutex); |
88e081bf | 6104 | return ret; |
0ca1f7ce YZ |
6105 | } |
6106 | ||
7709cde3 JB |
6107 | /** |
6108 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
8b62f87b JB |
6109 | * @inode: the inode to release the reservation for. |
6110 | * @num_bytes: the number of bytes we are releasing. | |
43b18595 | 6111 | * @qgroup_free: free qgroup reservation or convert it to per-trans reservation |
7709cde3 JB |
6112 | * |
6113 | * This will release the metadata reservation for an inode. This can be called | |
6114 | * once we complete IO for a given set of bytes to release their metadata | |
8b62f87b | 6115 | * reservations, or on error for the same reason. |
7709cde3 | 6116 | */ |
43b18595 QW |
6117 | void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes, |
6118 | bool qgroup_free) | |
0ca1f7ce | 6119 | { |
691fa059 | 6120 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
0ca1f7ce | 6121 | |
0b246afa | 6122 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
691fa059 | 6123 | spin_lock(&inode->lock); |
69fe2d75 JB |
6124 | inode->csum_bytes -= num_bytes; |
6125 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
691fa059 | 6126 | spin_unlock(&inode->lock); |
0ca1f7ce | 6127 | |
0b246afa | 6128 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
6129 | return; |
6130 | ||
43b18595 | 6131 | btrfs_inode_rsv_release(inode, qgroup_free); |
0ca1f7ce YZ |
6132 | } |
6133 | ||
8b62f87b JB |
6134 | /** |
6135 | * btrfs_delalloc_release_extents - release our outstanding_extents | |
6136 | * @inode: the inode to balance the reservation for. | |
6137 | * @num_bytes: the number of bytes we originally reserved with | |
43b18595 | 6138 | * @qgroup_free: do we need to free qgroup meta reservation or convert them. |
8b62f87b JB |
6139 | * |
6140 | * When we reserve space we increase outstanding_extents for the extents we may | |
6141 | * add. Once we've set the range as delalloc or created our ordered extents we | |
6142 | * have outstanding_extents to track the real usage, so we use this to free our | |
6143 | * temporarily tracked outstanding_extents. This _must_ be used in conjunction | |
6144 | * with btrfs_delalloc_reserve_metadata. | |
6145 | */ | |
43b18595 QW |
6146 | void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes, |
6147 | bool qgroup_free) | |
8b62f87b JB |
6148 | { |
6149 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); | |
6150 | unsigned num_extents; | |
8b62f87b JB |
6151 | |
6152 | spin_lock(&inode->lock); | |
6153 | num_extents = count_max_extents(num_bytes); | |
6154 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
69fe2d75 | 6155 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); |
8b62f87b JB |
6156 | spin_unlock(&inode->lock); |
6157 | ||
8b62f87b JB |
6158 | if (btrfs_is_testing(fs_info)) |
6159 | return; | |
6160 | ||
43b18595 | 6161 | btrfs_inode_rsv_release(inode, qgroup_free); |
8b62f87b JB |
6162 | } |
6163 | ||
1ada3a62 | 6164 | /** |
7cf5b976 | 6165 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
6166 | * delalloc |
6167 | * @inode: inode we're writing to | |
6168 | * @start: start range we are writing to | |
6169 | * @len: how long the range we are writing to | |
364ecf36 QW |
6170 | * @reserved: mandatory parameter, record actually reserved qgroup ranges of |
6171 | * current reservation. | |
1ada3a62 | 6172 | * |
1ada3a62 QW |
6173 | * This will do the following things |
6174 | * | |
6175 | * o reserve space in data space info for num bytes | |
6176 | * and reserve precious corresponding qgroup space | |
6177 | * (Done in check_data_free_space) | |
6178 | * | |
6179 | * o reserve space for metadata space, based on the number of outstanding | |
6180 | * extents and how much csums will be needed | |
6181 | * also reserve metadata space in a per root over-reserve method. | |
6182 | * o add to the inodes->delalloc_bytes | |
6183 | * o add it to the fs_info's delalloc inodes list. | |
6184 | * (Above 3 all done in delalloc_reserve_metadata) | |
6185 | * | |
6186 | * Return 0 for success | |
6187 | * Return <0 for error(-ENOSPC or -EQUOT) | |
6188 | */ | |
364ecf36 QW |
6189 | int btrfs_delalloc_reserve_space(struct inode *inode, |
6190 | struct extent_changeset **reserved, u64 start, u64 len) | |
1ada3a62 QW |
6191 | { |
6192 | int ret; | |
6193 | ||
364ecf36 | 6194 | ret = btrfs_check_data_free_space(inode, reserved, start, len); |
1ada3a62 QW |
6195 | if (ret < 0) |
6196 | return ret; | |
9f3db423 | 6197 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len); |
1ada3a62 | 6198 | if (ret < 0) |
bc42bda2 | 6199 | btrfs_free_reserved_data_space(inode, *reserved, start, len); |
1ada3a62 QW |
6200 | return ret; |
6201 | } | |
6202 | ||
7709cde3 | 6203 | /** |
7cf5b976 | 6204 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
6205 | * @inode: inode we're releasing space for |
6206 | * @start: start position of the space already reserved | |
6207 | * @len: the len of the space already reserved | |
8b62f87b | 6208 | * @release_bytes: the len of the space we consumed or didn't use |
1ada3a62 QW |
6209 | * |
6210 | * This function will release the metadata space that was not used and will | |
6211 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
6212 | * list if there are no delalloc bytes left. | |
6213 | * Also it will handle the qgroup reserved space. | |
6214 | */ | |
bc42bda2 | 6215 | void btrfs_delalloc_release_space(struct inode *inode, |
8b62f87b | 6216 | struct extent_changeset *reserved, |
43b18595 | 6217 | u64 start, u64 len, bool qgroup_free) |
1ada3a62 | 6218 | { |
43b18595 | 6219 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free); |
bc42bda2 | 6220 | btrfs_free_reserved_data_space(inode, reserved, start, len); |
6324fbf3 CM |
6221 | } |
6222 | ||
ce93ec54 | 6223 | static int update_block_group(struct btrfs_trans_handle *trans, |
6202df69 | 6224 | struct btrfs_fs_info *info, u64 bytenr, |
ce93ec54 | 6225 | u64 num_bytes, int alloc) |
9078a3e1 | 6226 | { |
0af3d00b | 6227 | struct btrfs_block_group_cache *cache = NULL; |
db94535d | 6228 | u64 total = num_bytes; |
9078a3e1 | 6229 | u64 old_val; |
db94535d | 6230 | u64 byte_in_group; |
0af3d00b | 6231 | int factor; |
3e1ad54f | 6232 | |
5d4f98a2 | 6233 | /* block accounting for super block */ |
eb73c1b7 | 6234 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 6235 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
6236 | if (alloc) |
6237 | old_val += num_bytes; | |
6238 | else | |
6239 | old_val -= num_bytes; | |
6c41761f | 6240 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 6241 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 6242 | |
d397712b | 6243 | while (total) { |
db94535d | 6244 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 6245 | if (!cache) |
79787eaa | 6246 | return -ENOENT; |
b742bb82 YZ |
6247 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
6248 | BTRFS_BLOCK_GROUP_RAID1 | | |
6249 | BTRFS_BLOCK_GROUP_RAID10)) | |
6250 | factor = 2; | |
6251 | else | |
6252 | factor = 1; | |
9d66e233 JB |
6253 | /* |
6254 | * If this block group has free space cache written out, we | |
6255 | * need to make sure to load it if we are removing space. This | |
6256 | * is because we need the unpinning stage to actually add the | |
6257 | * space back to the block group, otherwise we will leak space. | |
6258 | */ | |
6259 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 6260 | cache_block_group(cache, 1); |
0af3d00b | 6261 | |
db94535d CM |
6262 | byte_in_group = bytenr - cache->key.objectid; |
6263 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 6264 | |
25179201 | 6265 | spin_lock(&cache->space_info->lock); |
c286ac48 | 6266 | spin_lock(&cache->lock); |
0af3d00b | 6267 | |
6202df69 | 6268 | if (btrfs_test_opt(info, SPACE_CACHE) && |
0af3d00b JB |
6269 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
6270 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
6271 | ||
9078a3e1 | 6272 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 6273 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 6274 | if (alloc) { |
db94535d | 6275 | old_val += num_bytes; |
11833d66 YZ |
6276 | btrfs_set_block_group_used(&cache->item, old_val); |
6277 | cache->reserved -= num_bytes; | |
11833d66 | 6278 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
6279 | cache->space_info->bytes_used += num_bytes; |
6280 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 6281 | spin_unlock(&cache->lock); |
25179201 | 6282 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 6283 | } else { |
db94535d | 6284 | old_val -= num_bytes; |
ae0ab003 FM |
6285 | btrfs_set_block_group_used(&cache->item, old_val); |
6286 | cache->pinned += num_bytes; | |
6287 | cache->space_info->bytes_pinned += num_bytes; | |
6288 | cache->space_info->bytes_used -= num_bytes; | |
6289 | cache->space_info->disk_used -= num_bytes * factor; | |
6290 | spin_unlock(&cache->lock); | |
6291 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 6292 | |
0b246afa | 6293 | trace_btrfs_space_reservation(info, "pinned", |
c51e7bb1 JB |
6294 | cache->space_info->flags, |
6295 | num_bytes, 1); | |
d7eae340 OS |
6296 | percpu_counter_add(&cache->space_info->total_bytes_pinned, |
6297 | num_bytes); | |
ae0ab003 FM |
6298 | set_extent_dirty(info->pinned_extents, |
6299 | bytenr, bytenr + num_bytes - 1, | |
6300 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 6301 | } |
1bbc621e CM |
6302 | |
6303 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
6304 | if (list_empty(&cache->dirty_list)) { | |
6305 | list_add_tail(&cache->dirty_list, | |
6306 | &trans->transaction->dirty_bgs); | |
6307 | trans->transaction->num_dirty_bgs++; | |
6308 | btrfs_get_block_group(cache); | |
6309 | } | |
6310 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
6311 | ||
036a9348 FM |
6312 | /* |
6313 | * No longer have used bytes in this block group, queue it for | |
6314 | * deletion. We do this after adding the block group to the | |
6315 | * dirty list to avoid races between cleaner kthread and space | |
6316 | * cache writeout. | |
6317 | */ | |
6318 | if (!alloc && old_val == 0) { | |
6319 | spin_lock(&info->unused_bgs_lock); | |
6320 | if (list_empty(&cache->bg_list)) { | |
6321 | btrfs_get_block_group(cache); | |
6322 | list_add_tail(&cache->bg_list, | |
6323 | &info->unused_bgs); | |
6324 | } | |
6325 | spin_unlock(&info->unused_bgs_lock); | |
6326 | } | |
6327 | ||
fa9c0d79 | 6328 | btrfs_put_block_group(cache); |
db94535d CM |
6329 | total -= num_bytes; |
6330 | bytenr += num_bytes; | |
9078a3e1 CM |
6331 | } |
6332 | return 0; | |
6333 | } | |
6324fbf3 | 6334 | |
2ff7e61e | 6335 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 6336 | { |
0f9dd46c | 6337 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 6338 | u64 bytenr; |
0f9dd46c | 6339 | |
0b246afa JM |
6340 | spin_lock(&fs_info->block_group_cache_lock); |
6341 | bytenr = fs_info->first_logical_byte; | |
6342 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
6343 | |
6344 | if (bytenr < (u64)-1) | |
6345 | return bytenr; | |
6346 | ||
0b246afa | 6347 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 6348 | if (!cache) |
a061fc8d | 6349 | return 0; |
0f9dd46c | 6350 | |
d2fb3437 | 6351 | bytenr = cache->key.objectid; |
fa9c0d79 | 6352 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
6353 | |
6354 | return bytenr; | |
a061fc8d CM |
6355 | } |
6356 | ||
2ff7e61e | 6357 | static int pin_down_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6358 | struct btrfs_block_group_cache *cache, |
6359 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 6360 | { |
11833d66 YZ |
6361 | spin_lock(&cache->space_info->lock); |
6362 | spin_lock(&cache->lock); | |
6363 | cache->pinned += num_bytes; | |
6364 | cache->space_info->bytes_pinned += num_bytes; | |
6365 | if (reserved) { | |
6366 | cache->reserved -= num_bytes; | |
6367 | cache->space_info->bytes_reserved -= num_bytes; | |
6368 | } | |
6369 | spin_unlock(&cache->lock); | |
6370 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 6371 | |
0b246afa | 6372 | trace_btrfs_space_reservation(fs_info, "pinned", |
c51e7bb1 | 6373 | cache->space_info->flags, num_bytes, 1); |
4da8b76d | 6374 | percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes); |
0b246afa | 6375 | set_extent_dirty(fs_info->pinned_extents, bytenr, |
f0486c68 YZ |
6376 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
6377 | return 0; | |
6378 | } | |
68b38550 | 6379 | |
f0486c68 YZ |
6380 | /* |
6381 | * this function must be called within transaction | |
6382 | */ | |
2ff7e61e | 6383 | int btrfs_pin_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6384 | u64 bytenr, u64 num_bytes, int reserved) |
6385 | { | |
6386 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6387 | |
0b246afa | 6388 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
79787eaa | 6389 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 6390 | |
2ff7e61e | 6391 | pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
6392 | |
6393 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6394 | return 0; |
6395 | } | |
6396 | ||
f0486c68 | 6397 | /* |
e688b725 CM |
6398 | * this function must be called within transaction |
6399 | */ | |
2ff7e61e | 6400 | int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info, |
e688b725 CM |
6401 | u64 bytenr, u64 num_bytes) |
6402 | { | |
6403 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6404 | int ret; |
e688b725 | 6405 | |
0b246afa | 6406 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
b50c6e25 JB |
6407 | if (!cache) |
6408 | return -EINVAL; | |
e688b725 CM |
6409 | |
6410 | /* | |
6411 | * pull in the free space cache (if any) so that our pin | |
6412 | * removes the free space from the cache. We have load_only set | |
6413 | * to one because the slow code to read in the free extents does check | |
6414 | * the pinned extents. | |
6415 | */ | |
f6373bf3 | 6416 | cache_block_group(cache, 1); |
e688b725 | 6417 | |
2ff7e61e | 6418 | pin_down_extent(fs_info, cache, bytenr, num_bytes, 0); |
e688b725 CM |
6419 | |
6420 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6421 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6422 | btrfs_put_block_group(cache); |
b50c6e25 | 6423 | return ret; |
e688b725 CM |
6424 | } |
6425 | ||
2ff7e61e JM |
6426 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
6427 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
6428 | { |
6429 | int ret; | |
6430 | struct btrfs_block_group_cache *block_group; | |
6431 | struct btrfs_caching_control *caching_ctl; | |
6432 | ||
0b246afa | 6433 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
6434 | if (!block_group) |
6435 | return -EINVAL; | |
6436 | ||
6437 | cache_block_group(block_group, 0); | |
6438 | caching_ctl = get_caching_control(block_group); | |
6439 | ||
6440 | if (!caching_ctl) { | |
6441 | /* Logic error */ | |
6442 | BUG_ON(!block_group_cache_done(block_group)); | |
6443 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6444 | } else { | |
6445 | mutex_lock(&caching_ctl->mutex); | |
6446 | ||
6447 | if (start >= caching_ctl->progress) { | |
2ff7e61e | 6448 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6449 | } else if (start + num_bytes <= caching_ctl->progress) { |
6450 | ret = btrfs_remove_free_space(block_group, | |
6451 | start, num_bytes); | |
6452 | } else { | |
6453 | num_bytes = caching_ctl->progress - start; | |
6454 | ret = btrfs_remove_free_space(block_group, | |
6455 | start, num_bytes); | |
6456 | if (ret) | |
6457 | goto out_lock; | |
6458 | ||
6459 | num_bytes = (start + num_bytes) - | |
6460 | caching_ctl->progress; | |
6461 | start = caching_ctl->progress; | |
2ff7e61e | 6462 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6463 | } |
6464 | out_lock: | |
6465 | mutex_unlock(&caching_ctl->mutex); | |
6466 | put_caching_control(caching_ctl); | |
6467 | } | |
6468 | btrfs_put_block_group(block_group); | |
6469 | return ret; | |
6470 | } | |
6471 | ||
2ff7e61e | 6472 | int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info, |
8c2a1a30 JB |
6473 | struct extent_buffer *eb) |
6474 | { | |
6475 | struct btrfs_file_extent_item *item; | |
6476 | struct btrfs_key key; | |
6477 | int found_type; | |
6478 | int i; | |
6479 | ||
2ff7e61e | 6480 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
6481 | return 0; |
6482 | ||
6483 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6484 | btrfs_item_key_to_cpu(eb, &key, i); | |
6485 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6486 | continue; | |
6487 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6488 | found_type = btrfs_file_extent_type(eb, item); | |
6489 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6490 | continue; | |
6491 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6492 | continue; | |
6493 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6494 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
2ff7e61e | 6495 | __exclude_logged_extent(fs_info, key.objectid, key.offset); |
8c2a1a30 JB |
6496 | } |
6497 | ||
6498 | return 0; | |
6499 | } | |
6500 | ||
9cfa3e34 FM |
6501 | static void |
6502 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6503 | { | |
6504 | atomic_inc(&bg->reservations); | |
6505 | } | |
6506 | ||
6507 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
6508 | const u64 start) | |
6509 | { | |
6510 | struct btrfs_block_group_cache *bg; | |
6511 | ||
6512 | bg = btrfs_lookup_block_group(fs_info, start); | |
6513 | ASSERT(bg); | |
6514 | if (atomic_dec_and_test(&bg->reservations)) | |
6515 | wake_up_atomic_t(&bg->reservations); | |
6516 | btrfs_put_block_group(bg); | |
6517 | } | |
6518 | ||
9cfa3e34 FM |
6519 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) |
6520 | { | |
6521 | struct btrfs_space_info *space_info = bg->space_info; | |
6522 | ||
6523 | ASSERT(bg->ro); | |
6524 | ||
6525 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
6526 | return; | |
6527 | ||
6528 | /* | |
6529 | * Our block group is read only but before we set it to read only, | |
6530 | * some task might have had allocated an extent from it already, but it | |
6531 | * has not yet created a respective ordered extent (and added it to a | |
6532 | * root's list of ordered extents). | |
6533 | * Therefore wait for any task currently allocating extents, since the | |
6534 | * block group's reservations counter is incremented while a read lock | |
6535 | * on the groups' semaphore is held and decremented after releasing | |
6536 | * the read access on that semaphore and creating the ordered extent. | |
6537 | */ | |
6538 | down_write(&space_info->groups_sem); | |
6539 | up_write(&space_info->groups_sem); | |
6540 | ||
5e4def20 | 6541 | wait_on_atomic_t(&bg->reservations, atomic_t_wait, |
9cfa3e34 FM |
6542 | TASK_UNINTERRUPTIBLE); |
6543 | } | |
6544 | ||
fb25e914 | 6545 | /** |
4824f1f4 | 6546 | * btrfs_add_reserved_bytes - update the block_group and space info counters |
fb25e914 | 6547 | * @cache: The cache we are manipulating |
18513091 WX |
6548 | * @ram_bytes: The number of bytes of file content, and will be same to |
6549 | * @num_bytes except for the compress path. | |
fb25e914 | 6550 | * @num_bytes: The number of bytes in question |
e570fd27 | 6551 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 | 6552 | * |
745699ef XW |
6553 | * This is called by the allocator when it reserves space. If this is a |
6554 | * reservation and the block group has become read only we cannot make the | |
6555 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
f0486c68 | 6556 | */ |
4824f1f4 | 6557 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 6558 | u64 ram_bytes, u64 num_bytes, int delalloc) |
11833d66 | 6559 | { |
fb25e914 | 6560 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6561 | int ret = 0; |
79787eaa | 6562 | |
fb25e914 JB |
6563 | spin_lock(&space_info->lock); |
6564 | spin_lock(&cache->lock); | |
4824f1f4 WX |
6565 | if (cache->ro) { |
6566 | ret = -EAGAIN; | |
fb25e914 | 6567 | } else { |
4824f1f4 WX |
6568 | cache->reserved += num_bytes; |
6569 | space_info->bytes_reserved += num_bytes; | |
e570fd27 | 6570 | |
18513091 WX |
6571 | trace_btrfs_space_reservation(cache->fs_info, |
6572 | "space_info", space_info->flags, | |
6573 | ram_bytes, 0); | |
6574 | space_info->bytes_may_use -= ram_bytes; | |
e570fd27 | 6575 | if (delalloc) |
4824f1f4 | 6576 | cache->delalloc_bytes += num_bytes; |
324ae4df | 6577 | } |
fb25e914 JB |
6578 | spin_unlock(&cache->lock); |
6579 | spin_unlock(&space_info->lock); | |
f0486c68 | 6580 | return ret; |
324ae4df | 6581 | } |
9078a3e1 | 6582 | |
4824f1f4 WX |
6583 | /** |
6584 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
6585 | * @cache: The cache we are manipulating | |
6586 | * @num_bytes: The number of bytes in question | |
6587 | * @delalloc: The blocks are allocated for the delalloc write | |
6588 | * | |
6589 | * This is called by somebody who is freeing space that was never actually used | |
6590 | * on disk. For example if you reserve some space for a new leaf in transaction | |
6591 | * A and before transaction A commits you free that leaf, you call this with | |
6592 | * reserve set to 0 in order to clear the reservation. | |
6593 | */ | |
6594 | ||
6595 | static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, | |
6596 | u64 num_bytes, int delalloc) | |
6597 | { | |
6598 | struct btrfs_space_info *space_info = cache->space_info; | |
6599 | int ret = 0; | |
6600 | ||
6601 | spin_lock(&space_info->lock); | |
6602 | spin_lock(&cache->lock); | |
6603 | if (cache->ro) | |
6604 | space_info->bytes_readonly += num_bytes; | |
6605 | cache->reserved -= num_bytes; | |
6606 | space_info->bytes_reserved -= num_bytes; | |
6607 | ||
6608 | if (delalloc) | |
6609 | cache->delalloc_bytes -= num_bytes; | |
6610 | spin_unlock(&cache->lock); | |
6611 | spin_unlock(&space_info->lock); | |
6612 | return ret; | |
6613 | } | |
8b74c03e | 6614 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 6615 | { |
11833d66 YZ |
6616 | struct btrfs_caching_control *next; |
6617 | struct btrfs_caching_control *caching_ctl; | |
6618 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6619 | |
9e351cc8 | 6620 | down_write(&fs_info->commit_root_sem); |
25179201 | 6621 | |
11833d66 YZ |
6622 | list_for_each_entry_safe(caching_ctl, next, |
6623 | &fs_info->caching_block_groups, list) { | |
6624 | cache = caching_ctl->block_group; | |
6625 | if (block_group_cache_done(cache)) { | |
6626 | cache->last_byte_to_unpin = (u64)-1; | |
6627 | list_del_init(&caching_ctl->list); | |
6628 | put_caching_control(caching_ctl); | |
e8569813 | 6629 | } else { |
11833d66 | 6630 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6631 | } |
e8569813 | 6632 | } |
11833d66 YZ |
6633 | |
6634 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6635 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6636 | else | |
6637 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6638 | ||
9e351cc8 | 6639 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6640 | |
6641 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6642 | } |
6643 | ||
c759c4e1 JB |
6644 | /* |
6645 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6646 | * what it should be based on the mount options. | |
6647 | */ | |
6648 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
6649 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
6650 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
6651 | { |
6652 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
6653 | |
6654 | *empty_cluster = 0; | |
6655 | if (btrfs_mixed_space_info(space_info)) | |
6656 | return ret; | |
6657 | ||
c759c4e1 | 6658 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 6659 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
6660 | if (btrfs_test_opt(fs_info, SSD)) |
6661 | *empty_cluster = SZ_2M; | |
6662 | else | |
ee22184b | 6663 | *empty_cluster = SZ_64K; |
583b7231 HK |
6664 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
6665 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
6666 | *empty_cluster = SZ_2M; | |
0b246afa | 6667 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
6668 | } |
6669 | ||
6670 | return ret; | |
6671 | } | |
6672 | ||
2ff7e61e JM |
6673 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
6674 | u64 start, u64 end, | |
678886bd | 6675 | const bool return_free_space) |
ccd467d6 | 6676 | { |
11833d66 | 6677 | struct btrfs_block_group_cache *cache = NULL; |
7b398f8e JB |
6678 | struct btrfs_space_info *space_info; |
6679 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6680 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6681 | u64 len; |
c759c4e1 JB |
6682 | u64 total_unpinned = 0; |
6683 | u64 empty_cluster = 0; | |
7b398f8e | 6684 | bool readonly; |
ccd467d6 | 6685 | |
11833d66 | 6686 | while (start <= end) { |
7b398f8e | 6687 | readonly = false; |
11833d66 YZ |
6688 | if (!cache || |
6689 | start >= cache->key.objectid + cache->key.offset) { | |
6690 | if (cache) | |
6691 | btrfs_put_block_group(cache); | |
c759c4e1 | 6692 | total_unpinned = 0; |
11833d66 | 6693 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6694 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 6695 | |
2ff7e61e | 6696 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
6697 | cache->space_info, |
6698 | &empty_cluster); | |
6699 | empty_cluster <<= 1; | |
11833d66 YZ |
6700 | } |
6701 | ||
6702 | len = cache->key.objectid + cache->key.offset - start; | |
6703 | len = min(len, end + 1 - start); | |
6704 | ||
6705 | if (start < cache->last_byte_to_unpin) { | |
6706 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6707 | if (return_free_space) |
6708 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6709 | } |
6710 | ||
f0486c68 | 6711 | start += len; |
c759c4e1 | 6712 | total_unpinned += len; |
7b398f8e | 6713 | space_info = cache->space_info; |
f0486c68 | 6714 | |
c759c4e1 JB |
6715 | /* |
6716 | * If this space cluster has been marked as fragmented and we've | |
6717 | * unpinned enough in this block group to potentially allow a | |
6718 | * cluster to be created inside of it go ahead and clear the | |
6719 | * fragmented check. | |
6720 | */ | |
6721 | if (cluster && cluster->fragmented && | |
6722 | total_unpinned > empty_cluster) { | |
6723 | spin_lock(&cluster->lock); | |
6724 | cluster->fragmented = 0; | |
6725 | spin_unlock(&cluster->lock); | |
6726 | } | |
6727 | ||
7b398f8e | 6728 | spin_lock(&space_info->lock); |
11833d66 YZ |
6729 | spin_lock(&cache->lock); |
6730 | cache->pinned -= len; | |
7b398f8e | 6731 | space_info->bytes_pinned -= len; |
c51e7bb1 JB |
6732 | |
6733 | trace_btrfs_space_reservation(fs_info, "pinned", | |
6734 | space_info->flags, len, 0); | |
4f4db217 | 6735 | space_info->max_extent_size = 0; |
d288db5d | 6736 | percpu_counter_add(&space_info->total_bytes_pinned, -len); |
7b398f8e JB |
6737 | if (cache->ro) { |
6738 | space_info->bytes_readonly += len; | |
6739 | readonly = true; | |
6740 | } | |
11833d66 | 6741 | spin_unlock(&cache->lock); |
957780eb JB |
6742 | if (!readonly && return_free_space && |
6743 | global_rsv->space_info == space_info) { | |
6744 | u64 to_add = len; | |
92ac58ec | 6745 | |
7b398f8e JB |
6746 | spin_lock(&global_rsv->lock); |
6747 | if (!global_rsv->full) { | |
957780eb JB |
6748 | to_add = min(len, global_rsv->size - |
6749 | global_rsv->reserved); | |
6750 | global_rsv->reserved += to_add; | |
6751 | space_info->bytes_may_use += to_add; | |
7b398f8e JB |
6752 | if (global_rsv->reserved >= global_rsv->size) |
6753 | global_rsv->full = 1; | |
957780eb JB |
6754 | trace_btrfs_space_reservation(fs_info, |
6755 | "space_info", | |
6756 | space_info->flags, | |
6757 | to_add, 1); | |
6758 | len -= to_add; | |
7b398f8e JB |
6759 | } |
6760 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
6761 | /* Add to any tickets we may have */ |
6762 | if (len) | |
6763 | space_info_add_new_bytes(fs_info, space_info, | |
6764 | len); | |
7b398f8e JB |
6765 | } |
6766 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6767 | } |
11833d66 YZ |
6768 | |
6769 | if (cache) | |
6770 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6771 | return 0; |
6772 | } | |
6773 | ||
5ead2dd0 | 6774 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 6775 | { |
5ead2dd0 | 6776 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e33e17ee JM |
6777 | struct btrfs_block_group_cache *block_group, *tmp; |
6778 | struct list_head *deleted_bgs; | |
11833d66 | 6779 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6780 | u64 start; |
6781 | u64 end; | |
a28ec197 | 6782 | int ret; |
a28ec197 | 6783 | |
11833d66 YZ |
6784 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6785 | unpin = &fs_info->freed_extents[1]; | |
6786 | else | |
6787 | unpin = &fs_info->freed_extents[0]; | |
6788 | ||
e33e17ee | 6789 | while (!trans->aborted) { |
d4b450cd | 6790 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6791 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 6792 | EXTENT_DIRTY, NULL); |
d4b450cd FM |
6793 | if (ret) { |
6794 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6795 | break; |
d4b450cd | 6796 | } |
1f3c79a2 | 6797 | |
0b246afa | 6798 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 6799 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 6800 | end + 1 - start, NULL); |
1f3c79a2 | 6801 | |
af6f8f60 | 6802 | clear_extent_dirty(unpin, start, end); |
2ff7e61e | 6803 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 6804 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
b9473439 | 6805 | cond_resched(); |
a28ec197 | 6806 | } |
817d52f8 | 6807 | |
e33e17ee JM |
6808 | /* |
6809 | * Transaction is finished. We don't need the lock anymore. We | |
6810 | * do need to clean up the block groups in case of a transaction | |
6811 | * abort. | |
6812 | */ | |
6813 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6814 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6815 | u64 trimmed = 0; | |
6816 | ||
6817 | ret = -EROFS; | |
6818 | if (!trans->aborted) | |
2ff7e61e | 6819 | ret = btrfs_discard_extent(fs_info, |
e33e17ee JM |
6820 | block_group->key.objectid, |
6821 | block_group->key.offset, | |
6822 | &trimmed); | |
6823 | ||
6824 | list_del_init(&block_group->bg_list); | |
6825 | btrfs_put_block_group_trimming(block_group); | |
6826 | btrfs_put_block_group(block_group); | |
6827 | ||
6828 | if (ret) { | |
6829 | const char *errstr = btrfs_decode_error(ret); | |
6830 | btrfs_warn(fs_info, | |
913e1535 | 6831 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
6832 | ret, errstr); |
6833 | } | |
6834 | } | |
6835 | ||
e20d96d6 CM |
6836 | return 0; |
6837 | } | |
6838 | ||
5d4f98a2 | 6839 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 6840 | struct btrfs_fs_info *info, |
c682f9b3 | 6841 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
6842 | u64 root_objectid, u64 owner_objectid, |
6843 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 6844 | struct btrfs_delayed_extent_op *extent_op) |
a28ec197 | 6845 | { |
e2fa7227 | 6846 | struct btrfs_key key; |
5d4f98a2 | 6847 | struct btrfs_path *path; |
1261ec42 | 6848 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 6849 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6850 | struct btrfs_extent_item *ei; |
6851 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6852 | int ret; |
5d4f98a2 | 6853 | int is_data; |
952fccac CM |
6854 | int extent_slot = 0; |
6855 | int found_extent = 0; | |
6856 | int num_to_del = 1; | |
5d4f98a2 YZ |
6857 | u32 item_size; |
6858 | u64 refs; | |
c682f9b3 QW |
6859 | u64 bytenr = node->bytenr; |
6860 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6861 | int last_ref = 0; |
0b246afa | 6862 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 6863 | |
5caf2a00 | 6864 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6865 | if (!path) |
6866 | return -ENOMEM; | |
5f26f772 | 6867 | |
e4058b54 | 6868 | path->reada = READA_FORWARD; |
b9473439 | 6869 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6870 | |
6871 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6872 | BUG_ON(!is_data && refs_to_drop != 1); | |
6873 | ||
3173a18f | 6874 | if (is_data) |
897ca819 | 6875 | skinny_metadata = false; |
3173a18f | 6876 | |
87bde3cd | 6877 | ret = lookup_extent_backref(trans, info, path, &iref, |
5d4f98a2 YZ |
6878 | bytenr, num_bytes, parent, |
6879 | root_objectid, owner_objectid, | |
6880 | owner_offset); | |
7bb86316 | 6881 | if (ret == 0) { |
952fccac | 6882 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6883 | while (extent_slot >= 0) { |
6884 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6885 | extent_slot); |
5d4f98a2 | 6886 | if (key.objectid != bytenr) |
952fccac | 6887 | break; |
5d4f98a2 YZ |
6888 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6889 | key.offset == num_bytes) { | |
952fccac CM |
6890 | found_extent = 1; |
6891 | break; | |
6892 | } | |
3173a18f JB |
6893 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
6894 | key.offset == owner_objectid) { | |
6895 | found_extent = 1; | |
6896 | break; | |
6897 | } | |
952fccac CM |
6898 | if (path->slots[0] - extent_slot > 5) |
6899 | break; | |
5d4f98a2 | 6900 | extent_slot--; |
952fccac | 6901 | } |
5d4f98a2 YZ |
6902 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
6903 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
6904 | if (found_extent && item_size < sizeof(*ei)) | |
6905 | found_extent = 0; | |
6906 | #endif | |
31840ae1 | 6907 | if (!found_extent) { |
5d4f98a2 | 6908 | BUG_ON(iref); |
87bde3cd JM |
6909 | ret = remove_extent_backref(trans, info, path, NULL, |
6910 | refs_to_drop, | |
fcebe456 | 6911 | is_data, &last_ref); |
005d6427 | 6912 | if (ret) { |
66642832 | 6913 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6914 | goto out; |
6915 | } | |
b3b4aa74 | 6916 | btrfs_release_path(path); |
b9473439 | 6917 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6918 | |
6919 | key.objectid = bytenr; | |
6920 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6921 | key.offset = num_bytes; | |
6922 | ||
3173a18f JB |
6923 | if (!is_data && skinny_metadata) { |
6924 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6925 | key.offset = owner_objectid; | |
6926 | } | |
6927 | ||
31840ae1 ZY |
6928 | ret = btrfs_search_slot(trans, extent_root, |
6929 | &key, path, -1, 1); | |
3173a18f JB |
6930 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
6931 | /* | |
6932 | * Couldn't find our skinny metadata item, | |
6933 | * see if we have ye olde extent item. | |
6934 | */ | |
6935 | path->slots[0]--; | |
6936 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6937 | path->slots[0]); | |
6938 | if (key.objectid == bytenr && | |
6939 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6940 | key.offset == num_bytes) | |
6941 | ret = 0; | |
6942 | } | |
6943 | ||
6944 | if (ret > 0 && skinny_metadata) { | |
6945 | skinny_metadata = false; | |
9ce49a0b | 6946 | key.objectid = bytenr; |
3173a18f JB |
6947 | key.type = BTRFS_EXTENT_ITEM_KEY; |
6948 | key.offset = num_bytes; | |
6949 | btrfs_release_path(path); | |
6950 | ret = btrfs_search_slot(trans, extent_root, | |
6951 | &key, path, -1, 1); | |
6952 | } | |
6953 | ||
f3465ca4 | 6954 | if (ret) { |
5d163e0e JM |
6955 | btrfs_err(info, |
6956 | "umm, got %d back from search, was looking for %llu", | |
6957 | ret, bytenr); | |
b783e62d | 6958 | if (ret > 0) |
a4f78750 | 6959 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 6960 | } |
005d6427 | 6961 | if (ret < 0) { |
66642832 | 6962 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6963 | goto out; |
6964 | } | |
31840ae1 ZY |
6965 | extent_slot = path->slots[0]; |
6966 | } | |
fae7f21c | 6967 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 6968 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
6969 | btrfs_err(info, |
6970 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
6971 | bytenr, parent, root_objectid, owner_objectid, |
6972 | owner_offset); | |
66642832 | 6973 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 6974 | goto out; |
79787eaa | 6975 | } else { |
66642832 | 6976 | btrfs_abort_transaction(trans, ret); |
005d6427 | 6977 | goto out; |
7bb86316 | 6978 | } |
5f39d397 CM |
6979 | |
6980 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
6981 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6982 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
6983 | if (item_size < sizeof(*ei)) { | |
6984 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
87bde3cd JM |
6985 | ret = convert_extent_item_v0(trans, info, path, owner_objectid, |
6986 | 0); | |
005d6427 | 6987 | if (ret < 0) { |
66642832 | 6988 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6989 | goto out; |
6990 | } | |
5d4f98a2 | 6991 | |
b3b4aa74 | 6992 | btrfs_release_path(path); |
5d4f98a2 YZ |
6993 | path->leave_spinning = 1; |
6994 | ||
6995 | key.objectid = bytenr; | |
6996 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6997 | key.offset = num_bytes; | |
6998 | ||
6999 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
7000 | -1, 1); | |
7001 | if (ret) { | |
5d163e0e JM |
7002 | btrfs_err(info, |
7003 | "umm, got %d back from search, was looking for %llu", | |
c1c9ff7c | 7004 | ret, bytenr); |
a4f78750 | 7005 | btrfs_print_leaf(path->nodes[0]); |
5d4f98a2 | 7006 | } |
005d6427 | 7007 | if (ret < 0) { |
66642832 | 7008 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7009 | goto out; |
7010 | } | |
7011 | ||
5d4f98a2 YZ |
7012 | extent_slot = path->slots[0]; |
7013 | leaf = path->nodes[0]; | |
7014 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
7015 | } | |
7016 | #endif | |
7017 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 7018 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 7019 | struct btrfs_extent_item); |
3173a18f JB |
7020 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
7021 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
7022 | struct btrfs_tree_block_info *bi; |
7023 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
7024 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
7025 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
7026 | } | |
56bec294 | 7027 | |
5d4f98a2 | 7028 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 7029 | if (refs < refs_to_drop) { |
5d163e0e JM |
7030 | btrfs_err(info, |
7031 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
7032 | refs_to_drop, refs, bytenr); | |
32b02538 | 7033 | ret = -EINVAL; |
66642832 | 7034 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
7035 | goto out; |
7036 | } | |
56bec294 | 7037 | refs -= refs_to_drop; |
5f39d397 | 7038 | |
5d4f98a2 YZ |
7039 | if (refs > 0) { |
7040 | if (extent_op) | |
7041 | __run_delayed_extent_op(extent_op, leaf, ei); | |
7042 | /* | |
7043 | * In the case of inline back ref, reference count will | |
7044 | * be updated by remove_extent_backref | |
952fccac | 7045 | */ |
5d4f98a2 YZ |
7046 | if (iref) { |
7047 | BUG_ON(!found_extent); | |
7048 | } else { | |
7049 | btrfs_set_extent_refs(leaf, ei, refs); | |
7050 | btrfs_mark_buffer_dirty(leaf); | |
7051 | } | |
7052 | if (found_extent) { | |
87bde3cd | 7053 | ret = remove_extent_backref(trans, info, path, |
5d4f98a2 | 7054 | iref, refs_to_drop, |
fcebe456 | 7055 | is_data, &last_ref); |
005d6427 | 7056 | if (ret) { |
66642832 | 7057 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7058 | goto out; |
7059 | } | |
952fccac | 7060 | } |
5d4f98a2 | 7061 | } else { |
5d4f98a2 YZ |
7062 | if (found_extent) { |
7063 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 7064 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
7065 | if (iref) { |
7066 | BUG_ON(path->slots[0] != extent_slot); | |
7067 | } else { | |
7068 | BUG_ON(path->slots[0] != extent_slot + 1); | |
7069 | path->slots[0] = extent_slot; | |
7070 | num_to_del = 2; | |
7071 | } | |
78fae27e | 7072 | } |
b9473439 | 7073 | |
fcebe456 | 7074 | last_ref = 1; |
952fccac CM |
7075 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
7076 | num_to_del); | |
005d6427 | 7077 | if (ret) { |
66642832 | 7078 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7079 | goto out; |
7080 | } | |
b3b4aa74 | 7081 | btrfs_release_path(path); |
21af804c | 7082 | |
5d4f98a2 | 7083 | if (is_data) { |
5b4aacef | 7084 | ret = btrfs_del_csums(trans, info, bytenr, num_bytes); |
005d6427 | 7085 | if (ret) { |
66642832 | 7086 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7087 | goto out; |
7088 | } | |
459931ec CM |
7089 | } |
7090 | ||
0b246afa | 7091 | ret = add_to_free_space_tree(trans, info, bytenr, num_bytes); |
1e144fb8 | 7092 | if (ret) { |
66642832 | 7093 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
7094 | goto out; |
7095 | } | |
7096 | ||
0b246afa | 7097 | ret = update_block_group(trans, info, bytenr, num_bytes, 0); |
005d6427 | 7098 | if (ret) { |
66642832 | 7099 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7100 | goto out; |
7101 | } | |
a28ec197 | 7102 | } |
fcebe456 JB |
7103 | btrfs_release_path(path); |
7104 | ||
79787eaa | 7105 | out: |
5caf2a00 | 7106 | btrfs_free_path(path); |
a28ec197 CM |
7107 | return ret; |
7108 | } | |
7109 | ||
1887be66 | 7110 | /* |
f0486c68 | 7111 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
7112 | * delayed ref for that extent as well. This searches the delayed ref tree for |
7113 | * a given extent, and if there are no other delayed refs to be processed, it | |
7114 | * removes it from the tree. | |
7115 | */ | |
7116 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 7117 | u64 bytenr) |
1887be66 CM |
7118 | { |
7119 | struct btrfs_delayed_ref_head *head; | |
7120 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 7121 | int ret = 0; |
1887be66 CM |
7122 | |
7123 | delayed_refs = &trans->transaction->delayed_refs; | |
7124 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 7125 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 7126 | if (!head) |
cf93da7b | 7127 | goto out_delayed_unlock; |
1887be66 | 7128 | |
d7df2c79 | 7129 | spin_lock(&head->lock); |
0e0adbcf | 7130 | if (!RB_EMPTY_ROOT(&head->ref_tree)) |
1887be66 CM |
7131 | goto out; |
7132 | ||
5d4f98a2 YZ |
7133 | if (head->extent_op) { |
7134 | if (!head->must_insert_reserved) | |
7135 | goto out; | |
78a6184a | 7136 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
7137 | head->extent_op = NULL; |
7138 | } | |
7139 | ||
1887be66 CM |
7140 | /* |
7141 | * waiting for the lock here would deadlock. If someone else has it | |
7142 | * locked they are already in the process of dropping it anyway | |
7143 | */ | |
7144 | if (!mutex_trylock(&head->mutex)) | |
7145 | goto out; | |
7146 | ||
7147 | /* | |
7148 | * at this point we have a head with no other entries. Go | |
7149 | * ahead and process it. | |
7150 | */ | |
c46effa6 | 7151 | rb_erase(&head->href_node, &delayed_refs->href_root); |
d278850e | 7152 | RB_CLEAR_NODE(&head->href_node); |
d7df2c79 | 7153 | atomic_dec(&delayed_refs->num_entries); |
1887be66 CM |
7154 | |
7155 | /* | |
7156 | * we don't take a ref on the node because we're removing it from the | |
7157 | * tree, so we just steal the ref the tree was holding. | |
7158 | */ | |
c3e69d58 | 7159 | delayed_refs->num_heads--; |
d7df2c79 | 7160 | if (head->processing == 0) |
c3e69d58 | 7161 | delayed_refs->num_heads_ready--; |
d7df2c79 JB |
7162 | head->processing = 0; |
7163 | spin_unlock(&head->lock); | |
1887be66 CM |
7164 | spin_unlock(&delayed_refs->lock); |
7165 | ||
f0486c68 YZ |
7166 | BUG_ON(head->extent_op); |
7167 | if (head->must_insert_reserved) | |
7168 | ret = 1; | |
7169 | ||
7170 | mutex_unlock(&head->mutex); | |
d278850e | 7171 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 7172 | return ret; |
1887be66 | 7173 | out: |
d7df2c79 | 7174 | spin_unlock(&head->lock); |
cf93da7b CM |
7175 | |
7176 | out_delayed_unlock: | |
1887be66 CM |
7177 | spin_unlock(&delayed_refs->lock); |
7178 | return 0; | |
7179 | } | |
7180 | ||
f0486c68 YZ |
7181 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
7182 | struct btrfs_root *root, | |
7183 | struct extent_buffer *buf, | |
5581a51a | 7184 | u64 parent, int last_ref) |
f0486c68 | 7185 | { |
0b246afa | 7186 | struct btrfs_fs_info *fs_info = root->fs_info; |
b150a4f1 | 7187 | int pin = 1; |
f0486c68 YZ |
7188 | int ret; |
7189 | ||
7190 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
d7eae340 OS |
7191 | int old_ref_mod, new_ref_mod; |
7192 | ||
fd708b81 JB |
7193 | btrfs_ref_tree_mod(root, buf->start, buf->len, parent, |
7194 | root->root_key.objectid, | |
7195 | btrfs_header_level(buf), 0, | |
7196 | BTRFS_DROP_DELAYED_REF); | |
7be07912 OS |
7197 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start, |
7198 | buf->len, parent, | |
0b246afa JM |
7199 | root->root_key.objectid, |
7200 | btrfs_header_level(buf), | |
7be07912 | 7201 | BTRFS_DROP_DELAYED_REF, NULL, |
d7eae340 | 7202 | &old_ref_mod, &new_ref_mod); |
79787eaa | 7203 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 7204 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
7205 | } |
7206 | ||
0a16c7d7 | 7207 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
7208 | struct btrfs_block_group_cache *cache; |
7209 | ||
f0486c68 | 7210 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 7211 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 7212 | if (!ret) |
37be25bc | 7213 | goto out; |
f0486c68 YZ |
7214 | } |
7215 | ||
4da8b76d | 7216 | pin = 0; |
0b246afa | 7217 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 7218 | |
f0486c68 | 7219 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
2ff7e61e JM |
7220 | pin_down_extent(fs_info, cache, buf->start, |
7221 | buf->len, 1); | |
6219872d | 7222 | btrfs_put_block_group(cache); |
37be25bc | 7223 | goto out; |
f0486c68 YZ |
7224 | } |
7225 | ||
7226 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
7227 | ||
7228 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 7229 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 7230 | btrfs_put_block_group(cache); |
71ff6437 | 7231 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
7232 | } |
7233 | out: | |
b150a4f1 | 7234 | if (pin) |
0b246afa | 7235 | add_pinned_bytes(fs_info, buf->len, btrfs_header_level(buf), |
b150a4f1 JB |
7236 | root->root_key.objectid); |
7237 | ||
0a16c7d7 OS |
7238 | if (last_ref) { |
7239 | /* | |
7240 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
7241 | * matter anymore. | |
7242 | */ | |
7243 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
7244 | } | |
f0486c68 YZ |
7245 | } |
7246 | ||
79787eaa | 7247 | /* Can return -ENOMEM */ |
2ff7e61e | 7248 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
84f7d8e6 | 7249 | struct btrfs_root *root, |
66d7e7f0 | 7250 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, |
b06c4bf5 | 7251 | u64 owner, u64 offset) |
925baedd | 7252 | { |
84f7d8e6 | 7253 | struct btrfs_fs_info *fs_info = root->fs_info; |
d7eae340 | 7254 | int old_ref_mod, new_ref_mod; |
925baedd CM |
7255 | int ret; |
7256 | ||
f5ee5c9a | 7257 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 7258 | return 0; |
fccb84c9 | 7259 | |
fd708b81 JB |
7260 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) |
7261 | btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, | |
7262 | root_objectid, owner, offset, | |
7263 | BTRFS_DROP_DELAYED_REF); | |
7264 | ||
56bec294 CM |
7265 | /* |
7266 | * tree log blocks never actually go into the extent allocation | |
7267 | * tree, just update pinning info and exit early. | |
56bec294 | 7268 | */ |
5d4f98a2 YZ |
7269 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
7270 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 7271 | /* unlocks the pinned mutex */ |
2ff7e61e | 7272 | btrfs_pin_extent(fs_info, bytenr, num_bytes, 1); |
d7eae340 | 7273 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 7274 | ret = 0; |
5d4f98a2 | 7275 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 | 7276 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
7be07912 OS |
7277 | num_bytes, parent, |
7278 | root_objectid, (int)owner, | |
7279 | BTRFS_DROP_DELAYED_REF, NULL, | |
d7eae340 | 7280 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 7281 | } else { |
66d7e7f0 | 7282 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
7be07912 OS |
7283 | num_bytes, parent, |
7284 | root_objectid, owner, offset, | |
7285 | 0, BTRFS_DROP_DELAYED_REF, | |
d7eae340 | 7286 | &old_ref_mod, &new_ref_mod); |
56bec294 | 7287 | } |
d7eae340 OS |
7288 | |
7289 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) | |
7290 | add_pinned_bytes(fs_info, num_bytes, owner, root_objectid); | |
7291 | ||
925baedd CM |
7292 | return ret; |
7293 | } | |
7294 | ||
817d52f8 JB |
7295 | /* |
7296 | * when we wait for progress in the block group caching, its because | |
7297 | * our allocation attempt failed at least once. So, we must sleep | |
7298 | * and let some progress happen before we try again. | |
7299 | * | |
7300 | * This function will sleep at least once waiting for new free space to | |
7301 | * show up, and then it will check the block group free space numbers | |
7302 | * for our min num_bytes. Another option is to have it go ahead | |
7303 | * and look in the rbtree for a free extent of a given size, but this | |
7304 | * is a good start. | |
36cce922 JB |
7305 | * |
7306 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
7307 | * any of the information in this block group. | |
817d52f8 | 7308 | */ |
36cce922 | 7309 | static noinline void |
817d52f8 JB |
7310 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
7311 | u64 num_bytes) | |
7312 | { | |
11833d66 | 7313 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 7314 | |
11833d66 YZ |
7315 | caching_ctl = get_caching_control(cache); |
7316 | if (!caching_ctl) | |
36cce922 | 7317 | return; |
817d52f8 | 7318 | |
11833d66 | 7319 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 7320 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
7321 | |
7322 | put_caching_control(caching_ctl); | |
11833d66 YZ |
7323 | } |
7324 | ||
7325 | static noinline int | |
7326 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
7327 | { | |
7328 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 7329 | int ret = 0; |
11833d66 YZ |
7330 | |
7331 | caching_ctl = get_caching_control(cache); | |
7332 | if (!caching_ctl) | |
36cce922 | 7333 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
7334 | |
7335 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
7336 | if (cache->cached == BTRFS_CACHE_ERROR) |
7337 | ret = -EIO; | |
11833d66 | 7338 | put_caching_control(caching_ctl); |
36cce922 | 7339 | return ret; |
817d52f8 JB |
7340 | } |
7341 | ||
6ab0a202 JM |
7342 | static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = { |
7343 | [BTRFS_RAID_RAID10] = "raid10", | |
7344 | [BTRFS_RAID_RAID1] = "raid1", | |
7345 | [BTRFS_RAID_DUP] = "dup", | |
7346 | [BTRFS_RAID_RAID0] = "raid0", | |
7347 | [BTRFS_RAID_SINGLE] = "single", | |
7348 | [BTRFS_RAID_RAID5] = "raid5", | |
7349 | [BTRFS_RAID_RAID6] = "raid6", | |
7350 | }; | |
7351 | ||
1b8e5df6 | 7352 | static const char *get_raid_name(enum btrfs_raid_types type) |
6ab0a202 JM |
7353 | { |
7354 | if (type >= BTRFS_NR_RAID_TYPES) | |
7355 | return NULL; | |
7356 | ||
7357 | return btrfs_raid_type_names[type]; | |
7358 | } | |
7359 | ||
817d52f8 | 7360 | enum btrfs_loop_type { |
285ff5af JB |
7361 | LOOP_CACHING_NOWAIT = 0, |
7362 | LOOP_CACHING_WAIT = 1, | |
7363 | LOOP_ALLOC_CHUNK = 2, | |
7364 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
7365 | }; |
7366 | ||
e570fd27 MX |
7367 | static inline void |
7368 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
7369 | int delalloc) | |
7370 | { | |
7371 | if (delalloc) | |
7372 | down_read(&cache->data_rwsem); | |
7373 | } | |
7374 | ||
7375 | static inline void | |
7376 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
7377 | int delalloc) | |
7378 | { | |
7379 | btrfs_get_block_group(cache); | |
7380 | if (delalloc) | |
7381 | down_read(&cache->data_rwsem); | |
7382 | } | |
7383 | ||
7384 | static struct btrfs_block_group_cache * | |
7385 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
7386 | struct btrfs_free_cluster *cluster, | |
7387 | int delalloc) | |
7388 | { | |
89771cc9 | 7389 | struct btrfs_block_group_cache *used_bg = NULL; |
6719afdc | 7390 | |
e570fd27 | 7391 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
7392 | while (1) { |
7393 | used_bg = cluster->block_group; | |
7394 | if (!used_bg) | |
7395 | return NULL; | |
7396 | ||
7397 | if (used_bg == block_group) | |
e570fd27 MX |
7398 | return used_bg; |
7399 | ||
6719afdc | 7400 | btrfs_get_block_group(used_bg); |
e570fd27 | 7401 | |
6719afdc GU |
7402 | if (!delalloc) |
7403 | return used_bg; | |
e570fd27 | 7404 | |
6719afdc GU |
7405 | if (down_read_trylock(&used_bg->data_rwsem)) |
7406 | return used_bg; | |
e570fd27 | 7407 | |
6719afdc | 7408 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 7409 | |
e321f8a8 LB |
7410 | /* We should only have one-level nested. */ |
7411 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 7412 | |
6719afdc GU |
7413 | spin_lock(&cluster->refill_lock); |
7414 | if (used_bg == cluster->block_group) | |
7415 | return used_bg; | |
e570fd27 | 7416 | |
6719afdc GU |
7417 | up_read(&used_bg->data_rwsem); |
7418 | btrfs_put_block_group(used_bg); | |
7419 | } | |
e570fd27 MX |
7420 | } |
7421 | ||
7422 | static inline void | |
7423 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7424 | int delalloc) | |
7425 | { | |
7426 | if (delalloc) | |
7427 | up_read(&cache->data_rwsem); | |
7428 | btrfs_put_block_group(cache); | |
7429 | } | |
7430 | ||
fec577fb CM |
7431 | /* |
7432 | * walks the btree of allocated extents and find a hole of a given size. | |
7433 | * The key ins is changed to record the hole: | |
a4820398 | 7434 | * ins->objectid == start position |
62e2749e | 7435 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7436 | * ins->offset == the size of the hole. |
fec577fb | 7437 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7438 | * |
7439 | * If there is no suitable free space, we will record the max size of | |
7440 | * the free space extent currently. | |
fec577fb | 7441 | */ |
87bde3cd | 7442 | static noinline int find_free_extent(struct btrfs_fs_info *fs_info, |
18513091 WX |
7443 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
7444 | u64 hint_byte, struct btrfs_key *ins, | |
7445 | u64 flags, int delalloc) | |
fec577fb | 7446 | { |
80eb234a | 7447 | int ret = 0; |
0b246afa | 7448 | struct btrfs_root *root = fs_info->extent_root; |
fa9c0d79 | 7449 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7450 | struct btrfs_block_group_cache *block_group = NULL; |
81c9ad23 | 7451 | u64 search_start = 0; |
a4820398 | 7452 | u64 max_extent_size = 0; |
c759c4e1 | 7453 | u64 empty_cluster = 0; |
80eb234a | 7454 | struct btrfs_space_info *space_info; |
fa9c0d79 | 7455 | int loop = 0; |
3e72ee88 | 7456 | int index = btrfs_bg_flags_to_raid_index(flags); |
0a24325e | 7457 | bool failed_cluster_refill = false; |
1cdda9b8 | 7458 | bool failed_alloc = false; |
67377734 | 7459 | bool use_cluster = true; |
60d2adbb | 7460 | bool have_caching_bg = false; |
13a0db5a | 7461 | bool orig_have_caching_bg = false; |
a5e681d9 | 7462 | bool full_search = false; |
fec577fb | 7463 | |
0b246afa | 7464 | WARN_ON(num_bytes < fs_info->sectorsize); |
962a298f | 7465 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7466 | ins->objectid = 0; |
7467 | ins->offset = 0; | |
b1a4d965 | 7468 | |
71ff6437 | 7469 | trace_find_free_extent(fs_info, num_bytes, empty_size, flags); |
3f7de037 | 7470 | |
0b246afa | 7471 | space_info = __find_space_info(fs_info, flags); |
1b1d1f66 | 7472 | if (!space_info) { |
0b246afa | 7473 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7474 | return -ENOSPC; |
7475 | } | |
2552d17e | 7476 | |
67377734 | 7477 | /* |
4f4db217 JB |
7478 | * If our free space is heavily fragmented we may not be able to make |
7479 | * big contiguous allocations, so instead of doing the expensive search | |
7480 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7481 | * can go ahead and search for a more manageable chunk. | |
7482 | * | |
7483 | * If our max_extent_size is large enough for our allocation simply | |
7484 | * disable clustering since we will likely not be able to find enough | |
7485 | * space to create a cluster and induce latency trying. | |
67377734 | 7486 | */ |
4f4db217 JB |
7487 | if (unlikely(space_info->max_extent_size)) { |
7488 | spin_lock(&space_info->lock); | |
7489 | if (space_info->max_extent_size && | |
7490 | num_bytes > space_info->max_extent_size) { | |
7491 | ins->offset = space_info->max_extent_size; | |
7492 | spin_unlock(&space_info->lock); | |
7493 | return -ENOSPC; | |
7494 | } else if (space_info->max_extent_size) { | |
7495 | use_cluster = false; | |
7496 | } | |
7497 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7498 | } |
0f9dd46c | 7499 | |
2ff7e61e | 7500 | last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster); |
239b14b3 | 7501 | if (last_ptr) { |
fa9c0d79 CM |
7502 | spin_lock(&last_ptr->lock); |
7503 | if (last_ptr->block_group) | |
7504 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7505 | if (last_ptr->fragmented) { |
7506 | /* | |
7507 | * We still set window_start so we can keep track of the | |
7508 | * last place we found an allocation to try and save | |
7509 | * some time. | |
7510 | */ | |
7511 | hint_byte = last_ptr->window_start; | |
7512 | use_cluster = false; | |
7513 | } | |
fa9c0d79 | 7514 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7515 | } |
fa9c0d79 | 7516 | |
2ff7e61e | 7517 | search_start = max(search_start, first_logical_byte(fs_info, 0)); |
239b14b3 | 7518 | search_start = max(search_start, hint_byte); |
2552d17e | 7519 | if (search_start == hint_byte) { |
0b246afa | 7520 | block_group = btrfs_lookup_block_group(fs_info, search_start); |
817d52f8 JB |
7521 | /* |
7522 | * we don't want to use the block group if it doesn't match our | |
7523 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7524 | * |
7525 | * However if we are re-searching with an ideal block group | |
7526 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7527 | */ |
b6919a58 | 7528 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7529 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7530 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7531 | if (list_empty(&block_group->list) || |
7532 | block_group->ro) { | |
7533 | /* | |
7534 | * someone is removing this block group, | |
7535 | * we can't jump into the have_block_group | |
7536 | * target because our list pointers are not | |
7537 | * valid | |
7538 | */ | |
7539 | btrfs_put_block_group(block_group); | |
7540 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7541 | } else { |
3e72ee88 QW |
7542 | index = btrfs_bg_flags_to_raid_index( |
7543 | block_group->flags); | |
e570fd27 | 7544 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7545 | goto have_block_group; |
ccf0e725 | 7546 | } |
2552d17e | 7547 | } else if (block_group) { |
fa9c0d79 | 7548 | btrfs_put_block_group(block_group); |
2552d17e | 7549 | } |
42e70e7a | 7550 | } |
2552d17e | 7551 | search: |
60d2adbb | 7552 | have_caching_bg = false; |
3e72ee88 | 7553 | if (index == 0 || index == btrfs_bg_flags_to_raid_index(flags)) |
a5e681d9 | 7554 | full_search = true; |
80eb234a | 7555 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
7556 | list_for_each_entry(block_group, &space_info->block_groups[index], |
7557 | list) { | |
6226cb0a | 7558 | u64 offset; |
817d52f8 | 7559 | int cached; |
8a1413a2 | 7560 | |
14443937 JM |
7561 | /* If the block group is read-only, we can skip it entirely. */ |
7562 | if (unlikely(block_group->ro)) | |
7563 | continue; | |
7564 | ||
e570fd27 | 7565 | btrfs_grab_block_group(block_group, delalloc); |
2552d17e | 7566 | search_start = block_group->key.objectid; |
42e70e7a | 7567 | |
83a50de9 CM |
7568 | /* |
7569 | * this can happen if we end up cycling through all the | |
7570 | * raid types, but we want to make sure we only allocate | |
7571 | * for the proper type. | |
7572 | */ | |
b6919a58 | 7573 | if (!block_group_bits(block_group, flags)) { |
83a50de9 CM |
7574 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
7575 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
7576 | BTRFS_BLOCK_GROUP_RAID5 | |
7577 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
7578 | BTRFS_BLOCK_GROUP_RAID10; |
7579 | ||
7580 | /* | |
7581 | * if they asked for extra copies and this block group | |
7582 | * doesn't provide them, bail. This does allow us to | |
7583 | * fill raid0 from raid1. | |
7584 | */ | |
b6919a58 | 7585 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
7586 | goto loop; |
7587 | } | |
7588 | ||
2552d17e | 7589 | have_block_group: |
291c7d2f JB |
7590 | cached = block_group_cache_done(block_group); |
7591 | if (unlikely(!cached)) { | |
a5e681d9 | 7592 | have_caching_bg = true; |
f6373bf3 | 7593 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7594 | BUG_ON(ret < 0); |
7595 | ret = 0; | |
817d52f8 JB |
7596 | } |
7597 | ||
36cce922 JB |
7598 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7599 | goto loop; | |
0f9dd46c | 7600 | |
0a24325e | 7601 | /* |
062c05c4 AO |
7602 | * Ok we want to try and use the cluster allocator, so |
7603 | * lets look there | |
0a24325e | 7604 | */ |
c759c4e1 | 7605 | if (last_ptr && use_cluster) { |
215a63d1 | 7606 | struct btrfs_block_group_cache *used_block_group; |
8de972b4 | 7607 | unsigned long aligned_cluster; |
fa9c0d79 CM |
7608 | /* |
7609 | * the refill lock keeps out other | |
7610 | * people trying to start a new cluster | |
7611 | */ | |
e570fd27 MX |
7612 | used_block_group = btrfs_lock_cluster(block_group, |
7613 | last_ptr, | |
7614 | delalloc); | |
7615 | if (!used_block_group) | |
44fb5511 | 7616 | goto refill_cluster; |
274bd4fb | 7617 | |
e570fd27 MX |
7618 | if (used_block_group != block_group && |
7619 | (used_block_group->ro || | |
7620 | !block_group_bits(used_block_group, flags))) | |
7621 | goto release_cluster; | |
44fb5511 | 7622 | |
274bd4fb | 7623 | offset = btrfs_alloc_from_cluster(used_block_group, |
a4820398 MX |
7624 | last_ptr, |
7625 | num_bytes, | |
7626 | used_block_group->key.objectid, | |
7627 | &max_extent_size); | |
fa9c0d79 CM |
7628 | if (offset) { |
7629 | /* we have a block, we're done */ | |
7630 | spin_unlock(&last_ptr->refill_lock); | |
71ff6437 | 7631 | trace_btrfs_reserve_extent_cluster(fs_info, |
89d4346a MX |
7632 | used_block_group, |
7633 | search_start, num_bytes); | |
215a63d1 | 7634 | if (used_block_group != block_group) { |
e570fd27 MX |
7635 | btrfs_release_block_group(block_group, |
7636 | delalloc); | |
215a63d1 MX |
7637 | block_group = used_block_group; |
7638 | } | |
fa9c0d79 CM |
7639 | goto checks; |
7640 | } | |
7641 | ||
274bd4fb | 7642 | WARN_ON(last_ptr->block_group != used_block_group); |
e570fd27 | 7643 | release_cluster: |
062c05c4 AO |
7644 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
7645 | * set up a new clusters, so lets just skip it | |
7646 | * and let the allocator find whatever block | |
7647 | * it can find. If we reach this point, we | |
7648 | * will have tried the cluster allocator | |
7649 | * plenty of times and not have found | |
7650 | * anything, so we are likely way too | |
7651 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
7652 | * anything. |
7653 | * | |
7654 | * However, if the cluster is taken from the | |
7655 | * current block group, release the cluster | |
7656 | * first, so that we stand a better chance of | |
7657 | * succeeding in the unclustered | |
7658 | * allocation. */ | |
7659 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
e570fd27 | 7660 | used_block_group != block_group) { |
062c05c4 | 7661 | spin_unlock(&last_ptr->refill_lock); |
e570fd27 MX |
7662 | btrfs_release_block_group(used_block_group, |
7663 | delalloc); | |
062c05c4 AO |
7664 | goto unclustered_alloc; |
7665 | } | |
7666 | ||
fa9c0d79 CM |
7667 | /* |
7668 | * this cluster didn't work out, free it and | |
7669 | * start over | |
7670 | */ | |
7671 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7672 | ||
e570fd27 MX |
7673 | if (used_block_group != block_group) |
7674 | btrfs_release_block_group(used_block_group, | |
7675 | delalloc); | |
7676 | refill_cluster: | |
a5f6f719 AO |
7677 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
7678 | spin_unlock(&last_ptr->refill_lock); | |
7679 | goto unclustered_alloc; | |
7680 | } | |
7681 | ||
8de972b4 CM |
7682 | aligned_cluster = max_t(unsigned long, |
7683 | empty_cluster + empty_size, | |
7684 | block_group->full_stripe_len); | |
7685 | ||
fa9c0d79 | 7686 | /* allocate a cluster in this block group */ |
2ff7e61e | 7687 | ret = btrfs_find_space_cluster(fs_info, block_group, |
00361589 JB |
7688 | last_ptr, search_start, |
7689 | num_bytes, | |
7690 | aligned_cluster); | |
fa9c0d79 CM |
7691 | if (ret == 0) { |
7692 | /* | |
7693 | * now pull our allocation out of this | |
7694 | * cluster | |
7695 | */ | |
7696 | offset = btrfs_alloc_from_cluster(block_group, | |
a4820398 MX |
7697 | last_ptr, |
7698 | num_bytes, | |
7699 | search_start, | |
7700 | &max_extent_size); | |
fa9c0d79 CM |
7701 | if (offset) { |
7702 | /* we found one, proceed */ | |
7703 | spin_unlock(&last_ptr->refill_lock); | |
71ff6437 | 7704 | trace_btrfs_reserve_extent_cluster(fs_info, |
3f7de037 JB |
7705 | block_group, search_start, |
7706 | num_bytes); | |
fa9c0d79 CM |
7707 | goto checks; |
7708 | } | |
0a24325e JB |
7709 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
7710 | && !failed_cluster_refill) { | |
817d52f8 JB |
7711 | spin_unlock(&last_ptr->refill_lock); |
7712 | ||
0a24325e | 7713 | failed_cluster_refill = true; |
817d52f8 JB |
7714 | wait_block_group_cache_progress(block_group, |
7715 | num_bytes + empty_cluster + empty_size); | |
7716 | goto have_block_group; | |
fa9c0d79 | 7717 | } |
817d52f8 | 7718 | |
fa9c0d79 CM |
7719 | /* |
7720 | * at this point we either didn't find a cluster | |
7721 | * or we weren't able to allocate a block from our | |
7722 | * cluster. Free the cluster we've been trying | |
7723 | * to use, and go to the next block group | |
7724 | */ | |
0a24325e | 7725 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 7726 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 7727 | goto loop; |
fa9c0d79 CM |
7728 | } |
7729 | ||
062c05c4 | 7730 | unclustered_alloc: |
c759c4e1 JB |
7731 | /* |
7732 | * We are doing an unclustered alloc, set the fragmented flag so | |
7733 | * we don't bother trying to setup a cluster again until we get | |
7734 | * more space. | |
7735 | */ | |
7736 | if (unlikely(last_ptr)) { | |
7737 | spin_lock(&last_ptr->lock); | |
7738 | last_ptr->fragmented = 1; | |
7739 | spin_unlock(&last_ptr->lock); | |
7740 | } | |
0c9b36e0 LB |
7741 | if (cached) { |
7742 | struct btrfs_free_space_ctl *ctl = | |
7743 | block_group->free_space_ctl; | |
7744 | ||
7745 | spin_lock(&ctl->tree_lock); | |
7746 | if (ctl->free_space < | |
7747 | num_bytes + empty_cluster + empty_size) { | |
7748 | if (ctl->free_space > max_extent_size) | |
7749 | max_extent_size = ctl->free_space; | |
7750 | spin_unlock(&ctl->tree_lock); | |
7751 | goto loop; | |
7752 | } | |
7753 | spin_unlock(&ctl->tree_lock); | |
a5f6f719 | 7754 | } |
a5f6f719 | 7755 | |
6226cb0a | 7756 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
a4820398 MX |
7757 | num_bytes, empty_size, |
7758 | &max_extent_size); | |
1cdda9b8 JB |
7759 | /* |
7760 | * If we didn't find a chunk, and we haven't failed on this | |
7761 | * block group before, and this block group is in the middle of | |
7762 | * caching and we are ok with waiting, then go ahead and wait | |
7763 | * for progress to be made, and set failed_alloc to true. | |
7764 | * | |
7765 | * If failed_alloc is true then we've already waited on this | |
7766 | * block group once and should move on to the next block group. | |
7767 | */ | |
7768 | if (!offset && !failed_alloc && !cached && | |
7769 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 7770 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
7771 | num_bytes + empty_size); |
7772 | failed_alloc = true; | |
817d52f8 | 7773 | goto have_block_group; |
1cdda9b8 JB |
7774 | } else if (!offset) { |
7775 | goto loop; | |
817d52f8 | 7776 | } |
fa9c0d79 | 7777 | checks: |
0b246afa | 7778 | search_start = ALIGN(offset, fs_info->stripesize); |
25179201 | 7779 | |
2552d17e JB |
7780 | /* move on to the next group */ |
7781 | if (search_start + num_bytes > | |
215a63d1 MX |
7782 | block_group->key.objectid + block_group->key.offset) { |
7783 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 7784 | goto loop; |
6226cb0a | 7785 | } |
f5a31e16 | 7786 | |
f0486c68 | 7787 | if (offset < search_start) |
215a63d1 | 7788 | btrfs_add_free_space(block_group, offset, |
f0486c68 YZ |
7789 | search_start - offset); |
7790 | BUG_ON(offset > search_start); | |
2552d17e | 7791 | |
18513091 WX |
7792 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
7793 | num_bytes, delalloc); | |
f0486c68 | 7794 | if (ret == -EAGAIN) { |
215a63d1 | 7795 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 7796 | goto loop; |
0f9dd46c | 7797 | } |
9cfa3e34 | 7798 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 7799 | |
f0486c68 | 7800 | /* we are all good, lets return */ |
2552d17e JB |
7801 | ins->objectid = search_start; |
7802 | ins->offset = num_bytes; | |
d2fb3437 | 7803 | |
71ff6437 | 7804 | trace_btrfs_reserve_extent(fs_info, block_group, |
3f7de037 | 7805 | search_start, num_bytes); |
e570fd27 | 7806 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7807 | break; |
7808 | loop: | |
0a24325e | 7809 | failed_cluster_refill = false; |
1cdda9b8 | 7810 | failed_alloc = false; |
3e72ee88 QW |
7811 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != |
7812 | index); | |
e570fd27 | 7813 | btrfs_release_block_group(block_group, delalloc); |
14443937 | 7814 | cond_resched(); |
2552d17e JB |
7815 | } |
7816 | up_read(&space_info->groups_sem); | |
7817 | ||
13a0db5a | 7818 | if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg |
7819 | && !orig_have_caching_bg) | |
7820 | orig_have_caching_bg = true; | |
7821 | ||
60d2adbb MX |
7822 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
7823 | goto search; | |
7824 | ||
b742bb82 YZ |
7825 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
7826 | goto search; | |
7827 | ||
285ff5af | 7828 | /* |
ccf0e725 JB |
7829 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
7830 | * caching kthreads as we move along | |
817d52f8 JB |
7831 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
7832 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7833 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7834 | * again | |
fa9c0d79 | 7835 | */ |
723bda20 | 7836 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 7837 | index = 0; |
a5e681d9 JB |
7838 | if (loop == LOOP_CACHING_NOWAIT) { |
7839 | /* | |
7840 | * We want to skip the LOOP_CACHING_WAIT step if we | |
01327610 | 7841 | * don't have any uncached bgs and we've already done a |
a5e681d9 JB |
7842 | * full search through. |
7843 | */ | |
13a0db5a | 7844 | if (orig_have_caching_bg || !full_search) |
a5e681d9 JB |
7845 | loop = LOOP_CACHING_WAIT; |
7846 | else | |
7847 | loop = LOOP_ALLOC_CHUNK; | |
7848 | } else { | |
7849 | loop++; | |
7850 | } | |
7851 | ||
817d52f8 | 7852 | if (loop == LOOP_ALLOC_CHUNK) { |
00361589 | 7853 | struct btrfs_trans_handle *trans; |
f017f15f WS |
7854 | int exist = 0; |
7855 | ||
7856 | trans = current->journal_info; | |
7857 | if (trans) | |
7858 | exist = 1; | |
7859 | else | |
7860 | trans = btrfs_join_transaction(root); | |
00361589 | 7861 | |
00361589 JB |
7862 | if (IS_ERR(trans)) { |
7863 | ret = PTR_ERR(trans); | |
7864 | goto out; | |
7865 | } | |
7866 | ||
2ff7e61e | 7867 | ret = do_chunk_alloc(trans, fs_info, flags, |
ea658bad | 7868 | CHUNK_ALLOC_FORCE); |
a5e681d9 JB |
7869 | |
7870 | /* | |
7871 | * If we can't allocate a new chunk we've already looped | |
7872 | * through at least once, move on to the NO_EMPTY_SIZE | |
7873 | * case. | |
7874 | */ | |
7875 | if (ret == -ENOSPC) | |
7876 | loop = LOOP_NO_EMPTY_SIZE; | |
7877 | ||
ea658bad JB |
7878 | /* |
7879 | * Do not bail out on ENOSPC since we | |
7880 | * can do more things. | |
7881 | */ | |
00361589 | 7882 | if (ret < 0 && ret != -ENOSPC) |
66642832 | 7883 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
7884 | else |
7885 | ret = 0; | |
f017f15f | 7886 | if (!exist) |
3a45bb20 | 7887 | btrfs_end_transaction(trans); |
00361589 | 7888 | if (ret) |
ea658bad | 7889 | goto out; |
2552d17e JB |
7890 | } |
7891 | ||
723bda20 | 7892 | if (loop == LOOP_NO_EMPTY_SIZE) { |
a5e681d9 JB |
7893 | /* |
7894 | * Don't loop again if we already have no empty_size and | |
7895 | * no empty_cluster. | |
7896 | */ | |
7897 | if (empty_size == 0 && | |
7898 | empty_cluster == 0) { | |
7899 | ret = -ENOSPC; | |
7900 | goto out; | |
7901 | } | |
723bda20 JB |
7902 | empty_size = 0; |
7903 | empty_cluster = 0; | |
fa9c0d79 | 7904 | } |
723bda20 JB |
7905 | |
7906 | goto search; | |
2552d17e JB |
7907 | } else if (!ins->objectid) { |
7908 | ret = -ENOSPC; | |
d82a6f1d | 7909 | } else if (ins->objectid) { |
c759c4e1 JB |
7910 | if (!use_cluster && last_ptr) { |
7911 | spin_lock(&last_ptr->lock); | |
7912 | last_ptr->window_start = ins->objectid; | |
7913 | spin_unlock(&last_ptr->lock); | |
7914 | } | |
80eb234a | 7915 | ret = 0; |
be744175 | 7916 | } |
79787eaa | 7917 | out: |
4f4db217 JB |
7918 | if (ret == -ENOSPC) { |
7919 | spin_lock(&space_info->lock); | |
7920 | space_info->max_extent_size = max_extent_size; | |
7921 | spin_unlock(&space_info->lock); | |
a4820398 | 7922 | ins->offset = max_extent_size; |
4f4db217 | 7923 | } |
0f70abe2 | 7924 | return ret; |
fec577fb | 7925 | } |
ec44a35c | 7926 | |
ab8d0fc4 JM |
7927 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
7928 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 7929 | int dump_block_groups) |
0f9dd46c JB |
7930 | { |
7931 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 7932 | int index = 0; |
0f9dd46c | 7933 | |
9ed74f2d | 7934 | spin_lock(&info->lock); |
ab8d0fc4 JM |
7935 | btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull", |
7936 | info->flags, | |
4136135b LB |
7937 | info->total_bytes - btrfs_space_info_used(info, true), |
7938 | info->full ? "" : "not "); | |
ab8d0fc4 JM |
7939 | btrfs_info(fs_info, |
7940 | "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu", | |
7941 | info->total_bytes, info->bytes_used, info->bytes_pinned, | |
7942 | info->bytes_reserved, info->bytes_may_use, | |
7943 | info->bytes_readonly); | |
9ed74f2d JB |
7944 | spin_unlock(&info->lock); |
7945 | ||
7946 | if (!dump_block_groups) | |
7947 | return; | |
0f9dd46c | 7948 | |
80eb234a | 7949 | down_read(&info->groups_sem); |
b742bb82 YZ |
7950 | again: |
7951 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 7952 | spin_lock(&cache->lock); |
ab8d0fc4 JM |
7953 | btrfs_info(fs_info, |
7954 | "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s", | |
7955 | cache->key.objectid, cache->key.offset, | |
7956 | btrfs_block_group_used(&cache->item), cache->pinned, | |
7957 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
7958 | btrfs_dump_free_space(cache, bytes); |
7959 | spin_unlock(&cache->lock); | |
7960 | } | |
b742bb82 YZ |
7961 | if (++index < BTRFS_NR_RAID_TYPES) |
7962 | goto again; | |
80eb234a | 7963 | up_read(&info->groups_sem); |
0f9dd46c | 7964 | } |
e8569813 | 7965 | |
6f47c706 NB |
7966 | /* |
7967 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a | |
7968 | * hole that is at least as big as @num_bytes. | |
7969 | * | |
7970 | * @root - The root that will contain this extent | |
7971 | * | |
7972 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
7973 | * is used for accounting purposes. This value differs | |
7974 | * from @num_bytes only in the case of compressed extents. | |
7975 | * | |
7976 | * @num_bytes - Number of bytes to allocate on-disk. | |
7977 | * | |
7978 | * @min_alloc_size - Indicates the minimum amount of space that the | |
7979 | * allocator should try to satisfy. In some cases | |
7980 | * @num_bytes may be larger than what is required and if | |
7981 | * the filesystem is fragmented then allocation fails. | |
7982 | * However, the presence of @min_alloc_size gives a | |
7983 | * chance to try and satisfy the smaller allocation. | |
7984 | * | |
7985 | * @empty_size - A hint that you plan on doing more COW. This is the | |
7986 | * size in bytes the allocator should try to find free | |
7987 | * next to the block it returns. This is just a hint and | |
7988 | * may be ignored by the allocator. | |
7989 | * | |
7990 | * @hint_byte - Hint to the allocator to start searching above the byte | |
7991 | * address passed. It might be ignored. | |
7992 | * | |
7993 | * @ins - This key is modified to record the found hole. It will | |
7994 | * have the following values: | |
7995 | * ins->objectid == start position | |
7996 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
7997 | * ins->offset == the size of the hole. | |
7998 | * | |
7999 | * @is_data - Boolean flag indicating whether an extent is | |
8000 | * allocated for data (true) or metadata (false) | |
8001 | * | |
8002 | * @delalloc - Boolean flag indicating whether this allocation is for | |
8003 | * delalloc or not. If 'true' data_rwsem of block groups | |
8004 | * is going to be acquired. | |
8005 | * | |
8006 | * | |
8007 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
8008 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
8009 | * largest available hole the allocator managed to find. | |
8010 | */ | |
18513091 | 8011 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
8012 | u64 num_bytes, u64 min_alloc_size, |
8013 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 8014 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 8015 | { |
ab8d0fc4 | 8016 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 8017 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 8018 | u64 flags; |
fec577fb | 8019 | int ret; |
925baedd | 8020 | |
1b86826d | 8021 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 8022 | again: |
0b246afa | 8023 | WARN_ON(num_bytes < fs_info->sectorsize); |
87bde3cd | 8024 | ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size, |
18513091 | 8025 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 8026 | if (!ret && !is_data) { |
ab8d0fc4 | 8027 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 8028 | } else if (ret == -ENOSPC) { |
a4820398 MX |
8029 | if (!final_tried && ins->offset) { |
8030 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 8031 | num_bytes = round_down(num_bytes, |
0b246afa | 8032 | fs_info->sectorsize); |
9e622d6b | 8033 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 8034 | ram_bytes = num_bytes; |
9e622d6b MX |
8035 | if (num_bytes == min_alloc_size) |
8036 | final_tried = true; | |
8037 | goto again; | |
ab8d0fc4 | 8038 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
8039 | struct btrfs_space_info *sinfo; |
8040 | ||
ab8d0fc4 | 8041 | sinfo = __find_space_info(fs_info, flags); |
0b246afa | 8042 | btrfs_err(fs_info, |
5d163e0e JM |
8043 | "allocation failed flags %llu, wanted %llu", |
8044 | flags, num_bytes); | |
53804280 | 8045 | if (sinfo) |
ab8d0fc4 | 8046 | dump_space_info(fs_info, sinfo, num_bytes, 1); |
9e622d6b | 8047 | } |
925baedd | 8048 | } |
0f9dd46c JB |
8049 | |
8050 | return ret; | |
e6dcd2dc CM |
8051 | } |
8052 | ||
2ff7e61e | 8053 | static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
8054 | u64 start, u64 len, |
8055 | int pin, int delalloc) | |
65b51a00 | 8056 | { |
0f9dd46c | 8057 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 8058 | int ret = 0; |
0f9dd46c | 8059 | |
0b246afa | 8060 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 8061 | if (!cache) { |
0b246afa JM |
8062 | btrfs_err(fs_info, "Unable to find block group for %llu", |
8063 | start); | |
0f9dd46c JB |
8064 | return -ENOSPC; |
8065 | } | |
1f3c79a2 | 8066 | |
e688b725 | 8067 | if (pin) |
2ff7e61e | 8068 | pin_down_extent(fs_info, cache, start, len, 1); |
e688b725 | 8069 | else { |
0b246afa | 8070 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 8071 | ret = btrfs_discard_extent(fs_info, start, len, NULL); |
e688b725 | 8072 | btrfs_add_free_space(cache, start, len); |
4824f1f4 | 8073 | btrfs_free_reserved_bytes(cache, len, delalloc); |
71ff6437 | 8074 | trace_btrfs_reserved_extent_free(fs_info, start, len); |
e688b725 | 8075 | } |
31193213 | 8076 | |
fa9c0d79 | 8077 | btrfs_put_block_group(cache); |
e6dcd2dc CM |
8078 | return ret; |
8079 | } | |
8080 | ||
2ff7e61e | 8081 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 | 8082 | u64 start, u64 len, int delalloc) |
e688b725 | 8083 | { |
2ff7e61e | 8084 | return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc); |
e688b725 CM |
8085 | } |
8086 | ||
2ff7e61e | 8087 | int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info, |
e688b725 CM |
8088 | u64 start, u64 len) |
8089 | { | |
2ff7e61e | 8090 | return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0); |
e688b725 CM |
8091 | } |
8092 | ||
5d4f98a2 | 8093 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 8094 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8095 | u64 parent, u64 root_objectid, |
8096 | u64 flags, u64 owner, u64 offset, | |
8097 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
8098 | { |
8099 | int ret; | |
e6dcd2dc | 8100 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 8101 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 8102 | struct btrfs_path *path; |
5d4f98a2 YZ |
8103 | struct extent_buffer *leaf; |
8104 | int type; | |
8105 | u32 size; | |
26b8003f | 8106 | |
5d4f98a2 YZ |
8107 | if (parent > 0) |
8108 | type = BTRFS_SHARED_DATA_REF_KEY; | |
8109 | else | |
8110 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 8111 | |
5d4f98a2 | 8112 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
8113 | |
8114 | path = btrfs_alloc_path(); | |
db5b493a TI |
8115 | if (!path) |
8116 | return -ENOMEM; | |
47e4bb98 | 8117 | |
b9473439 | 8118 | path->leave_spinning = 1; |
5d4f98a2 YZ |
8119 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
8120 | ins, size); | |
79787eaa JM |
8121 | if (ret) { |
8122 | btrfs_free_path(path); | |
8123 | return ret; | |
8124 | } | |
0f9dd46c | 8125 | |
5d4f98a2 YZ |
8126 | leaf = path->nodes[0]; |
8127 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 8128 | struct btrfs_extent_item); |
5d4f98a2 YZ |
8129 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
8130 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8131 | btrfs_set_extent_flags(leaf, extent_item, | |
8132 | flags | BTRFS_EXTENT_FLAG_DATA); | |
8133 | ||
8134 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8135 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
8136 | if (parent > 0) { | |
8137 | struct btrfs_shared_data_ref *ref; | |
8138 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
8139 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8140 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
8141 | } else { | |
8142 | struct btrfs_extent_data_ref *ref; | |
8143 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
8144 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
8145 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
8146 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
8147 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
8148 | } | |
47e4bb98 CM |
8149 | |
8150 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 8151 | btrfs_free_path(path); |
f510cfec | 8152 | |
1e144fb8 OS |
8153 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
8154 | ins->offset); | |
8155 | if (ret) | |
8156 | return ret; | |
8157 | ||
6202df69 | 8158 | ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1); |
79787eaa | 8159 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8160 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8161 | ins->objectid, ins->offset); |
f5947066 CM |
8162 | BUG(); |
8163 | } | |
71ff6437 | 8164 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
8165 | return ret; |
8166 | } | |
8167 | ||
5d4f98a2 | 8168 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
2ff7e61e | 8169 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8170 | u64 parent, u64 root_objectid, |
8171 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 8172 | int level, struct btrfs_key *ins) |
e6dcd2dc CM |
8173 | { |
8174 | int ret; | |
5d4f98a2 YZ |
8175 | struct btrfs_extent_item *extent_item; |
8176 | struct btrfs_tree_block_info *block_info; | |
8177 | struct btrfs_extent_inline_ref *iref; | |
8178 | struct btrfs_path *path; | |
8179 | struct extent_buffer *leaf; | |
3173a18f | 8180 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
fcebe456 | 8181 | u64 num_bytes = ins->offset; |
0b246afa | 8182 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f JB |
8183 | |
8184 | if (!skinny_metadata) | |
8185 | size += sizeof(*block_info); | |
1c2308f8 | 8186 | |
5d4f98a2 | 8187 | path = btrfs_alloc_path(); |
857cc2fc | 8188 | if (!path) { |
2ff7e61e | 8189 | btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid, |
0b246afa | 8190 | fs_info->nodesize); |
d8926bb3 | 8191 | return -ENOMEM; |
857cc2fc | 8192 | } |
56bec294 | 8193 | |
5d4f98a2 YZ |
8194 | path->leave_spinning = 1; |
8195 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
8196 | ins, size); | |
79787eaa | 8197 | if (ret) { |
dd825259 | 8198 | btrfs_free_path(path); |
2ff7e61e | 8199 | btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid, |
0b246afa | 8200 | fs_info->nodesize); |
79787eaa JM |
8201 | return ret; |
8202 | } | |
5d4f98a2 YZ |
8203 | |
8204 | leaf = path->nodes[0]; | |
8205 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
8206 | struct btrfs_extent_item); | |
8207 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
8208 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8209 | btrfs_set_extent_flags(leaf, extent_item, | |
8210 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 8211 | |
3173a18f JB |
8212 | if (skinny_metadata) { |
8213 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
0b246afa | 8214 | num_bytes = fs_info->nodesize; |
3173a18f JB |
8215 | } else { |
8216 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
8217 | btrfs_set_tree_block_key(leaf, block_info, key); | |
8218 | btrfs_set_tree_block_level(leaf, block_info, level); | |
8219 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
8220 | } | |
5d4f98a2 | 8221 | |
5d4f98a2 YZ |
8222 | if (parent > 0) { |
8223 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
8224 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8225 | BTRFS_SHARED_BLOCK_REF_KEY); | |
8226 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8227 | } else { | |
8228 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8229 | BTRFS_TREE_BLOCK_REF_KEY); | |
8230 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
8231 | } | |
8232 | ||
8233 | btrfs_mark_buffer_dirty(leaf); | |
8234 | btrfs_free_path(path); | |
8235 | ||
1e144fb8 OS |
8236 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
8237 | num_bytes); | |
8238 | if (ret) | |
8239 | return ret; | |
8240 | ||
6202df69 JM |
8241 | ret = update_block_group(trans, fs_info, ins->objectid, |
8242 | fs_info->nodesize, 1); | |
79787eaa | 8243 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8244 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8245 | ins->objectid, ins->offset); |
5d4f98a2 YZ |
8246 | BUG(); |
8247 | } | |
0be5dc67 | 8248 | |
71ff6437 | 8249 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, |
0b246afa | 8250 | fs_info->nodesize); |
5d4f98a2 YZ |
8251 | return ret; |
8252 | } | |
8253 | ||
8254 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 8255 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
8256 | u64 offset, u64 ram_bytes, |
8257 | struct btrfs_key *ins) | |
5d4f98a2 | 8258 | { |
84f7d8e6 | 8259 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
8260 | int ret; |
8261 | ||
84f7d8e6 | 8262 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 8263 | |
fd708b81 JB |
8264 | btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0, |
8265 | root->root_key.objectid, owner, offset, | |
8266 | BTRFS_ADD_DELAYED_EXTENT); | |
8267 | ||
0b246afa | 8268 | ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid, |
84f7d8e6 JB |
8269 | ins->offset, 0, |
8270 | root->root_key.objectid, owner, | |
7be07912 OS |
8271 | offset, ram_bytes, |
8272 | BTRFS_ADD_DELAYED_EXTENT, NULL, NULL); | |
e6dcd2dc CM |
8273 | return ret; |
8274 | } | |
e02119d5 CM |
8275 | |
8276 | /* | |
8277 | * this is used by the tree logging recovery code. It records that | |
8278 | * an extent has been allocated and makes sure to clear the free | |
8279 | * space cache bits as well | |
8280 | */ | |
5d4f98a2 | 8281 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 8282 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8283 | u64 root_objectid, u64 owner, u64 offset, |
8284 | struct btrfs_key *ins) | |
e02119d5 CM |
8285 | { |
8286 | int ret; | |
8287 | struct btrfs_block_group_cache *block_group; | |
ed7a6948 | 8288 | struct btrfs_space_info *space_info; |
11833d66 | 8289 | |
8c2a1a30 JB |
8290 | /* |
8291 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 8292 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 8293 | */ |
0b246afa | 8294 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
8295 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
8296 | ins->offset); | |
b50c6e25 | 8297 | if (ret) |
8c2a1a30 | 8298 | return ret; |
11833d66 YZ |
8299 | } |
8300 | ||
0b246afa | 8301 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
8302 | if (!block_group) |
8303 | return -EINVAL; | |
8304 | ||
ed7a6948 WX |
8305 | space_info = block_group->space_info; |
8306 | spin_lock(&space_info->lock); | |
8307 | spin_lock(&block_group->lock); | |
8308 | space_info->bytes_reserved += ins->offset; | |
8309 | block_group->reserved += ins->offset; | |
8310 | spin_unlock(&block_group->lock); | |
8311 | spin_unlock(&space_info->lock); | |
8312 | ||
2ff7e61e | 8313 | ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid, |
5d4f98a2 | 8314 | 0, owner, offset, ins, 1); |
b50c6e25 | 8315 | btrfs_put_block_group(block_group); |
e02119d5 CM |
8316 | return ret; |
8317 | } | |
8318 | ||
48a3b636 ES |
8319 | static struct extent_buffer * |
8320 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
fe864576 | 8321 | u64 bytenr, int level) |
65b51a00 | 8322 | { |
0b246afa | 8323 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
8324 | struct extent_buffer *buf; |
8325 | ||
2ff7e61e | 8326 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8327 | if (IS_ERR(buf)) |
8328 | return buf; | |
8329 | ||
65b51a00 | 8330 | btrfs_set_header_generation(buf, trans->transid); |
85d4e461 | 8331 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 8332 | btrfs_tree_lock(buf); |
7c302b49 | 8333 | clean_tree_block(fs_info, buf); |
3083ee2e | 8334 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
8335 | |
8336 | btrfs_set_lock_blocking(buf); | |
4db8c528 | 8337 | set_extent_buffer_uptodate(buf); |
b4ce94de | 8338 | |
d0c803c4 | 8339 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 8340 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
8341 | /* |
8342 | * we allow two log transactions at a time, use different | |
8343 | * EXENT bit to differentiate dirty pages. | |
8344 | */ | |
656f30db | 8345 | if (buf->log_index == 0) |
8cef4e16 YZ |
8346 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
8347 | buf->start + buf->len - 1, GFP_NOFS); | |
8348 | else | |
8349 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 8350 | buf->start + buf->len - 1); |
d0c803c4 | 8351 | } else { |
656f30db | 8352 | buf->log_index = -1; |
d0c803c4 | 8353 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 8354 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 8355 | } |
64c12921 | 8356 | trans->dirty = true; |
b4ce94de | 8357 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
8358 | return buf; |
8359 | } | |
8360 | ||
f0486c68 YZ |
8361 | static struct btrfs_block_rsv * |
8362 | use_block_rsv(struct btrfs_trans_handle *trans, | |
8363 | struct btrfs_root *root, u32 blocksize) | |
8364 | { | |
0b246afa | 8365 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8366 | struct btrfs_block_rsv *block_rsv; |
0b246afa | 8367 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
f0486c68 | 8368 | int ret; |
d88033db | 8369 | bool global_updated = false; |
f0486c68 YZ |
8370 | |
8371 | block_rsv = get_block_rsv(trans, root); | |
8372 | ||
b586b323 MX |
8373 | if (unlikely(block_rsv->size == 0)) |
8374 | goto try_reserve; | |
d88033db | 8375 | again: |
f0486c68 YZ |
8376 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
8377 | if (!ret) | |
8378 | return block_rsv; | |
8379 | ||
b586b323 MX |
8380 | if (block_rsv->failfast) |
8381 | return ERR_PTR(ret); | |
8382 | ||
d88033db MX |
8383 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
8384 | global_updated = true; | |
0b246afa | 8385 | update_global_block_rsv(fs_info); |
d88033db MX |
8386 | goto again; |
8387 | } | |
8388 | ||
0b246afa | 8389 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
b586b323 MX |
8390 | static DEFINE_RATELIMIT_STATE(_rs, |
8391 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
8392 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
8393 | if (__ratelimit(&_rs)) | |
8394 | WARN(1, KERN_DEBUG | |
efe120a0 | 8395 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
8396 | } |
8397 | try_reserve: | |
8398 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
8399 | BTRFS_RESERVE_NO_FLUSH); | |
8400 | if (!ret) | |
8401 | return block_rsv; | |
8402 | /* | |
8403 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
8404 | * the global reserve if its space type is the same as the global |
8405 | * reservation. | |
b586b323 | 8406 | */ |
5881cfc9 MX |
8407 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
8408 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
8409 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
8410 | if (!ret) | |
8411 | return global_rsv; | |
8412 | } | |
8413 | return ERR_PTR(ret); | |
f0486c68 YZ |
8414 | } |
8415 | ||
8c2a3ca2 JB |
8416 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
8417 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
8418 | { |
8419 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
8c2a3ca2 | 8420 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
f0486c68 YZ |
8421 | } |
8422 | ||
fec577fb | 8423 | /* |
f0486c68 | 8424 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 8425 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 8426 | */ |
4d75f8a9 | 8427 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
8428 | struct btrfs_root *root, |
8429 | u64 parent, u64 root_objectid, | |
8430 | const struct btrfs_disk_key *key, | |
8431 | int level, u64 hint, | |
8432 | u64 empty_size) | |
fec577fb | 8433 | { |
0b246afa | 8434 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 8435 | struct btrfs_key ins; |
f0486c68 | 8436 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 8437 | struct extent_buffer *buf; |
67b7859e | 8438 | struct btrfs_delayed_extent_op *extent_op; |
f0486c68 YZ |
8439 | u64 flags = 0; |
8440 | int ret; | |
0b246afa JM |
8441 | u32 blocksize = fs_info->nodesize; |
8442 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 8443 | |
05653ef3 | 8444 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 8445 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 8446 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
fe864576 | 8447 | level); |
faa2dbf0 JB |
8448 | if (!IS_ERR(buf)) |
8449 | root->alloc_bytenr += blocksize; | |
8450 | return buf; | |
8451 | } | |
05653ef3 | 8452 | #endif |
fccb84c9 | 8453 | |
f0486c68 YZ |
8454 | block_rsv = use_block_rsv(trans, root, blocksize); |
8455 | if (IS_ERR(block_rsv)) | |
8456 | return ERR_CAST(block_rsv); | |
8457 | ||
18513091 | 8458 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 8459 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
8460 | if (ret) |
8461 | goto out_unuse; | |
55c69072 | 8462 | |
fe864576 | 8463 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level); |
67b7859e OS |
8464 | if (IS_ERR(buf)) { |
8465 | ret = PTR_ERR(buf); | |
8466 | goto out_free_reserved; | |
8467 | } | |
f0486c68 YZ |
8468 | |
8469 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8470 | if (parent == 0) | |
8471 | parent = ins.objectid; | |
8472 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8473 | } else | |
8474 | BUG_ON(parent > 0); | |
8475 | ||
8476 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 8477 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
8478 | if (!extent_op) { |
8479 | ret = -ENOMEM; | |
8480 | goto out_free_buf; | |
8481 | } | |
f0486c68 YZ |
8482 | if (key) |
8483 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8484 | else | |
8485 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8486 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
8487 | extent_op->update_key = skinny_metadata ? false : true; |
8488 | extent_op->update_flags = true; | |
8489 | extent_op->is_data = false; | |
b1c79e09 | 8490 | extent_op->level = level; |
f0486c68 | 8491 | |
fd708b81 JB |
8492 | btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent, |
8493 | root_objectid, level, 0, | |
8494 | BTRFS_ADD_DELAYED_EXTENT); | |
7be07912 OS |
8495 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid, |
8496 | ins.offset, parent, | |
8497 | root_objectid, level, | |
67b7859e | 8498 | BTRFS_ADD_DELAYED_EXTENT, |
7be07912 | 8499 | extent_op, NULL, NULL); |
67b7859e OS |
8500 | if (ret) |
8501 | goto out_free_delayed; | |
f0486c68 | 8502 | } |
fec577fb | 8503 | return buf; |
67b7859e OS |
8504 | |
8505 | out_free_delayed: | |
8506 | btrfs_free_delayed_extent_op(extent_op); | |
8507 | out_free_buf: | |
8508 | free_extent_buffer(buf); | |
8509 | out_free_reserved: | |
2ff7e61e | 8510 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 8511 | out_unuse: |
0b246afa | 8512 | unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 8513 | return ERR_PTR(ret); |
fec577fb | 8514 | } |
a28ec197 | 8515 | |
2c47e605 YZ |
8516 | struct walk_control { |
8517 | u64 refs[BTRFS_MAX_LEVEL]; | |
8518 | u64 flags[BTRFS_MAX_LEVEL]; | |
8519 | struct btrfs_key update_progress; | |
8520 | int stage; | |
8521 | int level; | |
8522 | int shared_level; | |
8523 | int update_ref; | |
8524 | int keep_locks; | |
1c4850e2 YZ |
8525 | int reada_slot; |
8526 | int reada_count; | |
66d7e7f0 | 8527 | int for_reloc; |
2c47e605 YZ |
8528 | }; |
8529 | ||
8530 | #define DROP_REFERENCE 1 | |
8531 | #define UPDATE_BACKREF 2 | |
8532 | ||
1c4850e2 YZ |
8533 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8534 | struct btrfs_root *root, | |
8535 | struct walk_control *wc, | |
8536 | struct btrfs_path *path) | |
6407bf6d | 8537 | { |
0b246afa | 8538 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8539 | u64 bytenr; |
8540 | u64 generation; | |
8541 | u64 refs; | |
94fcca9f | 8542 | u64 flags; |
5d4f98a2 | 8543 | u32 nritems; |
1c4850e2 YZ |
8544 | struct btrfs_key key; |
8545 | struct extent_buffer *eb; | |
6407bf6d | 8546 | int ret; |
1c4850e2 YZ |
8547 | int slot; |
8548 | int nread = 0; | |
6407bf6d | 8549 | |
1c4850e2 YZ |
8550 | if (path->slots[wc->level] < wc->reada_slot) { |
8551 | wc->reada_count = wc->reada_count * 2 / 3; | |
8552 | wc->reada_count = max(wc->reada_count, 2); | |
8553 | } else { | |
8554 | wc->reada_count = wc->reada_count * 3 / 2; | |
8555 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 8556 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 8557 | } |
7bb86316 | 8558 | |
1c4850e2 YZ |
8559 | eb = path->nodes[wc->level]; |
8560 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 8561 | |
1c4850e2 YZ |
8562 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8563 | if (nread >= wc->reada_count) | |
8564 | break; | |
bd56b302 | 8565 | |
2dd3e67b | 8566 | cond_resched(); |
1c4850e2 YZ |
8567 | bytenr = btrfs_node_blockptr(eb, slot); |
8568 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8569 | |
1c4850e2 YZ |
8570 | if (slot == path->slots[wc->level]) |
8571 | goto reada; | |
5d4f98a2 | 8572 | |
1c4850e2 YZ |
8573 | if (wc->stage == UPDATE_BACKREF && |
8574 | generation <= root->root_key.offset) | |
bd56b302 CM |
8575 | continue; |
8576 | ||
94fcca9f | 8577 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 8578 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
8579 | wc->level - 1, 1, &refs, |
8580 | &flags); | |
79787eaa JM |
8581 | /* We don't care about errors in readahead. */ |
8582 | if (ret < 0) | |
8583 | continue; | |
94fcca9f YZ |
8584 | BUG_ON(refs == 0); |
8585 | ||
1c4850e2 | 8586 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8587 | if (refs == 1) |
8588 | goto reada; | |
bd56b302 | 8589 | |
94fcca9f YZ |
8590 | if (wc->level == 1 && |
8591 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8592 | continue; | |
1c4850e2 YZ |
8593 | if (!wc->update_ref || |
8594 | generation <= root->root_key.offset) | |
8595 | continue; | |
8596 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8597 | ret = btrfs_comp_cpu_keys(&key, | |
8598 | &wc->update_progress); | |
8599 | if (ret < 0) | |
8600 | continue; | |
94fcca9f YZ |
8601 | } else { |
8602 | if (wc->level == 1 && | |
8603 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8604 | continue; | |
6407bf6d | 8605 | } |
1c4850e2 | 8606 | reada: |
2ff7e61e | 8607 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 8608 | nread++; |
20524f02 | 8609 | } |
1c4850e2 | 8610 | wc->reada_slot = slot; |
20524f02 | 8611 | } |
2c47e605 | 8612 | |
f82d02d9 | 8613 | /* |
2c016dc2 | 8614 | * helper to process tree block while walking down the tree. |
2c47e605 | 8615 | * |
2c47e605 YZ |
8616 | * when wc->stage == UPDATE_BACKREF, this function updates |
8617 | * back refs for pointers in the block. | |
8618 | * | |
8619 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8620 | */ |
2c47e605 | 8621 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8622 | struct btrfs_root *root, |
2c47e605 | 8623 | struct btrfs_path *path, |
94fcca9f | 8624 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8625 | { |
2ff7e61e | 8626 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
8627 | int level = wc->level; |
8628 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 8629 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
8630 | int ret; |
8631 | ||
2c47e605 YZ |
8632 | if (wc->stage == UPDATE_BACKREF && |
8633 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8634 | return 1; | |
f82d02d9 | 8635 | |
2c47e605 YZ |
8636 | /* |
8637 | * when reference count of tree block is 1, it won't increase | |
8638 | * again. once full backref flag is set, we never clear it. | |
8639 | */ | |
94fcca9f YZ |
8640 | if (lookup_info && |
8641 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8642 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 8643 | BUG_ON(!path->locks[level]); |
2ff7e61e | 8644 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8645 | eb->start, level, 1, |
2c47e605 YZ |
8646 | &wc->refs[level], |
8647 | &wc->flags[level]); | |
79787eaa JM |
8648 | BUG_ON(ret == -ENOMEM); |
8649 | if (ret) | |
8650 | return ret; | |
2c47e605 YZ |
8651 | BUG_ON(wc->refs[level] == 0); |
8652 | } | |
5d4f98a2 | 8653 | |
2c47e605 YZ |
8654 | if (wc->stage == DROP_REFERENCE) { |
8655 | if (wc->refs[level] > 1) | |
8656 | return 1; | |
f82d02d9 | 8657 | |
2c47e605 | 8658 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 8659 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8660 | path->locks[level] = 0; |
8661 | } | |
8662 | return 0; | |
8663 | } | |
f82d02d9 | 8664 | |
2c47e605 YZ |
8665 | /* wc->stage == UPDATE_BACKREF */ |
8666 | if (!(wc->flags[level] & flag)) { | |
8667 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 8668 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 8669 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 8670 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8671 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8672 | ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start, |
b1c79e09 JB |
8673 | eb->len, flag, |
8674 | btrfs_header_level(eb), 0); | |
79787eaa | 8675 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
8676 | wc->flags[level] |= flag; |
8677 | } | |
8678 | ||
8679 | /* | |
8680 | * the block is shared by multiple trees, so it's not good to | |
8681 | * keep the tree lock | |
8682 | */ | |
8683 | if (path->locks[level] && level > 0) { | |
bd681513 | 8684 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8685 | path->locks[level] = 0; |
8686 | } | |
8687 | return 0; | |
8688 | } | |
8689 | ||
1c4850e2 | 8690 | /* |
2c016dc2 | 8691 | * helper to process tree block pointer. |
1c4850e2 YZ |
8692 | * |
8693 | * when wc->stage == DROP_REFERENCE, this function checks | |
8694 | * reference count of the block pointed to. if the block | |
8695 | * is shared and we need update back refs for the subtree | |
8696 | * rooted at the block, this function changes wc->stage to | |
8697 | * UPDATE_BACKREF. if the block is shared and there is no | |
8698 | * need to update back, this function drops the reference | |
8699 | * to the block. | |
8700 | * | |
8701 | * NOTE: return value 1 means we should stop walking down. | |
8702 | */ | |
8703 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8704 | struct btrfs_root *root, | |
8705 | struct btrfs_path *path, | |
94fcca9f | 8706 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 8707 | { |
0b246afa | 8708 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8709 | u64 bytenr; |
8710 | u64 generation; | |
8711 | u64 parent; | |
8712 | u32 blocksize; | |
8713 | struct btrfs_key key; | |
581c1760 | 8714 | struct btrfs_key first_key; |
1c4850e2 YZ |
8715 | struct extent_buffer *next; |
8716 | int level = wc->level; | |
8717 | int reada = 0; | |
8718 | int ret = 0; | |
1152651a | 8719 | bool need_account = false; |
1c4850e2 YZ |
8720 | |
8721 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8722 | path->slots[level]); | |
8723 | /* | |
8724 | * if the lower level block was created before the snapshot | |
8725 | * was created, we know there is no need to update back refs | |
8726 | * for the subtree | |
8727 | */ | |
8728 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8729 | generation <= root->root_key.offset) { |
8730 | *lookup_info = 1; | |
1c4850e2 | 8731 | return 1; |
94fcca9f | 8732 | } |
1c4850e2 YZ |
8733 | |
8734 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
581c1760 QW |
8735 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
8736 | path->slots[level]); | |
0b246afa | 8737 | blocksize = fs_info->nodesize; |
1c4850e2 | 8738 | |
0b246afa | 8739 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 8740 | if (!next) { |
2ff7e61e | 8741 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8742 | if (IS_ERR(next)) |
8743 | return PTR_ERR(next); | |
8744 | ||
b2aaaa3b JB |
8745 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8746 | level - 1); | |
1c4850e2 YZ |
8747 | reada = 1; |
8748 | } | |
8749 | btrfs_tree_lock(next); | |
8750 | btrfs_set_lock_blocking(next); | |
8751 | ||
2ff7e61e | 8752 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
8753 | &wc->refs[level - 1], |
8754 | &wc->flags[level - 1]); | |
4867268c JB |
8755 | if (ret < 0) |
8756 | goto out_unlock; | |
79787eaa | 8757 | |
c2cf52eb | 8758 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 8759 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
8760 | ret = -EIO; |
8761 | goto out_unlock; | |
c2cf52eb | 8762 | } |
94fcca9f | 8763 | *lookup_info = 0; |
1c4850e2 | 8764 | |
94fcca9f | 8765 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8766 | if (wc->refs[level - 1] > 1) { |
1152651a | 8767 | need_account = true; |
94fcca9f YZ |
8768 | if (level == 1 && |
8769 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8770 | goto skip; | |
8771 | ||
1c4850e2 YZ |
8772 | if (!wc->update_ref || |
8773 | generation <= root->root_key.offset) | |
8774 | goto skip; | |
8775 | ||
8776 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8777 | path->slots[level]); | |
8778 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8779 | if (ret < 0) | |
8780 | goto skip; | |
8781 | ||
8782 | wc->stage = UPDATE_BACKREF; | |
8783 | wc->shared_level = level - 1; | |
8784 | } | |
94fcca9f YZ |
8785 | } else { |
8786 | if (level == 1 && | |
8787 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8788 | goto skip; | |
1c4850e2 YZ |
8789 | } |
8790 | ||
b9fab919 | 8791 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8792 | btrfs_tree_unlock(next); |
8793 | free_extent_buffer(next); | |
8794 | next = NULL; | |
94fcca9f | 8795 | *lookup_info = 1; |
1c4850e2 YZ |
8796 | } |
8797 | ||
8798 | if (!next) { | |
8799 | if (reada && level == 1) | |
8800 | reada_walk_down(trans, root, wc, path); | |
581c1760 QW |
8801 | next = read_tree_block(fs_info, bytenr, generation, level - 1, |
8802 | &first_key); | |
64c043de LB |
8803 | if (IS_ERR(next)) { |
8804 | return PTR_ERR(next); | |
8805 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8806 | free_extent_buffer(next); |
97d9a8a4 | 8807 | return -EIO; |
416bc658 | 8808 | } |
1c4850e2 YZ |
8809 | btrfs_tree_lock(next); |
8810 | btrfs_set_lock_blocking(next); | |
8811 | } | |
8812 | ||
8813 | level--; | |
4867268c JB |
8814 | ASSERT(level == btrfs_header_level(next)); |
8815 | if (level != btrfs_header_level(next)) { | |
8816 | btrfs_err(root->fs_info, "mismatched level"); | |
8817 | ret = -EIO; | |
8818 | goto out_unlock; | |
8819 | } | |
1c4850e2 YZ |
8820 | path->nodes[level] = next; |
8821 | path->slots[level] = 0; | |
bd681513 | 8822 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8823 | wc->level = level; |
8824 | if (wc->level == 1) | |
8825 | wc->reada_slot = 0; | |
8826 | return 0; | |
8827 | skip: | |
8828 | wc->refs[level - 1] = 0; | |
8829 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
8830 | if (wc->stage == DROP_REFERENCE) { |
8831 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8832 | parent = path->nodes[level]->start; | |
8833 | } else { | |
4867268c | 8834 | ASSERT(root->root_key.objectid == |
94fcca9f | 8835 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
8836 | if (root->root_key.objectid != |
8837 | btrfs_header_owner(path->nodes[level])) { | |
8838 | btrfs_err(root->fs_info, | |
8839 | "mismatched block owner"); | |
8840 | ret = -EIO; | |
8841 | goto out_unlock; | |
8842 | } | |
94fcca9f YZ |
8843 | parent = 0; |
8844 | } | |
1c4850e2 | 8845 | |
1152651a | 8846 | if (need_account) { |
33d1f05c QW |
8847 | ret = btrfs_qgroup_trace_subtree(trans, root, next, |
8848 | generation, level - 1); | |
1152651a | 8849 | if (ret) { |
0b246afa | 8850 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8851 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
8852 | ret); | |
1152651a MF |
8853 | } |
8854 | } | |
84f7d8e6 | 8855 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, |
2ff7e61e JM |
8856 | parent, root->root_key.objectid, |
8857 | level - 1, 0); | |
4867268c JB |
8858 | if (ret) |
8859 | goto out_unlock; | |
1c4850e2 | 8860 | } |
4867268c JB |
8861 | |
8862 | *lookup_info = 1; | |
8863 | ret = 1; | |
8864 | ||
8865 | out_unlock: | |
1c4850e2 YZ |
8866 | btrfs_tree_unlock(next); |
8867 | free_extent_buffer(next); | |
4867268c JB |
8868 | |
8869 | return ret; | |
1c4850e2 YZ |
8870 | } |
8871 | ||
2c47e605 | 8872 | /* |
2c016dc2 | 8873 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
8874 | * |
8875 | * when wc->stage == DROP_REFERENCE, this function drops | |
8876 | * reference count on the block. | |
8877 | * | |
8878 | * when wc->stage == UPDATE_BACKREF, this function changes | |
8879 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
8880 | * to UPDATE_BACKREF previously while processing the block. | |
8881 | * | |
8882 | * NOTE: return value 1 means we should stop walking up. | |
8883 | */ | |
8884 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
8885 | struct btrfs_root *root, | |
8886 | struct btrfs_path *path, | |
8887 | struct walk_control *wc) | |
8888 | { | |
0b246afa | 8889 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8890 | int ret; |
2c47e605 YZ |
8891 | int level = wc->level; |
8892 | struct extent_buffer *eb = path->nodes[level]; | |
8893 | u64 parent = 0; | |
8894 | ||
8895 | if (wc->stage == UPDATE_BACKREF) { | |
8896 | BUG_ON(wc->shared_level < level); | |
8897 | if (level < wc->shared_level) | |
8898 | goto out; | |
8899 | ||
2c47e605 YZ |
8900 | ret = find_next_key(path, level + 1, &wc->update_progress); |
8901 | if (ret > 0) | |
8902 | wc->update_ref = 0; | |
8903 | ||
8904 | wc->stage = DROP_REFERENCE; | |
8905 | wc->shared_level = -1; | |
8906 | path->slots[level] = 0; | |
8907 | ||
8908 | /* | |
8909 | * check reference count again if the block isn't locked. | |
8910 | * we should start walking down the tree again if reference | |
8911 | * count is one. | |
8912 | */ | |
8913 | if (!path->locks[level]) { | |
8914 | BUG_ON(level == 0); | |
8915 | btrfs_tree_lock(eb); | |
8916 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8917 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8918 | |
2ff7e61e | 8919 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8920 | eb->start, level, 1, |
2c47e605 YZ |
8921 | &wc->refs[level], |
8922 | &wc->flags[level]); | |
79787eaa JM |
8923 | if (ret < 0) { |
8924 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 8925 | path->locks[level] = 0; |
79787eaa JM |
8926 | return ret; |
8927 | } | |
2c47e605 YZ |
8928 | BUG_ON(wc->refs[level] == 0); |
8929 | if (wc->refs[level] == 1) { | |
bd681513 | 8930 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 8931 | path->locks[level] = 0; |
2c47e605 YZ |
8932 | return 1; |
8933 | } | |
f82d02d9 | 8934 | } |
2c47e605 | 8935 | } |
f82d02d9 | 8936 | |
2c47e605 YZ |
8937 | /* wc->stage == DROP_REFERENCE */ |
8938 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 8939 | |
2c47e605 YZ |
8940 | if (wc->refs[level] == 1) { |
8941 | if (level == 0) { | |
8942 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 8943 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 8944 | else |
e339a6b0 | 8945 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8946 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8947 | ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb); |
1152651a | 8948 | if (ret) { |
0b246afa | 8949 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8950 | "error %d accounting leaf items. Quota is out of sync, rescan required.", |
8951 | ret); | |
1152651a | 8952 | } |
2c47e605 YZ |
8953 | } |
8954 | /* make block locked assertion in clean_tree_block happy */ | |
8955 | if (!path->locks[level] && | |
8956 | btrfs_header_generation(eb) == trans->transid) { | |
8957 | btrfs_tree_lock(eb); | |
8958 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8959 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8960 | } |
7c302b49 | 8961 | clean_tree_block(fs_info, eb); |
2c47e605 YZ |
8962 | } |
8963 | ||
8964 | if (eb == root->node) { | |
8965 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8966 | parent = eb->start; | |
8967 | else | |
8968 | BUG_ON(root->root_key.objectid != | |
8969 | btrfs_header_owner(eb)); | |
8970 | } else { | |
8971 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8972 | parent = path->nodes[level + 1]->start; | |
8973 | else | |
8974 | BUG_ON(root->root_key.objectid != | |
8975 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 8976 | } |
f82d02d9 | 8977 | |
5581a51a | 8978 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
8979 | out: |
8980 | wc->refs[level] = 0; | |
8981 | wc->flags[level] = 0; | |
f0486c68 | 8982 | return 0; |
2c47e605 YZ |
8983 | } |
8984 | ||
8985 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
8986 | struct btrfs_root *root, | |
8987 | struct btrfs_path *path, | |
8988 | struct walk_control *wc) | |
8989 | { | |
2c47e605 | 8990 | int level = wc->level; |
94fcca9f | 8991 | int lookup_info = 1; |
2c47e605 YZ |
8992 | int ret; |
8993 | ||
8994 | while (level >= 0) { | |
94fcca9f | 8995 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
8996 | if (ret > 0) |
8997 | break; | |
8998 | ||
8999 | if (level == 0) | |
9000 | break; | |
9001 | ||
7a7965f8 YZ |
9002 | if (path->slots[level] >= |
9003 | btrfs_header_nritems(path->nodes[level])) | |
9004 | break; | |
9005 | ||
94fcca9f | 9006 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
9007 | if (ret > 0) { |
9008 | path->slots[level]++; | |
9009 | continue; | |
90d2c51d MX |
9010 | } else if (ret < 0) |
9011 | return ret; | |
1c4850e2 | 9012 | level = wc->level; |
f82d02d9 | 9013 | } |
f82d02d9 YZ |
9014 | return 0; |
9015 | } | |
9016 | ||
d397712b | 9017 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 9018 | struct btrfs_root *root, |
f82d02d9 | 9019 | struct btrfs_path *path, |
2c47e605 | 9020 | struct walk_control *wc, int max_level) |
20524f02 | 9021 | { |
2c47e605 | 9022 | int level = wc->level; |
20524f02 | 9023 | int ret; |
9f3a7427 | 9024 | |
2c47e605 YZ |
9025 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
9026 | while (level < max_level && path->nodes[level]) { | |
9027 | wc->level = level; | |
9028 | if (path->slots[level] + 1 < | |
9029 | btrfs_header_nritems(path->nodes[level])) { | |
9030 | path->slots[level]++; | |
20524f02 CM |
9031 | return 0; |
9032 | } else { | |
2c47e605 YZ |
9033 | ret = walk_up_proc(trans, root, path, wc); |
9034 | if (ret > 0) | |
9035 | return 0; | |
bd56b302 | 9036 | |
2c47e605 | 9037 | if (path->locks[level]) { |
bd681513 CM |
9038 | btrfs_tree_unlock_rw(path->nodes[level], |
9039 | path->locks[level]); | |
2c47e605 | 9040 | path->locks[level] = 0; |
f82d02d9 | 9041 | } |
2c47e605 YZ |
9042 | free_extent_buffer(path->nodes[level]); |
9043 | path->nodes[level] = NULL; | |
9044 | level++; | |
20524f02 CM |
9045 | } |
9046 | } | |
9047 | return 1; | |
9048 | } | |
9049 | ||
9aca1d51 | 9050 | /* |
2c47e605 YZ |
9051 | * drop a subvolume tree. |
9052 | * | |
9053 | * this function traverses the tree freeing any blocks that only | |
9054 | * referenced by the tree. | |
9055 | * | |
9056 | * when a shared tree block is found. this function decreases its | |
9057 | * reference count by one. if update_ref is true, this function | |
9058 | * also make sure backrefs for the shared block and all lower level | |
9059 | * blocks are properly updated. | |
9d1a2a3a DS |
9060 | * |
9061 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 9062 | */ |
2c536799 | 9063 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
9064 | struct btrfs_block_rsv *block_rsv, int update_ref, |
9065 | int for_reloc) | |
20524f02 | 9066 | { |
ab8d0fc4 | 9067 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 9068 | struct btrfs_path *path; |
2c47e605 | 9069 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 9070 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 9071 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
9072 | struct walk_control *wc; |
9073 | struct btrfs_key key; | |
9074 | int err = 0; | |
9075 | int ret; | |
9076 | int level; | |
d29a9f62 | 9077 | bool root_dropped = false; |
20524f02 | 9078 | |
ab8d0fc4 | 9079 | btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid); |
1152651a | 9080 | |
5caf2a00 | 9081 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
9082 | if (!path) { |
9083 | err = -ENOMEM; | |
9084 | goto out; | |
9085 | } | |
20524f02 | 9086 | |
2c47e605 | 9087 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
9088 | if (!wc) { |
9089 | btrfs_free_path(path); | |
cb1b69f4 TI |
9090 | err = -ENOMEM; |
9091 | goto out; | |
38a1a919 | 9092 | } |
2c47e605 | 9093 | |
a22285a6 | 9094 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9095 | if (IS_ERR(trans)) { |
9096 | err = PTR_ERR(trans); | |
9097 | goto out_free; | |
9098 | } | |
98d5dc13 | 9099 | |
3fd0a558 YZ |
9100 | if (block_rsv) |
9101 | trans->block_rsv = block_rsv; | |
2c47e605 | 9102 | |
9f3a7427 | 9103 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 9104 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
9105 | path->nodes[level] = btrfs_lock_root_node(root); |
9106 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 9107 | path->slots[level] = 0; |
bd681513 | 9108 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9109 | memset(&wc->update_progress, 0, |
9110 | sizeof(wc->update_progress)); | |
9f3a7427 | 9111 | } else { |
9f3a7427 | 9112 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
9113 | memcpy(&wc->update_progress, &key, |
9114 | sizeof(wc->update_progress)); | |
9115 | ||
6702ed49 | 9116 | level = root_item->drop_level; |
2c47e605 | 9117 | BUG_ON(level == 0); |
6702ed49 | 9118 | path->lowest_level = level; |
2c47e605 YZ |
9119 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
9120 | path->lowest_level = 0; | |
9121 | if (ret < 0) { | |
9122 | err = ret; | |
79787eaa | 9123 | goto out_end_trans; |
9f3a7427 | 9124 | } |
1c4850e2 | 9125 | WARN_ON(ret > 0); |
2c47e605 | 9126 | |
7d9eb12c CM |
9127 | /* |
9128 | * unlock our path, this is safe because only this | |
9129 | * function is allowed to delete this snapshot | |
9130 | */ | |
5d4f98a2 | 9131 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
9132 | |
9133 | level = btrfs_header_level(root->node); | |
9134 | while (1) { | |
9135 | btrfs_tree_lock(path->nodes[level]); | |
9136 | btrfs_set_lock_blocking(path->nodes[level]); | |
fec386ac | 9137 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9138 | |
2ff7e61e | 9139 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 9140 | path->nodes[level]->start, |
3173a18f | 9141 | level, 1, &wc->refs[level], |
2c47e605 | 9142 | &wc->flags[level]); |
79787eaa JM |
9143 | if (ret < 0) { |
9144 | err = ret; | |
9145 | goto out_end_trans; | |
9146 | } | |
2c47e605 YZ |
9147 | BUG_ON(wc->refs[level] == 0); |
9148 | ||
9149 | if (level == root_item->drop_level) | |
9150 | break; | |
9151 | ||
9152 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 9153 | path->locks[level] = 0; |
2c47e605 YZ |
9154 | WARN_ON(wc->refs[level] != 1); |
9155 | level--; | |
9156 | } | |
9f3a7427 | 9157 | } |
2c47e605 YZ |
9158 | |
9159 | wc->level = level; | |
9160 | wc->shared_level = -1; | |
9161 | wc->stage = DROP_REFERENCE; | |
9162 | wc->update_ref = update_ref; | |
9163 | wc->keep_locks = 0; | |
66d7e7f0 | 9164 | wc->for_reloc = for_reloc; |
0b246afa | 9165 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 9166 | |
d397712b | 9167 | while (1) { |
9d1a2a3a | 9168 | |
2c47e605 YZ |
9169 | ret = walk_down_tree(trans, root, path, wc); |
9170 | if (ret < 0) { | |
9171 | err = ret; | |
20524f02 | 9172 | break; |
2c47e605 | 9173 | } |
9aca1d51 | 9174 | |
2c47e605 YZ |
9175 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
9176 | if (ret < 0) { | |
9177 | err = ret; | |
20524f02 | 9178 | break; |
2c47e605 YZ |
9179 | } |
9180 | ||
9181 | if (ret > 0) { | |
9182 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
9183 | break; |
9184 | } | |
2c47e605 YZ |
9185 | |
9186 | if (wc->stage == DROP_REFERENCE) { | |
9187 | level = wc->level; | |
9188 | btrfs_node_key(path->nodes[level], | |
9189 | &root_item->drop_progress, | |
9190 | path->slots[level]); | |
9191 | root_item->drop_level = level; | |
9192 | } | |
9193 | ||
9194 | BUG_ON(wc->level == 0); | |
3a45bb20 | 9195 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 9196 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
9197 | ret = btrfs_update_root(trans, tree_root, |
9198 | &root->root_key, | |
9199 | root_item); | |
79787eaa | 9200 | if (ret) { |
66642832 | 9201 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9202 | err = ret; |
9203 | goto out_end_trans; | |
9204 | } | |
2c47e605 | 9205 | |
3a45bb20 | 9206 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 9207 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
9208 | btrfs_debug(fs_info, |
9209 | "drop snapshot early exit"); | |
3c8f2422 JB |
9210 | err = -EAGAIN; |
9211 | goto out_free; | |
9212 | } | |
9213 | ||
a22285a6 | 9214 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9215 | if (IS_ERR(trans)) { |
9216 | err = PTR_ERR(trans); | |
9217 | goto out_free; | |
9218 | } | |
3fd0a558 YZ |
9219 | if (block_rsv) |
9220 | trans->block_rsv = block_rsv; | |
c3e69d58 | 9221 | } |
20524f02 | 9222 | } |
b3b4aa74 | 9223 | btrfs_release_path(path); |
79787eaa JM |
9224 | if (err) |
9225 | goto out_end_trans; | |
2c47e605 | 9226 | |
1cd5447e | 9227 | ret = btrfs_del_root(trans, fs_info, &root->root_key); |
79787eaa | 9228 | if (ret) { |
66642832 | 9229 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 9230 | err = ret; |
79787eaa JM |
9231 | goto out_end_trans; |
9232 | } | |
2c47e605 | 9233 | |
76dda93c | 9234 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
9235 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
9236 | NULL, NULL); | |
79787eaa | 9237 | if (ret < 0) { |
66642832 | 9238 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9239 | err = ret; |
9240 | goto out_end_trans; | |
9241 | } else if (ret > 0) { | |
84cd948c JB |
9242 | /* if we fail to delete the orphan item this time |
9243 | * around, it'll get picked up the next time. | |
9244 | * | |
9245 | * The most common failure here is just -ENOENT. | |
9246 | */ | |
9247 | btrfs_del_orphan_item(trans, tree_root, | |
9248 | root->root_key.objectid); | |
76dda93c YZ |
9249 | } |
9250 | } | |
9251 | ||
27cdeb70 | 9252 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 9253 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
9254 | } else { |
9255 | free_extent_buffer(root->node); | |
9256 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 9257 | btrfs_put_fs_root(root); |
76dda93c | 9258 | } |
d29a9f62 | 9259 | root_dropped = true; |
79787eaa | 9260 | out_end_trans: |
3a45bb20 | 9261 | btrfs_end_transaction_throttle(trans); |
79787eaa | 9262 | out_free: |
2c47e605 | 9263 | kfree(wc); |
5caf2a00 | 9264 | btrfs_free_path(path); |
cb1b69f4 | 9265 | out: |
d29a9f62 JB |
9266 | /* |
9267 | * So if we need to stop dropping the snapshot for whatever reason we | |
9268 | * need to make sure to add it back to the dead root list so that we | |
9269 | * keep trying to do the work later. This also cleans up roots if we | |
9270 | * don't have it in the radix (like when we recover after a power fail | |
9271 | * or unmount) so we don't leak memory. | |
9272 | */ | |
897ca819 | 9273 | if (!for_reloc && !root_dropped) |
d29a9f62 | 9274 | btrfs_add_dead_root(root); |
90515e7f | 9275 | if (err && err != -EAGAIN) |
ab8d0fc4 | 9276 | btrfs_handle_fs_error(fs_info, err, NULL); |
2c536799 | 9277 | return err; |
20524f02 | 9278 | } |
9078a3e1 | 9279 | |
2c47e605 YZ |
9280 | /* |
9281 | * drop subtree rooted at tree block 'node'. | |
9282 | * | |
9283 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 9284 | * only used by relocation code |
2c47e605 | 9285 | */ |
f82d02d9 YZ |
9286 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
9287 | struct btrfs_root *root, | |
9288 | struct extent_buffer *node, | |
9289 | struct extent_buffer *parent) | |
9290 | { | |
0b246afa | 9291 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 9292 | struct btrfs_path *path; |
2c47e605 | 9293 | struct walk_control *wc; |
f82d02d9 YZ |
9294 | int level; |
9295 | int parent_level; | |
9296 | int ret = 0; | |
9297 | int wret; | |
9298 | ||
2c47e605 YZ |
9299 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
9300 | ||
f82d02d9 | 9301 | path = btrfs_alloc_path(); |
db5b493a TI |
9302 | if (!path) |
9303 | return -ENOMEM; | |
f82d02d9 | 9304 | |
2c47e605 | 9305 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
9306 | if (!wc) { |
9307 | btrfs_free_path(path); | |
9308 | return -ENOMEM; | |
9309 | } | |
2c47e605 | 9310 | |
b9447ef8 | 9311 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
9312 | parent_level = btrfs_header_level(parent); |
9313 | extent_buffer_get(parent); | |
9314 | path->nodes[parent_level] = parent; | |
9315 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9316 | ||
b9447ef8 | 9317 | btrfs_assert_tree_locked(node); |
f82d02d9 | 9318 | level = btrfs_header_level(node); |
f82d02d9 YZ |
9319 | path->nodes[level] = node; |
9320 | path->slots[level] = 0; | |
bd681513 | 9321 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9322 | |
9323 | wc->refs[parent_level] = 1; | |
9324 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9325 | wc->level = level; | |
9326 | wc->shared_level = -1; | |
9327 | wc->stage = DROP_REFERENCE; | |
9328 | wc->update_ref = 0; | |
9329 | wc->keep_locks = 1; | |
66d7e7f0 | 9330 | wc->for_reloc = 1; |
0b246afa | 9331 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
9332 | |
9333 | while (1) { | |
2c47e605 YZ |
9334 | wret = walk_down_tree(trans, root, path, wc); |
9335 | if (wret < 0) { | |
f82d02d9 | 9336 | ret = wret; |
f82d02d9 | 9337 | break; |
2c47e605 | 9338 | } |
f82d02d9 | 9339 | |
2c47e605 | 9340 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9341 | if (wret < 0) |
9342 | ret = wret; | |
9343 | if (wret != 0) | |
9344 | break; | |
9345 | } | |
9346 | ||
2c47e605 | 9347 | kfree(wc); |
f82d02d9 YZ |
9348 | btrfs_free_path(path); |
9349 | return ret; | |
9350 | } | |
9351 | ||
6202df69 | 9352 | static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c CM |
9353 | { |
9354 | u64 num_devices; | |
fc67c450 | 9355 | u64 stripped; |
e4d8ec0f | 9356 | |
fc67c450 ID |
9357 | /* |
9358 | * if restripe for this chunk_type is on pick target profile and | |
9359 | * return, otherwise do the usual balance | |
9360 | */ | |
6202df69 | 9361 | stripped = get_restripe_target(fs_info, flags); |
fc67c450 ID |
9362 | if (stripped) |
9363 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9364 | |
6202df69 | 9365 | num_devices = fs_info->fs_devices->rw_devices; |
cd02dca5 | 9366 | |
fc67c450 | 9367 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 9368 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
9369 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
9370 | ||
ec44a35c CM |
9371 | if (num_devices == 1) { |
9372 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9373 | stripped = flags & ~stripped; | |
9374 | ||
9375 | /* turn raid0 into single device chunks */ | |
9376 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9377 | return stripped; | |
9378 | ||
9379 | /* turn mirroring into duplication */ | |
9380 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9381 | BTRFS_BLOCK_GROUP_RAID10)) | |
9382 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9383 | } else { |
9384 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9385 | if (flags & stripped) |
9386 | return flags; | |
9387 | ||
9388 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9389 | stripped = flags & ~stripped; | |
9390 | ||
9391 | /* switch duplicated blocks with raid1 */ | |
9392 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9393 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9394 | ||
e3176ca2 | 9395 | /* this is drive concat, leave it alone */ |
ec44a35c | 9396 | } |
e3176ca2 | 9397 | |
ec44a35c CM |
9398 | return flags; |
9399 | } | |
9400 | ||
868f401a | 9401 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9402 | { |
f0486c68 YZ |
9403 | struct btrfs_space_info *sinfo = cache->space_info; |
9404 | u64 num_bytes; | |
199c36ea | 9405 | u64 min_allocable_bytes; |
f0486c68 | 9406 | int ret = -ENOSPC; |
0ef3e66b | 9407 | |
199c36ea MX |
9408 | /* |
9409 | * We need some metadata space and system metadata space for | |
9410 | * allocating chunks in some corner cases until we force to set | |
9411 | * it to be readonly. | |
9412 | */ | |
9413 | if ((sinfo->flags & | |
9414 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9415 | !force) | |
ee22184b | 9416 | min_allocable_bytes = SZ_1M; |
199c36ea MX |
9417 | else |
9418 | min_allocable_bytes = 0; | |
9419 | ||
f0486c68 YZ |
9420 | spin_lock(&sinfo->lock); |
9421 | spin_lock(&cache->lock); | |
61cfea9b W |
9422 | |
9423 | if (cache->ro) { | |
868f401a | 9424 | cache->ro++; |
61cfea9b W |
9425 | ret = 0; |
9426 | goto out; | |
9427 | } | |
9428 | ||
f0486c68 YZ |
9429 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9430 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
9431 | ||
4136135b | 9432 | if (btrfs_space_info_used(sinfo, true) + num_bytes + |
37be25bc | 9433 | min_allocable_bytes <= sinfo->total_bytes) { |
f0486c68 | 9434 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9435 | cache->ro++; |
633c0aad | 9436 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9437 | ret = 0; |
9438 | } | |
61cfea9b | 9439 | out: |
f0486c68 YZ |
9440 | spin_unlock(&cache->lock); |
9441 | spin_unlock(&sinfo->lock); | |
9442 | return ret; | |
9443 | } | |
7d9eb12c | 9444 | |
5e00f193 | 9445 | int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info, |
f0486c68 | 9446 | struct btrfs_block_group_cache *cache) |
c286ac48 | 9447 | |
f0486c68 YZ |
9448 | { |
9449 | struct btrfs_trans_handle *trans; | |
9450 | u64 alloc_flags; | |
9451 | int ret; | |
7d9eb12c | 9452 | |
1bbc621e | 9453 | again: |
5e00f193 | 9454 | trans = btrfs_join_transaction(fs_info->extent_root); |
79787eaa JM |
9455 | if (IS_ERR(trans)) |
9456 | return PTR_ERR(trans); | |
5d4f98a2 | 9457 | |
1bbc621e CM |
9458 | /* |
9459 | * we're not allowed to set block groups readonly after the dirty | |
9460 | * block groups cache has started writing. If it already started, | |
9461 | * back off and let this transaction commit | |
9462 | */ | |
0b246afa | 9463 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c | 9464 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9465 | u64 transid = trans->transid; |
9466 | ||
0b246afa | 9467 | mutex_unlock(&fs_info->ro_block_group_mutex); |
3a45bb20 | 9468 | btrfs_end_transaction(trans); |
1bbc621e | 9469 | |
2ff7e61e | 9470 | ret = btrfs_wait_for_commit(fs_info, transid); |
1bbc621e CM |
9471 | if (ret) |
9472 | return ret; | |
9473 | goto again; | |
9474 | } | |
9475 | ||
153c35b6 CM |
9476 | /* |
9477 | * if we are changing raid levels, try to allocate a corresponding | |
9478 | * block group with the new raid level. | |
9479 | */ | |
0b246afa | 9480 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
153c35b6 | 9481 | if (alloc_flags != cache->flags) { |
2ff7e61e | 9482 | ret = do_chunk_alloc(trans, fs_info, alloc_flags, |
153c35b6 CM |
9483 | CHUNK_ALLOC_FORCE); |
9484 | /* | |
9485 | * ENOSPC is allowed here, we may have enough space | |
9486 | * already allocated at the new raid level to | |
9487 | * carry on | |
9488 | */ | |
9489 | if (ret == -ENOSPC) | |
9490 | ret = 0; | |
9491 | if (ret < 0) | |
9492 | goto out; | |
9493 | } | |
1bbc621e | 9494 | |
868f401a | 9495 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9496 | if (!ret) |
9497 | goto out; | |
2ff7e61e JM |
9498 | alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags); |
9499 | ret = do_chunk_alloc(trans, fs_info, alloc_flags, | |
0e4f8f88 | 9500 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9501 | if (ret < 0) |
9502 | goto out; | |
868f401a | 9503 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9504 | out: |
2f081088 | 9505 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 9506 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
34441361 | 9507 | mutex_lock(&fs_info->chunk_mutex); |
2ff7e61e | 9508 | check_system_chunk(trans, fs_info, alloc_flags); |
34441361 | 9509 | mutex_unlock(&fs_info->chunk_mutex); |
2f081088 | 9510 | } |
0b246afa | 9511 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2f081088 | 9512 | |
3a45bb20 | 9513 | btrfs_end_transaction(trans); |
f0486c68 YZ |
9514 | return ret; |
9515 | } | |
5d4f98a2 | 9516 | |
c87f08ca | 9517 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 9518 | struct btrfs_fs_info *fs_info, u64 type) |
c87f08ca | 9519 | { |
2ff7e61e JM |
9520 | u64 alloc_flags = get_alloc_profile(fs_info, type); |
9521 | ||
9522 | return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE); | |
c87f08ca CM |
9523 | } |
9524 | ||
6d07bcec MX |
9525 | /* |
9526 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9527 | * space_info. takes mirrors into account. |
6d07bcec | 9528 | */ |
633c0aad | 9529 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9530 | { |
9531 | struct btrfs_block_group_cache *block_group; | |
9532 | u64 free_bytes = 0; | |
9533 | int factor; | |
9534 | ||
01327610 | 9535 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
9536 | if (list_empty(&sinfo->ro_bgs)) |
9537 | return 0; | |
9538 | ||
9539 | spin_lock(&sinfo->lock); | |
9540 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9541 | spin_lock(&block_group->lock); |
9542 | ||
9543 | if (!block_group->ro) { | |
9544 | spin_unlock(&block_group->lock); | |
9545 | continue; | |
9546 | } | |
9547 | ||
9548 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9549 | BTRFS_BLOCK_GROUP_RAID10 | | |
9550 | BTRFS_BLOCK_GROUP_DUP)) | |
9551 | factor = 2; | |
9552 | else | |
9553 | factor = 1; | |
9554 | ||
9555 | free_bytes += (block_group->key.offset - | |
9556 | btrfs_block_group_used(&block_group->item)) * | |
9557 | factor; | |
9558 | ||
9559 | spin_unlock(&block_group->lock); | |
9560 | } | |
6d07bcec MX |
9561 | spin_unlock(&sinfo->lock); |
9562 | ||
9563 | return free_bytes; | |
9564 | } | |
9565 | ||
2ff7e61e | 9566 | void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9567 | { |
f0486c68 YZ |
9568 | struct btrfs_space_info *sinfo = cache->space_info; |
9569 | u64 num_bytes; | |
9570 | ||
9571 | BUG_ON(!cache->ro); | |
9572 | ||
9573 | spin_lock(&sinfo->lock); | |
9574 | spin_lock(&cache->lock); | |
868f401a Z |
9575 | if (!--cache->ro) { |
9576 | num_bytes = cache->key.offset - cache->reserved - | |
9577 | cache->pinned - cache->bytes_super - | |
9578 | btrfs_block_group_used(&cache->item); | |
9579 | sinfo->bytes_readonly -= num_bytes; | |
9580 | list_del_init(&cache->ro_list); | |
9581 | } | |
f0486c68 YZ |
9582 | spin_unlock(&cache->lock); |
9583 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9584 | } |
9585 | ||
ba1bf481 JB |
9586 | /* |
9587 | * checks to see if its even possible to relocate this block group. | |
9588 | * | |
9589 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9590 | * ok to go ahead and try. | |
9591 | */ | |
6bccf3ab | 9592 | int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr) |
1a40e23b | 9593 | { |
6bccf3ab | 9594 | struct btrfs_root *root = fs_info->extent_root; |
ba1bf481 JB |
9595 | struct btrfs_block_group_cache *block_group; |
9596 | struct btrfs_space_info *space_info; | |
0b246afa | 9597 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
ba1bf481 | 9598 | struct btrfs_device *device; |
6df9a95e | 9599 | struct btrfs_trans_handle *trans; |
cdcb725c | 9600 | u64 min_free; |
6719db6a JB |
9601 | u64 dev_min = 1; |
9602 | u64 dev_nr = 0; | |
4a5e98f5 | 9603 | u64 target; |
0305bc27 | 9604 | int debug; |
cdcb725c | 9605 | int index; |
ba1bf481 JB |
9606 | int full = 0; |
9607 | int ret = 0; | |
1a40e23b | 9608 | |
0b246afa | 9609 | debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG); |
0305bc27 | 9610 | |
0b246afa | 9611 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
1a40e23b | 9612 | |
ba1bf481 | 9613 | /* odd, couldn't find the block group, leave it alone */ |
0305bc27 QW |
9614 | if (!block_group) { |
9615 | if (debug) | |
0b246afa | 9616 | btrfs_warn(fs_info, |
0305bc27 QW |
9617 | "can't find block group for bytenr %llu", |
9618 | bytenr); | |
ba1bf481 | 9619 | return -1; |
0305bc27 | 9620 | } |
1a40e23b | 9621 | |
cdcb725c | 9622 | min_free = btrfs_block_group_used(&block_group->item); |
9623 | ||
ba1bf481 | 9624 | /* no bytes used, we're good */ |
cdcb725c | 9625 | if (!min_free) |
1a40e23b ZY |
9626 | goto out; |
9627 | ||
ba1bf481 JB |
9628 | space_info = block_group->space_info; |
9629 | spin_lock(&space_info->lock); | |
17d217fe | 9630 | |
ba1bf481 | 9631 | full = space_info->full; |
17d217fe | 9632 | |
ba1bf481 JB |
9633 | /* |
9634 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9635 | * relocate it unless we're able to allocate a new chunk below. |
9636 | * | |
9637 | * Otherwise, we need to make sure we have room in the space to handle | |
9638 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9639 | */ |
7ce618db | 9640 | if ((space_info->total_bytes != block_group->key.offset) && |
4136135b LB |
9641 | (btrfs_space_info_used(space_info, false) + min_free < |
9642 | space_info->total_bytes)) { | |
ba1bf481 JB |
9643 | spin_unlock(&space_info->lock); |
9644 | goto out; | |
17d217fe | 9645 | } |
ba1bf481 | 9646 | spin_unlock(&space_info->lock); |
ea8c2819 | 9647 | |
ba1bf481 JB |
9648 | /* |
9649 | * ok we don't have enough space, but maybe we have free space on our | |
9650 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9651 | * alloc devices and guess if we have enough space. if this block |
9652 | * group is going to be restriped, run checks against the target | |
9653 | * profile instead of the current one. | |
ba1bf481 JB |
9654 | */ |
9655 | ret = -1; | |
ea8c2819 | 9656 | |
cdcb725c | 9657 | /* |
9658 | * index: | |
9659 | * 0: raid10 | |
9660 | * 1: raid1 | |
9661 | * 2: dup | |
9662 | * 3: raid0 | |
9663 | * 4: single | |
9664 | */ | |
0b246afa | 9665 | target = get_restripe_target(fs_info, block_group->flags); |
4a5e98f5 | 9666 | if (target) { |
3e72ee88 | 9667 | index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9668 | } else { |
9669 | /* | |
9670 | * this is just a balance, so if we were marked as full | |
9671 | * we know there is no space for a new chunk | |
9672 | */ | |
0305bc27 QW |
9673 | if (full) { |
9674 | if (debug) | |
0b246afa JM |
9675 | btrfs_warn(fs_info, |
9676 | "no space to alloc new chunk for block group %llu", | |
9677 | block_group->key.objectid); | |
4a5e98f5 | 9678 | goto out; |
0305bc27 | 9679 | } |
4a5e98f5 | 9680 | |
3e72ee88 | 9681 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
4a5e98f5 ID |
9682 | } |
9683 | ||
e6ec716f | 9684 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9685 | dev_min = 4; |
6719db6a JB |
9686 | /* Divide by 2 */ |
9687 | min_free >>= 1; | |
e6ec716f | 9688 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9689 | dev_min = 2; |
e6ec716f | 9690 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9691 | /* Multiply by 2 */ |
9692 | min_free <<= 1; | |
e6ec716f | 9693 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9694 | dev_min = fs_devices->rw_devices; |
47c5713f | 9695 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9696 | } |
9697 | ||
6df9a95e JB |
9698 | /* We need to do this so that we can look at pending chunks */ |
9699 | trans = btrfs_join_transaction(root); | |
9700 | if (IS_ERR(trans)) { | |
9701 | ret = PTR_ERR(trans); | |
9702 | goto out; | |
9703 | } | |
9704 | ||
0b246afa | 9705 | mutex_lock(&fs_info->chunk_mutex); |
ba1bf481 | 9706 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
7bfc837d | 9707 | u64 dev_offset; |
56bec294 | 9708 | |
ba1bf481 JB |
9709 | /* |
9710 | * check to make sure we can actually find a chunk with enough | |
9711 | * space to fit our block group in. | |
9712 | */ | |
63a212ab | 9713 | if (device->total_bytes > device->bytes_used + min_free && |
401e29c1 | 9714 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
6df9a95e | 9715 | ret = find_free_dev_extent(trans, device, min_free, |
7bfc837d | 9716 | &dev_offset, NULL); |
ba1bf481 | 9717 | if (!ret) |
cdcb725c | 9718 | dev_nr++; |
9719 | ||
9720 | if (dev_nr >= dev_min) | |
73e48b27 | 9721 | break; |
cdcb725c | 9722 | |
ba1bf481 | 9723 | ret = -1; |
725c8463 | 9724 | } |
edbd8d4e | 9725 | } |
0305bc27 | 9726 | if (debug && ret == -1) |
0b246afa JM |
9727 | btrfs_warn(fs_info, |
9728 | "no space to allocate a new chunk for block group %llu", | |
9729 | block_group->key.objectid); | |
9730 | mutex_unlock(&fs_info->chunk_mutex); | |
3a45bb20 | 9731 | btrfs_end_transaction(trans); |
edbd8d4e | 9732 | out: |
ba1bf481 | 9733 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9734 | return ret; |
9735 | } | |
9736 | ||
6bccf3ab JM |
9737 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
9738 | struct btrfs_path *path, | |
9739 | struct btrfs_key *key) | |
0b86a832 | 9740 | { |
6bccf3ab | 9741 | struct btrfs_root *root = fs_info->extent_root; |
925baedd | 9742 | int ret = 0; |
0b86a832 CM |
9743 | struct btrfs_key found_key; |
9744 | struct extent_buffer *leaf; | |
9745 | int slot; | |
edbd8d4e | 9746 | |
0b86a832 CM |
9747 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9748 | if (ret < 0) | |
925baedd CM |
9749 | goto out; |
9750 | ||
d397712b | 9751 | while (1) { |
0b86a832 | 9752 | slot = path->slots[0]; |
edbd8d4e | 9753 | leaf = path->nodes[0]; |
0b86a832 CM |
9754 | if (slot >= btrfs_header_nritems(leaf)) { |
9755 | ret = btrfs_next_leaf(root, path); | |
9756 | if (ret == 0) | |
9757 | continue; | |
9758 | if (ret < 0) | |
925baedd | 9759 | goto out; |
0b86a832 | 9760 | break; |
edbd8d4e | 9761 | } |
0b86a832 | 9762 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9763 | |
0b86a832 | 9764 | if (found_key.objectid >= key->objectid && |
925baedd | 9765 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6fb37b75 LB |
9766 | struct extent_map_tree *em_tree; |
9767 | struct extent_map *em; | |
9768 | ||
9769 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
9770 | read_lock(&em_tree->lock); | |
9771 | em = lookup_extent_mapping(em_tree, found_key.objectid, | |
9772 | found_key.offset); | |
9773 | read_unlock(&em_tree->lock); | |
9774 | if (!em) { | |
0b246afa | 9775 | btrfs_err(fs_info, |
6fb37b75 LB |
9776 | "logical %llu len %llu found bg but no related chunk", |
9777 | found_key.objectid, found_key.offset); | |
9778 | ret = -ENOENT; | |
9779 | } else { | |
9780 | ret = 0; | |
9781 | } | |
187ee58c | 9782 | free_extent_map(em); |
925baedd CM |
9783 | goto out; |
9784 | } | |
0b86a832 | 9785 | path->slots[0]++; |
edbd8d4e | 9786 | } |
925baedd | 9787 | out: |
0b86a832 | 9788 | return ret; |
edbd8d4e CM |
9789 | } |
9790 | ||
0af3d00b JB |
9791 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
9792 | { | |
9793 | struct btrfs_block_group_cache *block_group; | |
9794 | u64 last = 0; | |
9795 | ||
9796 | while (1) { | |
9797 | struct inode *inode; | |
9798 | ||
9799 | block_group = btrfs_lookup_first_block_group(info, last); | |
9800 | while (block_group) { | |
9801 | spin_lock(&block_group->lock); | |
9802 | if (block_group->iref) | |
9803 | break; | |
9804 | spin_unlock(&block_group->lock); | |
2ff7e61e | 9805 | block_group = next_block_group(info, block_group); |
0af3d00b JB |
9806 | } |
9807 | if (!block_group) { | |
9808 | if (last == 0) | |
9809 | break; | |
9810 | last = 0; | |
9811 | continue; | |
9812 | } | |
9813 | ||
9814 | inode = block_group->inode; | |
9815 | block_group->iref = 0; | |
9816 | block_group->inode = NULL; | |
9817 | spin_unlock(&block_group->lock); | |
f3bca802 | 9818 | ASSERT(block_group->io_ctl.inode == NULL); |
0af3d00b JB |
9819 | iput(inode); |
9820 | last = block_group->key.objectid + block_group->key.offset; | |
9821 | btrfs_put_block_group(block_group); | |
9822 | } | |
9823 | } | |
9824 | ||
5cdd7db6 FM |
9825 | /* |
9826 | * Must be called only after stopping all workers, since we could have block | |
9827 | * group caching kthreads running, and therefore they could race with us if we | |
9828 | * freed the block groups before stopping them. | |
9829 | */ | |
1a40e23b ZY |
9830 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9831 | { | |
9832 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 9833 | struct btrfs_space_info *space_info; |
11833d66 | 9834 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
9835 | struct rb_node *n; |
9836 | ||
9e351cc8 | 9837 | down_write(&info->commit_root_sem); |
11833d66 YZ |
9838 | while (!list_empty(&info->caching_block_groups)) { |
9839 | caching_ctl = list_entry(info->caching_block_groups.next, | |
9840 | struct btrfs_caching_control, list); | |
9841 | list_del(&caching_ctl->list); | |
9842 | put_caching_control(caching_ctl); | |
9843 | } | |
9e351cc8 | 9844 | up_write(&info->commit_root_sem); |
11833d66 | 9845 | |
47ab2a6c JB |
9846 | spin_lock(&info->unused_bgs_lock); |
9847 | while (!list_empty(&info->unused_bgs)) { | |
9848 | block_group = list_first_entry(&info->unused_bgs, | |
9849 | struct btrfs_block_group_cache, | |
9850 | bg_list); | |
9851 | list_del_init(&block_group->bg_list); | |
9852 | btrfs_put_block_group(block_group); | |
9853 | } | |
9854 | spin_unlock(&info->unused_bgs_lock); | |
9855 | ||
1a40e23b ZY |
9856 | spin_lock(&info->block_group_cache_lock); |
9857 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
9858 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
9859 | cache_node); | |
1a40e23b ZY |
9860 | rb_erase(&block_group->cache_node, |
9861 | &info->block_group_cache_tree); | |
01eacb27 | 9862 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
9863 | spin_unlock(&info->block_group_cache_lock); |
9864 | ||
80eb234a | 9865 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 9866 | list_del(&block_group->list); |
80eb234a | 9867 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 9868 | |
3c14874a JB |
9869 | /* |
9870 | * We haven't cached this block group, which means we could | |
9871 | * possibly have excluded extents on this block group. | |
9872 | */ | |
36cce922 JB |
9873 | if (block_group->cached == BTRFS_CACHE_NO || |
9874 | block_group->cached == BTRFS_CACHE_ERROR) | |
2ff7e61e | 9875 | free_excluded_extents(info, block_group); |
3c14874a | 9876 | |
817d52f8 | 9877 | btrfs_remove_free_space_cache(block_group); |
5cdd7db6 | 9878 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); |
f3bca802 LB |
9879 | ASSERT(list_empty(&block_group->dirty_list)); |
9880 | ASSERT(list_empty(&block_group->io_list)); | |
9881 | ASSERT(list_empty(&block_group->bg_list)); | |
9882 | ASSERT(atomic_read(&block_group->count) == 1); | |
11dfe35a | 9883 | btrfs_put_block_group(block_group); |
d899e052 YZ |
9884 | |
9885 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
9886 | } |
9887 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
9888 | |
9889 | /* now that all the block groups are freed, go through and | |
9890 | * free all the space_info structs. This is only called during | |
9891 | * the final stages of unmount, and so we know nobody is | |
9892 | * using them. We call synchronize_rcu() once before we start, | |
9893 | * just to be on the safe side. | |
9894 | */ | |
9895 | synchronize_rcu(); | |
9896 | ||
8929ecfa YZ |
9897 | release_global_block_rsv(info); |
9898 | ||
67871254 | 9899 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
9900 | int i; |
9901 | ||
4184ea7f CM |
9902 | space_info = list_entry(info->space_info.next, |
9903 | struct btrfs_space_info, | |
9904 | list); | |
d555b6c3 JB |
9905 | |
9906 | /* | |
9907 | * Do not hide this behind enospc_debug, this is actually | |
9908 | * important and indicates a real bug if this happens. | |
9909 | */ | |
9910 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
b069e0c3 | 9911 | space_info->bytes_reserved > 0 || |
d555b6c3 | 9912 | space_info->bytes_may_use > 0)) |
ab8d0fc4 | 9913 | dump_space_info(info, space_info, 0, 0); |
4184ea7f | 9914 | list_del(&space_info->list); |
6ab0a202 JM |
9915 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
9916 | struct kobject *kobj; | |
c1895442 JM |
9917 | kobj = space_info->block_group_kobjs[i]; |
9918 | space_info->block_group_kobjs[i] = NULL; | |
9919 | if (kobj) { | |
6ab0a202 JM |
9920 | kobject_del(kobj); |
9921 | kobject_put(kobj); | |
9922 | } | |
9923 | } | |
9924 | kobject_del(&space_info->kobj); | |
9925 | kobject_put(&space_info->kobj); | |
4184ea7f | 9926 | } |
1a40e23b ZY |
9927 | return 0; |
9928 | } | |
9929 | ||
75cb379d JM |
9930 | /* link_block_group will queue up kobjects to add when we're reclaim-safe */ |
9931 | void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info) | |
9932 | { | |
9933 | struct btrfs_space_info *space_info; | |
9934 | struct raid_kobject *rkobj; | |
9935 | LIST_HEAD(list); | |
9936 | int index; | |
9937 | int ret = 0; | |
9938 | ||
9939 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
9940 | list_splice_init(&fs_info->pending_raid_kobjs, &list); | |
9941 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
9942 | ||
9943 | list_for_each_entry(rkobj, &list, list) { | |
9944 | space_info = __find_space_info(fs_info, rkobj->flags); | |
9945 | index = btrfs_bg_flags_to_raid_index(rkobj->flags); | |
9946 | ||
9947 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
9948 | "%s", get_raid_name(index)); | |
9949 | if (ret) { | |
9950 | kobject_put(&rkobj->kobj); | |
9951 | break; | |
9952 | } | |
9953 | } | |
9954 | if (ret) | |
9955 | btrfs_warn(fs_info, | |
9956 | "failed to add kobject for block cache, ignoring"); | |
9957 | } | |
9958 | ||
c434d21c | 9959 | static void link_block_group(struct btrfs_block_group_cache *cache) |
b742bb82 | 9960 | { |
c434d21c | 9961 | struct btrfs_space_info *space_info = cache->space_info; |
75cb379d | 9962 | struct btrfs_fs_info *fs_info = cache->fs_info; |
3e72ee88 | 9963 | int index = btrfs_bg_flags_to_raid_index(cache->flags); |
ed55b6ac | 9964 | bool first = false; |
b742bb82 YZ |
9965 | |
9966 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
9967 | if (list_empty(&space_info->block_groups[index])) |
9968 | first = true; | |
9969 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
9970 | up_write(&space_info->groups_sem); | |
9971 | ||
9972 | if (first) { | |
75cb379d JM |
9973 | struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
9974 | if (!rkobj) { | |
9975 | btrfs_warn(cache->fs_info, | |
9976 | "couldn't alloc memory for raid level kobject"); | |
9977 | return; | |
6ab0a202 | 9978 | } |
75cb379d JM |
9979 | rkobj->flags = cache->flags; |
9980 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
9981 | ||
9982 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
9983 | list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs); | |
9984 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
c1895442 | 9985 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 9986 | } |
b742bb82 YZ |
9987 | } |
9988 | ||
920e4a58 | 9989 | static struct btrfs_block_group_cache * |
2ff7e61e JM |
9990 | btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info, |
9991 | u64 start, u64 size) | |
920e4a58 MX |
9992 | { |
9993 | struct btrfs_block_group_cache *cache; | |
9994 | ||
9995 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
9996 | if (!cache) | |
9997 | return NULL; | |
9998 | ||
9999 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
10000 | GFP_NOFS); | |
10001 | if (!cache->free_space_ctl) { | |
10002 | kfree(cache); | |
10003 | return NULL; | |
10004 | } | |
10005 | ||
10006 | cache->key.objectid = start; | |
10007 | cache->key.offset = size; | |
10008 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
10009 | ||
0b246afa | 10010 | cache->fs_info = fs_info; |
e4ff5fb5 | 10011 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); |
1e144fb8 OS |
10012 | set_free_space_tree_thresholds(cache); |
10013 | ||
920e4a58 MX |
10014 | atomic_set(&cache->count, 1); |
10015 | spin_lock_init(&cache->lock); | |
e570fd27 | 10016 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
10017 | INIT_LIST_HEAD(&cache->list); |
10018 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 10019 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 10020 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 10021 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 10022 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 10023 | btrfs_init_free_space_ctl(cache); |
04216820 | 10024 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 10025 | mutex_init(&cache->free_space_lock); |
0966a7b1 | 10026 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); |
920e4a58 MX |
10027 | |
10028 | return cache; | |
10029 | } | |
10030 | ||
5b4aacef | 10031 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
9078a3e1 CM |
10032 | { |
10033 | struct btrfs_path *path; | |
10034 | int ret; | |
9078a3e1 | 10035 | struct btrfs_block_group_cache *cache; |
6324fbf3 | 10036 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
10037 | struct btrfs_key key; |
10038 | struct btrfs_key found_key; | |
5f39d397 | 10039 | struct extent_buffer *leaf; |
0af3d00b JB |
10040 | int need_clear = 0; |
10041 | u64 cache_gen; | |
49303381 LB |
10042 | u64 feature; |
10043 | int mixed; | |
10044 | ||
10045 | feature = btrfs_super_incompat_flags(info->super_copy); | |
10046 | mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS); | |
96b5179d | 10047 | |
9078a3e1 | 10048 | key.objectid = 0; |
0b86a832 | 10049 | key.offset = 0; |
962a298f | 10050 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
10051 | path = btrfs_alloc_path(); |
10052 | if (!path) | |
10053 | return -ENOMEM; | |
e4058b54 | 10054 | path->reada = READA_FORWARD; |
9078a3e1 | 10055 | |
0b246afa JM |
10056 | cache_gen = btrfs_super_cache_generation(info->super_copy); |
10057 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
10058 | btrfs_super_generation(info->super_copy) != cache_gen) | |
0af3d00b | 10059 | need_clear = 1; |
0b246afa | 10060 | if (btrfs_test_opt(info, CLEAR_CACHE)) |
88c2ba3b | 10061 | need_clear = 1; |
0af3d00b | 10062 | |
d397712b | 10063 | while (1) { |
6bccf3ab | 10064 | ret = find_first_block_group(info, path, &key); |
b742bb82 YZ |
10065 | if (ret > 0) |
10066 | break; | |
0b86a832 CM |
10067 | if (ret != 0) |
10068 | goto error; | |
920e4a58 | 10069 | |
5f39d397 CM |
10070 | leaf = path->nodes[0]; |
10071 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 | 10072 | |
2ff7e61e | 10073 | cache = btrfs_create_block_group_cache(info, found_key.objectid, |
920e4a58 | 10074 | found_key.offset); |
9078a3e1 | 10075 | if (!cache) { |
0b86a832 | 10076 | ret = -ENOMEM; |
f0486c68 | 10077 | goto error; |
9078a3e1 | 10078 | } |
96303081 | 10079 | |
cf7c1ef6 LB |
10080 | if (need_clear) { |
10081 | /* | |
10082 | * When we mount with old space cache, we need to | |
10083 | * set BTRFS_DC_CLEAR and set dirty flag. | |
10084 | * | |
10085 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
10086 | * truncate the old free space cache inode and | |
10087 | * setup a new one. | |
10088 | * b) Setting 'dirty flag' makes sure that we flush | |
10089 | * the new space cache info onto disk. | |
10090 | */ | |
0b246afa | 10091 | if (btrfs_test_opt(info, SPACE_CACHE)) |
ce93ec54 | 10092 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 10093 | } |
0af3d00b | 10094 | |
5f39d397 CM |
10095 | read_extent_buffer(leaf, &cache->item, |
10096 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
10097 | sizeof(cache->item)); | |
920e4a58 | 10098 | cache->flags = btrfs_block_group_flags(&cache->item); |
49303381 LB |
10099 | if (!mixed && |
10100 | ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && | |
10101 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
10102 | btrfs_err(info, | |
10103 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
10104 | cache->key.objectid); | |
10105 | ret = -EINVAL; | |
10106 | goto error; | |
10107 | } | |
0b86a832 | 10108 | |
9078a3e1 | 10109 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 10110 | btrfs_release_path(path); |
34d52cb6 | 10111 | |
3c14874a JB |
10112 | /* |
10113 | * We need to exclude the super stripes now so that the space | |
10114 | * info has super bytes accounted for, otherwise we'll think | |
10115 | * we have more space than we actually do. | |
10116 | */ | |
2ff7e61e | 10117 | ret = exclude_super_stripes(info, cache); |
835d974f JB |
10118 | if (ret) { |
10119 | /* | |
10120 | * We may have excluded something, so call this just in | |
10121 | * case. | |
10122 | */ | |
2ff7e61e | 10123 | free_excluded_extents(info, cache); |
920e4a58 | 10124 | btrfs_put_block_group(cache); |
835d974f JB |
10125 | goto error; |
10126 | } | |
3c14874a | 10127 | |
817d52f8 JB |
10128 | /* |
10129 | * check for two cases, either we are full, and therefore | |
10130 | * don't need to bother with the caching work since we won't | |
10131 | * find any space, or we are empty, and we can just add all | |
10132 | * the space in and be done with it. This saves us _alot_ of | |
10133 | * time, particularly in the full case. | |
10134 | */ | |
10135 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 10136 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10137 | cache->cached = BTRFS_CACHE_FINISHED; |
2ff7e61e | 10138 | free_excluded_extents(info, cache); |
817d52f8 | 10139 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 10140 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10141 | cache->cached = BTRFS_CACHE_FINISHED; |
0b246afa | 10142 | add_new_free_space(cache, info, |
817d52f8 JB |
10143 | found_key.objectid, |
10144 | found_key.objectid + | |
10145 | found_key.offset); | |
2ff7e61e | 10146 | free_excluded_extents(info, cache); |
817d52f8 | 10147 | } |
96b5179d | 10148 | |
0b246afa | 10149 | ret = btrfs_add_block_group_cache(info, cache); |
8c579fe7 JB |
10150 | if (ret) { |
10151 | btrfs_remove_free_space_cache(cache); | |
10152 | btrfs_put_block_group(cache); | |
10153 | goto error; | |
10154 | } | |
10155 | ||
0b246afa | 10156 | trace_btrfs_add_block_group(info, cache, 0); |
d2006e6d NB |
10157 | update_space_info(info, cache->flags, found_key.offset, |
10158 | btrfs_block_group_used(&cache->item), | |
10159 | cache->bytes_super, &space_info); | |
8c579fe7 | 10160 | |
6324fbf3 | 10161 | cache->space_info = space_info; |
1b2da372 | 10162 | |
c434d21c | 10163 | link_block_group(cache); |
0f9dd46c | 10164 | |
0b246afa | 10165 | set_avail_alloc_bits(info, cache->flags); |
2ff7e61e | 10166 | if (btrfs_chunk_readonly(info, cache->key.objectid)) { |
868f401a | 10167 | inc_block_group_ro(cache, 1); |
47ab2a6c JB |
10168 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
10169 | spin_lock(&info->unused_bgs_lock); | |
10170 | /* Should always be true but just in case. */ | |
10171 | if (list_empty(&cache->bg_list)) { | |
10172 | btrfs_get_block_group(cache); | |
10173 | list_add_tail(&cache->bg_list, | |
10174 | &info->unused_bgs); | |
10175 | } | |
10176 | spin_unlock(&info->unused_bgs_lock); | |
10177 | } | |
9078a3e1 | 10178 | } |
b742bb82 | 10179 | |
0b246afa | 10180 | list_for_each_entry_rcu(space_info, &info->space_info, list) { |
2ff7e61e | 10181 | if (!(get_alloc_profile(info, space_info->flags) & |
b742bb82 YZ |
10182 | (BTRFS_BLOCK_GROUP_RAID10 | |
10183 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
10184 | BTRFS_BLOCK_GROUP_RAID5 | |
10185 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
10186 | BTRFS_BLOCK_GROUP_DUP))) |
10187 | continue; | |
10188 | /* | |
10189 | * avoid allocating from un-mirrored block group if there are | |
10190 | * mirrored block groups. | |
10191 | */ | |
1095cc0d | 10192 | list_for_each_entry(cache, |
10193 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
10194 | list) | |
868f401a | 10195 | inc_block_group_ro(cache, 1); |
1095cc0d | 10196 | list_for_each_entry(cache, |
10197 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
10198 | list) | |
868f401a | 10199 | inc_block_group_ro(cache, 1); |
9078a3e1 | 10200 | } |
f0486c68 | 10201 | |
75cb379d | 10202 | btrfs_add_raid_kobjects(info); |
f0486c68 | 10203 | init_global_block_rsv(info); |
0b86a832 CM |
10204 | ret = 0; |
10205 | error: | |
9078a3e1 | 10206 | btrfs_free_path(path); |
0b86a832 | 10207 | return ret; |
9078a3e1 | 10208 | } |
6324fbf3 | 10209 | |
6c686b35 | 10210 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) |
ea658bad | 10211 | { |
6c686b35 | 10212 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ea658bad | 10213 | struct btrfs_block_group_cache *block_group, *tmp; |
0b246afa | 10214 | struct btrfs_root *extent_root = fs_info->extent_root; |
ea658bad JB |
10215 | struct btrfs_block_group_item item; |
10216 | struct btrfs_key key; | |
10217 | int ret = 0; | |
d9a0540a | 10218 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
ea658bad | 10219 | |
d9a0540a | 10220 | trans->can_flush_pending_bgs = false; |
47ab2a6c | 10221 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { |
ea658bad | 10222 | if (ret) |
c92f6be3 | 10223 | goto next; |
ea658bad JB |
10224 | |
10225 | spin_lock(&block_group->lock); | |
10226 | memcpy(&item, &block_group->item, sizeof(item)); | |
10227 | memcpy(&key, &block_group->key, sizeof(key)); | |
10228 | spin_unlock(&block_group->lock); | |
10229 | ||
10230 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
10231 | sizeof(item)); | |
10232 | if (ret) | |
66642832 | 10233 | btrfs_abort_transaction(trans, ret); |
0b246afa JM |
10234 | ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid, |
10235 | key.offset); | |
6df9a95e | 10236 | if (ret) |
66642832 | 10237 | btrfs_abort_transaction(trans, ret); |
0b246afa | 10238 | add_block_group_free_space(trans, fs_info, block_group); |
1e144fb8 | 10239 | /* already aborted the transaction if it failed. */ |
c92f6be3 FM |
10240 | next: |
10241 | list_del_init(&block_group->bg_list); | |
ea658bad | 10242 | } |
d9a0540a | 10243 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
ea658bad JB |
10244 | } |
10245 | ||
6324fbf3 | 10246 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 10247 | struct btrfs_fs_info *fs_info, u64 bytes_used, |
0174484d | 10248 | u64 type, u64 chunk_offset, u64 size) |
6324fbf3 | 10249 | { |
6324fbf3 | 10250 | struct btrfs_block_group_cache *cache; |
0b246afa | 10251 | int ret; |
6324fbf3 | 10252 | |
0b246afa | 10253 | btrfs_set_log_full_commit(fs_info, trans); |
e02119d5 | 10254 | |
2ff7e61e | 10255 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size); |
0f9dd46c JB |
10256 | if (!cache) |
10257 | return -ENOMEM; | |
34d52cb6 | 10258 | |
6324fbf3 | 10259 | btrfs_set_block_group_used(&cache->item, bytes_used); |
0174484d NB |
10260 | btrfs_set_block_group_chunk_objectid(&cache->item, |
10261 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
6324fbf3 CM |
10262 | btrfs_set_block_group_flags(&cache->item, type); |
10263 | ||
920e4a58 | 10264 | cache->flags = type; |
11833d66 | 10265 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10266 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 10267 | cache->needs_free_space = 1; |
2ff7e61e | 10268 | ret = exclude_super_stripes(fs_info, cache); |
835d974f JB |
10269 | if (ret) { |
10270 | /* | |
10271 | * We may have excluded something, so call this just in | |
10272 | * case. | |
10273 | */ | |
2ff7e61e | 10274 | free_excluded_extents(fs_info, cache); |
920e4a58 | 10275 | btrfs_put_block_group(cache); |
835d974f JB |
10276 | return ret; |
10277 | } | |
96303081 | 10278 | |
0b246afa | 10279 | add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size); |
817d52f8 | 10280 | |
2ff7e61e | 10281 | free_excluded_extents(fs_info, cache); |
11833d66 | 10282 | |
d0bd4560 | 10283 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 10284 | if (btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
10285 | u64 new_bytes_used = size - bytes_used; |
10286 | ||
10287 | bytes_used += new_bytes_used >> 1; | |
2ff7e61e | 10288 | fragment_free_space(cache); |
d0bd4560 JB |
10289 | } |
10290 | #endif | |
2e6e5183 | 10291 | /* |
2be12ef7 NB |
10292 | * Ensure the corresponding space_info object is created and |
10293 | * assigned to our block group. We want our bg to be added to the rbtree | |
10294 | * with its ->space_info set. | |
2e6e5183 | 10295 | */ |
2be12ef7 | 10296 | cache->space_info = __find_space_info(fs_info, cache->flags); |
dc2d3005 | 10297 | ASSERT(cache->space_info); |
2e6e5183 | 10298 | |
0b246afa | 10299 | ret = btrfs_add_block_group_cache(fs_info, cache); |
8c579fe7 JB |
10300 | if (ret) { |
10301 | btrfs_remove_free_space_cache(cache); | |
10302 | btrfs_put_block_group(cache); | |
10303 | return ret; | |
10304 | } | |
10305 | ||
2e6e5183 FM |
10306 | /* |
10307 | * Now that our block group has its ->space_info set and is inserted in | |
10308 | * the rbtree, update the space info's counters. | |
10309 | */ | |
0b246afa | 10310 | trace_btrfs_add_block_group(fs_info, cache, 1); |
d2006e6d | 10311 | update_space_info(fs_info, cache->flags, size, bytes_used, |
e40edf2d | 10312 | cache->bytes_super, &cache->space_info); |
0b246afa | 10313 | update_global_block_rsv(fs_info); |
1b2da372 | 10314 | |
c434d21c | 10315 | link_block_group(cache); |
6324fbf3 | 10316 | |
47ab2a6c | 10317 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
6324fbf3 | 10318 | |
0b246afa | 10319 | set_avail_alloc_bits(fs_info, type); |
6324fbf3 CM |
10320 | return 0; |
10321 | } | |
1a40e23b | 10322 | |
10ea00f5 ID |
10323 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
10324 | { | |
899c81ea ID |
10325 | u64 extra_flags = chunk_to_extended(flags) & |
10326 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 10327 | |
de98ced9 | 10328 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10329 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
10330 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10331 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10332 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10333 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10334 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 10335 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10336 | } |
10337 | ||
1a40e23b | 10338 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
6bccf3ab | 10339 | struct btrfs_fs_info *fs_info, u64 group_start, |
04216820 | 10340 | struct extent_map *em) |
1a40e23b | 10341 | { |
6bccf3ab | 10342 | struct btrfs_root *root = fs_info->extent_root; |
1a40e23b ZY |
10343 | struct btrfs_path *path; |
10344 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 10345 | struct btrfs_free_cluster *cluster; |
0b246afa | 10346 | struct btrfs_root *tree_root = fs_info->tree_root; |
1a40e23b | 10347 | struct btrfs_key key; |
0af3d00b | 10348 | struct inode *inode; |
c1895442 | 10349 | struct kobject *kobj = NULL; |
1a40e23b | 10350 | int ret; |
10ea00f5 | 10351 | int index; |
89a55897 | 10352 | int factor; |
4f69cb98 | 10353 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 10354 | bool remove_em; |
1a40e23b | 10355 | |
6bccf3ab | 10356 | block_group = btrfs_lookup_block_group(fs_info, group_start); |
1a40e23b | 10357 | BUG_ON(!block_group); |
c146afad | 10358 | BUG_ON(!block_group->ro); |
1a40e23b | 10359 | |
9f7c43c9 | 10360 | /* |
10361 | * Free the reserved super bytes from this block group before | |
10362 | * remove it. | |
10363 | */ | |
2ff7e61e | 10364 | free_excluded_extents(fs_info, block_group); |
fd708b81 JB |
10365 | btrfs_free_ref_tree_range(fs_info, block_group->key.objectid, |
10366 | block_group->key.offset); | |
9f7c43c9 | 10367 | |
1a40e23b | 10368 | memcpy(&key, &block_group->key, sizeof(key)); |
3e72ee88 | 10369 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
89a55897 JB |
10370 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
10371 | BTRFS_BLOCK_GROUP_RAID1 | | |
10372 | BTRFS_BLOCK_GROUP_RAID10)) | |
10373 | factor = 2; | |
10374 | else | |
10375 | factor = 1; | |
1a40e23b | 10376 | |
44fb5511 | 10377 | /* make sure this block group isn't part of an allocation cluster */ |
0b246afa | 10378 | cluster = &fs_info->data_alloc_cluster; |
44fb5511 CM |
10379 | spin_lock(&cluster->refill_lock); |
10380 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10381 | spin_unlock(&cluster->refill_lock); | |
10382 | ||
10383 | /* | |
10384 | * make sure this block group isn't part of a metadata | |
10385 | * allocation cluster | |
10386 | */ | |
0b246afa | 10387 | cluster = &fs_info->meta_alloc_cluster; |
44fb5511 CM |
10388 | spin_lock(&cluster->refill_lock); |
10389 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10390 | spin_unlock(&cluster->refill_lock); | |
10391 | ||
1a40e23b | 10392 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10393 | if (!path) { |
10394 | ret = -ENOMEM; | |
10395 | goto out; | |
10396 | } | |
1a40e23b | 10397 | |
1bbc621e CM |
10398 | /* |
10399 | * get the inode first so any iput calls done for the io_list | |
10400 | * aren't the final iput (no unlinks allowed now) | |
10401 | */ | |
77ab86bf | 10402 | inode = lookup_free_space_inode(fs_info, block_group, path); |
1bbc621e CM |
10403 | |
10404 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10405 | /* | |
10406 | * make sure our free spache cache IO is done before remove the | |
10407 | * free space inode | |
10408 | */ | |
10409 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10410 | if (!list_empty(&block_group->io_list)) { | |
10411 | list_del_init(&block_group->io_list); | |
10412 | ||
10413 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10414 | ||
10415 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
afdb5718 | 10416 | btrfs_wait_cache_io(trans, block_group, path); |
1bbc621e CM |
10417 | btrfs_put_block_group(block_group); |
10418 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10419 | } | |
10420 | ||
10421 | if (!list_empty(&block_group->dirty_list)) { | |
10422 | list_del_init(&block_group->dirty_list); | |
10423 | btrfs_put_block_group(block_group); | |
10424 | } | |
10425 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10426 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10427 | ||
0af3d00b | 10428 | if (!IS_ERR(inode)) { |
73f2e545 | 10429 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
79787eaa JM |
10430 | if (ret) { |
10431 | btrfs_add_delayed_iput(inode); | |
10432 | goto out; | |
10433 | } | |
0af3d00b JB |
10434 | clear_nlink(inode); |
10435 | /* One for the block groups ref */ | |
10436 | spin_lock(&block_group->lock); | |
10437 | if (block_group->iref) { | |
10438 | block_group->iref = 0; | |
10439 | block_group->inode = NULL; | |
10440 | spin_unlock(&block_group->lock); | |
10441 | iput(inode); | |
10442 | } else { | |
10443 | spin_unlock(&block_group->lock); | |
10444 | } | |
10445 | /* One for our lookup ref */ | |
455757c3 | 10446 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10447 | } |
10448 | ||
10449 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10450 | key.offset = block_group->key.objectid; | |
10451 | key.type = 0; | |
10452 | ||
10453 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10454 | if (ret < 0) | |
10455 | goto out; | |
10456 | if (ret > 0) | |
b3b4aa74 | 10457 | btrfs_release_path(path); |
0af3d00b JB |
10458 | if (ret == 0) { |
10459 | ret = btrfs_del_item(trans, tree_root, path); | |
10460 | if (ret) | |
10461 | goto out; | |
b3b4aa74 | 10462 | btrfs_release_path(path); |
0af3d00b JB |
10463 | } |
10464 | ||
0b246afa | 10465 | spin_lock(&fs_info->block_group_cache_lock); |
1a40e23b | 10466 | rb_erase(&block_group->cache_node, |
0b246afa | 10467 | &fs_info->block_group_cache_tree); |
292cbd51 | 10468 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd | 10469 | |
0b246afa JM |
10470 | if (fs_info->first_logical_byte == block_group->key.objectid) |
10471 | fs_info->first_logical_byte = (u64)-1; | |
10472 | spin_unlock(&fs_info->block_group_cache_lock); | |
817d52f8 | 10473 | |
80eb234a | 10474 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10475 | /* |
10476 | * we must use list_del_init so people can check to see if they | |
10477 | * are still on the list after taking the semaphore | |
10478 | */ | |
10479 | list_del_init(&block_group->list); | |
6ab0a202 | 10480 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10481 | kobj = block_group->space_info->block_group_kobjs[index]; |
10482 | block_group->space_info->block_group_kobjs[index] = NULL; | |
0b246afa | 10483 | clear_avail_alloc_bits(fs_info, block_group->flags); |
6ab0a202 | 10484 | } |
80eb234a | 10485 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
10486 | if (kobj) { |
10487 | kobject_del(kobj); | |
10488 | kobject_put(kobj); | |
10489 | } | |
1a40e23b | 10490 | |
4f69cb98 FM |
10491 | if (block_group->has_caching_ctl) |
10492 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10493 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10494 | wait_block_group_cache_done(block_group); |
4f69cb98 | 10495 | if (block_group->has_caching_ctl) { |
0b246afa | 10496 | down_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10497 | if (!caching_ctl) { |
10498 | struct btrfs_caching_control *ctl; | |
10499 | ||
10500 | list_for_each_entry(ctl, | |
0b246afa | 10501 | &fs_info->caching_block_groups, list) |
4f69cb98 FM |
10502 | if (ctl->block_group == block_group) { |
10503 | caching_ctl = ctl; | |
1e4f4714 | 10504 | refcount_inc(&caching_ctl->count); |
4f69cb98 FM |
10505 | break; |
10506 | } | |
10507 | } | |
10508 | if (caching_ctl) | |
10509 | list_del_init(&caching_ctl->list); | |
0b246afa | 10510 | up_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10511 | if (caching_ctl) { |
10512 | /* Once for the caching bgs list and once for us. */ | |
10513 | put_caching_control(caching_ctl); | |
10514 | put_caching_control(caching_ctl); | |
10515 | } | |
10516 | } | |
817d52f8 | 10517 | |
ce93ec54 JB |
10518 | spin_lock(&trans->transaction->dirty_bgs_lock); |
10519 | if (!list_empty(&block_group->dirty_list)) { | |
1bbc621e CM |
10520 | WARN_ON(1); |
10521 | } | |
10522 | if (!list_empty(&block_group->io_list)) { | |
10523 | WARN_ON(1); | |
ce93ec54 JB |
10524 | } |
10525 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
817d52f8 JB |
10526 | btrfs_remove_free_space_cache(block_group); |
10527 | ||
c146afad | 10528 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10529 | list_del_init(&block_group->ro_list); |
18d018ad | 10530 | |
0b246afa | 10531 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
18d018ad ZL |
10532 | WARN_ON(block_group->space_info->total_bytes |
10533 | < block_group->key.offset); | |
10534 | WARN_ON(block_group->space_info->bytes_readonly | |
10535 | < block_group->key.offset); | |
10536 | WARN_ON(block_group->space_info->disk_total | |
10537 | < block_group->key.offset * factor); | |
10538 | } | |
c146afad YZ |
10539 | block_group->space_info->total_bytes -= block_group->key.offset; |
10540 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10541 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10542 | |
c146afad | 10543 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10544 | |
0af3d00b JB |
10545 | memcpy(&key, &block_group->key, sizeof(key)); |
10546 | ||
34441361 | 10547 | mutex_lock(&fs_info->chunk_mutex); |
495e64f4 FM |
10548 | if (!list_empty(&em->list)) { |
10549 | /* We're in the transaction->pending_chunks list. */ | |
10550 | free_extent_map(em); | |
10551 | } | |
04216820 FM |
10552 | spin_lock(&block_group->lock); |
10553 | block_group->removed = 1; | |
10554 | /* | |
10555 | * At this point trimming can't start on this block group, because we | |
10556 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10557 | * so no one can't find it anymore and even if someone already got this | |
10558 | * block group before we removed it from the rbtree, they have already | |
10559 | * incremented block_group->trimming - if they didn't, they won't find | |
10560 | * any free space entries because we already removed them all when we | |
10561 | * called btrfs_remove_free_space_cache(). | |
10562 | * | |
10563 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10564 | * to prevent the same logical address range and physical device space | |
10565 | * ranges from being reused for a new block group. This is because our | |
10566 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10567 | * completely transactionless, so while it is trimming a range the | |
10568 | * currently running transaction might finish and a new one start, | |
10569 | * allowing for new block groups to be created that can reuse the same | |
10570 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10571 | * |
10572 | * There may also be an implicit trim operation if the file system | |
10573 | * is mounted with -odiscard. The same protections must remain | |
10574 | * in place until the extents have been discarded completely when | |
10575 | * the transaction commit has completed. | |
04216820 FM |
10576 | */ |
10577 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
10578 | /* | |
10579 | * Make sure a trimmer task always sees the em in the pinned_chunks list | |
10580 | * if it sees block_group->removed == 1 (needs to lock block_group->lock | |
10581 | * before checking block_group->removed). | |
10582 | */ | |
10583 | if (!remove_em) { | |
10584 | /* | |
10585 | * Our em might be in trans->transaction->pending_chunks which | |
10586 | * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks), | |
10587 | * and so is the fs_info->pinned_chunks list. | |
10588 | * | |
10589 | * So at this point we must be holding the chunk_mutex to avoid | |
10590 | * any races with chunk allocation (more specifically at | |
10591 | * volumes.c:contains_pending_extent()), to ensure it always | |
10592 | * sees the em, either in the pending_chunks list or in the | |
10593 | * pinned_chunks list. | |
10594 | */ | |
0b246afa | 10595 | list_move_tail(&em->list, &fs_info->pinned_chunks); |
04216820 FM |
10596 | } |
10597 | spin_unlock(&block_group->lock); | |
04216820 FM |
10598 | |
10599 | if (remove_em) { | |
10600 | struct extent_map_tree *em_tree; | |
10601 | ||
0b246afa | 10602 | em_tree = &fs_info->mapping_tree.map_tree; |
04216820 | 10603 | write_lock(&em_tree->lock); |
8dbcd10f FM |
10604 | /* |
10605 | * The em might be in the pending_chunks list, so make sure the | |
10606 | * chunk mutex is locked, since remove_extent_mapping() will | |
10607 | * delete us from that list. | |
10608 | */ | |
04216820 FM |
10609 | remove_extent_mapping(em_tree, em); |
10610 | write_unlock(&em_tree->lock); | |
10611 | /* once for the tree */ | |
10612 | free_extent_map(em); | |
10613 | } | |
10614 | ||
34441361 | 10615 | mutex_unlock(&fs_info->chunk_mutex); |
8dbcd10f | 10616 | |
0b246afa | 10617 | ret = remove_block_group_free_space(trans, fs_info, block_group); |
1e144fb8 OS |
10618 | if (ret) |
10619 | goto out; | |
10620 | ||
fa9c0d79 CM |
10621 | btrfs_put_block_group(block_group); |
10622 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10623 | |
10624 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10625 | if (ret > 0) | |
10626 | ret = -EIO; | |
10627 | if (ret < 0) | |
10628 | goto out; | |
10629 | ||
10630 | ret = btrfs_del_item(trans, root, path); | |
10631 | out: | |
10632 | btrfs_free_path(path); | |
10633 | return ret; | |
10634 | } | |
acce952b | 10635 | |
8eab77ff | 10636 | struct btrfs_trans_handle * |
7fd01182 FM |
10637 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10638 | const u64 chunk_offset) | |
8eab77ff | 10639 | { |
7fd01182 FM |
10640 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; |
10641 | struct extent_map *em; | |
10642 | struct map_lookup *map; | |
10643 | unsigned int num_items; | |
10644 | ||
10645 | read_lock(&em_tree->lock); | |
10646 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10647 | read_unlock(&em_tree->lock); | |
10648 | ASSERT(em && em->start == chunk_offset); | |
10649 | ||
8eab77ff | 10650 | /* |
7fd01182 FM |
10651 | * We need to reserve 3 + N units from the metadata space info in order |
10652 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10653 | * btrfs_remove_block_group()), which are used for: | |
10654 | * | |
8eab77ff FM |
10655 | * 1 unit for adding the free space inode's orphan (located in the tree |
10656 | * of tree roots). | |
7fd01182 FM |
10657 | * 1 unit for deleting the block group item (located in the extent |
10658 | * tree). | |
10659 | * 1 unit for deleting the free space item (located in tree of tree | |
10660 | * roots). | |
10661 | * N units for deleting N device extent items corresponding to each | |
10662 | * stripe (located in the device tree). | |
10663 | * | |
10664 | * In order to remove a block group we also need to reserve units in the | |
10665 | * system space info in order to update the chunk tree (update one or | |
10666 | * more device items and remove one chunk item), but this is done at | |
10667 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10668 | */ |
95617d69 | 10669 | map = em->map_lookup; |
7fd01182 FM |
10670 | num_items = 3 + map->num_stripes; |
10671 | free_extent_map(em); | |
10672 | ||
8eab77ff | 10673 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10674 | num_items, 1); |
8eab77ff FM |
10675 | } |
10676 | ||
47ab2a6c JB |
10677 | /* |
10678 | * Process the unused_bgs list and remove any that don't have any allocated | |
10679 | * space inside of them. | |
10680 | */ | |
10681 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10682 | { | |
10683 | struct btrfs_block_group_cache *block_group; | |
10684 | struct btrfs_space_info *space_info; | |
47ab2a6c JB |
10685 | struct btrfs_trans_handle *trans; |
10686 | int ret = 0; | |
10687 | ||
afcdd129 | 10688 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) |
47ab2a6c JB |
10689 | return; |
10690 | ||
10691 | spin_lock(&fs_info->unused_bgs_lock); | |
10692 | while (!list_empty(&fs_info->unused_bgs)) { | |
10693 | u64 start, end; | |
e33e17ee | 10694 | int trimming; |
47ab2a6c JB |
10695 | |
10696 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10697 | struct btrfs_block_group_cache, | |
10698 | bg_list); | |
47ab2a6c | 10699 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10700 | |
10701 | space_info = block_group->space_info; | |
10702 | ||
47ab2a6c JB |
10703 | if (ret || btrfs_mixed_space_info(space_info)) { |
10704 | btrfs_put_block_group(block_group); | |
10705 | continue; | |
10706 | } | |
10707 | spin_unlock(&fs_info->unused_bgs_lock); | |
10708 | ||
d5f2e33b | 10709 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10710 | |
47ab2a6c JB |
10711 | /* Don't want to race with allocators so take the groups_sem */ |
10712 | down_write(&space_info->groups_sem); | |
10713 | spin_lock(&block_group->lock); | |
10714 | if (block_group->reserved || | |
10715 | btrfs_block_group_used(&block_group->item) || | |
19c4d2f9 | 10716 | block_group->ro || |
aefbe9a6 | 10717 | list_is_singular(&block_group->list)) { |
47ab2a6c JB |
10718 | /* |
10719 | * We want to bail if we made new allocations or have | |
10720 | * outstanding allocations in this block group. We do | |
10721 | * the ro check in case balance is currently acting on | |
10722 | * this block group. | |
10723 | */ | |
10724 | spin_unlock(&block_group->lock); | |
10725 | up_write(&space_info->groups_sem); | |
10726 | goto next; | |
10727 | } | |
10728 | spin_unlock(&block_group->lock); | |
10729 | ||
10730 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 10731 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
10732 | up_write(&space_info->groups_sem); |
10733 | if (ret < 0) { | |
10734 | ret = 0; | |
10735 | goto next; | |
10736 | } | |
10737 | ||
10738 | /* | |
10739 | * Want to do this before we do anything else so we can recover | |
10740 | * properly if we fail to join the transaction. | |
10741 | */ | |
7fd01182 FM |
10742 | trans = btrfs_start_trans_remove_block_group(fs_info, |
10743 | block_group->key.objectid); | |
47ab2a6c | 10744 | if (IS_ERR(trans)) { |
2ff7e61e | 10745 | btrfs_dec_block_group_ro(block_group); |
47ab2a6c JB |
10746 | ret = PTR_ERR(trans); |
10747 | goto next; | |
10748 | } | |
10749 | ||
10750 | /* | |
10751 | * We could have pending pinned extents for this block group, | |
10752 | * just delete them, we don't care about them anymore. | |
10753 | */ | |
10754 | start = block_group->key.objectid; | |
10755 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
10756 | /* |
10757 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
10758 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
10759 | * another task might be running finish_extent_commit() for the | |
10760 | * previous transaction N - 1, and have seen a range belonging | |
10761 | * to the block group in freed_extents[] before we were able to | |
10762 | * clear the whole block group range from freed_extents[]. This | |
10763 | * means that task can lookup for the block group after we | |
10764 | * unpinned it from freed_extents[] and removed it, leading to | |
10765 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
10766 | */ | |
10767 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 10768 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
91166212 | 10769 | EXTENT_DIRTY); |
758eb51e | 10770 | if (ret) { |
d4b450cd | 10771 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10772 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10773 | goto end_trans; |
10774 | } | |
10775 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
91166212 | 10776 | EXTENT_DIRTY); |
758eb51e | 10777 | if (ret) { |
d4b450cd | 10778 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10779 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10780 | goto end_trans; |
10781 | } | |
d4b450cd | 10782 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
10783 | |
10784 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
10785 | spin_lock(&space_info->lock); |
10786 | spin_lock(&block_group->lock); | |
10787 | ||
10788 | space_info->bytes_pinned -= block_group->pinned; | |
10789 | space_info->bytes_readonly += block_group->pinned; | |
10790 | percpu_counter_add(&space_info->total_bytes_pinned, | |
10791 | -block_group->pinned); | |
47ab2a6c JB |
10792 | block_group->pinned = 0; |
10793 | ||
c30666d4 ZL |
10794 | spin_unlock(&block_group->lock); |
10795 | spin_unlock(&space_info->lock); | |
10796 | ||
e33e17ee | 10797 | /* DISCARD can flip during remount */ |
0b246afa | 10798 | trimming = btrfs_test_opt(fs_info, DISCARD); |
e33e17ee JM |
10799 | |
10800 | /* Implicit trim during transaction commit. */ | |
10801 | if (trimming) | |
10802 | btrfs_get_block_group_trimming(block_group); | |
10803 | ||
47ab2a6c JB |
10804 | /* |
10805 | * Btrfs_remove_chunk will abort the transaction if things go | |
10806 | * horribly wrong. | |
10807 | */ | |
5b4aacef | 10808 | ret = btrfs_remove_chunk(trans, fs_info, |
47ab2a6c | 10809 | block_group->key.objectid); |
e33e17ee JM |
10810 | |
10811 | if (ret) { | |
10812 | if (trimming) | |
10813 | btrfs_put_block_group_trimming(block_group); | |
10814 | goto end_trans; | |
10815 | } | |
10816 | ||
10817 | /* | |
10818 | * If we're not mounted with -odiscard, we can just forget | |
10819 | * about this block group. Otherwise we'll need to wait | |
10820 | * until transaction commit to do the actual discard. | |
10821 | */ | |
10822 | if (trimming) { | |
348a0013 FM |
10823 | spin_lock(&fs_info->unused_bgs_lock); |
10824 | /* | |
10825 | * A concurrent scrub might have added us to the list | |
10826 | * fs_info->unused_bgs, so use a list_move operation | |
10827 | * to add the block group to the deleted_bgs list. | |
10828 | */ | |
e33e17ee JM |
10829 | list_move(&block_group->bg_list, |
10830 | &trans->transaction->deleted_bgs); | |
348a0013 | 10831 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
10832 | btrfs_get_block_group(block_group); |
10833 | } | |
758eb51e | 10834 | end_trans: |
3a45bb20 | 10835 | btrfs_end_transaction(trans); |
47ab2a6c | 10836 | next: |
d5f2e33b | 10837 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
10838 | btrfs_put_block_group(block_group); |
10839 | spin_lock(&fs_info->unused_bgs_lock); | |
10840 | } | |
10841 | spin_unlock(&fs_info->unused_bgs_lock); | |
10842 | } | |
10843 | ||
c59021f8 | 10844 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
10845 | { | |
10846 | struct btrfs_space_info *space_info; | |
1aba86d6 | 10847 | struct btrfs_super_block *disk_super; |
10848 | u64 features; | |
10849 | u64 flags; | |
10850 | int mixed = 0; | |
c59021f8 | 10851 | int ret; |
10852 | ||
6c41761f | 10853 | disk_super = fs_info->super_copy; |
1aba86d6 | 10854 | if (!btrfs_super_root(disk_super)) |
0dc924c5 | 10855 | return -EINVAL; |
c59021f8 | 10856 | |
1aba86d6 | 10857 | features = btrfs_super_incompat_flags(disk_super); |
10858 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
10859 | mixed = 1; | |
c59021f8 | 10860 | |
1aba86d6 | 10861 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
2be12ef7 | 10862 | ret = create_space_info(fs_info, flags, &space_info); |
c59021f8 | 10863 | if (ret) |
1aba86d6 | 10864 | goto out; |
c59021f8 | 10865 | |
1aba86d6 | 10866 | if (mixed) { |
10867 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
2be12ef7 | 10868 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 10869 | } else { |
10870 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
2be12ef7 | 10871 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 10872 | if (ret) |
10873 | goto out; | |
10874 | ||
10875 | flags = BTRFS_BLOCK_GROUP_DATA; | |
2be12ef7 | 10876 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 10877 | } |
10878 | out: | |
c59021f8 | 10879 | return ret; |
10880 | } | |
10881 | ||
2ff7e61e JM |
10882 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
10883 | u64 start, u64 end) | |
acce952b | 10884 | { |
2ff7e61e | 10885 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 10886 | } |
10887 | ||
499f377f JM |
10888 | /* |
10889 | * It used to be that old block groups would be left around forever. | |
10890 | * Iterating over them would be enough to trim unused space. Since we | |
10891 | * now automatically remove them, we also need to iterate over unallocated | |
10892 | * space. | |
10893 | * | |
10894 | * We don't want a transaction for this since the discard may take a | |
10895 | * substantial amount of time. We don't require that a transaction be | |
10896 | * running, but we do need to take a running transaction into account | |
10897 | * to ensure that we're not discarding chunks that were released in | |
10898 | * the current transaction. | |
10899 | * | |
10900 | * Holding the chunks lock will prevent other threads from allocating | |
10901 | * or releasing chunks, but it won't prevent a running transaction | |
10902 | * from committing and releasing the memory that the pending chunks | |
10903 | * list head uses. For that, we need to take a reference to the | |
10904 | * transaction. | |
10905 | */ | |
10906 | static int btrfs_trim_free_extents(struct btrfs_device *device, | |
10907 | u64 minlen, u64 *trimmed) | |
10908 | { | |
10909 | u64 start = 0, len = 0; | |
10910 | int ret; | |
10911 | ||
10912 | *trimmed = 0; | |
10913 | ||
10914 | /* Not writeable = nothing to do. */ | |
ebbede42 | 10915 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
10916 | return 0; |
10917 | ||
10918 | /* No free space = nothing to do. */ | |
10919 | if (device->total_bytes <= device->bytes_used) | |
10920 | return 0; | |
10921 | ||
10922 | ret = 0; | |
10923 | ||
10924 | while (1) { | |
fb456252 | 10925 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
10926 | struct btrfs_transaction *trans; |
10927 | u64 bytes; | |
10928 | ||
10929 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
10930 | if (ret) | |
10931 | return ret; | |
10932 | ||
10933 | down_read(&fs_info->commit_root_sem); | |
10934 | ||
10935 | spin_lock(&fs_info->trans_lock); | |
10936 | trans = fs_info->running_transaction; | |
10937 | if (trans) | |
9b64f57d | 10938 | refcount_inc(&trans->use_count); |
499f377f JM |
10939 | spin_unlock(&fs_info->trans_lock); |
10940 | ||
10941 | ret = find_free_dev_extent_start(trans, device, minlen, start, | |
10942 | &start, &len); | |
10943 | if (trans) | |
10944 | btrfs_put_transaction(trans); | |
10945 | ||
10946 | if (ret) { | |
10947 | up_read(&fs_info->commit_root_sem); | |
10948 | mutex_unlock(&fs_info->chunk_mutex); | |
10949 | if (ret == -ENOSPC) | |
10950 | ret = 0; | |
10951 | break; | |
10952 | } | |
10953 | ||
10954 | ret = btrfs_issue_discard(device->bdev, start, len, &bytes); | |
10955 | up_read(&fs_info->commit_root_sem); | |
10956 | mutex_unlock(&fs_info->chunk_mutex); | |
10957 | ||
10958 | if (ret) | |
10959 | break; | |
10960 | ||
10961 | start += len; | |
10962 | *trimmed += bytes; | |
10963 | ||
10964 | if (fatal_signal_pending(current)) { | |
10965 | ret = -ERESTARTSYS; | |
10966 | break; | |
10967 | } | |
10968 | ||
10969 | cond_resched(); | |
10970 | } | |
10971 | ||
10972 | return ret; | |
10973 | } | |
10974 | ||
2ff7e61e | 10975 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 10976 | { |
f7039b1d | 10977 | struct btrfs_block_group_cache *cache = NULL; |
499f377f JM |
10978 | struct btrfs_device *device; |
10979 | struct list_head *devices; | |
f7039b1d LD |
10980 | u64 group_trimmed; |
10981 | u64 start; | |
10982 | u64 end; | |
10983 | u64 trimmed = 0; | |
2cac13e4 | 10984 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
10985 | int ret = 0; |
10986 | ||
2cac13e4 LB |
10987 | /* |
10988 | * try to trim all FS space, our block group may start from non-zero. | |
10989 | */ | |
10990 | if (range->len == total_bytes) | |
10991 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
10992 | else | |
10993 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
10994 | |
10995 | while (cache) { | |
10996 | if (cache->key.objectid >= (range->start + range->len)) { | |
10997 | btrfs_put_block_group(cache); | |
10998 | break; | |
10999 | } | |
11000 | ||
11001 | start = max(range->start, cache->key.objectid); | |
11002 | end = min(range->start + range->len, | |
11003 | cache->key.objectid + cache->key.offset); | |
11004 | ||
11005 | if (end - start >= range->minlen) { | |
11006 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 11007 | ret = cache_block_group(cache, 0); |
1be41b78 JB |
11008 | if (ret) { |
11009 | btrfs_put_block_group(cache); | |
11010 | break; | |
11011 | } | |
11012 | ret = wait_block_group_cache_done(cache); | |
11013 | if (ret) { | |
11014 | btrfs_put_block_group(cache); | |
11015 | break; | |
11016 | } | |
f7039b1d LD |
11017 | } |
11018 | ret = btrfs_trim_block_group(cache, | |
11019 | &group_trimmed, | |
11020 | start, | |
11021 | end, | |
11022 | range->minlen); | |
11023 | ||
11024 | trimmed += group_trimmed; | |
11025 | if (ret) { | |
11026 | btrfs_put_block_group(cache); | |
11027 | break; | |
11028 | } | |
11029 | } | |
11030 | ||
2ff7e61e | 11031 | cache = next_block_group(fs_info, cache); |
f7039b1d LD |
11032 | } |
11033 | ||
0b246afa JM |
11034 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
11035 | devices = &fs_info->fs_devices->alloc_list; | |
499f377f JM |
11036 | list_for_each_entry(device, devices, dev_alloc_list) { |
11037 | ret = btrfs_trim_free_extents(device, range->minlen, | |
11038 | &group_trimmed); | |
11039 | if (ret) | |
11040 | break; | |
11041 | ||
11042 | trimmed += group_trimmed; | |
11043 | } | |
0b246afa | 11044 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 11045 | |
f7039b1d LD |
11046 | range->len = trimmed; |
11047 | return ret; | |
11048 | } | |
8257b2dc MX |
11049 | |
11050 | /* | |
ea14b57f | 11051 | * btrfs_{start,end}_write_no_snapshotting() are similar to |
9ea24bbe FM |
11052 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing |
11053 | * data into the page cache through nocow before the subvolume is snapshoted, | |
11054 | * but flush the data into disk after the snapshot creation, or to prevent | |
ea14b57f | 11055 | * operations while snapshotting is ongoing and that cause the snapshot to be |
9ea24bbe | 11056 | * inconsistent (writes followed by expanding truncates for example). |
8257b2dc | 11057 | */ |
ea14b57f | 11058 | void btrfs_end_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc MX |
11059 | { |
11060 | percpu_counter_dec(&root->subv_writers->counter); | |
11061 | /* | |
a83342aa | 11062 | * Make sure counter is updated before we wake up waiters. |
8257b2dc MX |
11063 | */ |
11064 | smp_mb(); | |
11065 | if (waitqueue_active(&root->subv_writers->wait)) | |
11066 | wake_up(&root->subv_writers->wait); | |
11067 | } | |
11068 | ||
ea14b57f | 11069 | int btrfs_start_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc | 11070 | { |
ea14b57f | 11071 | if (atomic_read(&root->will_be_snapshotted)) |
8257b2dc MX |
11072 | return 0; |
11073 | ||
11074 | percpu_counter_inc(&root->subv_writers->counter); | |
11075 | /* | |
11076 | * Make sure counter is updated before we check for snapshot creation. | |
11077 | */ | |
11078 | smp_mb(); | |
ea14b57f DS |
11079 | if (atomic_read(&root->will_be_snapshotted)) { |
11080 | btrfs_end_write_no_snapshotting(root); | |
8257b2dc MX |
11081 | return 0; |
11082 | } | |
11083 | return 1; | |
11084 | } | |
0bc19f90 | 11085 | |
0bc19f90 ZL |
11086 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) |
11087 | { | |
11088 | while (true) { | |
11089 | int ret; | |
11090 | ||
ea14b57f | 11091 | ret = btrfs_start_write_no_snapshotting(root); |
0bc19f90 ZL |
11092 | if (ret) |
11093 | break; | |
5e4def20 | 11094 | wait_on_atomic_t(&root->will_be_snapshotted, atomic_t_wait, |
0bc19f90 ZL |
11095 | TASK_UNINTERRUPTIBLE); |
11096 | } | |
11097 | } |