Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 | 4 | */ |
c1d7c514 | 5 | |
ec6b910f | 6 | #include <linux/sched.h> |
f361bf4a | 7 | #include <linux/sched/signal.h> |
edbd8d4e | 8 | #include <linux/pagemap.h> |
ec44a35c | 9 | #include <linux/writeback.h> |
21af804c | 10 | #include <linux/blkdev.h> |
b7a9f29f | 11 | #include <linux/sort.h> |
4184ea7f | 12 | #include <linux/rcupdate.h> |
817d52f8 | 13 | #include <linux/kthread.h> |
5a0e3ad6 | 14 | #include <linux/slab.h> |
dff51cd1 | 15 | #include <linux/ratelimit.h> |
b150a4f1 | 16 | #include <linux/percpu_counter.h> |
69fe2d75 | 17 | #include <linux/lockdep.h> |
9678c543 | 18 | #include <linux/crc32c.h> |
995946dd | 19 | #include "tree-log.h" |
fec577fb CM |
20 | #include "disk-io.h" |
21 | #include "print-tree.h" | |
0b86a832 | 22 | #include "volumes.h" |
53b381b3 | 23 | #include "raid56.h" |
925baedd | 24 | #include "locking.h" |
fa9c0d79 | 25 | #include "free-space-cache.h" |
1e144fb8 | 26 | #include "free-space-tree.h" |
3fed40cc | 27 | #include "math.h" |
6ab0a202 | 28 | #include "sysfs.h" |
fcebe456 | 29 | #include "qgroup.h" |
fd708b81 | 30 | #include "ref-verify.h" |
fec577fb | 31 | |
709c0486 AJ |
32 | #undef SCRAMBLE_DELAYED_REFS |
33 | ||
9e622d6b MX |
34 | /* |
35 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
36 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
37 | * if we really need one. | |
38 | * | |
0e4f8f88 CM |
39 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
40 | * if we have very few chunks already allocated. This is | |
41 | * used as part of the clustering code to help make sure | |
42 | * we have a good pool of storage to cluster in, without | |
43 | * filling the FS with empty chunks | |
44 | * | |
9e622d6b MX |
45 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
46 | * | |
0e4f8f88 CM |
47 | */ |
48 | enum { | |
49 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
50 | CHUNK_ALLOC_LIMITED = 1, |
51 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
52 | }; |
53 | ||
9f9b8e8d QW |
54 | /* |
55 | * Declare a helper function to detect underflow of various space info members | |
56 | */ | |
57 | #define DECLARE_SPACE_INFO_UPDATE(name) \ | |
58 | static inline void update_##name(struct btrfs_space_info *sinfo, \ | |
59 | s64 bytes) \ | |
60 | { \ | |
61 | if (bytes < 0 && sinfo->name < -bytes) { \ | |
62 | WARN_ON(1); \ | |
63 | sinfo->name = 0; \ | |
64 | return; \ | |
65 | } \ | |
66 | sinfo->name += bytes; \ | |
67 | } | |
68 | ||
69 | DECLARE_SPACE_INFO_UPDATE(bytes_may_use); | |
70 | ||
5d4f98a2 | 71 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
72 | struct btrfs_delayed_ref_node *node, u64 parent, |
73 | u64 root_objectid, u64 owner_objectid, | |
74 | u64 owner_offset, int refs_to_drop, | |
75 | struct btrfs_delayed_extent_op *extra_op); | |
5d4f98a2 YZ |
76 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
77 | struct extent_buffer *leaf, | |
78 | struct btrfs_extent_item *ei); | |
79 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
80 | u64 parent, u64 root_objectid, |
81 | u64 flags, u64 owner, u64 offset, | |
82 | struct btrfs_key *ins, int ref_mod); | |
83 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
4e6bd4e0 | 84 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 85 | struct btrfs_delayed_extent_op *extent_op); |
01458828 | 86 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, 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 | |
9e715da8 | 237 | static void free_excluded_extents(struct btrfs_block_group_cache *cache) |
11833d66 | 238 | { |
9e715da8 | 239 | struct btrfs_fs_info *fs_info = cache->fs_info; |
11833d66 | 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 | ||
3c4da657 | 251 | static int exclude_super_stripes(struct btrfs_block_group_cache *cache) |
817d52f8 | 252 | { |
3c4da657 | 253 | struct btrfs_fs_info *fs_info = cache->fs_info; |
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, |
63a9c7b9 | 271 | bytenr, &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 | 357 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
4457c1c7 | 358 | u64 start, u64 end) |
0f9dd46c | 359 | { |
4457c1c7 | 360 | struct btrfs_fs_info *info = block_group->fs_info; |
817d52f8 | 361 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
362 | int ret; |
363 | ||
364 | while (start < end) { | |
11833d66 | 365 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 366 | &extent_start, &extent_end, |
e6138876 JB |
367 | EXTENT_DIRTY | EXTENT_UPTODATE, |
368 | NULL); | |
0f9dd46c JB |
369 | if (ret) |
370 | break; | |
371 | ||
06b2331f | 372 | if (extent_start <= start) { |
0f9dd46c JB |
373 | start = extent_end + 1; |
374 | } else if (extent_start > start && extent_start < end) { | |
375 | size = extent_start - start; | |
817d52f8 | 376 | total_added += size; |
ea6a478e JB |
377 | ret = btrfs_add_free_space(block_group, start, |
378 | size); | |
79787eaa | 379 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
380 | start = extent_end + 1; |
381 | } else { | |
382 | break; | |
383 | } | |
384 | } | |
385 | ||
386 | if (start < end) { | |
387 | size = end - start; | |
817d52f8 | 388 | total_added += size; |
ea6a478e | 389 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 390 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
391 | } |
392 | ||
817d52f8 | 393 | return total_added; |
0f9dd46c JB |
394 | } |
395 | ||
73fa48b6 | 396 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) |
e37c9e69 | 397 | { |
0b246afa JM |
398 | struct btrfs_block_group_cache *block_group = caching_ctl->block_group; |
399 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
400 | struct btrfs_root *extent_root = fs_info->extent_root; | |
e37c9e69 | 401 | struct btrfs_path *path; |
5f39d397 | 402 | struct extent_buffer *leaf; |
11833d66 | 403 | struct btrfs_key key; |
817d52f8 | 404 | u64 total_found = 0; |
11833d66 YZ |
405 | u64 last = 0; |
406 | u32 nritems; | |
73fa48b6 | 407 | int ret; |
d0bd4560 | 408 | bool wakeup = true; |
f510cfec | 409 | |
e37c9e69 CM |
410 | path = btrfs_alloc_path(); |
411 | if (!path) | |
73fa48b6 | 412 | return -ENOMEM; |
7d7d6068 | 413 | |
817d52f8 | 414 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 415 | |
d0bd4560 JB |
416 | #ifdef CONFIG_BTRFS_DEBUG |
417 | /* | |
418 | * If we're fragmenting we don't want to make anybody think we can | |
419 | * allocate from this block group until we've had a chance to fragment | |
420 | * the free space. | |
421 | */ | |
2ff7e61e | 422 | if (btrfs_should_fragment_free_space(block_group)) |
d0bd4560 JB |
423 | wakeup = false; |
424 | #endif | |
5cd57b2c | 425 | /* |
817d52f8 JB |
426 | * We don't want to deadlock with somebody trying to allocate a new |
427 | * extent for the extent root while also trying to search the extent | |
428 | * root to add free space. So we skip locking and search the commit | |
429 | * root, since its read-only | |
5cd57b2c CM |
430 | */ |
431 | path->skip_locking = 1; | |
817d52f8 | 432 | path->search_commit_root = 1; |
e4058b54 | 433 | path->reada = READA_FORWARD; |
817d52f8 | 434 | |
e4404d6e | 435 | key.objectid = last; |
e37c9e69 | 436 | key.offset = 0; |
11833d66 | 437 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 438 | |
52ee28d2 | 439 | next: |
11833d66 | 440 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 441 | if (ret < 0) |
73fa48b6 | 442 | goto out; |
a512bbf8 | 443 | |
11833d66 YZ |
444 | leaf = path->nodes[0]; |
445 | nritems = btrfs_header_nritems(leaf); | |
446 | ||
d397712b | 447 | while (1) { |
7841cb28 | 448 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 449 | last = (u64)-1; |
817d52f8 | 450 | break; |
f25784b3 | 451 | } |
817d52f8 | 452 | |
11833d66 YZ |
453 | if (path->slots[0] < nritems) { |
454 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
455 | } else { | |
456 | ret = find_next_key(path, 0, &key); | |
457 | if (ret) | |
e37c9e69 | 458 | break; |
817d52f8 | 459 | |
c9ea7b24 | 460 | if (need_resched() || |
9e351cc8 | 461 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
d0bd4560 JB |
462 | if (wakeup) |
463 | caching_ctl->progress = last; | |
ff5714cc | 464 | btrfs_release_path(path); |
9e351cc8 | 465 | up_read(&fs_info->commit_root_sem); |
589d8ade | 466 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 467 | cond_resched(); |
73fa48b6 OS |
468 | mutex_lock(&caching_ctl->mutex); |
469 | down_read(&fs_info->commit_root_sem); | |
470 | goto next; | |
589d8ade | 471 | } |
0a3896d0 JB |
472 | |
473 | ret = btrfs_next_leaf(extent_root, path); | |
474 | if (ret < 0) | |
73fa48b6 | 475 | goto out; |
0a3896d0 JB |
476 | if (ret) |
477 | break; | |
589d8ade JB |
478 | leaf = path->nodes[0]; |
479 | nritems = btrfs_header_nritems(leaf); | |
480 | continue; | |
11833d66 | 481 | } |
817d52f8 | 482 | |
52ee28d2 LB |
483 | if (key.objectid < last) { |
484 | key.objectid = last; | |
485 | key.offset = 0; | |
486 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
487 | ||
d0bd4560 JB |
488 | if (wakeup) |
489 | caching_ctl->progress = last; | |
52ee28d2 LB |
490 | btrfs_release_path(path); |
491 | goto next; | |
492 | } | |
493 | ||
11833d66 YZ |
494 | if (key.objectid < block_group->key.objectid) { |
495 | path->slots[0]++; | |
817d52f8 | 496 | continue; |
e37c9e69 | 497 | } |
0f9dd46c | 498 | |
e37c9e69 | 499 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 500 | block_group->key.offset) |
e37c9e69 | 501 | break; |
7d7d6068 | 502 | |
3173a18f JB |
503 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
504 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
4457c1c7 | 505 | total_found += add_new_free_space(block_group, last, |
817d52f8 | 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 | |
4457c1c7 | 523 | total_found += add_new_free_space(block_group, last, |
817d52f8 JB |
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); |
9e715da8 | 575 | free_excluded_extents(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); |
9e715da8 | 683 | free_excluded_extents(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 | 758 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes, |
29d2b84c | 759 | bool metadata, u64 root_objectid) |
0d9f824d OS |
760 | { |
761 | struct btrfs_space_info *space_info; | |
762 | u64 flags; | |
763 | ||
29d2b84c | 764 | if (metadata) { |
0d9f824d OS |
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); |
dec59fa3 EL |
775 | percpu_counter_add_batch(&space_info->total_bytes_pinned, num_bytes, |
776 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
0d9f824d OS |
777 | } |
778 | ||
4184ea7f CM |
779 | /* |
780 | * after adding space to the filesystem, we need to clear the full flags | |
781 | * on all the space infos. | |
782 | */ | |
783 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
784 | { | |
785 | struct list_head *head = &info->space_info; | |
786 | struct btrfs_space_info *found; | |
787 | ||
788 | rcu_read_lock(); | |
789 | list_for_each_entry_rcu(found, head, list) | |
790 | found->full = 0; | |
791 | rcu_read_unlock(); | |
792 | } | |
793 | ||
1a4ed8fd | 794 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 795 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
796 | { |
797 | int ret; | |
798 | struct btrfs_key key; | |
31840ae1 | 799 | struct btrfs_path *path; |
e02119d5 | 800 | |
31840ae1 | 801 | path = btrfs_alloc_path(); |
d8926bb3 MF |
802 | if (!path) |
803 | return -ENOMEM; | |
804 | ||
e02119d5 CM |
805 | key.objectid = start; |
806 | key.offset = len; | |
3173a18f | 807 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 808 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 809 | btrfs_free_path(path); |
7bb86316 CM |
810 | return ret; |
811 | } | |
812 | ||
a22285a6 | 813 | /* |
3173a18f | 814 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
815 | * |
816 | * the head node for delayed ref is used to store the sum of all the | |
817 | * reference count modifications queued up in the rbtree. the head | |
818 | * node may also store the extent flags to set. This way you can check | |
819 | * to see what the reference count and extent flags would be if all of | |
820 | * the delayed refs are not processed. | |
821 | */ | |
822 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 823 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 824 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
825 | { |
826 | struct btrfs_delayed_ref_head *head; | |
827 | struct btrfs_delayed_ref_root *delayed_refs; | |
828 | struct btrfs_path *path; | |
829 | struct btrfs_extent_item *ei; | |
830 | struct extent_buffer *leaf; | |
831 | struct btrfs_key key; | |
832 | u32 item_size; | |
833 | u64 num_refs; | |
834 | u64 extent_flags; | |
835 | int ret; | |
836 | ||
3173a18f JB |
837 | /* |
838 | * If we don't have skinny metadata, don't bother doing anything | |
839 | * different | |
840 | */ | |
0b246afa JM |
841 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
842 | offset = fs_info->nodesize; | |
3173a18f JB |
843 | metadata = 0; |
844 | } | |
845 | ||
a22285a6 YZ |
846 | path = btrfs_alloc_path(); |
847 | if (!path) | |
848 | return -ENOMEM; | |
849 | ||
a22285a6 YZ |
850 | if (!trans) { |
851 | path->skip_locking = 1; | |
852 | path->search_commit_root = 1; | |
853 | } | |
639eefc8 FDBM |
854 | |
855 | search_again: | |
856 | key.objectid = bytenr; | |
857 | key.offset = offset; | |
858 | if (metadata) | |
859 | key.type = BTRFS_METADATA_ITEM_KEY; | |
860 | else | |
861 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
862 | ||
0b246afa | 863 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
864 | if (ret < 0) |
865 | goto out_free; | |
866 | ||
3173a18f | 867 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
868 | if (path->slots[0]) { |
869 | path->slots[0]--; | |
870 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
871 | path->slots[0]); | |
872 | if (key.objectid == bytenr && | |
873 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 874 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
875 | ret = 0; |
876 | } | |
3173a18f JB |
877 | } |
878 | ||
a22285a6 YZ |
879 | if (ret == 0) { |
880 | leaf = path->nodes[0]; | |
881 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
882 | if (item_size >= sizeof(*ei)) { | |
883 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
884 | struct btrfs_extent_item); | |
885 | num_refs = btrfs_extent_refs(leaf, ei); | |
886 | extent_flags = btrfs_extent_flags(leaf, ei); | |
887 | } else { | |
ba3c2b19 NB |
888 | ret = -EINVAL; |
889 | btrfs_print_v0_err(fs_info); | |
890 | if (trans) | |
891 | btrfs_abort_transaction(trans, ret); | |
892 | else | |
893 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
894 | ||
895 | goto out_free; | |
a22285a6 | 896 | } |
ba3c2b19 | 897 | |
a22285a6 YZ |
898 | BUG_ON(num_refs == 0); |
899 | } else { | |
900 | num_refs = 0; | |
901 | extent_flags = 0; | |
902 | ret = 0; | |
903 | } | |
904 | ||
905 | if (!trans) | |
906 | goto out; | |
907 | ||
908 | delayed_refs = &trans->transaction->delayed_refs; | |
909 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 910 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
911 | if (head) { |
912 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 913 | refcount_inc(&head->refs); |
a22285a6 YZ |
914 | spin_unlock(&delayed_refs->lock); |
915 | ||
b3b4aa74 | 916 | btrfs_release_path(path); |
a22285a6 | 917 | |
8cc33e5c DS |
918 | /* |
919 | * Mutex was contended, block until it's released and try | |
920 | * again | |
921 | */ | |
a22285a6 YZ |
922 | mutex_lock(&head->mutex); |
923 | mutex_unlock(&head->mutex); | |
d278850e | 924 | btrfs_put_delayed_ref_head(head); |
639eefc8 | 925 | goto search_again; |
a22285a6 | 926 | } |
d7df2c79 | 927 | spin_lock(&head->lock); |
a22285a6 YZ |
928 | if (head->extent_op && head->extent_op->update_flags) |
929 | extent_flags |= head->extent_op->flags_to_set; | |
930 | else | |
931 | BUG_ON(num_refs == 0); | |
932 | ||
d278850e | 933 | num_refs += head->ref_mod; |
d7df2c79 | 934 | spin_unlock(&head->lock); |
a22285a6 YZ |
935 | mutex_unlock(&head->mutex); |
936 | } | |
937 | spin_unlock(&delayed_refs->lock); | |
938 | out: | |
939 | WARN_ON(num_refs == 0); | |
940 | if (refs) | |
941 | *refs = num_refs; | |
942 | if (flags) | |
943 | *flags = extent_flags; | |
944 | out_free: | |
945 | btrfs_free_path(path); | |
946 | return ret; | |
947 | } | |
948 | ||
d8d5f3e1 CM |
949 | /* |
950 | * Back reference rules. Back refs have three main goals: | |
951 | * | |
952 | * 1) differentiate between all holders of references to an extent so that | |
953 | * when a reference is dropped we can make sure it was a valid reference | |
954 | * before freeing the extent. | |
955 | * | |
956 | * 2) Provide enough information to quickly find the holders of an extent | |
957 | * if we notice a given block is corrupted or bad. | |
958 | * | |
959 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
960 | * maintenance. This is actually the same as #2, but with a slightly | |
961 | * different use case. | |
962 | * | |
5d4f98a2 YZ |
963 | * There are two kinds of back refs. The implicit back refs is optimized |
964 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
965 | * back refs of this kind provide information about the block's owner tree | |
966 | * and the pointer's key. These information allow us to find the block by | |
967 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
968 | * referenced by their owner trees. The location of tree block is recorded | |
969 | * in the back refs. Actually the full back refs is generic, and can be | |
970 | * used in all cases the implicit back refs is used. The major shortcoming | |
971 | * of the full back refs is its overhead. Every time a tree block gets | |
972 | * COWed, we have to update back refs entry for all pointers in it. | |
973 | * | |
974 | * For a newly allocated tree block, we use implicit back refs for | |
975 | * pointers in it. This means most tree related operations only involve | |
976 | * implicit back refs. For a tree block created in old transaction, the | |
977 | * only way to drop a reference to it is COW it. So we can detect the | |
978 | * event that tree block loses its owner tree's reference and do the | |
979 | * back refs conversion. | |
980 | * | |
01327610 | 981 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
982 | * |
983 | * The reference count of the block is one and the tree is the block's | |
984 | * owner tree. Nothing to do in this case. | |
985 | * | |
986 | * The reference count of the block is one and the tree is not the | |
987 | * block's owner tree. In this case, full back refs is used for pointers | |
988 | * in the block. Remove these full back refs, add implicit back refs for | |
989 | * every pointers in the new block. | |
990 | * | |
991 | * The reference count of the block is greater than one and the tree is | |
992 | * the block's owner tree. In this case, implicit back refs is used for | |
993 | * pointers in the block. Add full back refs for every pointers in the | |
994 | * block, increase lower level extents' reference counts. The original | |
995 | * implicit back refs are entailed to the new block. | |
996 | * | |
997 | * The reference count of the block is greater than one and the tree is | |
998 | * not the block's owner tree. Add implicit back refs for every pointer in | |
999 | * the new block, increase lower level extents' reference count. | |
1000 | * | |
1001 | * Back Reference Key composing: | |
1002 | * | |
1003 | * The key objectid corresponds to the first byte in the extent, | |
1004 | * The key type is used to differentiate between types of back refs. | |
1005 | * There are different meanings of the key offset for different types | |
1006 | * of back refs. | |
1007 | * | |
d8d5f3e1 CM |
1008 | * File extents can be referenced by: |
1009 | * | |
1010 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 1011 | * - different files inside a single subvolume |
d8d5f3e1 CM |
1012 | * - different offsets inside a file (bookend extents in file.c) |
1013 | * | |
5d4f98a2 | 1014 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
1015 | * |
1016 | * - Objectid of the subvolume root | |
d8d5f3e1 | 1017 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
1018 | * - original offset in the file |
1019 | * - how many bookend extents | |
d8d5f3e1 | 1020 | * |
5d4f98a2 YZ |
1021 | * The key offset for the implicit back refs is hash of the first |
1022 | * three fields. | |
d8d5f3e1 | 1023 | * |
5d4f98a2 | 1024 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 1025 | * |
5d4f98a2 | 1026 | * - number of pointers in the tree leaf |
d8d5f3e1 | 1027 | * |
5d4f98a2 YZ |
1028 | * The key offset for the implicit back refs is the first byte of |
1029 | * the tree leaf | |
d8d5f3e1 | 1030 | * |
5d4f98a2 YZ |
1031 | * When a file extent is allocated, The implicit back refs is used. |
1032 | * the fields are filled in: | |
d8d5f3e1 | 1033 | * |
5d4f98a2 | 1034 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 1035 | * |
5d4f98a2 YZ |
1036 | * When a file extent is removed file truncation, we find the |
1037 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 1038 | * |
5d4f98a2 | 1039 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 1040 | * |
5d4f98a2 | 1041 | * Btree extents can be referenced by: |
d8d5f3e1 | 1042 | * |
5d4f98a2 | 1043 | * - Different subvolumes |
d8d5f3e1 | 1044 | * |
5d4f98a2 YZ |
1045 | * Both the implicit back refs and the full back refs for tree blocks |
1046 | * only consist of key. The key offset for the implicit back refs is | |
1047 | * objectid of block's owner tree. The key offset for the full back refs | |
1048 | * is the first byte of parent block. | |
d8d5f3e1 | 1049 | * |
5d4f98a2 YZ |
1050 | * When implicit back refs is used, information about the lowest key and |
1051 | * level of the tree block are required. These information are stored in | |
1052 | * tree block info structure. | |
d8d5f3e1 | 1053 | */ |
31840ae1 | 1054 | |
167ce953 LB |
1055 | /* |
1056 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
1057 | * is_data == BTRFS_REF_TYPE_DATA, data type is requried, | |
1058 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. | |
1059 | */ | |
1060 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
1061 | struct btrfs_extent_inline_ref *iref, | |
1062 | enum btrfs_inline_ref_type is_data) | |
1063 | { | |
1064 | int type = btrfs_extent_inline_ref_type(eb, iref); | |
64ecdb64 | 1065 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 LB |
1066 | |
1067 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
1068 | type == BTRFS_SHARED_BLOCK_REF_KEY || | |
1069 | type == BTRFS_SHARED_DATA_REF_KEY || | |
1070 | type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1071 | if (is_data == BTRFS_REF_TYPE_BLOCK) { | |
64ecdb64 | 1072 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
167ce953 | 1073 | return type; |
64ecdb64 LB |
1074 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
1075 | ASSERT(eb->fs_info); | |
1076 | /* | |
1077 | * Every shared one has parent tree | |
1078 | * block, which must be aligned to | |
1079 | * nodesize. | |
1080 | */ | |
1081 | if (offset && | |
1082 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1083 | return type; | |
1084 | } | |
167ce953 | 1085 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 1086 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 1087 | return type; |
64ecdb64 LB |
1088 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
1089 | ASSERT(eb->fs_info); | |
1090 | /* | |
1091 | * Every shared one has parent tree | |
1092 | * block, which must be aligned to | |
1093 | * nodesize. | |
1094 | */ | |
1095 | if (offset && | |
1096 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1097 | return type; | |
1098 | } | |
167ce953 LB |
1099 | } else { |
1100 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
1101 | return type; | |
1102 | } | |
1103 | } | |
1104 | ||
1105 | btrfs_print_leaf((struct extent_buffer *)eb); | |
1106 | btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d", | |
1107 | eb->start, type); | |
1108 | WARN_ON(1); | |
1109 | ||
1110 | return BTRFS_REF_TYPE_INVALID; | |
1111 | } | |
1112 | ||
5d4f98a2 YZ |
1113 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
1114 | { | |
1115 | u32 high_crc = ~(u32)0; | |
1116 | u32 low_crc = ~(u32)0; | |
1117 | __le64 lenum; | |
1118 | ||
1119 | lenum = cpu_to_le64(root_objectid); | |
9678c543 | 1120 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1121 | lenum = cpu_to_le64(owner); |
9678c543 | 1122 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1123 | lenum = cpu_to_le64(offset); |
9678c543 | 1124 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1125 | |
1126 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1127 | } | |
1128 | ||
1129 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1130 | struct btrfs_extent_data_ref *ref) | |
1131 | { | |
1132 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1133 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1134 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1135 | } | |
1136 | ||
1137 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1138 | struct btrfs_extent_data_ref *ref, | |
1139 | u64 root_objectid, u64 owner, u64 offset) | |
1140 | { | |
1141 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1142 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1143 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1144 | return 0; | |
1145 | return 1; | |
1146 | } | |
1147 | ||
1148 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1149 | struct btrfs_path *path, |
1150 | u64 bytenr, u64 parent, | |
1151 | u64 root_objectid, | |
1152 | u64 owner, u64 offset) | |
1153 | { | |
bd1d53ef | 1154 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 YZ |
1155 | struct btrfs_key key; |
1156 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1157 | struct extent_buffer *leaf; |
5d4f98a2 | 1158 | u32 nritems; |
74493f7a | 1159 | int ret; |
5d4f98a2 YZ |
1160 | int recow; |
1161 | int err = -ENOENT; | |
74493f7a | 1162 | |
31840ae1 | 1163 | key.objectid = bytenr; |
5d4f98a2 YZ |
1164 | if (parent) { |
1165 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1166 | key.offset = parent; | |
1167 | } else { | |
1168 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1169 | key.offset = hash_extent_data_ref(root_objectid, | |
1170 | owner, offset); | |
1171 | } | |
1172 | again: | |
1173 | recow = 0; | |
1174 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1175 | if (ret < 0) { | |
1176 | err = ret; | |
1177 | goto fail; | |
1178 | } | |
31840ae1 | 1179 | |
5d4f98a2 YZ |
1180 | if (parent) { |
1181 | if (!ret) | |
1182 | return 0; | |
5d4f98a2 | 1183 | goto fail; |
31840ae1 ZY |
1184 | } |
1185 | ||
1186 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1187 | nritems = btrfs_header_nritems(leaf); |
1188 | while (1) { | |
1189 | if (path->slots[0] >= nritems) { | |
1190 | ret = btrfs_next_leaf(root, path); | |
1191 | if (ret < 0) | |
1192 | err = ret; | |
1193 | if (ret) | |
1194 | goto fail; | |
1195 | ||
1196 | leaf = path->nodes[0]; | |
1197 | nritems = btrfs_header_nritems(leaf); | |
1198 | recow = 1; | |
1199 | } | |
1200 | ||
1201 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1202 | if (key.objectid != bytenr || | |
1203 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1204 | goto fail; | |
1205 | ||
1206 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1207 | struct btrfs_extent_data_ref); | |
1208 | ||
1209 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1210 | owner, offset)) { | |
1211 | if (recow) { | |
b3b4aa74 | 1212 | btrfs_release_path(path); |
5d4f98a2 YZ |
1213 | goto again; |
1214 | } | |
1215 | err = 0; | |
1216 | break; | |
1217 | } | |
1218 | path->slots[0]++; | |
31840ae1 | 1219 | } |
5d4f98a2 YZ |
1220 | fail: |
1221 | return err; | |
31840ae1 ZY |
1222 | } |
1223 | ||
5d4f98a2 | 1224 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1225 | struct btrfs_path *path, |
1226 | u64 bytenr, u64 parent, | |
1227 | u64 root_objectid, u64 owner, | |
1228 | u64 offset, int refs_to_add) | |
31840ae1 | 1229 | { |
62b895af | 1230 | struct btrfs_root *root = trans->fs_info->extent_root; |
31840ae1 ZY |
1231 | struct btrfs_key key; |
1232 | struct extent_buffer *leaf; | |
5d4f98a2 | 1233 | u32 size; |
31840ae1 ZY |
1234 | u32 num_refs; |
1235 | int ret; | |
74493f7a | 1236 | |
74493f7a | 1237 | key.objectid = bytenr; |
5d4f98a2 YZ |
1238 | if (parent) { |
1239 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1240 | key.offset = parent; | |
1241 | size = sizeof(struct btrfs_shared_data_ref); | |
1242 | } else { | |
1243 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1244 | key.offset = hash_extent_data_ref(root_objectid, | |
1245 | owner, offset); | |
1246 | size = sizeof(struct btrfs_extent_data_ref); | |
1247 | } | |
74493f7a | 1248 | |
5d4f98a2 YZ |
1249 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1250 | if (ret && ret != -EEXIST) | |
1251 | goto fail; | |
1252 | ||
1253 | leaf = path->nodes[0]; | |
1254 | if (parent) { | |
1255 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1256 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1257 | struct btrfs_shared_data_ref); |
1258 | if (ret == 0) { | |
1259 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1260 | } else { | |
1261 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1262 | num_refs += refs_to_add; | |
1263 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1264 | } |
5d4f98a2 YZ |
1265 | } else { |
1266 | struct btrfs_extent_data_ref *ref; | |
1267 | while (ret == -EEXIST) { | |
1268 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1269 | struct btrfs_extent_data_ref); | |
1270 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1271 | owner, offset)) | |
1272 | break; | |
b3b4aa74 | 1273 | btrfs_release_path(path); |
5d4f98a2 YZ |
1274 | key.offset++; |
1275 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1276 | size); | |
1277 | if (ret && ret != -EEXIST) | |
1278 | goto fail; | |
31840ae1 | 1279 | |
5d4f98a2 YZ |
1280 | leaf = path->nodes[0]; |
1281 | } | |
1282 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1283 | struct btrfs_extent_data_ref); | |
1284 | if (ret == 0) { | |
1285 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1286 | root_objectid); | |
1287 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1288 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1289 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1290 | } else { | |
1291 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1292 | num_refs += refs_to_add; | |
1293 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1294 | } |
31840ae1 | 1295 | } |
5d4f98a2 YZ |
1296 | btrfs_mark_buffer_dirty(leaf); |
1297 | ret = 0; | |
1298 | fail: | |
b3b4aa74 | 1299 | btrfs_release_path(path); |
7bb86316 | 1300 | return ret; |
74493f7a CM |
1301 | } |
1302 | ||
5d4f98a2 | 1303 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 | 1304 | struct btrfs_path *path, |
fcebe456 | 1305 | int refs_to_drop, int *last_ref) |
31840ae1 | 1306 | { |
5d4f98a2 YZ |
1307 | struct btrfs_key key; |
1308 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1309 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1310 | struct extent_buffer *leaf; |
5d4f98a2 | 1311 | u32 num_refs = 0; |
31840ae1 ZY |
1312 | int ret = 0; |
1313 | ||
1314 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1315 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1316 | ||
1317 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1318 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1319 | struct btrfs_extent_data_ref); | |
1320 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1321 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1322 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1323 | struct btrfs_shared_data_ref); | |
1324 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
6d8ff4e4 | 1325 | } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { |
ba3c2b19 NB |
1326 | btrfs_print_v0_err(trans->fs_info); |
1327 | btrfs_abort_transaction(trans, -EINVAL); | |
1328 | return -EINVAL; | |
5d4f98a2 YZ |
1329 | } else { |
1330 | BUG(); | |
1331 | } | |
1332 | ||
56bec294 CM |
1333 | BUG_ON(num_refs < refs_to_drop); |
1334 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1335 | |
31840ae1 | 1336 | if (num_refs == 0) { |
e9f6290d | 1337 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
fcebe456 | 1338 | *last_ref = 1; |
31840ae1 | 1339 | } else { |
5d4f98a2 YZ |
1340 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1341 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1342 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1343 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
31840ae1 ZY |
1344 | btrfs_mark_buffer_dirty(leaf); |
1345 | } | |
31840ae1 ZY |
1346 | return ret; |
1347 | } | |
1348 | ||
9ed0dea0 | 1349 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1350 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1351 | { |
5d4f98a2 YZ |
1352 | struct btrfs_key key; |
1353 | struct extent_buffer *leaf; | |
1354 | struct btrfs_extent_data_ref *ref1; | |
1355 | struct btrfs_shared_data_ref *ref2; | |
1356 | u32 num_refs = 0; | |
3de28d57 | 1357 | int type; |
5d4f98a2 YZ |
1358 | |
1359 | leaf = path->nodes[0]; | |
1360 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
ba3c2b19 NB |
1361 | |
1362 | BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); | |
5d4f98a2 | 1363 | if (iref) { |
3de28d57 LB |
1364 | /* |
1365 | * If type is invalid, we should have bailed out earlier than | |
1366 | * this call. | |
1367 | */ | |
1368 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
1369 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
1370 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
1371 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
1372 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1373 | } else { | |
1374 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1375 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1376 | } | |
1377 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1378 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1379 | struct btrfs_extent_data_ref); | |
1380 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1381 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1382 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1383 | struct btrfs_shared_data_ref); | |
1384 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
5d4f98a2 YZ |
1385 | } else { |
1386 | WARN_ON(1); | |
1387 | } | |
1388 | return num_refs; | |
1389 | } | |
15916de8 | 1390 | |
5d4f98a2 | 1391 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1392 | struct btrfs_path *path, |
1393 | u64 bytenr, u64 parent, | |
1394 | u64 root_objectid) | |
1f3c79a2 | 1395 | { |
b8582eea | 1396 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 | 1397 | struct btrfs_key key; |
1f3c79a2 | 1398 | int ret; |
1f3c79a2 | 1399 | |
5d4f98a2 YZ |
1400 | key.objectid = bytenr; |
1401 | if (parent) { | |
1402 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1403 | key.offset = parent; | |
1404 | } else { | |
1405 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1406 | key.offset = root_objectid; | |
1f3c79a2 LH |
1407 | } |
1408 | ||
5d4f98a2 YZ |
1409 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1410 | if (ret > 0) | |
1411 | ret = -ENOENT; | |
5d4f98a2 | 1412 | return ret; |
1f3c79a2 LH |
1413 | } |
1414 | ||
5d4f98a2 | 1415 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1416 | struct btrfs_path *path, |
1417 | u64 bytenr, u64 parent, | |
1418 | u64 root_objectid) | |
31840ae1 | 1419 | { |
5d4f98a2 | 1420 | struct btrfs_key key; |
31840ae1 | 1421 | int ret; |
31840ae1 | 1422 | |
5d4f98a2 YZ |
1423 | key.objectid = bytenr; |
1424 | if (parent) { | |
1425 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1426 | key.offset = parent; | |
1427 | } else { | |
1428 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1429 | key.offset = root_objectid; | |
1430 | } | |
1431 | ||
10728404 | 1432 | ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root, |
87bde3cd | 1433 | path, &key, 0); |
b3b4aa74 | 1434 | btrfs_release_path(path); |
31840ae1 ZY |
1435 | return ret; |
1436 | } | |
1437 | ||
5d4f98a2 | 1438 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1439 | { |
5d4f98a2 YZ |
1440 | int type; |
1441 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1442 | if (parent > 0) | |
1443 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1444 | else | |
1445 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1446 | } else { | |
1447 | if (parent > 0) | |
1448 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1449 | else | |
1450 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1451 | } | |
1452 | return type; | |
31840ae1 | 1453 | } |
56bec294 | 1454 | |
2c47e605 YZ |
1455 | static int find_next_key(struct btrfs_path *path, int level, |
1456 | struct btrfs_key *key) | |
56bec294 | 1457 | |
02217ed2 | 1458 | { |
2c47e605 | 1459 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1460 | if (!path->nodes[level]) |
1461 | break; | |
5d4f98a2 YZ |
1462 | if (path->slots[level] + 1 >= |
1463 | btrfs_header_nritems(path->nodes[level])) | |
1464 | continue; | |
1465 | if (level == 0) | |
1466 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1467 | path->slots[level] + 1); | |
1468 | else | |
1469 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1470 | path->slots[level] + 1); | |
1471 | return 0; | |
1472 | } | |
1473 | return 1; | |
1474 | } | |
037e6390 | 1475 | |
5d4f98a2 YZ |
1476 | /* |
1477 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1478 | * to the address of inline back ref, and 0 is returned. | |
1479 | * | |
1480 | * if back ref isn't found, *ref_ret is set to the address where it | |
1481 | * should be inserted, and -ENOENT is returned. | |
1482 | * | |
1483 | * if insert is true and there are too many inline back refs, the path | |
1484 | * points to the extent item, and -EAGAIN is returned. | |
1485 | * | |
1486 | * NOTE: inline back refs are ordered in the same way that back ref | |
1487 | * items in the tree are ordered. | |
1488 | */ | |
1489 | static noinline_for_stack | |
1490 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1491 | struct btrfs_path *path, |
1492 | struct btrfs_extent_inline_ref **ref_ret, | |
1493 | u64 bytenr, u64 num_bytes, | |
1494 | u64 parent, u64 root_objectid, | |
1495 | u64 owner, u64 offset, int insert) | |
1496 | { | |
867cc1fb | 1497 | struct btrfs_fs_info *fs_info = trans->fs_info; |
87bde3cd | 1498 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1499 | struct btrfs_key key; |
1500 | struct extent_buffer *leaf; | |
1501 | struct btrfs_extent_item *ei; | |
1502 | struct btrfs_extent_inline_ref *iref; | |
1503 | u64 flags; | |
1504 | u64 item_size; | |
1505 | unsigned long ptr; | |
1506 | unsigned long end; | |
1507 | int extra_size; | |
1508 | int type; | |
1509 | int want; | |
1510 | int ret; | |
1511 | int err = 0; | |
0b246afa | 1512 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 1513 | int needed; |
26b8003f | 1514 | |
db94535d | 1515 | key.objectid = bytenr; |
31840ae1 | 1516 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1517 | key.offset = num_bytes; |
31840ae1 | 1518 | |
5d4f98a2 YZ |
1519 | want = extent_ref_type(parent, owner); |
1520 | if (insert) { | |
1521 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1522 | path->keep_locks = 1; |
5d4f98a2 YZ |
1523 | } else |
1524 | extra_size = -1; | |
3173a18f JB |
1525 | |
1526 | /* | |
16d1c062 NB |
1527 | * Owner is our level, so we can just add one to get the level for the |
1528 | * block we are interested in. | |
3173a18f JB |
1529 | */ |
1530 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1531 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1532 | key.offset = owner; | |
1533 | } | |
1534 | ||
1535 | again: | |
5d4f98a2 | 1536 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1537 | if (ret < 0) { |
5d4f98a2 YZ |
1538 | err = ret; |
1539 | goto out; | |
1540 | } | |
3173a18f JB |
1541 | |
1542 | /* | |
1543 | * We may be a newly converted file system which still has the old fat | |
1544 | * extent entries for metadata, so try and see if we have one of those. | |
1545 | */ | |
1546 | if (ret > 0 && skinny_metadata) { | |
1547 | skinny_metadata = false; | |
1548 | if (path->slots[0]) { | |
1549 | path->slots[0]--; | |
1550 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1551 | path->slots[0]); | |
1552 | if (key.objectid == bytenr && | |
1553 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1554 | key.offset == num_bytes) | |
1555 | ret = 0; | |
1556 | } | |
1557 | if (ret) { | |
9ce49a0b | 1558 | key.objectid = bytenr; |
3173a18f JB |
1559 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1560 | key.offset = num_bytes; | |
1561 | btrfs_release_path(path); | |
1562 | goto again; | |
1563 | } | |
1564 | } | |
1565 | ||
79787eaa JM |
1566 | if (ret && !insert) { |
1567 | err = -ENOENT; | |
1568 | goto out; | |
fae7f21c | 1569 | } else if (WARN_ON(ret)) { |
492104c8 | 1570 | err = -EIO; |
492104c8 | 1571 | goto out; |
79787eaa | 1572 | } |
5d4f98a2 YZ |
1573 | |
1574 | leaf = path->nodes[0]; | |
1575 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
6d8ff4e4 | 1576 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
1577 | err = -EINVAL; |
1578 | btrfs_print_v0_err(fs_info); | |
1579 | btrfs_abort_transaction(trans, err); | |
1580 | goto out; | |
1581 | } | |
5d4f98a2 | 1582 | |
5d4f98a2 YZ |
1583 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1584 | flags = btrfs_extent_flags(leaf, ei); | |
1585 | ||
1586 | ptr = (unsigned long)(ei + 1); | |
1587 | end = (unsigned long)ei + item_size; | |
1588 | ||
3173a18f | 1589 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1590 | ptr += sizeof(struct btrfs_tree_block_info); |
1591 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1592 | } |
1593 | ||
3de28d57 LB |
1594 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
1595 | needed = BTRFS_REF_TYPE_DATA; | |
1596 | else | |
1597 | needed = BTRFS_REF_TYPE_BLOCK; | |
1598 | ||
5d4f98a2 YZ |
1599 | err = -ENOENT; |
1600 | while (1) { | |
1601 | if (ptr >= end) { | |
1602 | WARN_ON(ptr > end); | |
1603 | break; | |
1604 | } | |
1605 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
1606 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
1607 | if (type == BTRFS_REF_TYPE_INVALID) { | |
af431dcb | 1608 | err = -EUCLEAN; |
3de28d57 LB |
1609 | goto out; |
1610 | } | |
1611 | ||
5d4f98a2 YZ |
1612 | if (want < type) |
1613 | break; | |
1614 | if (want > type) { | |
1615 | ptr += btrfs_extent_inline_ref_size(type); | |
1616 | continue; | |
1617 | } | |
1618 | ||
1619 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1620 | struct btrfs_extent_data_ref *dref; | |
1621 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1622 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1623 | owner, offset)) { | |
1624 | err = 0; | |
1625 | break; | |
1626 | } | |
1627 | if (hash_extent_data_ref_item(leaf, dref) < | |
1628 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1629 | break; | |
1630 | } else { | |
1631 | u64 ref_offset; | |
1632 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1633 | if (parent > 0) { | |
1634 | if (parent == ref_offset) { | |
1635 | err = 0; | |
1636 | break; | |
1637 | } | |
1638 | if (ref_offset < parent) | |
1639 | break; | |
1640 | } else { | |
1641 | if (root_objectid == ref_offset) { | |
1642 | err = 0; | |
1643 | break; | |
1644 | } | |
1645 | if (ref_offset < root_objectid) | |
1646 | break; | |
1647 | } | |
1648 | } | |
1649 | ptr += btrfs_extent_inline_ref_size(type); | |
1650 | } | |
1651 | if (err == -ENOENT && insert) { | |
1652 | if (item_size + extra_size >= | |
1653 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1654 | err = -EAGAIN; | |
1655 | goto out; | |
1656 | } | |
1657 | /* | |
1658 | * To add new inline back ref, we have to make sure | |
1659 | * there is no corresponding back ref item. | |
1660 | * For simplicity, we just do not add new inline back | |
1661 | * ref if there is any kind of item for this block | |
1662 | */ | |
2c47e605 YZ |
1663 | if (find_next_key(path, 0, &key) == 0 && |
1664 | key.objectid == bytenr && | |
85d4198e | 1665 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1666 | err = -EAGAIN; |
1667 | goto out; | |
1668 | } | |
1669 | } | |
1670 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1671 | out: | |
85d4198e | 1672 | if (insert) { |
5d4f98a2 YZ |
1673 | path->keep_locks = 0; |
1674 | btrfs_unlock_up_safe(path, 1); | |
1675 | } | |
1676 | return err; | |
1677 | } | |
1678 | ||
1679 | /* | |
1680 | * helper to add new inline back ref | |
1681 | */ | |
1682 | static noinline_for_stack | |
87bde3cd | 1683 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1684 | struct btrfs_path *path, |
1685 | struct btrfs_extent_inline_ref *iref, | |
1686 | u64 parent, u64 root_objectid, | |
1687 | u64 owner, u64 offset, int refs_to_add, | |
1688 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1689 | { |
1690 | struct extent_buffer *leaf; | |
1691 | struct btrfs_extent_item *ei; | |
1692 | unsigned long ptr; | |
1693 | unsigned long end; | |
1694 | unsigned long item_offset; | |
1695 | u64 refs; | |
1696 | int size; | |
1697 | int type; | |
5d4f98a2 YZ |
1698 | |
1699 | leaf = path->nodes[0]; | |
1700 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1701 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1702 | ||
1703 | type = extent_ref_type(parent, owner); | |
1704 | size = btrfs_extent_inline_ref_size(type); | |
1705 | ||
87bde3cd | 1706 | btrfs_extend_item(fs_info, path, size); |
5d4f98a2 YZ |
1707 | |
1708 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1709 | refs = btrfs_extent_refs(leaf, ei); | |
1710 | refs += refs_to_add; | |
1711 | btrfs_set_extent_refs(leaf, ei, refs); | |
1712 | if (extent_op) | |
1713 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1714 | ||
1715 | ptr = (unsigned long)ei + item_offset; | |
1716 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1717 | if (ptr < end - size) | |
1718 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1719 | end - size - ptr); | |
1720 | ||
1721 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1722 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1723 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1724 | struct btrfs_extent_data_ref *dref; | |
1725 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1726 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1727 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1728 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1729 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1730 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1731 | struct btrfs_shared_data_ref *sref; | |
1732 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1733 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1734 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1735 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1736 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1737 | } else { | |
1738 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1739 | } | |
1740 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1741 | } |
1742 | ||
1743 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1744 | struct btrfs_path *path, |
1745 | struct btrfs_extent_inline_ref **ref_ret, | |
1746 | u64 bytenr, u64 num_bytes, u64 parent, | |
1747 | u64 root_objectid, u64 owner, u64 offset) | |
1748 | { | |
1749 | int ret; | |
1750 | ||
867cc1fb NB |
1751 | ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr, |
1752 | num_bytes, parent, root_objectid, | |
1753 | owner, offset, 0); | |
5d4f98a2 | 1754 | if (ret != -ENOENT) |
54aa1f4d | 1755 | return ret; |
5d4f98a2 | 1756 | |
b3b4aa74 | 1757 | btrfs_release_path(path); |
5d4f98a2 YZ |
1758 | *ref_ret = NULL; |
1759 | ||
1760 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
b8582eea NB |
1761 | ret = lookup_tree_block_ref(trans, path, bytenr, parent, |
1762 | root_objectid); | |
5d4f98a2 | 1763 | } else { |
bd1d53ef NB |
1764 | ret = lookup_extent_data_ref(trans, path, bytenr, parent, |
1765 | root_objectid, owner, offset); | |
b9473439 | 1766 | } |
5d4f98a2 YZ |
1767 | return ret; |
1768 | } | |
31840ae1 | 1769 | |
5d4f98a2 YZ |
1770 | /* |
1771 | * helper to update/remove inline back ref | |
1772 | */ | |
1773 | static noinline_for_stack | |
61a18f1c | 1774 | void update_inline_extent_backref(struct btrfs_path *path, |
143bede5 JM |
1775 | struct btrfs_extent_inline_ref *iref, |
1776 | int refs_to_mod, | |
fcebe456 JB |
1777 | struct btrfs_delayed_extent_op *extent_op, |
1778 | int *last_ref) | |
5d4f98a2 | 1779 | { |
61a18f1c NB |
1780 | struct extent_buffer *leaf = path->nodes[0]; |
1781 | struct btrfs_fs_info *fs_info = leaf->fs_info; | |
5d4f98a2 YZ |
1782 | struct btrfs_extent_item *ei; |
1783 | struct btrfs_extent_data_ref *dref = NULL; | |
1784 | struct btrfs_shared_data_ref *sref = NULL; | |
1785 | unsigned long ptr; | |
1786 | unsigned long end; | |
1787 | u32 item_size; | |
1788 | int size; | |
1789 | int type; | |
5d4f98a2 YZ |
1790 | u64 refs; |
1791 | ||
5d4f98a2 YZ |
1792 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1793 | refs = btrfs_extent_refs(leaf, ei); | |
1794 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1795 | refs += refs_to_mod; | |
1796 | btrfs_set_extent_refs(leaf, ei, refs); | |
1797 | if (extent_op) | |
1798 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1799 | ||
3de28d57 LB |
1800 | /* |
1801 | * If type is invalid, we should have bailed out after | |
1802 | * lookup_inline_extent_backref(). | |
1803 | */ | |
1804 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
1805 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
5d4f98a2 YZ |
1806 | |
1807 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1808 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1809 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1810 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1811 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1812 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1813 | } else { | |
1814 | refs = 1; | |
1815 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1816 | } |
31840ae1 | 1817 | |
5d4f98a2 YZ |
1818 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1819 | refs += refs_to_mod; | |
1820 | ||
1821 | if (refs > 0) { | |
1822 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1823 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1824 | else | |
1825 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1826 | } else { | |
fcebe456 | 1827 | *last_ref = 1; |
5d4f98a2 YZ |
1828 | size = btrfs_extent_inline_ref_size(type); |
1829 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1830 | ptr = (unsigned long)iref; | |
1831 | end = (unsigned long)ei + item_size; | |
1832 | if (ptr + size < end) | |
1833 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1834 | end - ptr - size); | |
1835 | item_size -= size; | |
87bde3cd | 1836 | btrfs_truncate_item(fs_info, path, item_size, 1); |
5d4f98a2 YZ |
1837 | } |
1838 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1839 | } |
1840 | ||
1841 | static noinline_for_stack | |
1842 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1843 | struct btrfs_path *path, |
1844 | u64 bytenr, u64 num_bytes, u64 parent, | |
1845 | u64 root_objectid, u64 owner, | |
1846 | u64 offset, int refs_to_add, | |
1847 | struct btrfs_delayed_extent_op *extent_op) | |
1848 | { | |
1849 | struct btrfs_extent_inline_ref *iref; | |
1850 | int ret; | |
1851 | ||
867cc1fb NB |
1852 | ret = lookup_inline_extent_backref(trans, path, &iref, bytenr, |
1853 | num_bytes, parent, root_objectid, | |
1854 | owner, offset, 1); | |
5d4f98a2 YZ |
1855 | if (ret == 0) { |
1856 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
61a18f1c NB |
1857 | update_inline_extent_backref(path, iref, refs_to_add, |
1858 | extent_op, NULL); | |
5d4f98a2 | 1859 | } else if (ret == -ENOENT) { |
a639cdeb | 1860 | setup_inline_extent_backref(trans->fs_info, path, iref, parent, |
143bede5 JM |
1861 | root_objectid, owner, offset, |
1862 | refs_to_add, extent_op); | |
1863 | ret = 0; | |
771ed689 | 1864 | } |
5d4f98a2 YZ |
1865 | return ret; |
1866 | } | |
31840ae1 | 1867 | |
5d4f98a2 | 1868 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1869 | struct btrfs_path *path, |
1870 | u64 bytenr, u64 parent, u64 root_objectid, | |
1871 | u64 owner, u64 offset, int refs_to_add) | |
1872 | { | |
1873 | int ret; | |
1874 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1875 | BUG_ON(refs_to_add != 1); | |
10728404 NB |
1876 | ret = insert_tree_block_ref(trans, path, bytenr, parent, |
1877 | root_objectid); | |
5d4f98a2 | 1878 | } else { |
62b895af NB |
1879 | ret = insert_extent_data_ref(trans, path, bytenr, parent, |
1880 | root_objectid, owner, offset, | |
1881 | refs_to_add); | |
5d4f98a2 YZ |
1882 | } |
1883 | return ret; | |
1884 | } | |
56bec294 | 1885 | |
5d4f98a2 | 1886 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1887 | struct btrfs_path *path, |
1888 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1889 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1890 | { |
143bede5 | 1891 | int ret = 0; |
b9473439 | 1892 | |
5d4f98a2 YZ |
1893 | BUG_ON(!is_data && refs_to_drop != 1); |
1894 | if (iref) { | |
61a18f1c NB |
1895 | update_inline_extent_backref(path, iref, -refs_to_drop, NULL, |
1896 | last_ref); | |
5d4f98a2 | 1897 | } else if (is_data) { |
e9f6290d | 1898 | ret = remove_extent_data_ref(trans, path, refs_to_drop, |
fcebe456 | 1899 | last_ref); |
5d4f98a2 | 1900 | } else { |
fcebe456 | 1901 | *last_ref = 1; |
87cc7a8a | 1902 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
5d4f98a2 YZ |
1903 | } |
1904 | return ret; | |
1905 | } | |
1906 | ||
86557861 | 1907 | #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len)) |
d04c6b88 JM |
1908 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1909 | u64 *discarded_bytes) | |
5d4f98a2 | 1910 | { |
86557861 JM |
1911 | int j, ret = 0; |
1912 | u64 bytes_left, end; | |
4d89d377 | 1913 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1914 | |
4d89d377 JM |
1915 | if (WARN_ON(start != aligned_start)) { |
1916 | len -= aligned_start - start; | |
1917 | len = round_down(len, 1 << 9); | |
1918 | start = aligned_start; | |
1919 | } | |
d04c6b88 | 1920 | |
4d89d377 | 1921 | *discarded_bytes = 0; |
86557861 JM |
1922 | |
1923 | if (!len) | |
1924 | return 0; | |
1925 | ||
1926 | end = start + len; | |
1927 | bytes_left = len; | |
1928 | ||
1929 | /* Skip any superblocks on this device. */ | |
1930 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1931 | u64 sb_start = btrfs_sb_offset(j); | |
1932 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1933 | u64 size = sb_start - start; | |
1934 | ||
1935 | if (!in_range(sb_start, start, bytes_left) && | |
1936 | !in_range(sb_end, start, bytes_left) && | |
1937 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1938 | continue; | |
1939 | ||
1940 | /* | |
1941 | * Superblock spans beginning of range. Adjust start and | |
1942 | * try again. | |
1943 | */ | |
1944 | if (sb_start <= start) { | |
1945 | start += sb_end - start; | |
1946 | if (start > end) { | |
1947 | bytes_left = 0; | |
1948 | break; | |
1949 | } | |
1950 | bytes_left = end - start; | |
1951 | continue; | |
1952 | } | |
1953 | ||
1954 | if (size) { | |
1955 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
1956 | GFP_NOFS, 0); | |
1957 | if (!ret) | |
1958 | *discarded_bytes += size; | |
1959 | else if (ret != -EOPNOTSUPP) | |
1960 | return ret; | |
1961 | } | |
1962 | ||
1963 | start = sb_end; | |
1964 | if (start > end) { | |
1965 | bytes_left = 0; | |
1966 | break; | |
1967 | } | |
1968 | bytes_left = end - start; | |
1969 | } | |
1970 | ||
1971 | if (bytes_left) { | |
1972 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
1973 | GFP_NOFS, 0); |
1974 | if (!ret) | |
86557861 | 1975 | *discarded_bytes += bytes_left; |
4d89d377 | 1976 | } |
d04c6b88 | 1977 | return ret; |
5d4f98a2 | 1978 | } |
5d4f98a2 | 1979 | |
2ff7e61e | 1980 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 1981 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1982 | { |
5d4f98a2 | 1983 | int ret; |
5378e607 | 1984 | u64 discarded_bytes = 0; |
a1d3c478 | 1985 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1986 | |
e244a0ae | 1987 | |
2999241d FM |
1988 | /* |
1989 | * Avoid races with device replace and make sure our bbio has devices | |
1990 | * associated to its stripes that don't go away while we are discarding. | |
1991 | */ | |
0b246afa | 1992 | btrfs_bio_counter_inc_blocked(fs_info); |
5d4f98a2 | 1993 | /* Tell the block device(s) that the sectors can be discarded */ |
0b246afa JM |
1994 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes, |
1995 | &bbio, 0); | |
79787eaa | 1996 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 1997 | if (!ret) { |
a1d3c478 | 1998 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
1999 | int i; |
2000 | ||
5d4f98a2 | 2001 | |
a1d3c478 | 2002 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 2003 | u64 bytes; |
38b5f68e AJ |
2004 | struct request_queue *req_q; |
2005 | ||
627e0873 FM |
2006 | if (!stripe->dev->bdev) { |
2007 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); | |
2008 | continue; | |
2009 | } | |
38b5f68e AJ |
2010 | req_q = bdev_get_queue(stripe->dev->bdev); |
2011 | if (!blk_queue_discard(req_q)) | |
d5e2003c JB |
2012 | continue; |
2013 | ||
5378e607 LD |
2014 | ret = btrfs_issue_discard(stripe->dev->bdev, |
2015 | stripe->physical, | |
d04c6b88 JM |
2016 | stripe->length, |
2017 | &bytes); | |
5378e607 | 2018 | if (!ret) |
d04c6b88 | 2019 | discarded_bytes += bytes; |
5378e607 | 2020 | else if (ret != -EOPNOTSUPP) |
79787eaa | 2021 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
2022 | |
2023 | /* | |
2024 | * Just in case we get back EOPNOTSUPP for some reason, | |
2025 | * just ignore the return value so we don't screw up | |
2026 | * people calling discard_extent. | |
2027 | */ | |
2028 | ret = 0; | |
5d4f98a2 | 2029 | } |
6e9606d2 | 2030 | btrfs_put_bbio(bbio); |
5d4f98a2 | 2031 | } |
0b246afa | 2032 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
2033 | |
2034 | if (actual_bytes) | |
2035 | *actual_bytes = discarded_bytes; | |
2036 | ||
5d4f98a2 | 2037 | |
53b381b3 DW |
2038 | if (ret == -EOPNOTSUPP) |
2039 | ret = 0; | |
5d4f98a2 | 2040 | return ret; |
5d4f98a2 YZ |
2041 | } |
2042 | ||
79787eaa | 2043 | /* Can return -ENOMEM */ |
5d4f98a2 | 2044 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
84f7d8e6 | 2045 | struct btrfs_root *root, |
5d4f98a2 | 2046 | u64 bytenr, u64 num_bytes, u64 parent, |
b06c4bf5 | 2047 | u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 | 2048 | { |
84f7d8e6 | 2049 | struct btrfs_fs_info *fs_info = root->fs_info; |
d7eae340 | 2050 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 2051 | int ret; |
66d7e7f0 | 2052 | |
5d4f98a2 YZ |
2053 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
2054 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
2055 | ||
fd708b81 JB |
2056 | btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid, |
2057 | owner, offset, BTRFS_ADD_DELAYED_REF); | |
2058 | ||
5d4f98a2 | 2059 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
44e1c47d | 2060 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, |
7be07912 OS |
2061 | num_bytes, parent, |
2062 | root_objectid, (int)owner, | |
2063 | BTRFS_ADD_DELAYED_REF, NULL, | |
d7eae340 | 2064 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 2065 | } else { |
88a979c6 | 2066 | ret = btrfs_add_delayed_data_ref(trans, bytenr, |
7be07912 OS |
2067 | num_bytes, parent, |
2068 | root_objectid, owner, offset, | |
d7eae340 OS |
2069 | 0, BTRFS_ADD_DELAYED_REF, |
2070 | &old_ref_mod, &new_ref_mod); | |
5d4f98a2 | 2071 | } |
d7eae340 | 2072 | |
29d2b84c NB |
2073 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) { |
2074 | bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID; | |
2075 | ||
2076 | add_pinned_bytes(fs_info, -num_bytes, metadata, root_objectid); | |
2077 | } | |
d7eae340 | 2078 | |
5d4f98a2 YZ |
2079 | return ret; |
2080 | } | |
2081 | ||
bd3c685e NB |
2082 | /* |
2083 | * __btrfs_inc_extent_ref - insert backreference for a given extent | |
2084 | * | |
2085 | * @trans: Handle of transaction | |
2086 | * | |
2087 | * @node: The delayed ref node used to get the bytenr/length for | |
2088 | * extent whose references are incremented. | |
2089 | * | |
2090 | * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/ | |
2091 | * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical | |
2092 | * bytenr of the parent block. Since new extents are always | |
2093 | * created with indirect references, this will only be the case | |
2094 | * when relocating a shared extent. In that case, root_objectid | |
2095 | * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must | |
2096 | * be 0 | |
2097 | * | |
2098 | * @root_objectid: The id of the root where this modification has originated, | |
2099 | * this can be either one of the well-known metadata trees or | |
2100 | * the subvolume id which references this extent. | |
2101 | * | |
2102 | * @owner: For data extents it is the inode number of the owning file. | |
2103 | * For metadata extents this parameter holds the level in the | |
2104 | * tree of the extent. | |
2105 | * | |
2106 | * @offset: For metadata extents the offset is ignored and is currently | |
2107 | * always passed as 0. For data extents it is the fileoffset | |
2108 | * this extent belongs to. | |
2109 | * | |
2110 | * @refs_to_add Number of references to add | |
2111 | * | |
2112 | * @extent_op Pointer to a structure, holding information necessary when | |
2113 | * updating a tree block's flags | |
2114 | * | |
2115 | */ | |
5d4f98a2 | 2116 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
c682f9b3 | 2117 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2118 | u64 parent, u64 root_objectid, |
2119 | u64 owner, u64 offset, int refs_to_add, | |
2120 | struct btrfs_delayed_extent_op *extent_op) | |
2121 | { | |
2122 | struct btrfs_path *path; | |
2123 | struct extent_buffer *leaf; | |
2124 | struct btrfs_extent_item *item; | |
fcebe456 | 2125 | struct btrfs_key key; |
c682f9b3 QW |
2126 | u64 bytenr = node->bytenr; |
2127 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2128 | u64 refs; |
2129 | int ret; | |
5d4f98a2 YZ |
2130 | |
2131 | path = btrfs_alloc_path(); | |
2132 | if (!path) | |
2133 | return -ENOMEM; | |
2134 | ||
e4058b54 | 2135 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2136 | path->leave_spinning = 1; |
2137 | /* this will setup the path even if it fails to insert the back ref */ | |
a639cdeb NB |
2138 | ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes, |
2139 | parent, root_objectid, owner, | |
2140 | offset, refs_to_add, extent_op); | |
0ed4792a | 2141 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2142 | goto out; |
fcebe456 JB |
2143 | |
2144 | /* | |
2145 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2146 | * inline extent ref, so just update the reference count and add a | |
2147 | * normal backref. | |
2148 | */ | |
5d4f98a2 | 2149 | leaf = path->nodes[0]; |
fcebe456 | 2150 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2151 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2152 | refs = btrfs_extent_refs(leaf, item); | |
2153 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2154 | if (extent_op) | |
2155 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2156 | |
5d4f98a2 | 2157 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2158 | btrfs_release_path(path); |
56bec294 | 2159 | |
e4058b54 | 2160 | path->reada = READA_FORWARD; |
b9473439 | 2161 | path->leave_spinning = 1; |
56bec294 | 2162 | /* now insert the actual backref */ |
37593410 NB |
2163 | ret = insert_extent_backref(trans, path, bytenr, parent, root_objectid, |
2164 | owner, offset, refs_to_add); | |
79787eaa | 2165 | if (ret) |
66642832 | 2166 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 2167 | out: |
56bec294 | 2168 | btrfs_free_path(path); |
30d133fc | 2169 | return ret; |
56bec294 CM |
2170 | } |
2171 | ||
5d4f98a2 | 2172 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2173 | struct btrfs_delayed_ref_node *node, |
2174 | struct btrfs_delayed_extent_op *extent_op, | |
2175 | int insert_reserved) | |
56bec294 | 2176 | { |
5d4f98a2 YZ |
2177 | int ret = 0; |
2178 | struct btrfs_delayed_data_ref *ref; | |
2179 | struct btrfs_key ins; | |
2180 | u64 parent = 0; | |
2181 | u64 ref_root = 0; | |
2182 | u64 flags = 0; | |
2183 | ||
2184 | ins.objectid = node->bytenr; | |
2185 | ins.offset = node->num_bytes; | |
2186 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2187 | ||
2188 | ref = btrfs_delayed_node_to_data_ref(node); | |
2bf98ef3 | 2189 | trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 2190 | |
5d4f98a2 YZ |
2191 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2192 | parent = ref->parent; | |
fcebe456 | 2193 | ref_root = ref->root; |
5d4f98a2 YZ |
2194 | |
2195 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2196 | if (extent_op) |
5d4f98a2 | 2197 | flags |= extent_op->flags_to_set; |
ef89b824 NB |
2198 | ret = alloc_reserved_file_extent(trans, parent, ref_root, |
2199 | flags, ref->objectid, | |
2200 | ref->offset, &ins, | |
2201 | node->ref_mod); | |
5d4f98a2 | 2202 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
2203 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
2204 | ref->objectid, ref->offset, | |
2205 | node->ref_mod, extent_op); | |
5d4f98a2 | 2206 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 2207 | ret = __btrfs_free_extent(trans, node, parent, |
5d4f98a2 YZ |
2208 | ref_root, ref->objectid, |
2209 | ref->offset, node->ref_mod, | |
c682f9b3 | 2210 | extent_op); |
5d4f98a2 YZ |
2211 | } else { |
2212 | BUG(); | |
2213 | } | |
2214 | return ret; | |
2215 | } | |
2216 | ||
2217 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2218 | struct extent_buffer *leaf, | |
2219 | struct btrfs_extent_item *ei) | |
2220 | { | |
2221 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2222 | if (extent_op->update_flags) { | |
2223 | flags |= extent_op->flags_to_set; | |
2224 | btrfs_set_extent_flags(leaf, ei, flags); | |
2225 | } | |
2226 | ||
2227 | if (extent_op->update_key) { | |
2228 | struct btrfs_tree_block_info *bi; | |
2229 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2230 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2231 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2232 | } | |
2233 | } | |
2234 | ||
2235 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
d278850e | 2236 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
2237 | struct btrfs_delayed_extent_op *extent_op) |
2238 | { | |
20b9a2d6 | 2239 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
2240 | struct btrfs_key key; |
2241 | struct btrfs_path *path; | |
2242 | struct btrfs_extent_item *ei; | |
2243 | struct extent_buffer *leaf; | |
2244 | u32 item_size; | |
56bec294 | 2245 | int ret; |
5d4f98a2 | 2246 | int err = 0; |
b1c79e09 | 2247 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2248 | |
79787eaa JM |
2249 | if (trans->aborted) |
2250 | return 0; | |
2251 | ||
0b246afa | 2252 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
2253 | metadata = 0; |
2254 | ||
5d4f98a2 YZ |
2255 | path = btrfs_alloc_path(); |
2256 | if (!path) | |
2257 | return -ENOMEM; | |
2258 | ||
d278850e | 2259 | key.objectid = head->bytenr; |
5d4f98a2 | 2260 | |
3173a18f | 2261 | if (metadata) { |
3173a18f | 2262 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2263 | key.offset = extent_op->level; |
3173a18f JB |
2264 | } else { |
2265 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 2266 | key.offset = head->num_bytes; |
3173a18f JB |
2267 | } |
2268 | ||
2269 | again: | |
e4058b54 | 2270 | path->reada = READA_FORWARD; |
5d4f98a2 | 2271 | path->leave_spinning = 1; |
0b246afa | 2272 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
2273 | if (ret < 0) { |
2274 | err = ret; | |
2275 | goto out; | |
2276 | } | |
2277 | if (ret > 0) { | |
3173a18f | 2278 | if (metadata) { |
55994887 FDBM |
2279 | if (path->slots[0] > 0) { |
2280 | path->slots[0]--; | |
2281 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2282 | path->slots[0]); | |
d278850e | 2283 | if (key.objectid == head->bytenr && |
55994887 | 2284 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 2285 | key.offset == head->num_bytes) |
55994887 FDBM |
2286 | ret = 0; |
2287 | } | |
2288 | if (ret > 0) { | |
2289 | btrfs_release_path(path); | |
2290 | metadata = 0; | |
3173a18f | 2291 | |
d278850e JB |
2292 | key.objectid = head->bytenr; |
2293 | key.offset = head->num_bytes; | |
55994887 FDBM |
2294 | key.type = BTRFS_EXTENT_ITEM_KEY; |
2295 | goto again; | |
2296 | } | |
2297 | } else { | |
2298 | err = -EIO; | |
2299 | goto out; | |
3173a18f | 2300 | } |
5d4f98a2 YZ |
2301 | } |
2302 | ||
2303 | leaf = path->nodes[0]; | |
2304 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ba3c2b19 | 2305 | |
6d8ff4e4 | 2306 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
2307 | err = -EINVAL; |
2308 | btrfs_print_v0_err(fs_info); | |
2309 | btrfs_abort_transaction(trans, err); | |
2310 | goto out; | |
2311 | } | |
2312 | ||
5d4f98a2 YZ |
2313 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2314 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2315 | |
5d4f98a2 YZ |
2316 | btrfs_mark_buffer_dirty(leaf); |
2317 | out: | |
2318 | btrfs_free_path(path); | |
2319 | return err; | |
56bec294 CM |
2320 | } |
2321 | ||
5d4f98a2 | 2322 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2323 | struct btrfs_delayed_ref_node *node, |
2324 | struct btrfs_delayed_extent_op *extent_op, | |
2325 | int insert_reserved) | |
56bec294 CM |
2326 | { |
2327 | int ret = 0; | |
5d4f98a2 | 2328 | struct btrfs_delayed_tree_ref *ref; |
5d4f98a2 YZ |
2329 | u64 parent = 0; |
2330 | u64 ref_root = 0; | |
56bec294 | 2331 | |
5d4f98a2 | 2332 | ref = btrfs_delayed_node_to_tree_ref(node); |
f97806f2 | 2333 | trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 2334 | |
5d4f98a2 YZ |
2335 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2336 | parent = ref->parent; | |
fcebe456 | 2337 | ref_root = ref->root; |
5d4f98a2 | 2338 | |
02794222 | 2339 | if (node->ref_mod != 1) { |
f97806f2 | 2340 | btrfs_err(trans->fs_info, |
02794222 LB |
2341 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
2342 | node->bytenr, node->ref_mod, node->action, ref_root, | |
2343 | parent); | |
2344 | return -EIO; | |
2345 | } | |
5d4f98a2 | 2346 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 2347 | BUG_ON(!extent_op || !extent_op->update_flags); |
21ebfbe7 | 2348 | ret = alloc_reserved_tree_block(trans, node, extent_op); |
5d4f98a2 | 2349 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
2350 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
2351 | ref->level, 0, 1, extent_op); | |
5d4f98a2 | 2352 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 2353 | ret = __btrfs_free_extent(trans, node, parent, ref_root, |
c682f9b3 | 2354 | ref->level, 0, 1, extent_op); |
5d4f98a2 YZ |
2355 | } else { |
2356 | BUG(); | |
2357 | } | |
56bec294 CM |
2358 | return ret; |
2359 | } | |
2360 | ||
2361 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 2362 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2363 | struct btrfs_delayed_ref_node *node, |
2364 | struct btrfs_delayed_extent_op *extent_op, | |
2365 | int insert_reserved) | |
56bec294 | 2366 | { |
79787eaa JM |
2367 | int ret = 0; |
2368 | ||
857cc2fc JB |
2369 | if (trans->aborted) { |
2370 | if (insert_reserved) | |
5fac7f9e | 2371 | btrfs_pin_extent(trans->fs_info, node->bytenr, |
857cc2fc | 2372 | node->num_bytes, 1); |
79787eaa | 2373 | return 0; |
857cc2fc | 2374 | } |
79787eaa | 2375 | |
5d4f98a2 YZ |
2376 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2377 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
f97806f2 | 2378 | ret = run_delayed_tree_ref(trans, node, extent_op, |
5d4f98a2 YZ |
2379 | insert_reserved); |
2380 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2381 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2bf98ef3 | 2382 | ret = run_delayed_data_ref(trans, node, extent_op, |
5d4f98a2 YZ |
2383 | insert_reserved); |
2384 | else | |
2385 | BUG(); | |
80ee54bf JB |
2386 | if (ret && insert_reserved) |
2387 | btrfs_pin_extent(trans->fs_info, node->bytenr, | |
2388 | node->num_bytes, 1); | |
5d4f98a2 | 2389 | return ret; |
56bec294 CM |
2390 | } |
2391 | ||
c6fc2454 | 2392 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2393 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2394 | { | |
cffc3374 FM |
2395 | struct btrfs_delayed_ref_node *ref; |
2396 | ||
e3d03965 | 2397 | if (RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
c6fc2454 | 2398 | return NULL; |
d7df2c79 | 2399 | |
cffc3374 FM |
2400 | /* |
2401 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2402 | * This is to prevent a ref count from going down to zero, which deletes | |
2403 | * the extent item from the extent tree, when there still are references | |
2404 | * to add, which would fail because they would not find the extent item. | |
2405 | */ | |
1d57ee94 WX |
2406 | if (!list_empty(&head->ref_add_list)) |
2407 | return list_first_entry(&head->ref_add_list, | |
2408 | struct btrfs_delayed_ref_node, add_list); | |
2409 | ||
e3d03965 | 2410 | ref = rb_entry(rb_first_cached(&head->ref_tree), |
0e0adbcf | 2411 | struct btrfs_delayed_ref_node, ref_node); |
1d57ee94 WX |
2412 | ASSERT(list_empty(&ref->add_list)); |
2413 | return ref; | |
56bec294 CM |
2414 | } |
2415 | ||
2eadaa22 JB |
2416 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
2417 | struct btrfs_delayed_ref_head *head) | |
2418 | { | |
2419 | spin_lock(&delayed_refs->lock); | |
2420 | head->processing = 0; | |
2421 | delayed_refs->num_heads_ready++; | |
2422 | spin_unlock(&delayed_refs->lock); | |
2423 | btrfs_delayed_ref_unlock(head); | |
2424 | } | |
2425 | ||
b00e6250 | 2426 | static int cleanup_extent_op(struct btrfs_trans_handle *trans, |
b00e6250 JB |
2427 | struct btrfs_delayed_ref_head *head) |
2428 | { | |
2429 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
2430 | int ret; | |
2431 | ||
2432 | if (!extent_op) | |
2433 | return 0; | |
2434 | head->extent_op = NULL; | |
2435 | if (head->must_insert_reserved) { | |
2436 | btrfs_free_delayed_extent_op(extent_op); | |
2437 | return 0; | |
2438 | } | |
2439 | spin_unlock(&head->lock); | |
20b9a2d6 | 2440 | ret = run_delayed_extent_op(trans, head, extent_op); |
b00e6250 JB |
2441 | btrfs_free_delayed_extent_op(extent_op); |
2442 | return ret ? ret : 1; | |
2443 | } | |
2444 | ||
194ab0bc | 2445 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
194ab0bc JB |
2446 | struct btrfs_delayed_ref_head *head) |
2447 | { | |
f9871edd NB |
2448 | |
2449 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
194ab0bc JB |
2450 | struct btrfs_delayed_ref_root *delayed_refs; |
2451 | int ret; | |
2452 | ||
2453 | delayed_refs = &trans->transaction->delayed_refs; | |
2454 | ||
c4d56d4a | 2455 | ret = cleanup_extent_op(trans, head); |
194ab0bc JB |
2456 | if (ret < 0) { |
2457 | unselect_delayed_ref_head(delayed_refs, head); | |
2458 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
2459 | return ret; | |
2460 | } else if (ret) { | |
2461 | return ret; | |
2462 | } | |
2463 | ||
2464 | /* | |
2465 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
2466 | * and then re-check to make sure nobody got added. | |
2467 | */ | |
2468 | spin_unlock(&head->lock); | |
2469 | spin_lock(&delayed_refs->lock); | |
2470 | spin_lock(&head->lock); | |
e3d03965 | 2471 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) { |
194ab0bc JB |
2472 | spin_unlock(&head->lock); |
2473 | spin_unlock(&delayed_refs->lock); | |
2474 | return 1; | |
2475 | } | |
194ab0bc | 2476 | delayed_refs->num_heads--; |
5c9d028b | 2477 | rb_erase_cached(&head->href_node, &delayed_refs->href_root); |
d278850e | 2478 | RB_CLEAR_NODE(&head->href_node); |
c1103f7a | 2479 | spin_unlock(&head->lock); |
1e7a1421 | 2480 | spin_unlock(&delayed_refs->lock); |
c1103f7a JB |
2481 | atomic_dec(&delayed_refs->num_entries); |
2482 | ||
d278850e | 2483 | trace_run_delayed_ref_head(fs_info, head, 0); |
c1103f7a JB |
2484 | |
2485 | if (head->total_ref_mod < 0) { | |
5e388e95 NB |
2486 | struct btrfs_space_info *space_info; |
2487 | u64 flags; | |
c1103f7a | 2488 | |
5e388e95 NB |
2489 | if (head->is_data) |
2490 | flags = BTRFS_BLOCK_GROUP_DATA; | |
2491 | else if (head->is_system) | |
2492 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
2493 | else | |
2494 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
2495 | space_info = __find_space_info(fs_info, flags); | |
2496 | ASSERT(space_info); | |
dec59fa3 EL |
2497 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
2498 | -head->num_bytes, | |
2499 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
c1103f7a JB |
2500 | |
2501 | if (head->is_data) { | |
2502 | spin_lock(&delayed_refs->lock); | |
d278850e | 2503 | delayed_refs->pending_csums -= head->num_bytes; |
c1103f7a JB |
2504 | spin_unlock(&delayed_refs->lock); |
2505 | } | |
2506 | } | |
2507 | ||
2508 | if (head->must_insert_reserved) { | |
d278850e JB |
2509 | btrfs_pin_extent(fs_info, head->bytenr, |
2510 | head->num_bytes, 1); | |
c1103f7a | 2511 | if (head->is_data) { |
d278850e JB |
2512 | ret = btrfs_del_csums(trans, fs_info, head->bytenr, |
2513 | head->num_bytes); | |
c1103f7a JB |
2514 | } |
2515 | } | |
2516 | ||
2517 | /* Also free its reserved qgroup space */ | |
2518 | btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root, | |
2519 | head->qgroup_reserved); | |
2520 | btrfs_delayed_ref_unlock(head); | |
d278850e | 2521 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
2522 | return 0; |
2523 | } | |
2524 | ||
b1cdbcb5 NB |
2525 | static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head( |
2526 | struct btrfs_trans_handle *trans) | |
2527 | { | |
2528 | struct btrfs_delayed_ref_root *delayed_refs = | |
2529 | &trans->transaction->delayed_refs; | |
2530 | struct btrfs_delayed_ref_head *head = NULL; | |
2531 | int ret; | |
2532 | ||
2533 | spin_lock(&delayed_refs->lock); | |
5637c74b | 2534 | head = btrfs_select_ref_head(delayed_refs); |
b1cdbcb5 NB |
2535 | if (!head) { |
2536 | spin_unlock(&delayed_refs->lock); | |
2537 | return head; | |
2538 | } | |
2539 | ||
2540 | /* | |
2541 | * Grab the lock that says we are going to process all the refs for | |
2542 | * this head | |
2543 | */ | |
9e920a6f | 2544 | ret = btrfs_delayed_ref_lock(delayed_refs, head); |
b1cdbcb5 NB |
2545 | spin_unlock(&delayed_refs->lock); |
2546 | ||
2547 | /* | |
2548 | * We may have dropped the spin lock to get the head mutex lock, and | |
2549 | * that might have given someone else time to free the head. If that's | |
2550 | * true, it has been removed from our list and we can move on. | |
2551 | */ | |
2552 | if (ret == -EAGAIN) | |
2553 | head = ERR_PTR(-EAGAIN); | |
2554 | ||
2555 | return head; | |
2556 | } | |
2557 | ||
e7261386 NB |
2558 | static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans, |
2559 | struct btrfs_delayed_ref_head *locked_ref, | |
2560 | unsigned long *run_refs) | |
2561 | { | |
2562 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2563 | struct btrfs_delayed_ref_root *delayed_refs; | |
2564 | struct btrfs_delayed_extent_op *extent_op; | |
2565 | struct btrfs_delayed_ref_node *ref; | |
2566 | int must_insert_reserved = 0; | |
2567 | int ret; | |
2568 | ||
2569 | delayed_refs = &trans->transaction->delayed_refs; | |
2570 | ||
0110a4c4 NB |
2571 | lockdep_assert_held(&locked_ref->mutex); |
2572 | lockdep_assert_held(&locked_ref->lock); | |
2573 | ||
e7261386 NB |
2574 | while ((ref = select_delayed_ref(locked_ref))) { |
2575 | if (ref->seq && | |
2576 | btrfs_check_delayed_seq(fs_info, ref->seq)) { | |
2577 | spin_unlock(&locked_ref->lock); | |
2578 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
2579 | return -EAGAIN; | |
2580 | } | |
2581 | ||
2582 | (*run_refs)++; | |
2583 | ref->in_tree = 0; | |
2584 | rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree); | |
2585 | RB_CLEAR_NODE(&ref->ref_node); | |
2586 | if (!list_empty(&ref->add_list)) | |
2587 | list_del(&ref->add_list); | |
2588 | /* | |
2589 | * When we play the delayed ref, also correct the ref_mod on | |
2590 | * head | |
2591 | */ | |
2592 | switch (ref->action) { | |
2593 | case BTRFS_ADD_DELAYED_REF: | |
2594 | case BTRFS_ADD_DELAYED_EXTENT: | |
2595 | locked_ref->ref_mod -= ref->ref_mod; | |
2596 | break; | |
2597 | case BTRFS_DROP_DELAYED_REF: | |
2598 | locked_ref->ref_mod += ref->ref_mod; | |
2599 | break; | |
2600 | default: | |
2601 | WARN_ON(1); | |
2602 | } | |
2603 | atomic_dec(&delayed_refs->num_entries); | |
2604 | ||
2605 | /* | |
2606 | * Record the must_insert_reserved flag before we drop the | |
2607 | * spin lock. | |
2608 | */ | |
2609 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2610 | locked_ref->must_insert_reserved = 0; | |
2611 | ||
2612 | extent_op = locked_ref->extent_op; | |
2613 | locked_ref->extent_op = NULL; | |
2614 | spin_unlock(&locked_ref->lock); | |
2615 | ||
2616 | ret = run_one_delayed_ref(trans, ref, extent_op, | |
2617 | must_insert_reserved); | |
2618 | ||
2619 | btrfs_free_delayed_extent_op(extent_op); | |
2620 | if (ret) { | |
2621 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
2622 | btrfs_put_delayed_ref(ref); | |
2623 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", | |
2624 | ret); | |
2625 | return ret; | |
2626 | } | |
2627 | ||
2628 | btrfs_put_delayed_ref(ref); | |
2629 | cond_resched(); | |
2630 | ||
2631 | spin_lock(&locked_ref->lock); | |
2632 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); | |
2633 | } | |
2634 | ||
2635 | return 0; | |
2636 | } | |
2637 | ||
79787eaa JM |
2638 | /* |
2639 | * Returns 0 on success or if called with an already aborted transaction. | |
2640 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2641 | */ | |
d7df2c79 | 2642 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
d7df2c79 | 2643 | unsigned long nr) |
56bec294 | 2644 | { |
0a1e458a | 2645 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 | 2646 | struct btrfs_delayed_ref_root *delayed_refs; |
56bec294 | 2647 | struct btrfs_delayed_ref_head *locked_ref = NULL; |
0a2b2a84 | 2648 | ktime_t start = ktime_get(); |
56bec294 | 2649 | int ret; |
d7df2c79 | 2650 | unsigned long count = 0; |
0a2b2a84 | 2651 | unsigned long actual_count = 0; |
56bec294 CM |
2652 | |
2653 | delayed_refs = &trans->transaction->delayed_refs; | |
0110a4c4 | 2654 | do { |
56bec294 | 2655 | if (!locked_ref) { |
b1cdbcb5 | 2656 | locked_ref = btrfs_obtain_ref_head(trans); |
0110a4c4 NB |
2657 | if (IS_ERR_OR_NULL(locked_ref)) { |
2658 | if (PTR_ERR(locked_ref) == -EAGAIN) { | |
2659 | continue; | |
2660 | } else { | |
2661 | break; | |
2662 | } | |
56bec294 | 2663 | } |
0110a4c4 | 2664 | count++; |
56bec294 | 2665 | } |
2c3cf7d5 FM |
2666 | /* |
2667 | * We need to try and merge add/drops of the same ref since we | |
2668 | * can run into issues with relocate dropping the implicit ref | |
2669 | * and then it being added back again before the drop can | |
2670 | * finish. If we merged anything we need to re-loop so we can | |
2671 | * get a good ref. | |
2672 | * Or we can get node references of the same type that weren't | |
2673 | * merged when created due to bumps in the tree mod seq, and | |
2674 | * we need to merge them to prevent adding an inline extent | |
2675 | * backref before dropping it (triggering a BUG_ON at | |
2676 | * insert_inline_extent_backref()). | |
2677 | */ | |
d7df2c79 | 2678 | spin_lock(&locked_ref->lock); |
be97f133 | 2679 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); |
ae1e206b | 2680 | |
0110a4c4 NB |
2681 | ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, |
2682 | &actual_count); | |
2683 | if (ret < 0 && ret != -EAGAIN) { | |
2684 | /* | |
2685 | * Error, btrfs_run_delayed_refs_for_head already | |
2686 | * unlocked everything so just bail out | |
2687 | */ | |
2688 | return ret; | |
2689 | } else if (!ret) { | |
2690 | /* | |
2691 | * Success, perform the usual cleanup of a processed | |
2692 | * head | |
2693 | */ | |
f9871edd | 2694 | ret = cleanup_ref_head(trans, locked_ref); |
194ab0bc | 2695 | if (ret > 0 ) { |
b00e6250 JB |
2696 | /* We dropped our lock, we need to loop. */ |
2697 | ret = 0; | |
d7df2c79 | 2698 | continue; |
194ab0bc JB |
2699 | } else if (ret) { |
2700 | return ret; | |
5d4f98a2 | 2701 | } |
22cd2e7d | 2702 | } |
1ce7a5ec | 2703 | |
b00e6250 | 2704 | /* |
0110a4c4 NB |
2705 | * Either success case or btrfs_run_delayed_refs_for_head |
2706 | * returned -EAGAIN, meaning we need to select another head | |
b00e6250 | 2707 | */ |
b00e6250 | 2708 | |
0110a4c4 | 2709 | locked_ref = NULL; |
c3e69d58 | 2710 | cond_resched(); |
0110a4c4 | 2711 | } while ((nr != -1 && count < nr) || locked_ref); |
0a2b2a84 JB |
2712 | |
2713 | /* | |
2714 | * We don't want to include ref heads since we can have empty ref heads | |
2715 | * and those will drastically skew our runtime down since we just do | |
2716 | * accounting, no actual extent tree updates. | |
2717 | */ | |
2718 | if (actual_count > 0) { | |
2719 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2720 | u64 avg; | |
2721 | ||
2722 | /* | |
2723 | * We weigh the current average higher than our current runtime | |
2724 | * to avoid large swings in the average. | |
2725 | */ | |
2726 | spin_lock(&delayed_refs->lock); | |
2727 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2728 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2729 | spin_unlock(&delayed_refs->lock); |
2730 | } | |
d7df2c79 | 2731 | return 0; |
c3e69d58 CM |
2732 | } |
2733 | ||
709c0486 AJ |
2734 | #ifdef SCRAMBLE_DELAYED_REFS |
2735 | /* | |
2736 | * Normally delayed refs get processed in ascending bytenr order. This | |
2737 | * correlates in most cases to the order added. To expose dependencies on this | |
2738 | * order, we start to process the tree in the middle instead of the beginning | |
2739 | */ | |
2740 | static u64 find_middle(struct rb_root *root) | |
2741 | { | |
2742 | struct rb_node *n = root->rb_node; | |
2743 | struct btrfs_delayed_ref_node *entry; | |
2744 | int alt = 1; | |
2745 | u64 middle; | |
2746 | u64 first = 0, last = 0; | |
2747 | ||
2748 | n = rb_first(root); | |
2749 | if (n) { | |
2750 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2751 | first = entry->bytenr; | |
2752 | } | |
2753 | n = rb_last(root); | |
2754 | if (n) { | |
2755 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2756 | last = entry->bytenr; | |
2757 | } | |
2758 | n = root->rb_node; | |
2759 | ||
2760 | while (n) { | |
2761 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2762 | WARN_ON(!entry->in_tree); | |
2763 | ||
2764 | middle = entry->bytenr; | |
2765 | ||
2766 | if (alt) | |
2767 | n = n->rb_left; | |
2768 | else | |
2769 | n = n->rb_right; | |
2770 | ||
2771 | alt = 1 - alt; | |
2772 | } | |
2773 | return middle; | |
2774 | } | |
2775 | #endif | |
2776 | ||
2ff7e61e | 2777 | static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads) |
1be41b78 JB |
2778 | { |
2779 | u64 num_bytes; | |
2780 | ||
2781 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2782 | sizeof(struct btrfs_extent_inline_ref)); | |
0b246afa | 2783 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1be41b78 JB |
2784 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
2785 | ||
2786 | /* | |
2787 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2788 | * closer to what we're really going to want to use. |
1be41b78 | 2789 | */ |
0b246afa | 2790 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info)); |
1be41b78 JB |
2791 | } |
2792 | ||
1262133b JB |
2793 | /* |
2794 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2795 | * would require to store the csums for that many bytes. | |
2796 | */ | |
2ff7e61e | 2797 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2798 | { |
2799 | u64 csum_size; | |
2800 | u64 num_csums_per_leaf; | |
2801 | u64 num_csums; | |
2802 | ||
0b246afa | 2803 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2804 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2805 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2806 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2807 | num_csums += num_csums_per_leaf - 1; |
2808 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2809 | return num_csums; | |
2810 | } | |
2811 | ||
af9b8a0e | 2812 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans) |
1be41b78 | 2813 | { |
af9b8a0e | 2814 | struct btrfs_fs_info *fs_info = trans->fs_info; |
1be41b78 JB |
2815 | struct btrfs_block_rsv *global_rsv; |
2816 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
1262133b | 2817 | u64 csum_bytes = trans->transaction->delayed_refs.pending_csums; |
165c8b02 | 2818 | unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs; |
cb723e49 | 2819 | u64 num_bytes, num_dirty_bgs_bytes; |
1be41b78 JB |
2820 | int ret = 0; |
2821 | ||
0b246afa | 2822 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
2ff7e61e | 2823 | num_heads = heads_to_leaves(fs_info, num_heads); |
1be41b78 | 2824 | if (num_heads > 1) |
0b246afa | 2825 | num_bytes += (num_heads - 1) * fs_info->nodesize; |
1be41b78 | 2826 | num_bytes <<= 1; |
2ff7e61e JM |
2827 | num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) * |
2828 | fs_info->nodesize; | |
0b246afa | 2829 | num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info, |
cb723e49 | 2830 | num_dirty_bgs); |
0b246afa | 2831 | global_rsv = &fs_info->global_block_rsv; |
1be41b78 JB |
2832 | |
2833 | /* | |
2834 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2835 | * wiggle room since running delayed refs can create more delayed refs. | |
2836 | */ | |
cb723e49 JB |
2837 | if (global_rsv->space_info->full) { |
2838 | num_dirty_bgs_bytes <<= 1; | |
1be41b78 | 2839 | num_bytes <<= 1; |
cb723e49 | 2840 | } |
1be41b78 JB |
2841 | |
2842 | spin_lock(&global_rsv->lock); | |
cb723e49 | 2843 | if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes) |
1be41b78 JB |
2844 | ret = 1; |
2845 | spin_unlock(&global_rsv->lock); | |
2846 | return ret; | |
2847 | } | |
2848 | ||
7c861627 | 2849 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans) |
0a2b2a84 | 2850 | { |
0a2b2a84 JB |
2851 | u64 num_entries = |
2852 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2853 | u64 avg_runtime; | |
a79b7d4b | 2854 | u64 val; |
0a2b2a84 JB |
2855 | |
2856 | smp_mb(); | |
7c861627 | 2857 | avg_runtime = trans->fs_info->avg_delayed_ref_runtime; |
a79b7d4b | 2858 | val = num_entries * avg_runtime; |
dc1a90c6 | 2859 | if (val >= NSEC_PER_SEC) |
0a2b2a84 | 2860 | return 1; |
a79b7d4b CM |
2861 | if (val >= NSEC_PER_SEC / 2) |
2862 | return 2; | |
0a2b2a84 | 2863 | |
af9b8a0e | 2864 | return btrfs_check_space_for_delayed_refs(trans); |
0a2b2a84 JB |
2865 | } |
2866 | ||
a79b7d4b CM |
2867 | struct async_delayed_refs { |
2868 | struct btrfs_root *root; | |
31b9655f | 2869 | u64 transid; |
a79b7d4b CM |
2870 | int count; |
2871 | int error; | |
2872 | int sync; | |
2873 | struct completion wait; | |
2874 | struct btrfs_work work; | |
2875 | }; | |
2876 | ||
2ff7e61e JM |
2877 | static inline struct async_delayed_refs * |
2878 | to_async_delayed_refs(struct btrfs_work *work) | |
2879 | { | |
2880 | return container_of(work, struct async_delayed_refs, work); | |
2881 | } | |
2882 | ||
a79b7d4b CM |
2883 | static void delayed_ref_async_start(struct btrfs_work *work) |
2884 | { | |
2ff7e61e | 2885 | struct async_delayed_refs *async = to_async_delayed_refs(work); |
a79b7d4b | 2886 | struct btrfs_trans_handle *trans; |
2ff7e61e | 2887 | struct btrfs_fs_info *fs_info = async->root->fs_info; |
a79b7d4b CM |
2888 | int ret; |
2889 | ||
0f873eca | 2890 | /* if the commit is already started, we don't need to wait here */ |
2ff7e61e | 2891 | if (btrfs_transaction_blocked(fs_info)) |
31b9655f | 2892 | goto done; |
31b9655f | 2893 | |
0f873eca CM |
2894 | trans = btrfs_join_transaction(async->root); |
2895 | if (IS_ERR(trans)) { | |
2896 | async->error = PTR_ERR(trans); | |
a79b7d4b CM |
2897 | goto done; |
2898 | } | |
2899 | ||
2900 | /* | |
01327610 | 2901 | * trans->sync means that when we call end_transaction, we won't |
a79b7d4b CM |
2902 | * wait on delayed refs |
2903 | */ | |
2904 | trans->sync = true; | |
0f873eca CM |
2905 | |
2906 | /* Don't bother flushing if we got into a different transaction */ | |
2907 | if (trans->transid > async->transid) | |
2908 | goto end; | |
2909 | ||
c79a70b1 | 2910 | ret = btrfs_run_delayed_refs(trans, async->count); |
a79b7d4b CM |
2911 | if (ret) |
2912 | async->error = ret; | |
0f873eca | 2913 | end: |
3a45bb20 | 2914 | ret = btrfs_end_transaction(trans); |
a79b7d4b CM |
2915 | if (ret && !async->error) |
2916 | async->error = ret; | |
2917 | done: | |
2918 | if (async->sync) | |
2919 | complete(&async->wait); | |
2920 | else | |
2921 | kfree(async); | |
2922 | } | |
2923 | ||
2ff7e61e | 2924 | int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info, |
31b9655f | 2925 | unsigned long count, u64 transid, int wait) |
a79b7d4b CM |
2926 | { |
2927 | struct async_delayed_refs *async; | |
2928 | int ret; | |
2929 | ||
2930 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
2931 | if (!async) | |
2932 | return -ENOMEM; | |
2933 | ||
0b246afa | 2934 | async->root = fs_info->tree_root; |
a79b7d4b CM |
2935 | async->count = count; |
2936 | async->error = 0; | |
31b9655f | 2937 | async->transid = transid; |
a79b7d4b CM |
2938 | if (wait) |
2939 | async->sync = 1; | |
2940 | else | |
2941 | async->sync = 0; | |
2942 | init_completion(&async->wait); | |
2943 | ||
9e0af237 LB |
2944 | btrfs_init_work(&async->work, btrfs_extent_refs_helper, |
2945 | delayed_ref_async_start, NULL, NULL); | |
a79b7d4b | 2946 | |
0b246afa | 2947 | btrfs_queue_work(fs_info->extent_workers, &async->work); |
a79b7d4b CM |
2948 | |
2949 | if (wait) { | |
2950 | wait_for_completion(&async->wait); | |
2951 | ret = async->error; | |
2952 | kfree(async); | |
2953 | return ret; | |
2954 | } | |
2955 | return 0; | |
2956 | } | |
2957 | ||
c3e69d58 CM |
2958 | /* |
2959 | * this starts processing the delayed reference count updates and | |
2960 | * extent insertions we have queued up so far. count can be | |
2961 | * 0, which means to process everything in the tree at the start | |
2962 | * of the run (but not newly added entries), or it can be some target | |
2963 | * number you'd like to process. | |
79787eaa JM |
2964 | * |
2965 | * Returns 0 on success or if called with an aborted transaction | |
2966 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2967 | */ |
2968 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
c79a70b1 | 2969 | unsigned long count) |
c3e69d58 | 2970 | { |
c79a70b1 | 2971 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 CM |
2972 | struct rb_node *node; |
2973 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2974 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2975 | int ret; |
2976 | int run_all = count == (unsigned long)-1; | |
c3e69d58 | 2977 | |
79787eaa JM |
2978 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2979 | if (trans->aborted) | |
2980 | return 0; | |
2981 | ||
0b246afa | 2982 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2983 | return 0; |
2984 | ||
c3e69d58 | 2985 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 2986 | if (count == 0) |
d7df2c79 | 2987 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 2988 | |
c3e69d58 | 2989 | again: |
709c0486 AJ |
2990 | #ifdef SCRAMBLE_DELAYED_REFS |
2991 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2992 | #endif | |
0a1e458a | 2993 | ret = __btrfs_run_delayed_refs(trans, count); |
d7df2c79 | 2994 | if (ret < 0) { |
66642832 | 2995 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 2996 | return ret; |
eb099670 | 2997 | } |
c3e69d58 | 2998 | |
56bec294 | 2999 | if (run_all) { |
d7df2c79 | 3000 | if (!list_empty(&trans->new_bgs)) |
6c686b35 | 3001 | btrfs_create_pending_block_groups(trans); |
ea658bad | 3002 | |
d7df2c79 | 3003 | spin_lock(&delayed_refs->lock); |
5c9d028b | 3004 | node = rb_first_cached(&delayed_refs->href_root); |
d7df2c79 JB |
3005 | if (!node) { |
3006 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 3007 | goto out; |
d7df2c79 | 3008 | } |
d278850e JB |
3009 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
3010 | href_node); | |
3011 | refcount_inc(&head->refs); | |
3012 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 3013 | |
d278850e JB |
3014 | /* Mutex was contended, block until it's released and retry. */ |
3015 | mutex_lock(&head->mutex); | |
3016 | mutex_unlock(&head->mutex); | |
56bec294 | 3017 | |
d278850e | 3018 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 3019 | cond_resched(); |
56bec294 | 3020 | goto again; |
5f39d397 | 3021 | } |
54aa1f4d | 3022 | out: |
a28ec197 CM |
3023 | return 0; |
3024 | } | |
3025 | ||
5d4f98a2 | 3026 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2ff7e61e | 3027 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 3028 | u64 bytenr, u64 num_bytes, u64 flags, |
b1c79e09 | 3029 | int level, int is_data) |
5d4f98a2 YZ |
3030 | { |
3031 | struct btrfs_delayed_extent_op *extent_op; | |
3032 | int ret; | |
3033 | ||
78a6184a | 3034 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
3035 | if (!extent_op) |
3036 | return -ENOMEM; | |
3037 | ||
3038 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
3039 | extent_op->update_flags = true; |
3040 | extent_op->update_key = false; | |
3041 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 3042 | extent_op->level = level; |
5d4f98a2 | 3043 | |
0b246afa | 3044 | ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr, |
66d7e7f0 | 3045 | num_bytes, extent_op); |
5d4f98a2 | 3046 | if (ret) |
78a6184a | 3047 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
3048 | return ret; |
3049 | } | |
3050 | ||
e4c3b2dc | 3051 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3052 | struct btrfs_path *path, |
3053 | u64 objectid, u64 offset, u64 bytenr) | |
3054 | { | |
3055 | struct btrfs_delayed_ref_head *head; | |
3056 | struct btrfs_delayed_ref_node *ref; | |
3057 | struct btrfs_delayed_data_ref *data_ref; | |
3058 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 3059 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 3060 | struct rb_node *node; |
5d4f98a2 YZ |
3061 | int ret = 0; |
3062 | ||
998ac6d2 | 3063 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 3064 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 3065 | if (cur_trans) |
3066 | refcount_inc(&cur_trans->use_count); | |
3067 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
3068 | if (!cur_trans) |
3069 | return 0; | |
3070 | ||
3071 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 3072 | spin_lock(&delayed_refs->lock); |
f72ad18e | 3073 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
3074 | if (!head) { |
3075 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 3076 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
3077 | return 0; |
3078 | } | |
5d4f98a2 YZ |
3079 | |
3080 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 3081 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
3082 | spin_unlock(&delayed_refs->lock); |
3083 | ||
b3b4aa74 | 3084 | btrfs_release_path(path); |
5d4f98a2 | 3085 | |
8cc33e5c DS |
3086 | /* |
3087 | * Mutex was contended, block until it's released and let | |
3088 | * caller try again | |
3089 | */ | |
5d4f98a2 YZ |
3090 | mutex_lock(&head->mutex); |
3091 | mutex_unlock(&head->mutex); | |
d278850e | 3092 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 3093 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
3094 | return -EAGAIN; |
3095 | } | |
d7df2c79 | 3096 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 3097 | |
d7df2c79 | 3098 | spin_lock(&head->lock); |
0e0adbcf JB |
3099 | /* |
3100 | * XXX: We should replace this with a proper search function in the | |
3101 | * future. | |
3102 | */ | |
e3d03965 LB |
3103 | for (node = rb_first_cached(&head->ref_tree); node; |
3104 | node = rb_next(node)) { | |
0e0adbcf | 3105 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
d7df2c79 JB |
3106 | /* If it's a shared ref we know a cross reference exists */ |
3107 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3108 | ret = 1; | |
3109 | break; | |
3110 | } | |
5d4f98a2 | 3111 | |
d7df2c79 | 3112 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 3113 | |
d7df2c79 JB |
3114 | /* |
3115 | * If our ref doesn't match the one we're currently looking at | |
3116 | * then we have a cross reference. | |
3117 | */ | |
3118 | if (data_ref->root != root->root_key.objectid || | |
3119 | data_ref->objectid != objectid || | |
3120 | data_ref->offset != offset) { | |
3121 | ret = 1; | |
3122 | break; | |
3123 | } | |
5d4f98a2 | 3124 | } |
d7df2c79 | 3125 | spin_unlock(&head->lock); |
5d4f98a2 | 3126 | mutex_unlock(&head->mutex); |
998ac6d2 | 3127 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
3128 | return ret; |
3129 | } | |
3130 | ||
e4c3b2dc | 3131 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3132 | struct btrfs_path *path, |
3133 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d | 3134 | { |
0b246afa JM |
3135 | struct btrfs_fs_info *fs_info = root->fs_info; |
3136 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 3137 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3138 | struct btrfs_extent_data_ref *ref; |
3139 | struct btrfs_extent_inline_ref *iref; | |
3140 | struct btrfs_extent_item *ei; | |
f321e491 | 3141 | struct btrfs_key key; |
5d4f98a2 | 3142 | u32 item_size; |
3de28d57 | 3143 | int type; |
be20aa9d | 3144 | int ret; |
925baedd | 3145 | |
be20aa9d | 3146 | key.objectid = bytenr; |
31840ae1 | 3147 | key.offset = (u64)-1; |
f321e491 | 3148 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 3149 | |
be20aa9d CM |
3150 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
3151 | if (ret < 0) | |
3152 | goto out; | |
79787eaa | 3153 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
3154 | |
3155 | ret = -ENOENT; | |
3156 | if (path->slots[0] == 0) | |
31840ae1 | 3157 | goto out; |
be20aa9d | 3158 | |
31840ae1 | 3159 | path->slots[0]--; |
f321e491 | 3160 | leaf = path->nodes[0]; |
5d4f98a2 | 3161 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3162 | |
5d4f98a2 | 3163 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3164 | goto out; |
f321e491 | 3165 | |
5d4f98a2 YZ |
3166 | ret = 1; |
3167 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
5d4f98a2 | 3168 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
bd09835d | 3169 | |
5d4f98a2 YZ |
3170 | if (item_size != sizeof(*ei) + |
3171 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3172 | goto out; | |
be20aa9d | 3173 | |
5d4f98a2 YZ |
3174 | if (btrfs_extent_generation(leaf, ei) <= |
3175 | btrfs_root_last_snapshot(&root->root_item)) | |
3176 | goto out; | |
3177 | ||
3178 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 LB |
3179 | |
3180 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
3181 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
3182 | goto out; |
3183 | ||
3184 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3185 | if (btrfs_extent_refs(leaf, ei) != | |
3186 | btrfs_extent_data_ref_count(leaf, ref) || | |
3187 | btrfs_extent_data_ref_root(leaf, ref) != | |
3188 | root->root_key.objectid || | |
3189 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3190 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3191 | goto out; | |
3192 | ||
3193 | ret = 0; | |
3194 | out: | |
3195 | return ret; | |
3196 | } | |
3197 | ||
e4c3b2dc LB |
3198 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
3199 | u64 bytenr) | |
5d4f98a2 YZ |
3200 | { |
3201 | struct btrfs_path *path; | |
3202 | int ret; | |
5d4f98a2 YZ |
3203 | |
3204 | path = btrfs_alloc_path(); | |
3205 | if (!path) | |
9132c4ff | 3206 | return -ENOMEM; |
5d4f98a2 YZ |
3207 | |
3208 | do { | |
e4c3b2dc | 3209 | ret = check_committed_ref(root, path, objectid, |
5d4f98a2 YZ |
3210 | offset, bytenr); |
3211 | if (ret && ret != -ENOENT) | |
f321e491 | 3212 | goto out; |
80ff3856 | 3213 | |
380fd066 MT |
3214 | ret = check_delayed_ref(root, path, objectid, offset, bytenr); |
3215 | } while (ret == -EAGAIN); | |
5d4f98a2 | 3216 | |
be20aa9d | 3217 | out: |
80ff3856 | 3218 | btrfs_free_path(path); |
f0486c68 YZ |
3219 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3220 | WARN_ON(ret > 0); | |
f321e491 | 3221 | return ret; |
be20aa9d | 3222 | } |
c5739bba | 3223 | |
5d4f98a2 | 3224 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3225 | struct btrfs_root *root, |
5d4f98a2 | 3226 | struct extent_buffer *buf, |
e339a6b0 | 3227 | int full_backref, int inc) |
31840ae1 | 3228 | { |
0b246afa | 3229 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 3230 | u64 bytenr; |
5d4f98a2 YZ |
3231 | u64 num_bytes; |
3232 | u64 parent; | |
31840ae1 | 3233 | u64 ref_root; |
31840ae1 | 3234 | u32 nritems; |
31840ae1 ZY |
3235 | struct btrfs_key key; |
3236 | struct btrfs_file_extent_item *fi; | |
3237 | int i; | |
3238 | int level; | |
3239 | int ret = 0; | |
2ff7e61e | 3240 | int (*process_func)(struct btrfs_trans_handle *, |
84f7d8e6 | 3241 | struct btrfs_root *, |
b06c4bf5 | 3242 | u64, u64, u64, u64, u64, u64); |
31840ae1 | 3243 | |
fccb84c9 | 3244 | |
0b246afa | 3245 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3246 | return 0; |
fccb84c9 | 3247 | |
31840ae1 | 3248 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3249 | nritems = btrfs_header_nritems(buf); |
3250 | level = btrfs_header_level(buf); | |
3251 | ||
27cdeb70 | 3252 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3253 | return 0; |
31840ae1 | 3254 | |
5d4f98a2 YZ |
3255 | if (inc) |
3256 | process_func = btrfs_inc_extent_ref; | |
3257 | else | |
3258 | process_func = btrfs_free_extent; | |
31840ae1 | 3259 | |
5d4f98a2 YZ |
3260 | if (full_backref) |
3261 | parent = buf->start; | |
3262 | else | |
3263 | parent = 0; | |
3264 | ||
3265 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3266 | if (level == 0) { |
5d4f98a2 | 3267 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3268 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3269 | continue; |
5d4f98a2 | 3270 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3271 | struct btrfs_file_extent_item); |
3272 | if (btrfs_file_extent_type(buf, fi) == | |
3273 | BTRFS_FILE_EXTENT_INLINE) | |
3274 | continue; | |
3275 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3276 | if (bytenr == 0) | |
3277 | continue; | |
5d4f98a2 YZ |
3278 | |
3279 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3280 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
84f7d8e6 | 3281 | ret = process_func(trans, root, bytenr, num_bytes, |
5d4f98a2 | 3282 | parent, ref_root, key.objectid, |
b06c4bf5 | 3283 | key.offset); |
31840ae1 ZY |
3284 | if (ret) |
3285 | goto fail; | |
3286 | } else { | |
5d4f98a2 | 3287 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 3288 | num_bytes = fs_info->nodesize; |
84f7d8e6 | 3289 | ret = process_func(trans, root, bytenr, num_bytes, |
b06c4bf5 | 3290 | parent, ref_root, level - 1, 0); |
31840ae1 ZY |
3291 | if (ret) |
3292 | goto fail; | |
3293 | } | |
3294 | } | |
3295 | return 0; | |
3296 | fail: | |
5d4f98a2 YZ |
3297 | return ret; |
3298 | } | |
3299 | ||
3300 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3301 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3302 | { |
e339a6b0 | 3303 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3304 | } |
3305 | ||
3306 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3307 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3308 | { |
e339a6b0 | 3309 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3310 | } |
3311 | ||
9078a3e1 | 3312 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 3313 | struct btrfs_fs_info *fs_info, |
9078a3e1 CM |
3314 | struct btrfs_path *path, |
3315 | struct btrfs_block_group_cache *cache) | |
3316 | { | |
3317 | int ret; | |
0b246afa | 3318 | struct btrfs_root *extent_root = fs_info->extent_root; |
5f39d397 CM |
3319 | unsigned long bi; |
3320 | struct extent_buffer *leaf; | |
9078a3e1 | 3321 | |
9078a3e1 | 3322 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3323 | if (ret) { |
3324 | if (ret > 0) | |
3325 | ret = -ENOENT; | |
54aa1f4d | 3326 | goto fail; |
df95e7f0 | 3327 | } |
5f39d397 CM |
3328 | |
3329 | leaf = path->nodes[0]; | |
3330 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3331 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3332 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3333 | fail: |
24b89d08 | 3334 | btrfs_release_path(path); |
df95e7f0 | 3335 | return ret; |
9078a3e1 CM |
3336 | |
3337 | } | |
3338 | ||
4a8c9a62 | 3339 | static struct btrfs_block_group_cache * |
2ff7e61e | 3340 | next_block_group(struct btrfs_fs_info *fs_info, |
4a8c9a62 YZ |
3341 | struct btrfs_block_group_cache *cache) |
3342 | { | |
3343 | struct rb_node *node; | |
292cbd51 | 3344 | |
0b246afa | 3345 | spin_lock(&fs_info->block_group_cache_lock); |
292cbd51 FM |
3346 | |
3347 | /* If our block group was removed, we need a full search. */ | |
3348 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3349 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3350 | ||
0b246afa | 3351 | spin_unlock(&fs_info->block_group_cache_lock); |
292cbd51 | 3352 | btrfs_put_block_group(cache); |
0b246afa | 3353 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; |
292cbd51 | 3354 | } |
4a8c9a62 YZ |
3355 | node = rb_next(&cache->cache_node); |
3356 | btrfs_put_block_group(cache); | |
3357 | if (node) { | |
3358 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3359 | cache_node); | |
11dfe35a | 3360 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3361 | } else |
3362 | cache = NULL; | |
0b246afa | 3363 | spin_unlock(&fs_info->block_group_cache_lock); |
4a8c9a62 YZ |
3364 | return cache; |
3365 | } | |
3366 | ||
0af3d00b JB |
3367 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3368 | struct btrfs_trans_handle *trans, | |
3369 | struct btrfs_path *path) | |
3370 | { | |
0b246afa JM |
3371 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
3372 | struct btrfs_root *root = fs_info->tree_root; | |
0af3d00b | 3373 | struct inode *inode = NULL; |
364ecf36 | 3374 | struct extent_changeset *data_reserved = NULL; |
0af3d00b | 3375 | u64 alloc_hint = 0; |
2b20982e | 3376 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3377 | u64 num_pages = 0; |
0af3d00b JB |
3378 | int retries = 0; |
3379 | int ret = 0; | |
3380 | ||
3381 | /* | |
3382 | * If this block group is smaller than 100 megs don't bother caching the | |
3383 | * block group. | |
3384 | */ | |
ee22184b | 3385 | if (block_group->key.offset < (100 * SZ_1M)) { |
0af3d00b JB |
3386 | spin_lock(&block_group->lock); |
3387 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3388 | spin_unlock(&block_group->lock); | |
3389 | return 0; | |
3390 | } | |
3391 | ||
0c0ef4bc JB |
3392 | if (trans->aborted) |
3393 | return 0; | |
0af3d00b | 3394 | again: |
77ab86bf | 3395 | inode = lookup_free_space_inode(fs_info, block_group, path); |
0af3d00b JB |
3396 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { |
3397 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3398 | btrfs_release_path(path); |
0af3d00b JB |
3399 | goto out; |
3400 | } | |
3401 | ||
3402 | if (IS_ERR(inode)) { | |
3403 | BUG_ON(retries); | |
3404 | retries++; | |
3405 | ||
3406 | if (block_group->ro) | |
3407 | goto out_free; | |
3408 | ||
77ab86bf JM |
3409 | ret = create_free_space_inode(fs_info, trans, block_group, |
3410 | path); | |
0af3d00b JB |
3411 | if (ret) |
3412 | goto out_free; | |
3413 | goto again; | |
3414 | } | |
3415 | ||
3416 | /* | |
3417 | * We want to set the generation to 0, that way if anything goes wrong | |
3418 | * from here on out we know not to trust this cache when we load up next | |
3419 | * time. | |
3420 | */ | |
3421 | BTRFS_I(inode)->generation = 0; | |
3422 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3423 | if (ret) { |
3424 | /* | |
3425 | * So theoretically we could recover from this, simply set the | |
3426 | * super cache generation to 0 so we know to invalidate the | |
3427 | * cache, but then we'd have to keep track of the block groups | |
3428 | * that fail this way so we know we _have_ to reset this cache | |
3429 | * before the next commit or risk reading stale cache. So to | |
3430 | * limit our exposure to horrible edge cases lets just abort the | |
3431 | * transaction, this only happens in really bad situations | |
3432 | * anyway. | |
3433 | */ | |
66642832 | 3434 | btrfs_abort_transaction(trans, ret); |
0c0ef4bc JB |
3435 | goto out_put; |
3436 | } | |
0af3d00b JB |
3437 | WARN_ON(ret); |
3438 | ||
8e138e0d JB |
3439 | /* We've already setup this transaction, go ahead and exit */ |
3440 | if (block_group->cache_generation == trans->transid && | |
3441 | i_size_read(inode)) { | |
3442 | dcs = BTRFS_DC_SETUP; | |
3443 | goto out_put; | |
3444 | } | |
3445 | ||
0af3d00b | 3446 | if (i_size_read(inode) > 0) { |
2ff7e61e | 3447 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
0b246afa | 3448 | &fs_info->global_block_rsv); |
7b61cd92 MX |
3449 | if (ret) |
3450 | goto out_put; | |
3451 | ||
77ab86bf | 3452 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); |
0af3d00b JB |
3453 | if (ret) |
3454 | goto out_put; | |
3455 | } | |
3456 | ||
3457 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3458 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
0b246afa | 3459 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cf7c1ef6 LB |
3460 | /* |
3461 | * don't bother trying to write stuff out _if_ | |
3462 | * a) we're not cached, | |
1a79c1f2 LB |
3463 | * b) we're with nospace_cache mount option, |
3464 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
cf7c1ef6 | 3465 | */ |
2b20982e | 3466 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3467 | spin_unlock(&block_group->lock); |
3468 | goto out_put; | |
3469 | } | |
3470 | spin_unlock(&block_group->lock); | |
3471 | ||
2968b1f4 JB |
3472 | /* |
3473 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3474 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3475 | */ | |
3476 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3477 | ret = -ENOSPC; | |
3478 | goto out_put; | |
3479 | } | |
3480 | ||
6fc823b1 JB |
3481 | /* |
3482 | * Try to preallocate enough space based on how big the block group is. | |
3483 | * Keep in mind this has to include any pinned space which could end up | |
3484 | * taking up quite a bit since it's not folded into the other space | |
3485 | * cache. | |
3486 | */ | |
ee22184b | 3487 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
0af3d00b JB |
3488 | if (!num_pages) |
3489 | num_pages = 1; | |
3490 | ||
0af3d00b | 3491 | num_pages *= 16; |
09cbfeaf | 3492 | num_pages *= PAGE_SIZE; |
0af3d00b | 3493 | |
364ecf36 | 3494 | ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages); |
0af3d00b JB |
3495 | if (ret) |
3496 | goto out_put; | |
3497 | ||
3498 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3499 | num_pages, num_pages, | |
3500 | &alloc_hint); | |
2968b1f4 JB |
3501 | /* |
3502 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3503 | * of metadata or split extents when writing the cache out, which means | |
3504 | * we can enospc if we are heavily fragmented in addition to just normal | |
3505 | * out of space conditions. So if we hit this just skip setting up any | |
3506 | * other block groups for this transaction, maybe we'll unpin enough | |
3507 | * space the next time around. | |
3508 | */ | |
2b20982e JB |
3509 | if (!ret) |
3510 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3511 | else if (ret == -ENOSPC) |
3512 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
c09544e0 | 3513 | |
0af3d00b JB |
3514 | out_put: |
3515 | iput(inode); | |
3516 | out_free: | |
b3b4aa74 | 3517 | btrfs_release_path(path); |
0af3d00b JB |
3518 | out: |
3519 | spin_lock(&block_group->lock); | |
e65cbb94 | 3520 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3521 | block_group->cache_generation = trans->transid; |
2b20982e | 3522 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3523 | spin_unlock(&block_group->lock); |
3524 | ||
364ecf36 | 3525 | extent_changeset_free(data_reserved); |
0af3d00b JB |
3526 | return ret; |
3527 | } | |
3528 | ||
dcdf7f6d | 3529 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans, |
2ff7e61e | 3530 | struct btrfs_fs_info *fs_info) |
dcdf7f6d JB |
3531 | { |
3532 | struct btrfs_block_group_cache *cache, *tmp; | |
3533 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3534 | struct btrfs_path *path; | |
3535 | ||
3536 | if (list_empty(&cur_trans->dirty_bgs) || | |
0b246afa | 3537 | !btrfs_test_opt(fs_info, SPACE_CACHE)) |
dcdf7f6d JB |
3538 | return 0; |
3539 | ||
3540 | path = btrfs_alloc_path(); | |
3541 | if (!path) | |
3542 | return -ENOMEM; | |
3543 | ||
3544 | /* Could add new block groups, use _safe just in case */ | |
3545 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3546 | dirty_list) { | |
3547 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3548 | cache_save_setup(cache, trans, path); | |
3549 | } | |
3550 | ||
3551 | btrfs_free_path(path); | |
3552 | return 0; | |
3553 | } | |
3554 | ||
1bbc621e CM |
3555 | /* |
3556 | * transaction commit does final block group cache writeback during a | |
3557 | * critical section where nothing is allowed to change the FS. This is | |
3558 | * required in order for the cache to actually match the block group, | |
3559 | * but can introduce a lot of latency into the commit. | |
3560 | * | |
3561 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3562 | * cache IO. There's a chance we'll have to redo some of it if the | |
3563 | * block group changes again during the commit, but it greatly reduces | |
3564 | * the commit latency by getting rid of the easy block groups while | |
3565 | * we're still allowing others to join the commit. | |
3566 | */ | |
21217054 | 3567 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) |
9078a3e1 | 3568 | { |
21217054 | 3569 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4a8c9a62 | 3570 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3571 | struct btrfs_transaction *cur_trans = trans->transaction; |
3572 | int ret = 0; | |
c9dc4c65 | 3573 | int should_put; |
1bbc621e CM |
3574 | struct btrfs_path *path = NULL; |
3575 | LIST_HEAD(dirty); | |
3576 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3577 | int num_started = 0; |
1bbc621e CM |
3578 | int loops = 0; |
3579 | ||
3580 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3581 | if (list_empty(&cur_trans->dirty_bgs)) { |
3582 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3583 | return 0; | |
1bbc621e | 3584 | } |
b58d1a9e | 3585 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3586 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3587 | |
1bbc621e | 3588 | again: |
1bbc621e CM |
3589 | /* |
3590 | * make sure all the block groups on our dirty list actually | |
3591 | * exist | |
3592 | */ | |
6c686b35 | 3593 | btrfs_create_pending_block_groups(trans); |
1bbc621e CM |
3594 | |
3595 | if (!path) { | |
3596 | path = btrfs_alloc_path(); | |
3597 | if (!path) | |
3598 | return -ENOMEM; | |
3599 | } | |
3600 | ||
b58d1a9e FM |
3601 | /* |
3602 | * cache_write_mutex is here only to save us from balance or automatic | |
3603 | * removal of empty block groups deleting this block group while we are | |
3604 | * writing out the cache | |
3605 | */ | |
3606 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e CM |
3607 | while (!list_empty(&dirty)) { |
3608 | cache = list_first_entry(&dirty, | |
3609 | struct btrfs_block_group_cache, | |
3610 | dirty_list); | |
1bbc621e CM |
3611 | /* |
3612 | * this can happen if something re-dirties a block | |
3613 | * group that is already under IO. Just wait for it to | |
3614 | * finish and then do it all again | |
3615 | */ | |
3616 | if (!list_empty(&cache->io_list)) { | |
3617 | list_del_init(&cache->io_list); | |
afdb5718 | 3618 | btrfs_wait_cache_io(trans, cache, path); |
1bbc621e CM |
3619 | btrfs_put_block_group(cache); |
3620 | } | |
3621 | ||
3622 | ||
3623 | /* | |
3624 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3625 | * if it should update the cache_state. Don't delete | |
3626 | * until after we wait. | |
3627 | * | |
3628 | * Since we're not running in the commit critical section | |
3629 | * we need the dirty_bgs_lock to protect from update_block_group | |
3630 | */ | |
3631 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3632 | list_del_init(&cache->dirty_list); | |
3633 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3634 | ||
3635 | should_put = 1; | |
3636 | ||
3637 | cache_save_setup(cache, trans, path); | |
3638 | ||
3639 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3640 | cache->io_ctl.inode = NULL; | |
0b246afa | 3641 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3642 | cache, path); |
1bbc621e CM |
3643 | if (ret == 0 && cache->io_ctl.inode) { |
3644 | num_started++; | |
3645 | should_put = 0; | |
3646 | ||
3647 | /* | |
45ae2c18 NB |
3648 | * The cache_write_mutex is protecting the |
3649 | * io_list, also refer to the definition of | |
3650 | * btrfs_transaction::io_bgs for more details | |
1bbc621e CM |
3651 | */ |
3652 | list_add_tail(&cache->io_list, io); | |
3653 | } else { | |
3654 | /* | |
3655 | * if we failed to write the cache, the | |
3656 | * generation will be bad and life goes on | |
3657 | */ | |
3658 | ret = 0; | |
3659 | } | |
3660 | } | |
ff1f8250 | 3661 | if (!ret) { |
2ff7e61e JM |
3662 | ret = write_one_cache_group(trans, fs_info, |
3663 | path, cache); | |
ff1f8250 FM |
3664 | /* |
3665 | * Our block group might still be attached to the list | |
3666 | * of new block groups in the transaction handle of some | |
3667 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3668 | * means its block group item isn't yet in the extent | |
3669 | * tree. If this happens ignore the error, as we will | |
3670 | * try again later in the critical section of the | |
3671 | * transaction commit. | |
3672 | */ | |
3673 | if (ret == -ENOENT) { | |
3674 | ret = 0; | |
3675 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3676 | if (list_empty(&cache->dirty_list)) { | |
3677 | list_add_tail(&cache->dirty_list, | |
3678 | &cur_trans->dirty_bgs); | |
3679 | btrfs_get_block_group(cache); | |
3680 | } | |
3681 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3682 | } else if (ret) { | |
66642832 | 3683 | btrfs_abort_transaction(trans, ret); |
ff1f8250 FM |
3684 | } |
3685 | } | |
1bbc621e CM |
3686 | |
3687 | /* if its not on the io list, we need to put the block group */ | |
3688 | if (should_put) | |
3689 | btrfs_put_block_group(cache); | |
3690 | ||
3691 | if (ret) | |
3692 | break; | |
b58d1a9e FM |
3693 | |
3694 | /* | |
3695 | * Avoid blocking other tasks for too long. It might even save | |
3696 | * us from writing caches for block groups that are going to be | |
3697 | * removed. | |
3698 | */ | |
3699 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3700 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3701 | } |
b58d1a9e | 3702 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3703 | |
3704 | /* | |
3705 | * go through delayed refs for all the stuff we've just kicked off | |
3706 | * and then loop back (just once) | |
3707 | */ | |
c79a70b1 | 3708 | ret = btrfs_run_delayed_refs(trans, 0); |
1bbc621e CM |
3709 | if (!ret && loops == 0) { |
3710 | loops++; | |
3711 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3712 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3713 | /* |
3714 | * dirty_bgs_lock protects us from concurrent block group | |
3715 | * deletes too (not just cache_write_mutex). | |
3716 | */ | |
3717 | if (!list_empty(&dirty)) { | |
3718 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3719 | goto again; | |
3720 | } | |
1bbc621e | 3721 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c79a1751 | 3722 | } else if (ret < 0) { |
2ff7e61e | 3723 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
1bbc621e CM |
3724 | } |
3725 | ||
3726 | btrfs_free_path(path); | |
3727 | return ret; | |
3728 | } | |
3729 | ||
3730 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3731 | struct btrfs_fs_info *fs_info) |
1bbc621e CM |
3732 | { |
3733 | struct btrfs_block_group_cache *cache; | |
3734 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3735 | int ret = 0; | |
3736 | int should_put; | |
3737 | struct btrfs_path *path; | |
3738 | struct list_head *io = &cur_trans->io_bgs; | |
3739 | int num_started = 0; | |
9078a3e1 CM |
3740 | |
3741 | path = btrfs_alloc_path(); | |
3742 | if (!path) | |
3743 | return -ENOMEM; | |
3744 | ||
ce93ec54 | 3745 | /* |
e44081ef FM |
3746 | * Even though we are in the critical section of the transaction commit, |
3747 | * we can still have concurrent tasks adding elements to this | |
3748 | * transaction's list of dirty block groups. These tasks correspond to | |
3749 | * endio free space workers started when writeback finishes for a | |
3750 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3751 | * allocate new block groups as a result of COWing nodes of the root | |
3752 | * tree when updating the free space inode. The writeback for the space | |
3753 | * caches is triggered by an earlier call to | |
3754 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3755 | * loop. | |
3756 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3757 | * delayed refs to make sure we have the best chance at doing this all |
3758 | * in one shot. | |
3759 | */ | |
e44081ef | 3760 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3761 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3762 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3763 | struct btrfs_block_group_cache, | |
3764 | dirty_list); | |
c9dc4c65 CM |
3765 | |
3766 | /* | |
3767 | * this can happen if cache_save_setup re-dirties a block | |
3768 | * group that is already under IO. Just wait for it to | |
3769 | * finish and then do it all again | |
3770 | */ | |
3771 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3772 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3773 | list_del_init(&cache->io_list); |
afdb5718 | 3774 | btrfs_wait_cache_io(trans, cache, path); |
c9dc4c65 | 3775 | btrfs_put_block_group(cache); |
e44081ef | 3776 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3777 | } |
3778 | ||
1bbc621e CM |
3779 | /* |
3780 | * don't remove from the dirty list until after we've waited | |
3781 | * on any pending IO | |
3782 | */ | |
ce93ec54 | 3783 | list_del_init(&cache->dirty_list); |
e44081ef | 3784 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3785 | should_put = 1; |
3786 | ||
1bbc621e | 3787 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3788 | |
ce93ec54 | 3789 | if (!ret) |
c79a70b1 | 3790 | ret = btrfs_run_delayed_refs(trans, |
2ff7e61e | 3791 | (unsigned long) -1); |
c9dc4c65 CM |
3792 | |
3793 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3794 | cache->io_ctl.inode = NULL; | |
0b246afa | 3795 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3796 | cache, path); |
c9dc4c65 CM |
3797 | if (ret == 0 && cache->io_ctl.inode) { |
3798 | num_started++; | |
3799 | should_put = 0; | |
1bbc621e | 3800 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3801 | } else { |
3802 | /* | |
3803 | * if we failed to write the cache, the | |
3804 | * generation will be bad and life goes on | |
3805 | */ | |
3806 | ret = 0; | |
3807 | } | |
3808 | } | |
ff1f8250 | 3809 | if (!ret) { |
2ff7e61e JM |
3810 | ret = write_one_cache_group(trans, fs_info, |
3811 | path, cache); | |
2bc0bb5f FM |
3812 | /* |
3813 | * One of the free space endio workers might have | |
3814 | * created a new block group while updating a free space | |
3815 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3816 | * and hasn't released its transaction handle yet, in | |
3817 | * which case the new block group is still attached to | |
3818 | * its transaction handle and its creation has not | |
3819 | * finished yet (no block group item in the extent tree | |
3820 | * yet, etc). If this is the case, wait for all free | |
3821 | * space endio workers to finish and retry. This is a | |
3822 | * a very rare case so no need for a more efficient and | |
3823 | * complex approach. | |
3824 | */ | |
3825 | if (ret == -ENOENT) { | |
3826 | wait_event(cur_trans->writer_wait, | |
3827 | atomic_read(&cur_trans->num_writers) == 1); | |
2ff7e61e JM |
3828 | ret = write_one_cache_group(trans, fs_info, |
3829 | path, cache); | |
2bc0bb5f | 3830 | } |
ff1f8250 | 3831 | if (ret) |
66642832 | 3832 | btrfs_abort_transaction(trans, ret); |
ff1f8250 | 3833 | } |
c9dc4c65 CM |
3834 | |
3835 | /* if its not on the io list, we need to put the block group */ | |
3836 | if (should_put) | |
3837 | btrfs_put_block_group(cache); | |
e44081ef | 3838 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3839 | } |
e44081ef | 3840 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3841 | |
45ae2c18 NB |
3842 | /* |
3843 | * Refer to the definition of io_bgs member for details why it's safe | |
3844 | * to use it without any locking | |
3845 | */ | |
1bbc621e CM |
3846 | while (!list_empty(io)) { |
3847 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3848 | io_list); |
3849 | list_del_init(&cache->io_list); | |
afdb5718 | 3850 | btrfs_wait_cache_io(trans, cache, path); |
0cb59c99 JB |
3851 | btrfs_put_block_group(cache); |
3852 | } | |
3853 | ||
9078a3e1 | 3854 | btrfs_free_path(path); |
ce93ec54 | 3855 | return ret; |
9078a3e1 CM |
3856 | } |
3857 | ||
2ff7e61e | 3858 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 YZ |
3859 | { |
3860 | struct btrfs_block_group_cache *block_group; | |
3861 | int readonly = 0; | |
3862 | ||
0b246afa | 3863 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
3864 | if (!block_group || block_group->ro) |
3865 | readonly = 1; | |
3866 | if (block_group) | |
fa9c0d79 | 3867 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3868 | return readonly; |
3869 | } | |
3870 | ||
f78c436c FM |
3871 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
3872 | { | |
3873 | struct btrfs_block_group_cache *bg; | |
3874 | bool ret = true; | |
3875 | ||
3876 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3877 | if (!bg) | |
3878 | return false; | |
3879 | ||
3880 | spin_lock(&bg->lock); | |
3881 | if (bg->ro) | |
3882 | ret = false; | |
3883 | else | |
3884 | atomic_inc(&bg->nocow_writers); | |
3885 | spin_unlock(&bg->lock); | |
3886 | ||
3887 | /* no put on block group, done by btrfs_dec_nocow_writers */ | |
3888 | if (!ret) | |
3889 | btrfs_put_block_group(bg); | |
3890 | ||
3891 | return ret; | |
3892 | ||
3893 | } | |
3894 | ||
3895 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
3896 | { | |
3897 | struct btrfs_block_group_cache *bg; | |
3898 | ||
3899 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3900 | ASSERT(bg); | |
3901 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
4625956a | 3902 | wake_up_var(&bg->nocow_writers); |
f78c436c FM |
3903 | /* |
3904 | * Once for our lookup and once for the lookup done by a previous call | |
3905 | * to btrfs_inc_nocow_writers() | |
3906 | */ | |
3907 | btrfs_put_block_group(bg); | |
3908 | btrfs_put_block_group(bg); | |
3909 | } | |
3910 | ||
f78c436c FM |
3911 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) |
3912 | { | |
4625956a | 3913 | wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers)); |
f78c436c FM |
3914 | } |
3915 | ||
6ab0a202 JM |
3916 | static const char *alloc_name(u64 flags) |
3917 | { | |
3918 | switch (flags) { | |
3919 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
3920 | return "mixed"; | |
3921 | case BTRFS_BLOCK_GROUP_METADATA: | |
3922 | return "metadata"; | |
3923 | case BTRFS_BLOCK_GROUP_DATA: | |
3924 | return "data"; | |
3925 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
3926 | return "system"; | |
3927 | default: | |
3928 | WARN_ON(1); | |
3929 | return "invalid-combination"; | |
3930 | }; | |
3931 | } | |
3932 | ||
4ca61683 | 3933 | static int create_space_info(struct btrfs_fs_info *info, u64 flags) |
2be12ef7 NB |
3934 | { |
3935 | ||
3936 | struct btrfs_space_info *space_info; | |
3937 | int i; | |
3938 | int ret; | |
3939 | ||
3940 | space_info = kzalloc(sizeof(*space_info), GFP_NOFS); | |
3941 | if (!space_info) | |
3942 | return -ENOMEM; | |
3943 | ||
3944 | ret = percpu_counter_init(&space_info->total_bytes_pinned, 0, | |
3945 | GFP_KERNEL); | |
3946 | if (ret) { | |
3947 | kfree(space_info); | |
3948 | return ret; | |
3949 | } | |
3950 | ||
3951 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
3952 | INIT_LIST_HEAD(&space_info->block_groups[i]); | |
3953 | init_rwsem(&space_info->groups_sem); | |
3954 | spin_lock_init(&space_info->lock); | |
3955 | space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3956 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
3957 | init_waitqueue_head(&space_info->wait); | |
3958 | INIT_LIST_HEAD(&space_info->ro_bgs); | |
3959 | INIT_LIST_HEAD(&space_info->tickets); | |
3960 | INIT_LIST_HEAD(&space_info->priority_tickets); | |
3961 | ||
3962 | ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype, | |
3963 | info->space_info_kobj, "%s", | |
3964 | alloc_name(space_info->flags)); | |
3965 | if (ret) { | |
3966 | percpu_counter_destroy(&space_info->total_bytes_pinned); | |
3967 | kfree(space_info); | |
3968 | return ret; | |
3969 | } | |
3970 | ||
2be12ef7 NB |
3971 | list_add_rcu(&space_info->list, &info->space_info); |
3972 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3973 | info->data_sinfo = space_info; | |
3974 | ||
3975 | return ret; | |
3976 | } | |
3977 | ||
d2006e6d | 3978 | static void update_space_info(struct btrfs_fs_info *info, u64 flags, |
593060d7 | 3979 | u64 total_bytes, u64 bytes_used, |
e40edf2d | 3980 | u64 bytes_readonly, |
593060d7 CM |
3981 | struct btrfs_space_info **space_info) |
3982 | { | |
3983 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3984 | int factor; |
3985 | ||
46df06b8 | 3986 | factor = btrfs_bg_type_to_factor(flags); |
593060d7 CM |
3987 | |
3988 | found = __find_space_info(info, flags); | |
d2006e6d NB |
3989 | ASSERT(found); |
3990 | spin_lock(&found->lock); | |
3991 | found->total_bytes += total_bytes; | |
3992 | found->disk_total += total_bytes * factor; | |
3993 | found->bytes_used += bytes_used; | |
3994 | found->disk_used += bytes_used * factor; | |
3995 | found->bytes_readonly += bytes_readonly; | |
3996 | if (total_bytes > 0) | |
3997 | found->full = 0; | |
3998 | space_info_add_new_bytes(info, found, total_bytes - | |
3999 | bytes_used - bytes_readonly); | |
4000 | spin_unlock(&found->lock); | |
4001 | *space_info = found; | |
593060d7 CM |
4002 | } |
4003 | ||
8790d502 CM |
4004 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
4005 | { | |
899c81ea ID |
4006 | u64 extra_flags = chunk_to_extended(flags) & |
4007 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 4008 | |
de98ced9 | 4009 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
4010 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
4011 | fs_info->avail_data_alloc_bits |= extra_flags; | |
4012 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
4013 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
4014 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
4015 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 4016 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 4017 | } |
593060d7 | 4018 | |
fc67c450 ID |
4019 | /* |
4020 | * returns target flags in extended format or 0 if restripe for this | |
4021 | * chunk_type is not in progress | |
c6664b42 | 4022 | * |
dccdb07b | 4023 | * should be called with balance_lock held |
fc67c450 ID |
4024 | */ |
4025 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
4026 | { | |
4027 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4028 | u64 target = 0; | |
4029 | ||
fc67c450 ID |
4030 | if (!bctl) |
4031 | return 0; | |
4032 | ||
4033 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
4034 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4035 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
4036 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
4037 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4038 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
4039 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
4040 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4041 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
4042 | } | |
4043 | ||
4044 | return target; | |
4045 | } | |
4046 | ||
a46d11a8 ID |
4047 | /* |
4048 | * @flags: available profiles in extended format (see ctree.h) | |
4049 | * | |
e4d8ec0f ID |
4050 | * Returns reduced profile in chunk format. If profile changing is in |
4051 | * progress (either running or paused) picks the target profile (if it's | |
4052 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 4053 | */ |
2ff7e61e | 4054 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c | 4055 | { |
0b246afa | 4056 | u64 num_devices = fs_info->fs_devices->rw_devices; |
fc67c450 | 4057 | u64 target; |
9c170b26 ZL |
4058 | u64 raid_type; |
4059 | u64 allowed = 0; | |
a061fc8d | 4060 | |
fc67c450 ID |
4061 | /* |
4062 | * see if restripe for this chunk_type is in progress, if so | |
4063 | * try to reduce to the target profile | |
4064 | */ | |
0b246afa JM |
4065 | spin_lock(&fs_info->balance_lock); |
4066 | target = get_restripe_target(fs_info, flags); | |
fc67c450 ID |
4067 | if (target) { |
4068 | /* pick target profile only if it's already available */ | |
4069 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
0b246afa | 4070 | spin_unlock(&fs_info->balance_lock); |
fc67c450 | 4071 | return extended_to_chunk(target); |
e4d8ec0f ID |
4072 | } |
4073 | } | |
0b246afa | 4074 | spin_unlock(&fs_info->balance_lock); |
e4d8ec0f | 4075 | |
53b381b3 | 4076 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
4077 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
4078 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
41a6e891 | 4079 | allowed |= btrfs_raid_array[raid_type].bg_flag; |
9c170b26 ZL |
4080 | } |
4081 | allowed &= flags; | |
4082 | ||
4083 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
4084 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
4085 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
4086 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
4087 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
4088 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
4089 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
4090 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
4091 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
4092 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
4093 | ||
4094 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
4095 | ||
4096 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
4097 | } |
4098 | ||
2ff7e61e | 4099 | static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
6a63209f | 4100 | { |
de98ced9 | 4101 | unsigned seq; |
f8213bdc | 4102 | u64 flags; |
de98ced9 MX |
4103 | |
4104 | do { | |
f8213bdc | 4105 | flags = orig_flags; |
0b246afa | 4106 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 MX |
4107 | |
4108 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
0b246afa | 4109 | flags |= fs_info->avail_data_alloc_bits; |
de98ced9 | 4110 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
0b246afa | 4111 | flags |= fs_info->avail_system_alloc_bits; |
de98ced9 | 4112 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
0b246afa JM |
4113 | flags |= fs_info->avail_metadata_alloc_bits; |
4114 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
6fef8df1 | 4115 | |
2ff7e61e | 4116 | return btrfs_reduce_alloc_profile(fs_info, flags); |
6a63209f JB |
4117 | } |
4118 | ||
1b86826d | 4119 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 4120 | { |
0b246afa | 4121 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 4122 | u64 flags; |
53b381b3 | 4123 | u64 ret; |
9ed74f2d | 4124 | |
b742bb82 YZ |
4125 | if (data) |
4126 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 4127 | else if (root == fs_info->chunk_root) |
b742bb82 | 4128 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 4129 | else |
b742bb82 | 4130 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 4131 | |
2ff7e61e | 4132 | ret = get_alloc_profile(fs_info, flags); |
53b381b3 | 4133 | return ret; |
6a63209f | 4134 | } |
9ed74f2d | 4135 | |
1b86826d JM |
4136 | u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info) |
4137 | { | |
4138 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA); | |
4139 | } | |
4140 | ||
4141 | u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info) | |
4142 | { | |
4143 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4144 | } | |
4145 | ||
4146 | u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info) | |
4147 | { | |
4148 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
4149 | } | |
4150 | ||
4136135b LB |
4151 | static u64 btrfs_space_info_used(struct btrfs_space_info *s_info, |
4152 | bool may_use_included) | |
4153 | { | |
4154 | ASSERT(s_info); | |
4155 | return s_info->bytes_used + s_info->bytes_reserved + | |
4156 | s_info->bytes_pinned + s_info->bytes_readonly + | |
4157 | (may_use_included ? s_info->bytes_may_use : 0); | |
4158 | } | |
4159 | ||
04f4f916 | 4160 | int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes) |
6a63209f | 4161 | { |
04f4f916 | 4162 | struct btrfs_root *root = inode->root; |
b4d7c3c9 | 4163 | struct btrfs_fs_info *fs_info = root->fs_info; |
1174cade | 4164 | struct btrfs_space_info *data_sinfo = fs_info->data_sinfo; |
ab6e2410 | 4165 | u64 used; |
94b947b2 | 4166 | int ret = 0; |
c99f1b0c ZL |
4167 | int need_commit = 2; |
4168 | int have_pinned_space; | |
6a63209f | 4169 | |
6a63209f | 4170 | /* make sure bytes are sectorsize aligned */ |
0b246afa | 4171 | bytes = ALIGN(bytes, fs_info->sectorsize); |
6a63209f | 4172 | |
9dced186 | 4173 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 4174 | need_commit = 0; |
9dced186 | 4175 | ASSERT(current->journal_info); |
0af3d00b JB |
4176 | } |
4177 | ||
6a63209f JB |
4178 | again: |
4179 | /* make sure we have enough space to handle the data first */ | |
4180 | spin_lock(&data_sinfo->lock); | |
4136135b | 4181 | used = btrfs_space_info_used(data_sinfo, true); |
ab6e2410 JB |
4182 | |
4183 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 4184 | struct btrfs_trans_handle *trans; |
9ed74f2d | 4185 | |
6a63209f JB |
4186 | /* |
4187 | * if we don't have enough free bytes in this space then we need | |
4188 | * to alloc a new chunk. | |
4189 | */ | |
b9fd47cd | 4190 | if (!data_sinfo->full) { |
6a63209f | 4191 | u64 alloc_target; |
9ed74f2d | 4192 | |
0e4f8f88 | 4193 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 4194 | spin_unlock(&data_sinfo->lock); |
1174cade | 4195 | |
1b86826d | 4196 | alloc_target = btrfs_data_alloc_profile(fs_info); |
9dced186 MX |
4197 | /* |
4198 | * It is ugly that we don't call nolock join | |
4199 | * transaction for the free space inode case here. | |
4200 | * But it is safe because we only do the data space | |
4201 | * reservation for the free space cache in the | |
4202 | * transaction context, the common join transaction | |
4203 | * just increase the counter of the current transaction | |
4204 | * handler, doesn't try to acquire the trans_lock of | |
4205 | * the fs. | |
4206 | */ | |
7a7eaa40 | 4207 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4208 | if (IS_ERR(trans)) |
4209 | return PTR_ERR(trans); | |
9ed74f2d | 4210 | |
01458828 | 4211 | ret = do_chunk_alloc(trans, alloc_target, |
0e4f8f88 | 4212 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 4213 | btrfs_end_transaction(trans); |
d52a5b5f MX |
4214 | if (ret < 0) { |
4215 | if (ret != -ENOSPC) | |
4216 | return ret; | |
c99f1b0c ZL |
4217 | else { |
4218 | have_pinned_space = 1; | |
d52a5b5f | 4219 | goto commit_trans; |
c99f1b0c | 4220 | } |
d52a5b5f | 4221 | } |
9ed74f2d | 4222 | |
6a63209f JB |
4223 | goto again; |
4224 | } | |
f2bb8f5c JB |
4225 | |
4226 | /* | |
b150a4f1 | 4227 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
4228 | * allocation, and no removed chunk in current transaction, |
4229 | * don't bother committing the transaction. | |
f2bb8f5c | 4230 | */ |
dec59fa3 | 4231 | have_pinned_space = __percpu_counter_compare( |
c99f1b0c | 4232 | &data_sinfo->total_bytes_pinned, |
dec59fa3 EL |
4233 | used + bytes - data_sinfo->total_bytes, |
4234 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
6a63209f | 4235 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4236 | |
4e06bdd6 | 4237 | /* commit the current transaction and try again */ |
d52a5b5f | 4238 | commit_trans: |
92e2f7e3 | 4239 | if (need_commit) { |
c99f1b0c | 4240 | need_commit--; |
b150a4f1 | 4241 | |
e1746e83 | 4242 | if (need_commit > 0) { |
82b3e53b | 4243 | btrfs_start_delalloc_roots(fs_info, -1); |
6374e57a | 4244 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, |
0b246afa | 4245 | (u64)-1); |
e1746e83 | 4246 | } |
9a4e7276 | 4247 | |
7a7eaa40 | 4248 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4249 | if (IS_ERR(trans)) |
4250 | return PTR_ERR(trans); | |
c99f1b0c | 4251 | if (have_pinned_space >= 0 || |
3204d33c JB |
4252 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4253 | &trans->transaction->flags) || | |
c99f1b0c | 4254 | need_commit > 0) { |
3a45bb20 | 4255 | ret = btrfs_commit_transaction(trans); |
94b947b2 ZL |
4256 | if (ret) |
4257 | return ret; | |
d7c15171 | 4258 | /* |
c2d6cb16 FM |
4259 | * The cleaner kthread might still be doing iput |
4260 | * operations. Wait for it to finish so that | |
4261 | * more space is released. | |
d7c15171 | 4262 | */ |
0b246afa JM |
4263 | mutex_lock(&fs_info->cleaner_delayed_iput_mutex); |
4264 | mutex_unlock(&fs_info->cleaner_delayed_iput_mutex); | |
94b947b2 ZL |
4265 | goto again; |
4266 | } else { | |
3a45bb20 | 4267 | btrfs_end_transaction(trans); |
94b947b2 | 4268 | } |
4e06bdd6 | 4269 | } |
9ed74f2d | 4270 | |
0b246afa | 4271 | trace_btrfs_space_reservation(fs_info, |
cab45e22 JM |
4272 | "space_info:enospc", |
4273 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4274 | return -ENOSPC; |
4275 | } | |
9f9b8e8d | 4276 | update_bytes_may_use(data_sinfo, bytes); |
0b246afa | 4277 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4278 | data_sinfo->flags, bytes, 1); |
6a63209f | 4279 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4280 | |
4559b0a7 | 4281 | return 0; |
9ed74f2d | 4282 | } |
6a63209f | 4283 | |
364ecf36 QW |
4284 | int btrfs_check_data_free_space(struct inode *inode, |
4285 | struct extent_changeset **reserved, u64 start, u64 len) | |
4ceff079 | 4286 | { |
0b246afa | 4287 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4288 | int ret; |
4289 | ||
4290 | /* align the range */ | |
0b246afa JM |
4291 | len = round_up(start + len, fs_info->sectorsize) - |
4292 | round_down(start, fs_info->sectorsize); | |
4293 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4294 | |
04f4f916 | 4295 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len); |
4ceff079 QW |
4296 | if (ret < 0) |
4297 | return ret; | |
4298 | ||
1e5ec2e7 | 4299 | /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */ |
364ecf36 | 4300 | ret = btrfs_qgroup_reserve_data(inode, reserved, start, len); |
7bc329c1 | 4301 | if (ret < 0) |
1e5ec2e7 | 4302 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
364ecf36 QW |
4303 | else |
4304 | ret = 0; | |
4ceff079 QW |
4305 | return ret; |
4306 | } | |
4307 | ||
4ceff079 QW |
4308 | /* |
4309 | * Called if we need to clear a data reservation for this inode | |
4310 | * Normally in a error case. | |
4311 | * | |
51773bec QW |
4312 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4313 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4314 | * Like clear_bit_hook(). | |
4ceff079 | 4315 | */ |
51773bec QW |
4316 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4317 | u64 len) | |
4ceff079 | 4318 | { |
0b246afa | 4319 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4320 | struct btrfs_space_info *data_sinfo; |
4321 | ||
4322 | /* Make sure the range is aligned to sectorsize */ | |
0b246afa JM |
4323 | len = round_up(start + len, fs_info->sectorsize) - |
4324 | round_down(start, fs_info->sectorsize); | |
4325 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4326 | |
0b246afa | 4327 | data_sinfo = fs_info->data_sinfo; |
4ceff079 | 4328 | spin_lock(&data_sinfo->lock); |
9f9b8e8d | 4329 | update_bytes_may_use(data_sinfo, -len); |
0b246afa | 4330 | trace_btrfs_space_reservation(fs_info, "space_info", |
4ceff079 QW |
4331 | data_sinfo->flags, len, 0); |
4332 | spin_unlock(&data_sinfo->lock); | |
4333 | } | |
4334 | ||
51773bec QW |
4335 | /* |
4336 | * Called if we need to clear a data reservation for this inode | |
4337 | * Normally in a error case. | |
4338 | * | |
01327610 | 4339 | * This one will handle the per-inode data rsv map for accurate reserved |
51773bec QW |
4340 | * space framework. |
4341 | */ | |
bc42bda2 QW |
4342 | void btrfs_free_reserved_data_space(struct inode *inode, |
4343 | struct extent_changeset *reserved, u64 start, u64 len) | |
51773bec | 4344 | { |
0c476a5d JM |
4345 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4346 | ||
4347 | /* Make sure the range is aligned to sectorsize */ | |
da17066c JM |
4348 | len = round_up(start + len, root->fs_info->sectorsize) - |
4349 | round_down(start, root->fs_info->sectorsize); | |
4350 | start = round_down(start, root->fs_info->sectorsize); | |
0c476a5d | 4351 | |
51773bec | 4352 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
bc42bda2 | 4353 | btrfs_qgroup_free_data(inode, reserved, start, len); |
51773bec QW |
4354 | } |
4355 | ||
97e728d4 | 4356 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4357 | { |
97e728d4 JB |
4358 | struct list_head *head = &info->space_info; |
4359 | struct btrfs_space_info *found; | |
e3ccfa98 | 4360 | |
97e728d4 JB |
4361 | rcu_read_lock(); |
4362 | list_for_each_entry_rcu(found, head, list) { | |
4363 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4364 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4365 | } |
97e728d4 | 4366 | rcu_read_unlock(); |
e3ccfa98 JB |
4367 | } |
4368 | ||
3c76cd84 MX |
4369 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
4370 | { | |
4371 | return (global->size << 1); | |
4372 | } | |
4373 | ||
2ff7e61e | 4374 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, |
698d0082 | 4375 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4376 | { |
0b246afa | 4377 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
8d8aafee | 4378 | u64 bytes_used = btrfs_space_info_used(sinfo, false); |
e5bc2458 | 4379 | u64 thresh; |
e3ccfa98 | 4380 | |
0e4f8f88 CM |
4381 | if (force == CHUNK_ALLOC_FORCE) |
4382 | return 1; | |
4383 | ||
fb25e914 JB |
4384 | /* |
4385 | * We need to take into account the global rsv because for all intents | |
4386 | * and purposes it's used space. Don't worry about locking the | |
4387 | * global_rsv, it doesn't change except when the transaction commits. | |
4388 | */ | |
54338b5c | 4389 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
8d8aafee | 4390 | bytes_used += calc_global_rsv_need_space(global_rsv); |
fb25e914 | 4391 | |
0e4f8f88 CM |
4392 | /* |
4393 | * in limited mode, we want to have some free space up to | |
4394 | * about 1% of the FS size. | |
4395 | */ | |
4396 | if (force == CHUNK_ALLOC_LIMITED) { | |
0b246afa | 4397 | thresh = btrfs_super_total_bytes(fs_info->super_copy); |
ee22184b | 4398 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
0e4f8f88 | 4399 | |
8d8aafee | 4400 | if (sinfo->total_bytes - bytes_used < thresh) |
0e4f8f88 CM |
4401 | return 1; |
4402 | } | |
0e4f8f88 | 4403 | |
8d8aafee | 4404 | if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) |
14ed0ca6 | 4405 | return 0; |
424499db | 4406 | return 1; |
32c00aff JB |
4407 | } |
4408 | ||
2ff7e61e | 4409 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4410 | { |
4411 | u64 num_dev; | |
4412 | ||
53b381b3 DW |
4413 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
4414 | BTRFS_BLOCK_GROUP_RAID0 | | |
4415 | BTRFS_BLOCK_GROUP_RAID5 | | |
4416 | BTRFS_BLOCK_GROUP_RAID6)) | |
0b246afa | 4417 | num_dev = fs_info->fs_devices->rw_devices; |
15d1ff81 LB |
4418 | else if (type & BTRFS_BLOCK_GROUP_RAID1) |
4419 | num_dev = 2; | |
4420 | else | |
4421 | num_dev = 1; /* DUP or single */ | |
4422 | ||
39c2d7fa | 4423 | return num_dev; |
15d1ff81 LB |
4424 | } |
4425 | ||
39c2d7fa FM |
4426 | /* |
4427 | * If @is_allocation is true, reserve space in the system space info necessary | |
4428 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4429 | * removing a chunk. | |
4430 | */ | |
451a2c13 | 4431 | void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) |
15d1ff81 | 4432 | { |
451a2c13 | 4433 | struct btrfs_fs_info *fs_info = trans->fs_info; |
15d1ff81 LB |
4434 | struct btrfs_space_info *info; |
4435 | u64 left; | |
4436 | u64 thresh; | |
4fbcdf66 | 4437 | int ret = 0; |
39c2d7fa | 4438 | u64 num_devs; |
4fbcdf66 FM |
4439 | |
4440 | /* | |
4441 | * Needed because we can end up allocating a system chunk and for an | |
4442 | * atomic and race free space reservation in the chunk block reserve. | |
4443 | */ | |
a32bf9a3 | 4444 | lockdep_assert_held(&fs_info->chunk_mutex); |
15d1ff81 | 4445 | |
0b246afa | 4446 | info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
15d1ff81 | 4447 | spin_lock(&info->lock); |
4136135b | 4448 | left = info->total_bytes - btrfs_space_info_used(info, true); |
15d1ff81 LB |
4449 | spin_unlock(&info->lock); |
4450 | ||
2ff7e61e | 4451 | num_devs = get_profile_num_devs(fs_info, type); |
39c2d7fa FM |
4452 | |
4453 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
0b246afa JM |
4454 | thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) + |
4455 | btrfs_calc_trans_metadata_size(fs_info, 1); | |
39c2d7fa | 4456 | |
0b246afa JM |
4457 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
4458 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
4459 | left, thresh, type); | |
4460 | dump_space_info(fs_info, info, 0, 0); | |
15d1ff81 LB |
4461 | } |
4462 | ||
4463 | if (left < thresh) { | |
1b86826d | 4464 | u64 flags = btrfs_system_alloc_profile(fs_info); |
15d1ff81 | 4465 | |
4fbcdf66 FM |
4466 | /* |
4467 | * Ignore failure to create system chunk. We might end up not | |
4468 | * needing it, as we might not need to COW all nodes/leafs from | |
4469 | * the paths we visit in the chunk tree (they were already COWed | |
4470 | * or created in the current transaction for example). | |
4471 | */ | |
c216b203 | 4472 | ret = btrfs_alloc_chunk(trans, flags); |
4fbcdf66 FM |
4473 | } |
4474 | ||
4475 | if (!ret) { | |
0b246afa JM |
4476 | ret = btrfs_block_rsv_add(fs_info->chunk_root, |
4477 | &fs_info->chunk_block_rsv, | |
4fbcdf66 FM |
4478 | thresh, BTRFS_RESERVE_NO_FLUSH); |
4479 | if (!ret) | |
4480 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4481 | } |
4482 | } | |
4483 | ||
28b737f6 LB |
4484 | /* |
4485 | * If force is CHUNK_ALLOC_FORCE: | |
4486 | * - return 1 if it successfully allocates a chunk, | |
4487 | * - return errors including -ENOSPC otherwise. | |
4488 | * If force is NOT CHUNK_ALLOC_FORCE: | |
4489 | * - return 0 if it doesn't need to allocate a new chunk, | |
4490 | * - return 1 if it successfully allocates a chunk, | |
4491 | * - return errors including -ENOSPC otherwise. | |
4492 | */ | |
01458828 NB |
4493 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, |
4494 | int force) | |
9ed74f2d | 4495 | { |
01458828 | 4496 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6324fbf3 | 4497 | struct btrfs_space_info *space_info; |
2556fbb0 NB |
4498 | bool wait_for_alloc = false; |
4499 | bool should_alloc = false; | |
9ed74f2d | 4500 | int ret = 0; |
9ed74f2d | 4501 | |
c6b305a8 JB |
4502 | /* Don't re-enter if we're already allocating a chunk */ |
4503 | if (trans->allocating_chunk) | |
4504 | return -ENOSPC; | |
4505 | ||
0b246afa | 4506 | space_info = __find_space_info(fs_info, flags); |
dc2d3005 | 4507 | ASSERT(space_info); |
9ed74f2d | 4508 | |
2556fbb0 NB |
4509 | do { |
4510 | spin_lock(&space_info->lock); | |
4511 | if (force < space_info->force_alloc) | |
4512 | force = space_info->force_alloc; | |
4513 | should_alloc = should_alloc_chunk(fs_info, space_info, force); | |
4514 | if (space_info->full) { | |
4515 | /* No more free physical space */ | |
4516 | if (should_alloc) | |
4517 | ret = -ENOSPC; | |
4518 | else | |
4519 | ret = 0; | |
4520 | spin_unlock(&space_info->lock); | |
4521 | return ret; | |
4522 | } else if (!should_alloc) { | |
4523 | spin_unlock(&space_info->lock); | |
4524 | return 0; | |
4525 | } else if (space_info->chunk_alloc) { | |
4526 | /* | |
4527 | * Someone is already allocating, so we need to block | |
4528 | * until this someone is finished and then loop to | |
4529 | * recheck if we should continue with our allocation | |
4530 | * attempt. | |
4531 | */ | |
4532 | wait_for_alloc = true; | |
4533 | spin_unlock(&space_info->lock); | |
4534 | mutex_lock(&fs_info->chunk_mutex); | |
4535 | mutex_unlock(&fs_info->chunk_mutex); | |
4536 | } else { | |
4537 | /* Proceed with allocation */ | |
4538 | space_info->chunk_alloc = 1; | |
4539 | wait_for_alloc = false; | |
4540 | spin_unlock(&space_info->lock); | |
4541 | } | |
6d74119f | 4542 | |
1e1c50a9 | 4543 | cond_resched(); |
2556fbb0 | 4544 | } while (wait_for_alloc); |
6d74119f | 4545 | |
2556fbb0 | 4546 | mutex_lock(&fs_info->chunk_mutex); |
c6b305a8 JB |
4547 | trans->allocating_chunk = true; |
4548 | ||
67377734 JB |
4549 | /* |
4550 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4551 | * allocating mixed chunks instead of individual chunks. | |
4552 | */ | |
4553 | if (btrfs_mixed_space_info(space_info)) | |
4554 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4555 | ||
97e728d4 JB |
4556 | /* |
4557 | * if we're doing a data chunk, go ahead and make sure that | |
4558 | * we keep a reasonable number of metadata chunks allocated in the | |
4559 | * FS as well. | |
4560 | */ | |
9ed74f2d | 4561 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4562 | fs_info->data_chunk_allocations++; |
4563 | if (!(fs_info->data_chunk_allocations % | |
4564 | fs_info->metadata_ratio)) | |
4565 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4566 | } |
4567 | ||
15d1ff81 LB |
4568 | /* |
4569 | * Check if we have enough space in SYSTEM chunk because we may need | |
4570 | * to update devices. | |
4571 | */ | |
451a2c13 | 4572 | check_system_chunk(trans, flags); |
15d1ff81 | 4573 | |
c216b203 | 4574 | ret = btrfs_alloc_chunk(trans, flags); |
c6b305a8 | 4575 | trans->allocating_chunk = false; |
92b8e897 | 4576 | |
9ed74f2d | 4577 | spin_lock(&space_info->lock); |
57f1642e NB |
4578 | if (ret < 0) { |
4579 | if (ret == -ENOSPC) | |
4580 | space_info->full = 1; | |
4581 | else | |
4582 | goto out; | |
4583 | } else { | |
424499db | 4584 | ret = 1; |
21a94f7a | 4585 | space_info->max_extent_size = 0; |
57f1642e | 4586 | } |
6d74119f | 4587 | |
0e4f8f88 | 4588 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4589 | out: |
6d74119f | 4590 | space_info->chunk_alloc = 0; |
9ed74f2d | 4591 | spin_unlock(&space_info->lock); |
a25c75d5 | 4592 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4593 | /* |
4594 | * When we allocate a new chunk we reserve space in the chunk block | |
4595 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4596 | * add new nodes/leafs to it if we end up needing to do it when | |
4597 | * inserting the chunk item and updating device items as part of the | |
4598 | * second phase of chunk allocation, performed by | |
4599 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4600 | * large number of new block groups to create in our transaction | |
4601 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4602 | * in extreme cases - like having a single transaction create many new | |
4603 | * block groups when starting to write out the free space caches of all | |
4604 | * the block groups that were made dirty during the lifetime of the | |
4605 | * transaction. | |
4606 | */ | |
5ce55557 | 4607 | if (trans->chunk_bytes_reserved >= (u64)SZ_2M) |
6c686b35 | 4608 | btrfs_create_pending_block_groups(trans); |
5ce55557 | 4609 | |
0f9dd46c | 4610 | return ret; |
6324fbf3 | 4611 | } |
9ed74f2d | 4612 | |
c1c4919b | 4613 | static int can_overcommit(struct btrfs_fs_info *fs_info, |
a80c8dcf | 4614 | struct btrfs_space_info *space_info, u64 bytes, |
c1c4919b JM |
4615 | enum btrfs_reserve_flush_enum flush, |
4616 | bool system_chunk) | |
a80c8dcf | 4617 | { |
0b246afa | 4618 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
957780eb | 4619 | u64 profile; |
3c76cd84 | 4620 | u64 space_size; |
a80c8dcf JB |
4621 | u64 avail; |
4622 | u64 used; | |
46df06b8 | 4623 | int factor; |
a80c8dcf | 4624 | |
957780eb JB |
4625 | /* Don't overcommit when in mixed mode. */ |
4626 | if (space_info->flags & BTRFS_BLOCK_GROUP_DATA) | |
4627 | return 0; | |
4628 | ||
c1c4919b JM |
4629 | if (system_chunk) |
4630 | profile = btrfs_system_alloc_profile(fs_info); | |
4631 | else | |
4632 | profile = btrfs_metadata_alloc_profile(fs_info); | |
4633 | ||
4136135b | 4634 | used = btrfs_space_info_used(space_info, false); |
96f1bb57 | 4635 | |
96f1bb57 JB |
4636 | /* |
4637 | * We only want to allow over committing if we have lots of actual space | |
4638 | * free, but if we don't have enough space to handle the global reserve | |
4639 | * space then we could end up having a real enospc problem when trying | |
4640 | * to allocate a chunk or some other such important allocation. | |
4641 | */ | |
3c76cd84 MX |
4642 | spin_lock(&global_rsv->lock); |
4643 | space_size = calc_global_rsv_need_space(global_rsv); | |
4644 | spin_unlock(&global_rsv->lock); | |
4645 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
4646 | return 0; |
4647 | ||
4648 | used += space_info->bytes_may_use; | |
a80c8dcf | 4649 | |
a5ed45f8 | 4650 | avail = atomic64_read(&fs_info->free_chunk_space); |
a80c8dcf JB |
4651 | |
4652 | /* | |
4653 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
4654 | * space is actually useable. For raid56, the space info used |
4655 | * doesn't include the parity drive, so we don't have to | |
4656 | * change the math | |
a80c8dcf | 4657 | */ |
46df06b8 DS |
4658 | factor = btrfs_bg_type_to_factor(profile); |
4659 | avail = div_u64(avail, factor); | |
a80c8dcf JB |
4660 | |
4661 | /* | |
561c294d MX |
4662 | * If we aren't flushing all things, let us overcommit up to |
4663 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4664 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 4665 | */ |
08e007d2 | 4666 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 4667 | avail >>= 3; |
a80c8dcf | 4668 | else |
14575aef | 4669 | avail >>= 1; |
a80c8dcf | 4670 | |
14575aef | 4671 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
4672 | return 1; |
4673 | return 0; | |
4674 | } | |
4675 | ||
2ff7e61e | 4676 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info, |
6c255e67 | 4677 | unsigned long nr_pages, int nr_items) |
da633a42 | 4678 | { |
0b246afa | 4679 | struct super_block *sb = fs_info->sb; |
da633a42 | 4680 | |
925a6efb JB |
4681 | if (down_read_trylock(&sb->s_umount)) { |
4682 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4683 | up_read(&sb->s_umount); | |
4684 | } else { | |
da633a42 MX |
4685 | /* |
4686 | * We needn't worry the filesystem going from r/w to r/o though | |
4687 | * we don't acquire ->s_umount mutex, because the filesystem | |
4688 | * should guarantee the delalloc inodes list be empty after | |
4689 | * the filesystem is readonly(all dirty pages are written to | |
4690 | * the disk). | |
4691 | */ | |
82b3e53b | 4692 | btrfs_start_delalloc_roots(fs_info, nr_items); |
98ad69cf | 4693 | if (!current->journal_info) |
0b246afa | 4694 | btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1); |
da633a42 MX |
4695 | } |
4696 | } | |
4697 | ||
6374e57a | 4698 | static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info, |
2ff7e61e | 4699 | u64 to_reclaim) |
18cd8ea6 MX |
4700 | { |
4701 | u64 bytes; | |
6374e57a | 4702 | u64 nr; |
18cd8ea6 | 4703 | |
2ff7e61e | 4704 | bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
6374e57a | 4705 | nr = div64_u64(to_reclaim, bytes); |
18cd8ea6 MX |
4706 | if (!nr) |
4707 | nr = 1; | |
4708 | return nr; | |
4709 | } | |
4710 | ||
ee22184b | 4711 | #define EXTENT_SIZE_PER_ITEM SZ_256K |
c61a16a7 | 4712 | |
9ed74f2d | 4713 | /* |
5da9d01b | 4714 | * shrink metadata reservation for delalloc |
9ed74f2d | 4715 | */ |
c1c4919b JM |
4716 | static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim, |
4717 | u64 orig, bool wait_ordered) | |
5da9d01b | 4718 | { |
0019f10d | 4719 | struct btrfs_space_info *space_info; |
663350ac | 4720 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4721 | u64 delalloc_bytes; |
5da9d01b | 4722 | u64 max_reclaim; |
6374e57a | 4723 | u64 items; |
b1953bce | 4724 | long time_left; |
d3ee29e3 MX |
4725 | unsigned long nr_pages; |
4726 | int loops; | |
5da9d01b | 4727 | |
c61a16a7 | 4728 | /* Calc the number of the pages we need flush for space reservation */ |
2ff7e61e | 4729 | items = calc_reclaim_items_nr(fs_info, to_reclaim); |
6374e57a | 4730 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; |
c61a16a7 | 4731 | |
663350ac | 4732 | trans = (struct btrfs_trans_handle *)current->journal_info; |
69fe2d75 | 4733 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
bf9022e0 | 4734 | |
963d678b | 4735 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4736 | &fs_info->delalloc_bytes); |
f4c738c2 | 4737 | if (delalloc_bytes == 0) { |
fdb5effd | 4738 | if (trans) |
f4c738c2 | 4739 | return; |
38c135af | 4740 | if (wait_ordered) |
0b246afa | 4741 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f4c738c2 | 4742 | return; |
fdb5effd JB |
4743 | } |
4744 | ||
d3ee29e3 | 4745 | loops = 0; |
f4c738c2 JB |
4746 | while (delalloc_bytes && loops < 3) { |
4747 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
09cbfeaf | 4748 | nr_pages = max_reclaim >> PAGE_SHIFT; |
2ff7e61e | 4749 | btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items); |
dea31f52 JB |
4750 | /* |
4751 | * We need to wait for the async pages to actually start before | |
4752 | * we do anything. | |
4753 | */ | |
0b246afa | 4754 | max_reclaim = atomic_read(&fs_info->async_delalloc_pages); |
9f3a074d MX |
4755 | if (!max_reclaim) |
4756 | goto skip_async; | |
4757 | ||
4758 | if (max_reclaim <= nr_pages) | |
4759 | max_reclaim = 0; | |
4760 | else | |
4761 | max_reclaim -= nr_pages; | |
dea31f52 | 4762 | |
0b246afa JM |
4763 | wait_event(fs_info->async_submit_wait, |
4764 | atomic_read(&fs_info->async_delalloc_pages) <= | |
9f3a074d MX |
4765 | (int)max_reclaim); |
4766 | skip_async: | |
0019f10d | 4767 | spin_lock(&space_info->lock); |
957780eb JB |
4768 | if (list_empty(&space_info->tickets) && |
4769 | list_empty(&space_info->priority_tickets)) { | |
4770 | spin_unlock(&space_info->lock); | |
4771 | break; | |
4772 | } | |
0019f10d | 4773 | spin_unlock(&space_info->lock); |
5da9d01b | 4774 | |
36e39c40 | 4775 | loops++; |
f104d044 | 4776 | if (wait_ordered && !trans) { |
0b246afa | 4777 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f104d044 | 4778 | } else { |
f4c738c2 | 4779 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4780 | if (time_left) |
4781 | break; | |
4782 | } | |
963d678b | 4783 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4784 | &fs_info->delalloc_bytes); |
5da9d01b | 4785 | } |
5da9d01b YZ |
4786 | } |
4787 | ||
996478ca JB |
4788 | struct reserve_ticket { |
4789 | u64 bytes; | |
4790 | int error; | |
4791 | struct list_head list; | |
4792 | wait_queue_head_t wait; | |
4793 | }; | |
4794 | ||
663350ac JB |
4795 | /** |
4796 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4797 | * @root - the root we're allocating for | |
4798 | * @bytes - the number of bytes we want to reserve | |
4799 | * @force - force the commit | |
8bb8ab2e | 4800 | * |
663350ac JB |
4801 | * This will check to make sure that committing the transaction will actually |
4802 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4803 | * will return -ENOSPC. | |
8bb8ab2e | 4804 | */ |
0c9ab349 | 4805 | static int may_commit_transaction(struct btrfs_fs_info *fs_info, |
996478ca | 4806 | struct btrfs_space_info *space_info) |
663350ac | 4807 | { |
996478ca | 4808 | struct reserve_ticket *ticket = NULL; |
0b246afa | 4809 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv; |
663350ac | 4810 | struct btrfs_trans_handle *trans; |
996478ca | 4811 | u64 bytes; |
663350ac JB |
4812 | |
4813 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4814 | if (trans) | |
4815 | return -EAGAIN; | |
4816 | ||
996478ca JB |
4817 | spin_lock(&space_info->lock); |
4818 | if (!list_empty(&space_info->priority_tickets)) | |
4819 | ticket = list_first_entry(&space_info->priority_tickets, | |
4820 | struct reserve_ticket, list); | |
4821 | else if (!list_empty(&space_info->tickets)) | |
4822 | ticket = list_first_entry(&space_info->tickets, | |
4823 | struct reserve_ticket, list); | |
4824 | bytes = (ticket) ? ticket->bytes : 0; | |
4825 | spin_unlock(&space_info->lock); | |
4826 | ||
4827 | if (!bytes) | |
4828 | return 0; | |
663350ac JB |
4829 | |
4830 | /* See if there is enough pinned space to make this reservation */ | |
dec59fa3 EL |
4831 | if (__percpu_counter_compare(&space_info->total_bytes_pinned, |
4832 | bytes, | |
4833 | BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0) | |
663350ac | 4834 | goto commit; |
663350ac JB |
4835 | |
4836 | /* | |
4837 | * See if there is some space in the delayed insertion reservation for | |
4838 | * this reservation. | |
4839 | */ | |
4840 | if (space_info != delayed_rsv->space_info) | |
4841 | return -ENOSPC; | |
4842 | ||
4843 | spin_lock(&delayed_rsv->lock); | |
996478ca JB |
4844 | if (delayed_rsv->size > bytes) |
4845 | bytes = 0; | |
4846 | else | |
4847 | bytes -= delayed_rsv->size; | |
057aac3e NB |
4848 | spin_unlock(&delayed_rsv->lock); |
4849 | ||
dec59fa3 EL |
4850 | if (__percpu_counter_compare(&space_info->total_bytes_pinned, |
4851 | bytes, | |
4852 | BTRFS_TOTAL_BYTES_PINNED_BATCH) < 0) { | |
663350ac JB |
4853 | return -ENOSPC; |
4854 | } | |
663350ac JB |
4855 | |
4856 | commit: | |
a9b3311e | 4857 | trans = btrfs_join_transaction(fs_info->extent_root); |
663350ac JB |
4858 | if (IS_ERR(trans)) |
4859 | return -ENOSPC; | |
4860 | ||
3a45bb20 | 4861 | return btrfs_commit_transaction(trans); |
663350ac JB |
4862 | } |
4863 | ||
e38ae7a0 NB |
4864 | /* |
4865 | * Try to flush some data based on policy set by @state. This is only advisory | |
4866 | * and may fail for various reasons. The caller is supposed to examine the | |
4867 | * state of @space_info to detect the outcome. | |
4868 | */ | |
4869 | static void flush_space(struct btrfs_fs_info *fs_info, | |
96c3f433 | 4870 | struct btrfs_space_info *space_info, u64 num_bytes, |
7bdd6277 | 4871 | int state) |
96c3f433 | 4872 | { |
a9b3311e | 4873 | struct btrfs_root *root = fs_info->extent_root; |
96c3f433 JB |
4874 | struct btrfs_trans_handle *trans; |
4875 | int nr; | |
f4c738c2 | 4876 | int ret = 0; |
96c3f433 JB |
4877 | |
4878 | switch (state) { | |
96c3f433 JB |
4879 | case FLUSH_DELAYED_ITEMS_NR: |
4880 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 | 4881 | if (state == FLUSH_DELAYED_ITEMS_NR) |
2ff7e61e | 4882 | nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2; |
18cd8ea6 | 4883 | else |
96c3f433 | 4884 | nr = -1; |
18cd8ea6 | 4885 | |
96c3f433 JB |
4886 | trans = btrfs_join_transaction(root); |
4887 | if (IS_ERR(trans)) { | |
4888 | ret = PTR_ERR(trans); | |
4889 | break; | |
4890 | } | |
e5c304e6 | 4891 | ret = btrfs_run_delayed_items_nr(trans, nr); |
3a45bb20 | 4892 | btrfs_end_transaction(trans); |
96c3f433 | 4893 | break; |
67b0fd63 JB |
4894 | case FLUSH_DELALLOC: |
4895 | case FLUSH_DELALLOC_WAIT: | |
7bdd6277 | 4896 | shrink_delalloc(fs_info, num_bytes * 2, num_bytes, |
67b0fd63 JB |
4897 | state == FLUSH_DELALLOC_WAIT); |
4898 | break; | |
ea658bad JB |
4899 | case ALLOC_CHUNK: |
4900 | trans = btrfs_join_transaction(root); | |
4901 | if (IS_ERR(trans)) { | |
4902 | ret = PTR_ERR(trans); | |
4903 | break; | |
4904 | } | |
01458828 | 4905 | ret = do_chunk_alloc(trans, |
1b86826d | 4906 | btrfs_metadata_alloc_profile(fs_info), |
ea658bad | 4907 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 4908 | btrfs_end_transaction(trans); |
eecba891 | 4909 | if (ret > 0 || ret == -ENOSPC) |
ea658bad JB |
4910 | ret = 0; |
4911 | break; | |
96c3f433 | 4912 | case COMMIT_TRANS: |
996478ca | 4913 | ret = may_commit_transaction(fs_info, space_info); |
96c3f433 JB |
4914 | break; |
4915 | default: | |
4916 | ret = -ENOSPC; | |
4917 | break; | |
4918 | } | |
4919 | ||
7bdd6277 NB |
4920 | trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state, |
4921 | ret); | |
e38ae7a0 | 4922 | return; |
96c3f433 | 4923 | } |
21c7e756 MX |
4924 | |
4925 | static inline u64 | |
c1c4919b JM |
4926 | btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info, |
4927 | struct btrfs_space_info *space_info, | |
4928 | bool system_chunk) | |
21c7e756 | 4929 | { |
957780eb | 4930 | struct reserve_ticket *ticket; |
21c7e756 MX |
4931 | u64 used; |
4932 | u64 expected; | |
957780eb | 4933 | u64 to_reclaim = 0; |
21c7e756 | 4934 | |
957780eb JB |
4935 | list_for_each_entry(ticket, &space_info->tickets, list) |
4936 | to_reclaim += ticket->bytes; | |
4937 | list_for_each_entry(ticket, &space_info->priority_tickets, list) | |
4938 | to_reclaim += ticket->bytes; | |
4939 | if (to_reclaim) | |
4940 | return to_reclaim; | |
21c7e756 | 4941 | |
e0af2484 | 4942 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); |
c1c4919b JM |
4943 | if (can_overcommit(fs_info, space_info, to_reclaim, |
4944 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
e0af2484 WX |
4945 | return 0; |
4946 | ||
0eee8a49 NB |
4947 | used = btrfs_space_info_used(space_info, true); |
4948 | ||
c1c4919b JM |
4949 | if (can_overcommit(fs_info, space_info, SZ_1M, |
4950 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
21c7e756 MX |
4951 | expected = div_factor_fine(space_info->total_bytes, 95); |
4952 | else | |
4953 | expected = div_factor_fine(space_info->total_bytes, 90); | |
4954 | ||
4955 | if (used > expected) | |
4956 | to_reclaim = used - expected; | |
4957 | else | |
4958 | to_reclaim = 0; | |
4959 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
4960 | space_info->bytes_reserved); | |
21c7e756 MX |
4961 | return to_reclaim; |
4962 | } | |
4963 | ||
c1c4919b JM |
4964 | static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info, |
4965 | struct btrfs_space_info *space_info, | |
4966 | u64 used, bool system_chunk) | |
21c7e756 | 4967 | { |
365c5313 JB |
4968 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
4969 | ||
4970 | /* If we're just plain full then async reclaim just slows us down. */ | |
baee8790 | 4971 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) |
365c5313 JB |
4972 | return 0; |
4973 | ||
c1c4919b JM |
4974 | if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
4975 | system_chunk)) | |
d38b349c JB |
4976 | return 0; |
4977 | ||
0b246afa JM |
4978 | return (used >= thresh && !btrfs_fs_closing(fs_info) && |
4979 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); | |
21c7e756 MX |
4980 | } |
4981 | ||
957780eb | 4982 | static void wake_all_tickets(struct list_head *head) |
21c7e756 | 4983 | { |
957780eb | 4984 | struct reserve_ticket *ticket; |
25ce459c | 4985 | |
957780eb JB |
4986 | while (!list_empty(head)) { |
4987 | ticket = list_first_entry(head, struct reserve_ticket, list); | |
4988 | list_del_init(&ticket->list); | |
4989 | ticket->error = -ENOSPC; | |
4990 | wake_up(&ticket->wait); | |
21c7e756 | 4991 | } |
21c7e756 MX |
4992 | } |
4993 | ||
957780eb JB |
4994 | /* |
4995 | * This is for normal flushers, we can wait all goddamned day if we want to. We | |
4996 | * will loop and continuously try to flush as long as we are making progress. | |
4997 | * We count progress as clearing off tickets each time we have to loop. | |
4998 | */ | |
21c7e756 MX |
4999 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) |
5000 | { | |
5001 | struct btrfs_fs_info *fs_info; | |
5002 | struct btrfs_space_info *space_info; | |
5003 | u64 to_reclaim; | |
5004 | int flush_state; | |
957780eb | 5005 | int commit_cycles = 0; |
ce129655 | 5006 | u64 last_tickets_id; |
21c7e756 MX |
5007 | |
5008 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
5009 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
5010 | ||
957780eb | 5011 | spin_lock(&space_info->lock); |
c1c4919b JM |
5012 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5013 | false); | |
957780eb JB |
5014 | if (!to_reclaim) { |
5015 | space_info->flush = 0; | |
5016 | spin_unlock(&space_info->lock); | |
21c7e756 | 5017 | return; |
957780eb | 5018 | } |
ce129655 | 5019 | last_tickets_id = space_info->tickets_id; |
957780eb | 5020 | spin_unlock(&space_info->lock); |
21c7e756 MX |
5021 | |
5022 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
957780eb | 5023 | do { |
e38ae7a0 | 5024 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
957780eb JB |
5025 | spin_lock(&space_info->lock); |
5026 | if (list_empty(&space_info->tickets)) { | |
5027 | space_info->flush = 0; | |
5028 | spin_unlock(&space_info->lock); | |
5029 | return; | |
5030 | } | |
c1c4919b JM |
5031 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, |
5032 | space_info, | |
5033 | false); | |
ce129655 | 5034 | if (last_tickets_id == space_info->tickets_id) { |
957780eb JB |
5035 | flush_state++; |
5036 | } else { | |
ce129655 | 5037 | last_tickets_id = space_info->tickets_id; |
957780eb JB |
5038 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
5039 | if (commit_cycles) | |
5040 | commit_cycles--; | |
5041 | } | |
5042 | ||
5043 | if (flush_state > COMMIT_TRANS) { | |
5044 | commit_cycles++; | |
5045 | if (commit_cycles > 2) { | |
5046 | wake_all_tickets(&space_info->tickets); | |
5047 | space_info->flush = 0; | |
5048 | } else { | |
5049 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5050 | } | |
5051 | } | |
5052 | spin_unlock(&space_info->lock); | |
5053 | } while (flush_state <= COMMIT_TRANS); | |
5054 | } | |
5055 | ||
5056 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
5057 | { | |
5058 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
5059 | } | |
5060 | ||
5061 | static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, | |
5062 | struct btrfs_space_info *space_info, | |
5063 | struct reserve_ticket *ticket) | |
5064 | { | |
5065 | u64 to_reclaim; | |
5066 | int flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5067 | ||
5068 | spin_lock(&space_info->lock); | |
c1c4919b JM |
5069 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5070 | false); | |
957780eb JB |
5071 | if (!to_reclaim) { |
5072 | spin_unlock(&space_info->lock); | |
5073 | return; | |
5074 | } | |
5075 | spin_unlock(&space_info->lock); | |
5076 | ||
21c7e756 | 5077 | do { |
7bdd6277 | 5078 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
21c7e756 | 5079 | flush_state++; |
957780eb JB |
5080 | spin_lock(&space_info->lock); |
5081 | if (ticket->bytes == 0) { | |
5082 | spin_unlock(&space_info->lock); | |
21c7e756 | 5083 | return; |
957780eb JB |
5084 | } |
5085 | spin_unlock(&space_info->lock); | |
5086 | ||
5087 | /* | |
5088 | * Priority flushers can't wait on delalloc without | |
5089 | * deadlocking. | |
5090 | */ | |
5091 | if (flush_state == FLUSH_DELALLOC || | |
5092 | flush_state == FLUSH_DELALLOC_WAIT) | |
5093 | flush_state = ALLOC_CHUNK; | |
365c5313 | 5094 | } while (flush_state < COMMIT_TRANS); |
21c7e756 MX |
5095 | } |
5096 | ||
957780eb JB |
5097 | static int wait_reserve_ticket(struct btrfs_fs_info *fs_info, |
5098 | struct btrfs_space_info *space_info, | |
5099 | struct reserve_ticket *ticket, u64 orig_bytes) | |
5100 | ||
21c7e756 | 5101 | { |
957780eb JB |
5102 | DEFINE_WAIT(wait); |
5103 | int ret = 0; | |
5104 | ||
5105 | spin_lock(&space_info->lock); | |
5106 | while (ticket->bytes > 0 && ticket->error == 0) { | |
5107 | ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE); | |
5108 | if (ret) { | |
5109 | ret = -EINTR; | |
5110 | break; | |
5111 | } | |
5112 | spin_unlock(&space_info->lock); | |
5113 | ||
5114 | schedule(); | |
5115 | ||
5116 | finish_wait(&ticket->wait, &wait); | |
5117 | spin_lock(&space_info->lock); | |
5118 | } | |
5119 | if (!ret) | |
5120 | ret = ticket->error; | |
5121 | if (!list_empty(&ticket->list)) | |
5122 | list_del_init(&ticket->list); | |
5123 | if (ticket->bytes && ticket->bytes < orig_bytes) { | |
5124 | u64 num_bytes = orig_bytes - ticket->bytes; | |
9f9b8e8d | 5125 | update_bytes_may_use(space_info, -num_bytes); |
957780eb JB |
5126 | trace_btrfs_space_reservation(fs_info, "space_info", |
5127 | space_info->flags, num_bytes, 0); | |
5128 | } | |
5129 | spin_unlock(&space_info->lock); | |
5130 | ||
5131 | return ret; | |
21c7e756 MX |
5132 | } |
5133 | ||
4a92b1b8 JB |
5134 | /** |
5135 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5136 | * @root - the root we're allocating for | |
957780eb | 5137 | * @space_info - the space info we want to allocate from |
4a92b1b8 | 5138 | * @orig_bytes - the number of bytes we want |
48fc7f7e | 5139 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 5140 | * |
01327610 | 5141 | * This will reserve orig_bytes number of bytes from the space info associated |
4a92b1b8 JB |
5142 | * with the block_rsv. If there is not enough space it will make an attempt to |
5143 | * flush out space to make room. It will do this by flushing delalloc if | |
5144 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5145 | * regain reservations will be made and this will fail if there is not enough | |
5146 | * space already. | |
8bb8ab2e | 5147 | */ |
c1c4919b | 5148 | static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info, |
957780eb JB |
5149 | struct btrfs_space_info *space_info, |
5150 | u64 orig_bytes, | |
c1c4919b JM |
5151 | enum btrfs_reserve_flush_enum flush, |
5152 | bool system_chunk) | |
9ed74f2d | 5153 | { |
957780eb | 5154 | struct reserve_ticket ticket; |
2bf64758 | 5155 | u64 used; |
8bb8ab2e | 5156 | int ret = 0; |
9ed74f2d | 5157 | |
957780eb | 5158 | ASSERT(orig_bytes); |
8ca17f0f | 5159 | ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL); |
fdb5effd | 5160 | |
8bb8ab2e | 5161 | spin_lock(&space_info->lock); |
fdb5effd | 5162 | ret = -ENOSPC; |
4136135b | 5163 | used = btrfs_space_info_used(space_info, true); |
9ed74f2d | 5164 | |
8bb8ab2e | 5165 | /* |
957780eb JB |
5166 | * If we have enough space then hooray, make our reservation and carry |
5167 | * on. If not see if we can overcommit, and if we can, hooray carry on. | |
5168 | * If not things get more complicated. | |
8bb8ab2e | 5169 | */ |
957780eb | 5170 | if (used + orig_bytes <= space_info->total_bytes) { |
9f9b8e8d | 5171 | update_bytes_may_use(space_info, orig_bytes); |
0b246afa JM |
5172 | trace_btrfs_space_reservation(fs_info, "space_info", |
5173 | space_info->flags, orig_bytes, 1); | |
957780eb | 5174 | ret = 0; |
c1c4919b JM |
5175 | } else if (can_overcommit(fs_info, space_info, orig_bytes, flush, |
5176 | system_chunk)) { | |
9f9b8e8d | 5177 | update_bytes_may_use(space_info, orig_bytes); |
0b246afa JM |
5178 | trace_btrfs_space_reservation(fs_info, "space_info", |
5179 | space_info->flags, orig_bytes, 1); | |
44734ed1 | 5180 | ret = 0; |
2bf64758 JB |
5181 | } |
5182 | ||
8bb8ab2e | 5183 | /* |
957780eb JB |
5184 | * If we couldn't make a reservation then setup our reservation ticket |
5185 | * and kick the async worker if it's not already running. | |
08e007d2 | 5186 | * |
957780eb JB |
5187 | * If we are a priority flusher then we just need to add our ticket to |
5188 | * the list and we will do our own flushing further down. | |
8bb8ab2e | 5189 | */ |
72bcd99d | 5190 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
957780eb JB |
5191 | ticket.bytes = orig_bytes; |
5192 | ticket.error = 0; | |
5193 | init_waitqueue_head(&ticket.wait); | |
5194 | if (flush == BTRFS_RESERVE_FLUSH_ALL) { | |
5195 | list_add_tail(&ticket.list, &space_info->tickets); | |
5196 | if (!space_info->flush) { | |
5197 | space_info->flush = 1; | |
0b246afa | 5198 | trace_btrfs_trigger_flush(fs_info, |
f376df2b JB |
5199 | space_info->flags, |
5200 | orig_bytes, flush, | |
5201 | "enospc"); | |
957780eb | 5202 | queue_work(system_unbound_wq, |
c1c4919b | 5203 | &fs_info->async_reclaim_work); |
957780eb JB |
5204 | } |
5205 | } else { | |
5206 | list_add_tail(&ticket.list, | |
5207 | &space_info->priority_tickets); | |
5208 | } | |
21c7e756 MX |
5209 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
5210 | used += orig_bytes; | |
f6acfd50 JB |
5211 | /* |
5212 | * We will do the space reservation dance during log replay, | |
5213 | * which means we won't have fs_info->fs_root set, so don't do | |
5214 | * the async reclaim as we will panic. | |
5215 | */ | |
0b246afa | 5216 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) && |
c1c4919b JM |
5217 | need_do_async_reclaim(fs_info, space_info, |
5218 | used, system_chunk) && | |
0b246afa JM |
5219 | !work_busy(&fs_info->async_reclaim_work)) { |
5220 | trace_btrfs_trigger_flush(fs_info, space_info->flags, | |
5221 | orig_bytes, flush, "preempt"); | |
21c7e756 | 5222 | queue_work(system_unbound_wq, |
0b246afa | 5223 | &fs_info->async_reclaim_work); |
f376df2b | 5224 | } |
8bb8ab2e | 5225 | } |
f0486c68 | 5226 | spin_unlock(&space_info->lock); |
08e007d2 | 5227 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
957780eb | 5228 | return ret; |
f0486c68 | 5229 | |
957780eb | 5230 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
0b246afa | 5231 | return wait_reserve_ticket(fs_info, space_info, &ticket, |
957780eb | 5232 | orig_bytes); |
08e007d2 | 5233 | |
957780eb | 5234 | ret = 0; |
0b246afa | 5235 | priority_reclaim_metadata_space(fs_info, space_info, &ticket); |
957780eb JB |
5236 | spin_lock(&space_info->lock); |
5237 | if (ticket.bytes) { | |
5238 | if (ticket.bytes < orig_bytes) { | |
5239 | u64 num_bytes = orig_bytes - ticket.bytes; | |
9f9b8e8d | 5240 | update_bytes_may_use(space_info, -num_bytes); |
0b246afa JM |
5241 | trace_btrfs_space_reservation(fs_info, "space_info", |
5242 | space_info->flags, | |
5243 | num_bytes, 0); | |
08e007d2 | 5244 | |
957780eb JB |
5245 | } |
5246 | list_del_init(&ticket.list); | |
5247 | ret = -ENOSPC; | |
5248 | } | |
5249 | spin_unlock(&space_info->lock); | |
5250 | ASSERT(list_empty(&ticket.list)); | |
5251 | return ret; | |
5252 | } | |
8bb8ab2e | 5253 | |
957780eb JB |
5254 | /** |
5255 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5256 | * @root - the root we're allocating for | |
5257 | * @block_rsv - the block_rsv we're allocating for | |
5258 | * @orig_bytes - the number of bytes we want | |
5259 | * @flush - whether or not we can flush to make our reservation | |
5260 | * | |
5261 | * This will reserve orgi_bytes number of bytes from the space info associated | |
5262 | * with the block_rsv. If there is not enough space it will make an attempt to | |
5263 | * flush out space to make room. It will do this by flushing delalloc if | |
5264 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5265 | * regain reservations will be made and this will fail if there is not enough | |
5266 | * space already. | |
5267 | */ | |
5268 | static int reserve_metadata_bytes(struct btrfs_root *root, | |
5269 | struct btrfs_block_rsv *block_rsv, | |
5270 | u64 orig_bytes, | |
5271 | enum btrfs_reserve_flush_enum flush) | |
5272 | { | |
0b246afa JM |
5273 | struct btrfs_fs_info *fs_info = root->fs_info; |
5274 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
957780eb | 5275 | int ret; |
c1c4919b | 5276 | bool system_chunk = (root == fs_info->chunk_root); |
957780eb | 5277 | |
c1c4919b JM |
5278 | ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info, |
5279 | orig_bytes, flush, system_chunk); | |
5d80366e JB |
5280 | if (ret == -ENOSPC && |
5281 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5d80366e JB |
5282 | if (block_rsv != global_rsv && |
5283 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5284 | ret = 0; | |
5285 | } | |
9a3daff3 | 5286 | if (ret == -ENOSPC) { |
0b246afa | 5287 | trace_btrfs_space_reservation(fs_info, "space_info:enospc", |
957780eb JB |
5288 | block_rsv->space_info->flags, |
5289 | orig_bytes, 1); | |
9a3daff3 NB |
5290 | |
5291 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) | |
5292 | dump_space_info(fs_info, block_rsv->space_info, | |
5293 | orig_bytes, 0); | |
5294 | } | |
f0486c68 YZ |
5295 | return ret; |
5296 | } | |
5297 | ||
79787eaa JM |
5298 | static struct btrfs_block_rsv *get_block_rsv( |
5299 | const struct btrfs_trans_handle *trans, | |
5300 | const struct btrfs_root *root) | |
f0486c68 | 5301 | { |
0b246afa | 5302 | struct btrfs_fs_info *fs_info = root->fs_info; |
4c13d758 JB |
5303 | struct btrfs_block_rsv *block_rsv = NULL; |
5304 | ||
e9cf439f | 5305 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa JM |
5306 | (root == fs_info->csum_root && trans->adding_csums) || |
5307 | (root == fs_info->uuid_root)) | |
f7a81ea4 SB |
5308 | block_rsv = trans->block_rsv; |
5309 | ||
4c13d758 | 5310 | if (!block_rsv) |
f0486c68 YZ |
5311 | block_rsv = root->block_rsv; |
5312 | ||
5313 | if (!block_rsv) | |
0b246afa | 5314 | block_rsv = &fs_info->empty_block_rsv; |
f0486c68 YZ |
5315 | |
5316 | return block_rsv; | |
5317 | } | |
5318 | ||
5319 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
5320 | u64 num_bytes) | |
5321 | { | |
5322 | int ret = -ENOSPC; | |
5323 | spin_lock(&block_rsv->lock); | |
5324 | if (block_rsv->reserved >= num_bytes) { | |
5325 | block_rsv->reserved -= num_bytes; | |
5326 | if (block_rsv->reserved < block_rsv->size) | |
5327 | block_rsv->full = 0; | |
5328 | ret = 0; | |
5329 | } | |
5330 | spin_unlock(&block_rsv->lock); | |
5331 | return ret; | |
5332 | } | |
5333 | ||
5334 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
3a584174 | 5335 | u64 num_bytes, bool update_size) |
f0486c68 YZ |
5336 | { |
5337 | spin_lock(&block_rsv->lock); | |
5338 | block_rsv->reserved += num_bytes; | |
5339 | if (update_size) | |
5340 | block_rsv->size += num_bytes; | |
5341 | else if (block_rsv->reserved >= block_rsv->size) | |
5342 | block_rsv->full = 1; | |
5343 | spin_unlock(&block_rsv->lock); | |
5344 | } | |
5345 | ||
d52be818 JB |
5346 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
5347 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5348 | int min_factor) | |
5349 | { | |
5350 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5351 | u64 min_bytes; | |
5352 | ||
5353 | if (global_rsv->space_info != dest->space_info) | |
5354 | return -ENOSPC; | |
5355 | ||
5356 | spin_lock(&global_rsv->lock); | |
5357 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5358 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5359 | spin_unlock(&global_rsv->lock); | |
5360 | return -ENOSPC; | |
5361 | } | |
5362 | global_rsv->reserved -= num_bytes; | |
5363 | if (global_rsv->reserved < global_rsv->size) | |
5364 | global_rsv->full = 0; | |
5365 | spin_unlock(&global_rsv->lock); | |
5366 | ||
3a584174 | 5367 | block_rsv_add_bytes(dest, num_bytes, true); |
d52be818 JB |
5368 | return 0; |
5369 | } | |
5370 | ||
957780eb JB |
5371 | /* |
5372 | * This is for space we already have accounted in space_info->bytes_may_use, so | |
5373 | * basically when we're returning space from block_rsv's. | |
5374 | */ | |
5375 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
5376 | struct btrfs_space_info *space_info, | |
5377 | u64 num_bytes) | |
5378 | { | |
5379 | struct reserve_ticket *ticket; | |
5380 | struct list_head *head; | |
5381 | u64 used; | |
5382 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH; | |
5383 | bool check_overcommit = false; | |
5384 | ||
5385 | spin_lock(&space_info->lock); | |
5386 | head = &space_info->priority_tickets; | |
5387 | ||
5388 | /* | |
5389 | * If we are over our limit then we need to check and see if we can | |
5390 | * overcommit, and if we can't then we just need to free up our space | |
5391 | * and not satisfy any requests. | |
5392 | */ | |
0eee8a49 | 5393 | used = btrfs_space_info_used(space_info, true); |
957780eb JB |
5394 | if (used - num_bytes >= space_info->total_bytes) |
5395 | check_overcommit = true; | |
5396 | again: | |
5397 | while (!list_empty(head) && num_bytes) { | |
5398 | ticket = list_first_entry(head, struct reserve_ticket, | |
5399 | list); | |
5400 | /* | |
5401 | * We use 0 bytes because this space is already reserved, so | |
5402 | * adding the ticket space would be a double count. | |
5403 | */ | |
5404 | if (check_overcommit && | |
c1c4919b | 5405 | !can_overcommit(fs_info, space_info, 0, flush, false)) |
957780eb JB |
5406 | break; |
5407 | if (num_bytes >= ticket->bytes) { | |
5408 | list_del_init(&ticket->list); | |
5409 | num_bytes -= ticket->bytes; | |
5410 | ticket->bytes = 0; | |
ce129655 | 5411 | space_info->tickets_id++; |
957780eb JB |
5412 | wake_up(&ticket->wait); |
5413 | } else { | |
5414 | ticket->bytes -= num_bytes; | |
5415 | num_bytes = 0; | |
5416 | } | |
5417 | } | |
5418 | ||
5419 | if (num_bytes && head == &space_info->priority_tickets) { | |
5420 | head = &space_info->tickets; | |
5421 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
5422 | goto again; | |
5423 | } | |
9f9b8e8d | 5424 | update_bytes_may_use(space_info, -num_bytes); |
957780eb JB |
5425 | trace_btrfs_space_reservation(fs_info, "space_info", |
5426 | space_info->flags, num_bytes, 0); | |
5427 | spin_unlock(&space_info->lock); | |
5428 | } | |
5429 | ||
5430 | /* | |
5431 | * This is for newly allocated space that isn't accounted in | |
5432 | * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent | |
5433 | * we use this helper. | |
5434 | */ | |
5435 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, | |
5436 | struct btrfs_space_info *space_info, | |
5437 | u64 num_bytes) | |
5438 | { | |
5439 | struct reserve_ticket *ticket; | |
5440 | struct list_head *head = &space_info->priority_tickets; | |
5441 | ||
5442 | again: | |
5443 | while (!list_empty(head) && num_bytes) { | |
5444 | ticket = list_first_entry(head, struct reserve_ticket, | |
5445 | list); | |
5446 | if (num_bytes >= ticket->bytes) { | |
5447 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5448 | space_info->flags, | |
5449 | ticket->bytes, 1); | |
5450 | list_del_init(&ticket->list); | |
5451 | num_bytes -= ticket->bytes; | |
9f9b8e8d | 5452 | update_bytes_may_use(space_info, ticket->bytes); |
957780eb | 5453 | ticket->bytes = 0; |
ce129655 | 5454 | space_info->tickets_id++; |
957780eb JB |
5455 | wake_up(&ticket->wait); |
5456 | } else { | |
5457 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5458 | space_info->flags, | |
5459 | num_bytes, 1); | |
9f9b8e8d | 5460 | update_bytes_may_use(space_info, num_bytes); |
957780eb JB |
5461 | ticket->bytes -= num_bytes; |
5462 | num_bytes = 0; | |
5463 | } | |
5464 | } | |
5465 | ||
5466 | if (num_bytes && head == &space_info->priority_tickets) { | |
5467 | head = &space_info->tickets; | |
5468 | goto again; | |
5469 | } | |
5470 | } | |
5471 | ||
69fe2d75 | 5472 | static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
8c2a3ca2 | 5473 | struct btrfs_block_rsv *block_rsv, |
ff6bc37e QW |
5474 | struct btrfs_block_rsv *dest, u64 num_bytes, |
5475 | u64 *qgroup_to_release_ret) | |
f0486c68 YZ |
5476 | { |
5477 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
ff6bc37e | 5478 | u64 qgroup_to_release = 0; |
69fe2d75 | 5479 | u64 ret; |
f0486c68 YZ |
5480 | |
5481 | spin_lock(&block_rsv->lock); | |
ff6bc37e | 5482 | if (num_bytes == (u64)-1) { |
f0486c68 | 5483 | num_bytes = block_rsv->size; |
ff6bc37e QW |
5484 | qgroup_to_release = block_rsv->qgroup_rsv_size; |
5485 | } | |
f0486c68 YZ |
5486 | block_rsv->size -= num_bytes; |
5487 | if (block_rsv->reserved >= block_rsv->size) { | |
5488 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5489 | block_rsv->reserved = block_rsv->size; | |
5490 | block_rsv->full = 1; | |
5491 | } else { | |
5492 | num_bytes = 0; | |
5493 | } | |
ff6bc37e QW |
5494 | if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) { |
5495 | qgroup_to_release = block_rsv->qgroup_rsv_reserved - | |
5496 | block_rsv->qgroup_rsv_size; | |
5497 | block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size; | |
5498 | } else { | |
5499 | qgroup_to_release = 0; | |
5500 | } | |
f0486c68 YZ |
5501 | spin_unlock(&block_rsv->lock); |
5502 | ||
69fe2d75 | 5503 | ret = num_bytes; |
f0486c68 YZ |
5504 | if (num_bytes > 0) { |
5505 | if (dest) { | |
e9e22899 JB |
5506 | spin_lock(&dest->lock); |
5507 | if (!dest->full) { | |
5508 | u64 bytes_to_add; | |
5509 | ||
5510 | bytes_to_add = dest->size - dest->reserved; | |
5511 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5512 | dest->reserved += bytes_to_add; | |
5513 | if (dest->reserved >= dest->size) | |
5514 | dest->full = 1; | |
5515 | num_bytes -= bytes_to_add; | |
5516 | } | |
5517 | spin_unlock(&dest->lock); | |
5518 | } | |
957780eb JB |
5519 | if (num_bytes) |
5520 | space_info_add_old_bytes(fs_info, space_info, | |
5521 | num_bytes); | |
9ed74f2d | 5522 | } |
ff6bc37e QW |
5523 | if (qgroup_to_release_ret) |
5524 | *qgroup_to_release_ret = qgroup_to_release; | |
69fe2d75 | 5525 | return ret; |
f0486c68 | 5526 | } |
4e06bdd6 | 5527 | |
25d609f8 JB |
5528 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src, |
5529 | struct btrfs_block_rsv *dst, u64 num_bytes, | |
3a584174 | 5530 | bool update_size) |
f0486c68 YZ |
5531 | { |
5532 | int ret; | |
9ed74f2d | 5533 | |
f0486c68 YZ |
5534 | ret = block_rsv_use_bytes(src, num_bytes); |
5535 | if (ret) | |
5536 | return ret; | |
9ed74f2d | 5537 | |
25d609f8 | 5538 | block_rsv_add_bytes(dst, num_bytes, update_size); |
9ed74f2d JB |
5539 | return 0; |
5540 | } | |
5541 | ||
66d8f3dd | 5542 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 5543 | { |
f0486c68 YZ |
5544 | memset(rsv, 0, sizeof(*rsv)); |
5545 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 5546 | rsv->type = type; |
f0486c68 YZ |
5547 | } |
5548 | ||
69fe2d75 JB |
5549 | void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info, |
5550 | struct btrfs_block_rsv *rsv, | |
5551 | unsigned short type) | |
5552 | { | |
5553 | btrfs_init_block_rsv(rsv, type); | |
5554 | rsv->space_info = __find_space_info(fs_info, | |
5555 | BTRFS_BLOCK_GROUP_METADATA); | |
5556 | } | |
5557 | ||
2ff7e61e | 5558 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info, |
66d8f3dd | 5559 | unsigned short type) |
f0486c68 YZ |
5560 | { |
5561 | struct btrfs_block_rsv *block_rsv; | |
9ed74f2d | 5562 | |
f0486c68 YZ |
5563 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
5564 | if (!block_rsv) | |
5565 | return NULL; | |
9ed74f2d | 5566 | |
69fe2d75 | 5567 | btrfs_init_metadata_block_rsv(fs_info, block_rsv, type); |
f0486c68 YZ |
5568 | return block_rsv; |
5569 | } | |
9ed74f2d | 5570 | |
2ff7e61e | 5571 | void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5572 | struct btrfs_block_rsv *rsv) |
5573 | { | |
2aaa6655 JB |
5574 | if (!rsv) |
5575 | return; | |
2ff7e61e | 5576 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
dabdb640 | 5577 | kfree(rsv); |
9ed74f2d JB |
5578 | } |
5579 | ||
08e007d2 MX |
5580 | int btrfs_block_rsv_add(struct btrfs_root *root, |
5581 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5582 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5583 | { |
f0486c68 | 5584 | int ret; |
9ed74f2d | 5585 | |
f0486c68 YZ |
5586 | if (num_bytes == 0) |
5587 | return 0; | |
8bb8ab2e | 5588 | |
61b520a9 | 5589 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
5a2cb25a | 5590 | if (!ret) |
3a584174 | 5591 | block_rsv_add_bytes(block_rsv, num_bytes, true); |
9ed74f2d | 5592 | |
f0486c68 | 5593 | return ret; |
f0486c68 | 5594 | } |
9ed74f2d | 5595 | |
2ff7e61e | 5596 | int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
5597 | { |
5598 | u64 num_bytes = 0; | |
f0486c68 | 5599 | int ret = -ENOSPC; |
9ed74f2d | 5600 | |
f0486c68 YZ |
5601 | if (!block_rsv) |
5602 | return 0; | |
9ed74f2d | 5603 | |
f0486c68 | 5604 | spin_lock(&block_rsv->lock); |
36ba022a JB |
5605 | num_bytes = div_factor(block_rsv->size, min_factor); |
5606 | if (block_rsv->reserved >= num_bytes) | |
5607 | ret = 0; | |
5608 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 5609 | |
36ba022a JB |
5610 | return ret; |
5611 | } | |
5612 | ||
08e007d2 MX |
5613 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
5614 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5615 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
5616 | { |
5617 | u64 num_bytes = 0; | |
5618 | int ret = -ENOSPC; | |
5619 | ||
5620 | if (!block_rsv) | |
5621 | return 0; | |
5622 | ||
5623 | spin_lock(&block_rsv->lock); | |
5624 | num_bytes = min_reserved; | |
13553e52 | 5625 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 5626 | ret = 0; |
13553e52 | 5627 | else |
f0486c68 | 5628 | num_bytes -= block_rsv->reserved; |
f0486c68 | 5629 | spin_unlock(&block_rsv->lock); |
13553e52 | 5630 | |
f0486c68 YZ |
5631 | if (!ret) |
5632 | return 0; | |
5633 | ||
aa38a711 | 5634 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 | 5635 | if (!ret) { |
3a584174 | 5636 | block_rsv_add_bytes(block_rsv, num_bytes, false); |
f0486c68 | 5637 | return 0; |
6a63209f | 5638 | } |
9ed74f2d | 5639 | |
13553e52 | 5640 | return ret; |
f0486c68 YZ |
5641 | } |
5642 | ||
69fe2d75 JB |
5643 | /** |
5644 | * btrfs_inode_rsv_refill - refill the inode block rsv. | |
5645 | * @inode - the inode we are refilling. | |
5646 | * @flush - the flusing restriction. | |
5647 | * | |
5648 | * Essentially the same as btrfs_block_rsv_refill, except it uses the | |
5649 | * block_rsv->size as the minimum size. We'll either refill the missing amount | |
5650 | * or return if we already have enough space. This will also handle the resreve | |
5651 | * tracepoint for the reserved amount. | |
5652 | */ | |
3f2dd7a0 QW |
5653 | static int btrfs_inode_rsv_refill(struct btrfs_inode *inode, |
5654 | enum btrfs_reserve_flush_enum flush) | |
69fe2d75 JB |
5655 | { |
5656 | struct btrfs_root *root = inode->root; | |
5657 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; | |
5658 | u64 num_bytes = 0; | |
ff6bc37e | 5659 | u64 qgroup_num_bytes = 0; |
69fe2d75 JB |
5660 | int ret = -ENOSPC; |
5661 | ||
5662 | spin_lock(&block_rsv->lock); | |
5663 | if (block_rsv->reserved < block_rsv->size) | |
5664 | num_bytes = block_rsv->size - block_rsv->reserved; | |
ff6bc37e QW |
5665 | if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size) |
5666 | qgroup_num_bytes = block_rsv->qgroup_rsv_size - | |
5667 | block_rsv->qgroup_rsv_reserved; | |
69fe2d75 JB |
5668 | spin_unlock(&block_rsv->lock); |
5669 | ||
5670 | if (num_bytes == 0) | |
5671 | return 0; | |
5672 | ||
ff6bc37e | 5673 | ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes, true); |
43b18595 QW |
5674 | if (ret) |
5675 | return ret; | |
69fe2d75 JB |
5676 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
5677 | if (!ret) { | |
3a584174 | 5678 | block_rsv_add_bytes(block_rsv, num_bytes, false); |
69fe2d75 JB |
5679 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
5680 | btrfs_ino(inode), num_bytes, 1); | |
ff6bc37e QW |
5681 | |
5682 | /* Don't forget to increase qgroup_rsv_reserved */ | |
5683 | spin_lock(&block_rsv->lock); | |
5684 | block_rsv->qgroup_rsv_reserved += qgroup_num_bytes; | |
5685 | spin_unlock(&block_rsv->lock); | |
5686 | } else | |
5687 | btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes); | |
69fe2d75 JB |
5688 | return ret; |
5689 | } | |
5690 | ||
5691 | /** | |
5692 | * btrfs_inode_rsv_release - release any excessive reservation. | |
5693 | * @inode - the inode we need to release from. | |
43b18595 QW |
5694 | * @qgroup_free - free or convert qgroup meta. |
5695 | * Unlike normal operation, qgroup meta reservation needs to know if we are | |
5696 | * freeing qgroup reservation or just converting it into per-trans. Normally | |
5697 | * @qgroup_free is true for error handling, and false for normal release. | |
69fe2d75 JB |
5698 | * |
5699 | * This is the same as btrfs_block_rsv_release, except that it handles the | |
5700 | * tracepoint for the reservation. | |
5701 | */ | |
43b18595 | 5702 | static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free) |
69fe2d75 JB |
5703 | { |
5704 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
5705 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5706 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; | |
5707 | u64 released = 0; | |
ff6bc37e | 5708 | u64 qgroup_to_release = 0; |
69fe2d75 JB |
5709 | |
5710 | /* | |
5711 | * Since we statically set the block_rsv->size we just want to say we | |
5712 | * are releasing 0 bytes, and then we'll just get the reservation over | |
5713 | * the size free'd. | |
5714 | */ | |
ff6bc37e QW |
5715 | released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0, |
5716 | &qgroup_to_release); | |
69fe2d75 JB |
5717 | if (released > 0) |
5718 | trace_btrfs_space_reservation(fs_info, "delalloc", | |
5719 | btrfs_ino(inode), released, 0); | |
43b18595 | 5720 | if (qgroup_free) |
ff6bc37e | 5721 | btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release); |
43b18595 | 5722 | else |
ff6bc37e QW |
5723 | btrfs_qgroup_convert_reserved_meta(inode->root, |
5724 | qgroup_to_release); | |
69fe2d75 JB |
5725 | } |
5726 | ||
2ff7e61e | 5727 | void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5728 | struct btrfs_block_rsv *block_rsv, |
5729 | u64 num_bytes) | |
5730 | { | |
0b246afa JM |
5731 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
5732 | ||
17504584 | 5733 | if (global_rsv == block_rsv || |
f0486c68 YZ |
5734 | block_rsv->space_info != global_rsv->space_info) |
5735 | global_rsv = NULL; | |
ff6bc37e | 5736 | block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes, NULL); |
6a63209f JB |
5737 | } |
5738 | ||
8929ecfa YZ |
5739 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5740 | { | |
5741 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5742 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5743 | u64 num_bytes; | |
6a63209f | 5744 | |
ae2e4728 JB |
5745 | /* |
5746 | * The global block rsv is based on the size of the extent tree, the | |
5747 | * checksum tree and the root tree. If the fs is empty we want to set | |
5748 | * it to a minimal amount for safety. | |
5749 | */ | |
5750 | num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) + | |
5751 | btrfs_root_used(&fs_info->csum_root->root_item) + | |
5752 | btrfs_root_used(&fs_info->tree_root->root_item); | |
5753 | num_bytes = max_t(u64, num_bytes, SZ_16M); | |
33b4d47f | 5754 | |
8929ecfa | 5755 | spin_lock(&sinfo->lock); |
1f699d38 | 5756 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5757 | |
ee22184b | 5758 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); |
4e06bdd6 | 5759 | |
fb4b10e5 | 5760 | if (block_rsv->reserved < block_rsv->size) { |
4136135b | 5761 | num_bytes = btrfs_space_info_used(sinfo, true); |
fb4b10e5 JB |
5762 | if (sinfo->total_bytes > num_bytes) { |
5763 | num_bytes = sinfo->total_bytes - num_bytes; | |
5764 | num_bytes = min(num_bytes, | |
5765 | block_rsv->size - block_rsv->reserved); | |
5766 | block_rsv->reserved += num_bytes; | |
9f9b8e8d | 5767 | update_bytes_may_use(sinfo, num_bytes); |
fb4b10e5 JB |
5768 | trace_btrfs_space_reservation(fs_info, "space_info", |
5769 | sinfo->flags, num_bytes, | |
5770 | 1); | |
5771 | } | |
5772 | } else if (block_rsv->reserved > block_rsv->size) { | |
8929ecfa | 5773 | num_bytes = block_rsv->reserved - block_rsv->size; |
9f9b8e8d | 5774 | update_bytes_may_use(sinfo, -num_bytes); |
8c2a3ca2 | 5775 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5776 | sinfo->flags, num_bytes, 0); |
8929ecfa | 5777 | block_rsv->reserved = block_rsv->size; |
8929ecfa | 5778 | } |
182608c8 | 5779 | |
fb4b10e5 JB |
5780 | if (block_rsv->reserved == block_rsv->size) |
5781 | block_rsv->full = 1; | |
5782 | else | |
5783 | block_rsv->full = 0; | |
5784 | ||
8929ecfa | 5785 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5786 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5787 | } |
5788 | ||
f0486c68 | 5789 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5790 | { |
f0486c68 | 5791 | struct btrfs_space_info *space_info; |
6a63209f | 5792 | |
f0486c68 YZ |
5793 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
5794 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 5795 | |
f0486c68 | 5796 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5797 | fs_info->global_block_rsv.space_info = space_info; |
f0486c68 YZ |
5798 | fs_info->trans_block_rsv.space_info = space_info; |
5799 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5800 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 5801 | |
8929ecfa YZ |
5802 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
5803 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
5804 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
5805 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5806 | if (fs_info->quota_root) |
5807 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5808 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5809 | |
8929ecfa | 5810 | update_global_block_rsv(fs_info); |
6a63209f JB |
5811 | } |
5812 | ||
8929ecfa | 5813 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5814 | { |
8c2a3ca2 | 5815 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
ff6bc37e | 5816 | (u64)-1, NULL); |
8929ecfa YZ |
5817 | WARN_ON(fs_info->trans_block_rsv.size > 0); |
5818 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5819 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5820 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5821 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5822 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
5823 | } |
5824 | ||
6a63209f | 5825 | |
4fbcdf66 FM |
5826 | /* |
5827 | * To be called after all the new block groups attached to the transaction | |
5828 | * handle have been created (btrfs_create_pending_block_groups()). | |
5829 | */ | |
5830 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5831 | { | |
64b63580 | 5832 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4fbcdf66 FM |
5833 | |
5834 | if (!trans->chunk_bytes_reserved) | |
5835 | return; | |
5836 | ||
5837 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5838 | ||
5839 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
ff6bc37e | 5840 | trans->chunk_bytes_reserved, NULL); |
4fbcdf66 FM |
5841 | trans->chunk_bytes_reserved = 0; |
5842 | } | |
5843 | ||
d5c12070 MX |
5844 | /* |
5845 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5846 | * root: the root of the parent directory | |
5847 | * rsv: block reservation | |
5848 | * items: the number of items that we need do reservation | |
a5b7f429 | 5849 | * use_global_rsv: allow fallback to the global block reservation |
d5c12070 MX |
5850 | * |
5851 | * This function is used to reserve the space for snapshot/subvolume | |
5852 | * creation and deletion. Those operations are different with the | |
5853 | * common file/directory operations, they change two fs/file trees | |
5854 | * and root tree, the number of items that the qgroup reserves is | |
5855 | * different with the free space reservation. So we can not use | |
01327610 | 5856 | * the space reservation mechanism in start_transaction(). |
d5c12070 MX |
5857 | */ |
5858 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
a5b7f429 | 5859 | struct btrfs_block_rsv *rsv, int items, |
ee3441b4 | 5860 | bool use_global_rsv) |
a22285a6 | 5861 | { |
a5b7f429 | 5862 | u64 qgroup_num_bytes = 0; |
d5c12070 MX |
5863 | u64 num_bytes; |
5864 | int ret; | |
0b246afa JM |
5865 | struct btrfs_fs_info *fs_info = root->fs_info; |
5866 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d5c12070 | 5867 | |
0b246afa | 5868 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
d5c12070 | 5869 | /* One for parent inode, two for dir entries */ |
a5b7f429 LF |
5870 | qgroup_num_bytes = 3 * fs_info->nodesize; |
5871 | ret = btrfs_qgroup_reserve_meta_prealloc(root, | |
5872 | qgroup_num_bytes, true); | |
d5c12070 MX |
5873 | if (ret) |
5874 | return ret; | |
d5c12070 MX |
5875 | } |
5876 | ||
0b246afa JM |
5877 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, items); |
5878 | rsv->space_info = __find_space_info(fs_info, | |
d5c12070 MX |
5879 | BTRFS_BLOCK_GROUP_METADATA); |
5880 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
5881 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
5882 | |
5883 | if (ret == -ENOSPC && use_global_rsv) | |
3a584174 | 5884 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true); |
ee3441b4 | 5885 | |
a5b7f429 LF |
5886 | if (ret && qgroup_num_bytes) |
5887 | btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes); | |
d5c12070 MX |
5888 | |
5889 | return ret; | |
5890 | } | |
5891 | ||
2ff7e61e | 5892 | void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info, |
7775c818 | 5893 | struct btrfs_block_rsv *rsv) |
d5c12070 | 5894 | { |
2ff7e61e | 5895 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
97e728d4 JB |
5896 | } |
5897 | ||
69fe2d75 JB |
5898 | static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info, |
5899 | struct btrfs_inode *inode) | |
9e0baf60 | 5900 | { |
69fe2d75 JB |
5901 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
5902 | u64 reserve_size = 0; | |
ff6bc37e | 5903 | u64 qgroup_rsv_size = 0; |
69fe2d75 JB |
5904 | u64 csum_leaves; |
5905 | unsigned outstanding_extents; | |
9e0baf60 | 5906 | |
69fe2d75 JB |
5907 | lockdep_assert_held(&inode->lock); |
5908 | outstanding_extents = inode->outstanding_extents; | |
5909 | if (outstanding_extents) | |
5910 | reserve_size = btrfs_calc_trans_metadata_size(fs_info, | |
5911 | outstanding_extents + 1); | |
5912 | csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, | |
5913 | inode->csum_bytes); | |
5914 | reserve_size += btrfs_calc_trans_metadata_size(fs_info, | |
5915 | csum_leaves); | |
ff6bc37e QW |
5916 | /* |
5917 | * For qgroup rsv, the calculation is very simple: | |
5918 | * account one nodesize for each outstanding extent | |
5919 | * | |
5920 | * This is overestimating in most cases. | |
5921 | */ | |
5922 | qgroup_rsv_size = outstanding_extents * fs_info->nodesize; | |
9e0baf60 | 5923 | |
69fe2d75 JB |
5924 | spin_lock(&block_rsv->lock); |
5925 | block_rsv->size = reserve_size; | |
ff6bc37e | 5926 | block_rsv->qgroup_rsv_size = qgroup_rsv_size; |
69fe2d75 | 5927 | spin_unlock(&block_rsv->lock); |
0ca1f7ce | 5928 | } |
c146afad | 5929 | |
9f3db423 | 5930 | int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 5931 | { |
3ffbd68c | 5932 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
69fe2d75 | 5933 | unsigned nr_extents; |
08e007d2 | 5934 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 5935 | int ret = 0; |
c64c2bd8 | 5936 | bool delalloc_lock = true; |
6324fbf3 | 5937 | |
c64c2bd8 JB |
5938 | /* If we are a free space inode we need to not flush since we will be in |
5939 | * the middle of a transaction commit. We also don't need the delalloc | |
5940 | * mutex since we won't race with anybody. We need this mostly to make | |
5941 | * lockdep shut its filthy mouth. | |
bac357dc JB |
5942 | * |
5943 | * If we have a transaction open (can happen if we call truncate_block | |
5944 | * from truncate), then we need FLUSH_LIMIT so we don't deadlock. | |
c64c2bd8 JB |
5945 | */ |
5946 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 5947 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 | 5948 | delalloc_lock = false; |
da07d4ab NB |
5949 | } else { |
5950 | if (current->journal_info) | |
5951 | flush = BTRFS_RESERVE_FLUSH_LIMIT; | |
c09544e0 | 5952 | |
da07d4ab NB |
5953 | if (btrfs_transaction_in_commit(fs_info)) |
5954 | schedule_timeout(1); | |
5955 | } | |
ec44a35c | 5956 | |
c64c2bd8 | 5957 | if (delalloc_lock) |
9f3db423 | 5958 | mutex_lock(&inode->delalloc_mutex); |
c64c2bd8 | 5959 | |
0b246afa | 5960 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
69fe2d75 JB |
5961 | |
5962 | /* Add our new extents and calculate the new rsv size. */ | |
9f3db423 | 5963 | spin_lock(&inode->lock); |
69fe2d75 | 5964 | nr_extents = count_max_extents(num_bytes); |
8b62f87b | 5965 | btrfs_mod_outstanding_extents(inode, nr_extents); |
69fe2d75 JB |
5966 | inode->csum_bytes += num_bytes; |
5967 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 5968 | spin_unlock(&inode->lock); |
57a45ced | 5969 | |
69fe2d75 | 5970 | ret = btrfs_inode_rsv_refill(inode, flush); |
43b18595 | 5971 | if (unlikely(ret)) |
88e081bf | 5972 | goto out_fail; |
25179201 | 5973 | |
c64c2bd8 | 5974 | if (delalloc_lock) |
9f3db423 | 5975 | mutex_unlock(&inode->delalloc_mutex); |
0ca1f7ce | 5976 | return 0; |
88e081bf WS |
5977 | |
5978 | out_fail: | |
9f3db423 | 5979 | spin_lock(&inode->lock); |
8b62f87b JB |
5980 | nr_extents = count_max_extents(num_bytes); |
5981 | btrfs_mod_outstanding_extents(inode, -nr_extents); | |
69fe2d75 JB |
5982 | inode->csum_bytes -= num_bytes; |
5983 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 5984 | spin_unlock(&inode->lock); |
88e081bf | 5985 | |
43b18595 | 5986 | btrfs_inode_rsv_release(inode, true); |
88e081bf | 5987 | if (delalloc_lock) |
9f3db423 | 5988 | mutex_unlock(&inode->delalloc_mutex); |
88e081bf | 5989 | return ret; |
0ca1f7ce YZ |
5990 | } |
5991 | ||
7709cde3 JB |
5992 | /** |
5993 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
8b62f87b JB |
5994 | * @inode: the inode to release the reservation for. |
5995 | * @num_bytes: the number of bytes we are releasing. | |
43b18595 | 5996 | * @qgroup_free: free qgroup reservation or convert it to per-trans reservation |
7709cde3 JB |
5997 | * |
5998 | * This will release the metadata reservation for an inode. This can be called | |
5999 | * once we complete IO for a given set of bytes to release their metadata | |
8b62f87b | 6000 | * reservations, or on error for the same reason. |
7709cde3 | 6001 | */ |
43b18595 QW |
6002 | void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes, |
6003 | bool qgroup_free) | |
0ca1f7ce | 6004 | { |
3ffbd68c | 6005 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
0ca1f7ce | 6006 | |
0b246afa | 6007 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
691fa059 | 6008 | spin_lock(&inode->lock); |
69fe2d75 JB |
6009 | inode->csum_bytes -= num_bytes; |
6010 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
691fa059 | 6011 | spin_unlock(&inode->lock); |
0ca1f7ce | 6012 | |
0b246afa | 6013 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
6014 | return; |
6015 | ||
43b18595 | 6016 | btrfs_inode_rsv_release(inode, qgroup_free); |
0ca1f7ce YZ |
6017 | } |
6018 | ||
8b62f87b JB |
6019 | /** |
6020 | * btrfs_delalloc_release_extents - release our outstanding_extents | |
6021 | * @inode: the inode to balance the reservation for. | |
6022 | * @num_bytes: the number of bytes we originally reserved with | |
43b18595 | 6023 | * @qgroup_free: do we need to free qgroup meta reservation or convert them. |
8b62f87b JB |
6024 | * |
6025 | * When we reserve space we increase outstanding_extents for the extents we may | |
6026 | * add. Once we've set the range as delalloc or created our ordered extents we | |
6027 | * have outstanding_extents to track the real usage, so we use this to free our | |
6028 | * temporarily tracked outstanding_extents. This _must_ be used in conjunction | |
6029 | * with btrfs_delalloc_reserve_metadata. | |
6030 | */ | |
43b18595 QW |
6031 | void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes, |
6032 | bool qgroup_free) | |
8b62f87b | 6033 | { |
3ffbd68c | 6034 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
8b62f87b | 6035 | unsigned num_extents; |
8b62f87b JB |
6036 | |
6037 | spin_lock(&inode->lock); | |
6038 | num_extents = count_max_extents(num_bytes); | |
6039 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
69fe2d75 | 6040 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); |
8b62f87b JB |
6041 | spin_unlock(&inode->lock); |
6042 | ||
8b62f87b JB |
6043 | if (btrfs_is_testing(fs_info)) |
6044 | return; | |
6045 | ||
43b18595 | 6046 | btrfs_inode_rsv_release(inode, qgroup_free); |
8b62f87b JB |
6047 | } |
6048 | ||
1ada3a62 | 6049 | /** |
7cf5b976 | 6050 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
6051 | * delalloc |
6052 | * @inode: inode we're writing to | |
6053 | * @start: start range we are writing to | |
6054 | * @len: how long the range we are writing to | |
364ecf36 QW |
6055 | * @reserved: mandatory parameter, record actually reserved qgroup ranges of |
6056 | * current reservation. | |
1ada3a62 | 6057 | * |
1ada3a62 QW |
6058 | * This will do the following things |
6059 | * | |
6060 | * o reserve space in data space info for num bytes | |
6061 | * and reserve precious corresponding qgroup space | |
6062 | * (Done in check_data_free_space) | |
6063 | * | |
6064 | * o reserve space for metadata space, based on the number of outstanding | |
6065 | * extents and how much csums will be needed | |
6066 | * also reserve metadata space in a per root over-reserve method. | |
6067 | * o add to the inodes->delalloc_bytes | |
6068 | * o add it to the fs_info's delalloc inodes list. | |
6069 | * (Above 3 all done in delalloc_reserve_metadata) | |
6070 | * | |
6071 | * Return 0 for success | |
6072 | * Return <0 for error(-ENOSPC or -EQUOT) | |
6073 | */ | |
364ecf36 QW |
6074 | int btrfs_delalloc_reserve_space(struct inode *inode, |
6075 | struct extent_changeset **reserved, u64 start, u64 len) | |
1ada3a62 QW |
6076 | { |
6077 | int ret; | |
6078 | ||
364ecf36 | 6079 | ret = btrfs_check_data_free_space(inode, reserved, start, len); |
1ada3a62 QW |
6080 | if (ret < 0) |
6081 | return ret; | |
9f3db423 | 6082 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len); |
1ada3a62 | 6083 | if (ret < 0) |
bc42bda2 | 6084 | btrfs_free_reserved_data_space(inode, *reserved, start, len); |
1ada3a62 QW |
6085 | return ret; |
6086 | } | |
6087 | ||
7709cde3 | 6088 | /** |
7cf5b976 | 6089 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
6090 | * @inode: inode we're releasing space for |
6091 | * @start: start position of the space already reserved | |
6092 | * @len: the len of the space already reserved | |
8b62f87b | 6093 | * @release_bytes: the len of the space we consumed or didn't use |
1ada3a62 QW |
6094 | * |
6095 | * This function will release the metadata space that was not used and will | |
6096 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
6097 | * list if there are no delalloc bytes left. | |
6098 | * Also it will handle the qgroup reserved space. | |
6099 | */ | |
bc42bda2 | 6100 | void btrfs_delalloc_release_space(struct inode *inode, |
8b62f87b | 6101 | struct extent_changeset *reserved, |
43b18595 | 6102 | u64 start, u64 len, bool qgroup_free) |
1ada3a62 | 6103 | { |
43b18595 | 6104 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free); |
bc42bda2 | 6105 | btrfs_free_reserved_data_space(inode, reserved, start, len); |
6324fbf3 CM |
6106 | } |
6107 | ||
ce93ec54 | 6108 | static int update_block_group(struct btrfs_trans_handle *trans, |
6202df69 | 6109 | struct btrfs_fs_info *info, u64 bytenr, |
ce93ec54 | 6110 | u64 num_bytes, int alloc) |
9078a3e1 | 6111 | { |
0af3d00b | 6112 | struct btrfs_block_group_cache *cache = NULL; |
db94535d | 6113 | u64 total = num_bytes; |
9078a3e1 | 6114 | u64 old_val; |
db94535d | 6115 | u64 byte_in_group; |
0af3d00b | 6116 | int factor; |
3e1ad54f | 6117 | |
5d4f98a2 | 6118 | /* block accounting for super block */ |
eb73c1b7 | 6119 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 6120 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
6121 | if (alloc) |
6122 | old_val += num_bytes; | |
6123 | else | |
6124 | old_val -= num_bytes; | |
6c41761f | 6125 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 6126 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 6127 | |
d397712b | 6128 | while (total) { |
db94535d | 6129 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 6130 | if (!cache) |
79787eaa | 6131 | return -ENOENT; |
46df06b8 DS |
6132 | factor = btrfs_bg_type_to_factor(cache->flags); |
6133 | ||
9d66e233 JB |
6134 | /* |
6135 | * If this block group has free space cache written out, we | |
6136 | * need to make sure to load it if we are removing space. This | |
6137 | * is because we need the unpinning stage to actually add the | |
6138 | * space back to the block group, otherwise we will leak space. | |
6139 | */ | |
6140 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 6141 | cache_block_group(cache, 1); |
0af3d00b | 6142 | |
db94535d CM |
6143 | byte_in_group = bytenr - cache->key.objectid; |
6144 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 6145 | |
25179201 | 6146 | spin_lock(&cache->space_info->lock); |
c286ac48 | 6147 | spin_lock(&cache->lock); |
0af3d00b | 6148 | |
6202df69 | 6149 | if (btrfs_test_opt(info, SPACE_CACHE) && |
0af3d00b JB |
6150 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
6151 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
6152 | ||
9078a3e1 | 6153 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 6154 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 6155 | if (alloc) { |
db94535d | 6156 | old_val += num_bytes; |
11833d66 YZ |
6157 | btrfs_set_block_group_used(&cache->item, old_val); |
6158 | cache->reserved -= num_bytes; | |
11833d66 | 6159 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
6160 | cache->space_info->bytes_used += num_bytes; |
6161 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 6162 | spin_unlock(&cache->lock); |
25179201 | 6163 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 6164 | } else { |
db94535d | 6165 | old_val -= num_bytes; |
ae0ab003 FM |
6166 | btrfs_set_block_group_used(&cache->item, old_val); |
6167 | cache->pinned += num_bytes; | |
6168 | cache->space_info->bytes_pinned += num_bytes; | |
6169 | cache->space_info->bytes_used -= num_bytes; | |
6170 | cache->space_info->disk_used -= num_bytes * factor; | |
6171 | spin_unlock(&cache->lock); | |
6172 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 6173 | |
0b246afa | 6174 | trace_btrfs_space_reservation(info, "pinned", |
c51e7bb1 JB |
6175 | cache->space_info->flags, |
6176 | num_bytes, 1); | |
dec59fa3 EL |
6177 | percpu_counter_add_batch(&cache->space_info->total_bytes_pinned, |
6178 | num_bytes, | |
6179 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
ae0ab003 FM |
6180 | set_extent_dirty(info->pinned_extents, |
6181 | bytenr, bytenr + num_bytes - 1, | |
6182 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 6183 | } |
1bbc621e CM |
6184 | |
6185 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
6186 | if (list_empty(&cache->dirty_list)) { | |
6187 | list_add_tail(&cache->dirty_list, | |
6188 | &trans->transaction->dirty_bgs); | |
bece2e82 | 6189 | trans->transaction->num_dirty_bgs++; |
1bbc621e CM |
6190 | btrfs_get_block_group(cache); |
6191 | } | |
6192 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
6193 | ||
036a9348 FM |
6194 | /* |
6195 | * No longer have used bytes in this block group, queue it for | |
6196 | * deletion. We do this after adding the block group to the | |
6197 | * dirty list to avoid races between cleaner kthread and space | |
6198 | * cache writeout. | |
6199 | */ | |
031f24da QW |
6200 | if (!alloc && old_val == 0) |
6201 | btrfs_mark_bg_unused(cache); | |
036a9348 | 6202 | |
fa9c0d79 | 6203 | btrfs_put_block_group(cache); |
db94535d CM |
6204 | total -= num_bytes; |
6205 | bytenr += num_bytes; | |
9078a3e1 CM |
6206 | } |
6207 | return 0; | |
6208 | } | |
6324fbf3 | 6209 | |
2ff7e61e | 6210 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 6211 | { |
0f9dd46c | 6212 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 6213 | u64 bytenr; |
0f9dd46c | 6214 | |
0b246afa JM |
6215 | spin_lock(&fs_info->block_group_cache_lock); |
6216 | bytenr = fs_info->first_logical_byte; | |
6217 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
6218 | |
6219 | if (bytenr < (u64)-1) | |
6220 | return bytenr; | |
6221 | ||
0b246afa | 6222 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 6223 | if (!cache) |
a061fc8d | 6224 | return 0; |
0f9dd46c | 6225 | |
d2fb3437 | 6226 | bytenr = cache->key.objectid; |
fa9c0d79 | 6227 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
6228 | |
6229 | return bytenr; | |
a061fc8d CM |
6230 | } |
6231 | ||
2ff7e61e | 6232 | static int pin_down_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6233 | struct btrfs_block_group_cache *cache, |
6234 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 6235 | { |
11833d66 YZ |
6236 | spin_lock(&cache->space_info->lock); |
6237 | spin_lock(&cache->lock); | |
6238 | cache->pinned += num_bytes; | |
6239 | cache->space_info->bytes_pinned += num_bytes; | |
6240 | if (reserved) { | |
6241 | cache->reserved -= num_bytes; | |
6242 | cache->space_info->bytes_reserved -= num_bytes; | |
6243 | } | |
6244 | spin_unlock(&cache->lock); | |
6245 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 6246 | |
0b246afa | 6247 | trace_btrfs_space_reservation(fs_info, "pinned", |
c51e7bb1 | 6248 | cache->space_info->flags, num_bytes, 1); |
dec59fa3 EL |
6249 | percpu_counter_add_batch(&cache->space_info->total_bytes_pinned, |
6250 | num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
0b246afa | 6251 | set_extent_dirty(fs_info->pinned_extents, bytenr, |
f0486c68 YZ |
6252 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
6253 | return 0; | |
6254 | } | |
68b38550 | 6255 | |
f0486c68 YZ |
6256 | /* |
6257 | * this function must be called within transaction | |
6258 | */ | |
2ff7e61e | 6259 | int btrfs_pin_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6260 | u64 bytenr, u64 num_bytes, int reserved) |
6261 | { | |
6262 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6263 | |
0b246afa | 6264 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
79787eaa | 6265 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 6266 | |
2ff7e61e | 6267 | pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
6268 | |
6269 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6270 | return 0; |
6271 | } | |
6272 | ||
f0486c68 | 6273 | /* |
e688b725 CM |
6274 | * this function must be called within transaction |
6275 | */ | |
2ff7e61e | 6276 | int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info, |
e688b725 CM |
6277 | u64 bytenr, u64 num_bytes) |
6278 | { | |
6279 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6280 | int ret; |
e688b725 | 6281 | |
0b246afa | 6282 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
b50c6e25 JB |
6283 | if (!cache) |
6284 | return -EINVAL; | |
e688b725 CM |
6285 | |
6286 | /* | |
6287 | * pull in the free space cache (if any) so that our pin | |
6288 | * removes the free space from the cache. We have load_only set | |
6289 | * to one because the slow code to read in the free extents does check | |
6290 | * the pinned extents. | |
6291 | */ | |
f6373bf3 | 6292 | cache_block_group(cache, 1); |
e688b725 | 6293 | |
2ff7e61e | 6294 | pin_down_extent(fs_info, cache, bytenr, num_bytes, 0); |
e688b725 CM |
6295 | |
6296 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6297 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6298 | btrfs_put_block_group(cache); |
b50c6e25 | 6299 | return ret; |
e688b725 CM |
6300 | } |
6301 | ||
2ff7e61e JM |
6302 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
6303 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
6304 | { |
6305 | int ret; | |
6306 | struct btrfs_block_group_cache *block_group; | |
6307 | struct btrfs_caching_control *caching_ctl; | |
6308 | ||
0b246afa | 6309 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
6310 | if (!block_group) |
6311 | return -EINVAL; | |
6312 | ||
6313 | cache_block_group(block_group, 0); | |
6314 | caching_ctl = get_caching_control(block_group); | |
6315 | ||
6316 | if (!caching_ctl) { | |
6317 | /* Logic error */ | |
6318 | BUG_ON(!block_group_cache_done(block_group)); | |
6319 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6320 | } else { | |
6321 | mutex_lock(&caching_ctl->mutex); | |
6322 | ||
6323 | if (start >= caching_ctl->progress) { | |
2ff7e61e | 6324 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6325 | } else if (start + num_bytes <= caching_ctl->progress) { |
6326 | ret = btrfs_remove_free_space(block_group, | |
6327 | start, num_bytes); | |
6328 | } else { | |
6329 | num_bytes = caching_ctl->progress - start; | |
6330 | ret = btrfs_remove_free_space(block_group, | |
6331 | start, num_bytes); | |
6332 | if (ret) | |
6333 | goto out_lock; | |
6334 | ||
6335 | num_bytes = (start + num_bytes) - | |
6336 | caching_ctl->progress; | |
6337 | start = caching_ctl->progress; | |
2ff7e61e | 6338 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6339 | } |
6340 | out_lock: | |
6341 | mutex_unlock(&caching_ctl->mutex); | |
6342 | put_caching_control(caching_ctl); | |
6343 | } | |
6344 | btrfs_put_block_group(block_group); | |
6345 | return ret; | |
6346 | } | |
6347 | ||
2ff7e61e | 6348 | int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info, |
8c2a1a30 JB |
6349 | struct extent_buffer *eb) |
6350 | { | |
6351 | struct btrfs_file_extent_item *item; | |
6352 | struct btrfs_key key; | |
6353 | int found_type; | |
6354 | int i; | |
b89311ef | 6355 | int ret = 0; |
8c2a1a30 | 6356 | |
2ff7e61e | 6357 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
6358 | return 0; |
6359 | ||
6360 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6361 | btrfs_item_key_to_cpu(eb, &key, i); | |
6362 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6363 | continue; | |
6364 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6365 | found_type = btrfs_file_extent_type(eb, item); | |
6366 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6367 | continue; | |
6368 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6369 | continue; | |
6370 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6371 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
b89311ef GJ |
6372 | ret = __exclude_logged_extent(fs_info, key.objectid, key.offset); |
6373 | if (ret) | |
6374 | break; | |
8c2a1a30 JB |
6375 | } |
6376 | ||
b89311ef | 6377 | return ret; |
8c2a1a30 JB |
6378 | } |
6379 | ||
9cfa3e34 FM |
6380 | static void |
6381 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6382 | { | |
6383 | atomic_inc(&bg->reservations); | |
6384 | } | |
6385 | ||
6386 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
6387 | const u64 start) | |
6388 | { | |
6389 | struct btrfs_block_group_cache *bg; | |
6390 | ||
6391 | bg = btrfs_lookup_block_group(fs_info, start); | |
6392 | ASSERT(bg); | |
6393 | if (atomic_dec_and_test(&bg->reservations)) | |
4625956a | 6394 | wake_up_var(&bg->reservations); |
9cfa3e34 FM |
6395 | btrfs_put_block_group(bg); |
6396 | } | |
6397 | ||
9cfa3e34 FM |
6398 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) |
6399 | { | |
6400 | struct btrfs_space_info *space_info = bg->space_info; | |
6401 | ||
6402 | ASSERT(bg->ro); | |
6403 | ||
6404 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
6405 | return; | |
6406 | ||
6407 | /* | |
6408 | * Our block group is read only but before we set it to read only, | |
6409 | * some task might have had allocated an extent from it already, but it | |
6410 | * has not yet created a respective ordered extent (and added it to a | |
6411 | * root's list of ordered extents). | |
6412 | * Therefore wait for any task currently allocating extents, since the | |
6413 | * block group's reservations counter is incremented while a read lock | |
6414 | * on the groups' semaphore is held and decremented after releasing | |
6415 | * the read access on that semaphore and creating the ordered extent. | |
6416 | */ | |
6417 | down_write(&space_info->groups_sem); | |
6418 | up_write(&space_info->groups_sem); | |
6419 | ||
4625956a | 6420 | wait_var_event(&bg->reservations, !atomic_read(&bg->reservations)); |
9cfa3e34 FM |
6421 | } |
6422 | ||
fb25e914 | 6423 | /** |
4824f1f4 | 6424 | * btrfs_add_reserved_bytes - update the block_group and space info counters |
fb25e914 | 6425 | * @cache: The cache we are manipulating |
18513091 WX |
6426 | * @ram_bytes: The number of bytes of file content, and will be same to |
6427 | * @num_bytes except for the compress path. | |
fb25e914 | 6428 | * @num_bytes: The number of bytes in question |
e570fd27 | 6429 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 | 6430 | * |
745699ef XW |
6431 | * This is called by the allocator when it reserves space. If this is a |
6432 | * reservation and the block group has become read only we cannot make the | |
6433 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
f0486c68 | 6434 | */ |
4824f1f4 | 6435 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 6436 | u64 ram_bytes, u64 num_bytes, int delalloc) |
11833d66 | 6437 | { |
fb25e914 | 6438 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6439 | int ret = 0; |
79787eaa | 6440 | |
fb25e914 JB |
6441 | spin_lock(&space_info->lock); |
6442 | spin_lock(&cache->lock); | |
4824f1f4 WX |
6443 | if (cache->ro) { |
6444 | ret = -EAGAIN; | |
fb25e914 | 6445 | } else { |
4824f1f4 WX |
6446 | cache->reserved += num_bytes; |
6447 | space_info->bytes_reserved += num_bytes; | |
9f9b8e8d | 6448 | update_bytes_may_use(space_info, -ram_bytes); |
e570fd27 | 6449 | if (delalloc) |
4824f1f4 | 6450 | cache->delalloc_bytes += num_bytes; |
324ae4df | 6451 | } |
fb25e914 JB |
6452 | spin_unlock(&cache->lock); |
6453 | spin_unlock(&space_info->lock); | |
f0486c68 | 6454 | return ret; |
324ae4df | 6455 | } |
9078a3e1 | 6456 | |
4824f1f4 WX |
6457 | /** |
6458 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
6459 | * @cache: The cache we are manipulating | |
6460 | * @num_bytes: The number of bytes in question | |
6461 | * @delalloc: The blocks are allocated for the delalloc write | |
6462 | * | |
6463 | * This is called by somebody who is freeing space that was never actually used | |
6464 | * on disk. For example if you reserve some space for a new leaf in transaction | |
6465 | * A and before transaction A commits you free that leaf, you call this with | |
6466 | * reserve set to 0 in order to clear the reservation. | |
6467 | */ | |
6468 | ||
556f3ca8 | 6469 | static void btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, |
6470 | u64 num_bytes, int delalloc) | |
4824f1f4 WX |
6471 | { |
6472 | struct btrfs_space_info *space_info = cache->space_info; | |
4824f1f4 WX |
6473 | |
6474 | spin_lock(&space_info->lock); | |
6475 | spin_lock(&cache->lock); | |
6476 | if (cache->ro) | |
6477 | space_info->bytes_readonly += num_bytes; | |
6478 | cache->reserved -= num_bytes; | |
6479 | space_info->bytes_reserved -= num_bytes; | |
21a94f7a | 6480 | space_info->max_extent_size = 0; |
4824f1f4 WX |
6481 | |
6482 | if (delalloc) | |
6483 | cache->delalloc_bytes -= num_bytes; | |
6484 | spin_unlock(&cache->lock); | |
6485 | spin_unlock(&space_info->lock); | |
4824f1f4 | 6486 | } |
8b74c03e | 6487 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 6488 | { |
11833d66 YZ |
6489 | struct btrfs_caching_control *next; |
6490 | struct btrfs_caching_control *caching_ctl; | |
6491 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6492 | |
9e351cc8 | 6493 | down_write(&fs_info->commit_root_sem); |
25179201 | 6494 | |
11833d66 YZ |
6495 | list_for_each_entry_safe(caching_ctl, next, |
6496 | &fs_info->caching_block_groups, list) { | |
6497 | cache = caching_ctl->block_group; | |
6498 | if (block_group_cache_done(cache)) { | |
6499 | cache->last_byte_to_unpin = (u64)-1; | |
6500 | list_del_init(&caching_ctl->list); | |
6501 | put_caching_control(caching_ctl); | |
e8569813 | 6502 | } else { |
11833d66 | 6503 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6504 | } |
e8569813 | 6505 | } |
11833d66 YZ |
6506 | |
6507 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6508 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6509 | else | |
6510 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6511 | ||
9e351cc8 | 6512 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6513 | |
6514 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6515 | } |
6516 | ||
c759c4e1 JB |
6517 | /* |
6518 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6519 | * what it should be based on the mount options. | |
6520 | */ | |
6521 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
6522 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
6523 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
6524 | { |
6525 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
6526 | |
6527 | *empty_cluster = 0; | |
6528 | if (btrfs_mixed_space_info(space_info)) | |
6529 | return ret; | |
6530 | ||
c759c4e1 | 6531 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 6532 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
6533 | if (btrfs_test_opt(fs_info, SSD)) |
6534 | *empty_cluster = SZ_2M; | |
6535 | else | |
ee22184b | 6536 | *empty_cluster = SZ_64K; |
583b7231 HK |
6537 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
6538 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
6539 | *empty_cluster = SZ_2M; | |
0b246afa | 6540 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
6541 | } |
6542 | ||
6543 | return ret; | |
6544 | } | |
6545 | ||
2ff7e61e JM |
6546 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
6547 | u64 start, u64 end, | |
678886bd | 6548 | const bool return_free_space) |
ccd467d6 | 6549 | { |
11833d66 | 6550 | struct btrfs_block_group_cache *cache = NULL; |
7b398f8e JB |
6551 | struct btrfs_space_info *space_info; |
6552 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6553 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6554 | u64 len; |
c759c4e1 JB |
6555 | u64 total_unpinned = 0; |
6556 | u64 empty_cluster = 0; | |
7b398f8e | 6557 | bool readonly; |
ccd467d6 | 6558 | |
11833d66 | 6559 | while (start <= end) { |
7b398f8e | 6560 | readonly = false; |
11833d66 YZ |
6561 | if (!cache || |
6562 | start >= cache->key.objectid + cache->key.offset) { | |
6563 | if (cache) | |
6564 | btrfs_put_block_group(cache); | |
c759c4e1 | 6565 | total_unpinned = 0; |
11833d66 | 6566 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6567 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 6568 | |
2ff7e61e | 6569 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
6570 | cache->space_info, |
6571 | &empty_cluster); | |
6572 | empty_cluster <<= 1; | |
11833d66 YZ |
6573 | } |
6574 | ||
6575 | len = cache->key.objectid + cache->key.offset - start; | |
6576 | len = min(len, end + 1 - start); | |
6577 | ||
6578 | if (start < cache->last_byte_to_unpin) { | |
6579 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6580 | if (return_free_space) |
6581 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6582 | } |
6583 | ||
f0486c68 | 6584 | start += len; |
c759c4e1 | 6585 | total_unpinned += len; |
7b398f8e | 6586 | space_info = cache->space_info; |
f0486c68 | 6587 | |
c759c4e1 JB |
6588 | /* |
6589 | * If this space cluster has been marked as fragmented and we've | |
6590 | * unpinned enough in this block group to potentially allow a | |
6591 | * cluster to be created inside of it go ahead and clear the | |
6592 | * fragmented check. | |
6593 | */ | |
6594 | if (cluster && cluster->fragmented && | |
6595 | total_unpinned > empty_cluster) { | |
6596 | spin_lock(&cluster->lock); | |
6597 | cluster->fragmented = 0; | |
6598 | spin_unlock(&cluster->lock); | |
6599 | } | |
6600 | ||
7b398f8e | 6601 | spin_lock(&space_info->lock); |
11833d66 YZ |
6602 | spin_lock(&cache->lock); |
6603 | cache->pinned -= len; | |
7b398f8e | 6604 | space_info->bytes_pinned -= len; |
c51e7bb1 JB |
6605 | |
6606 | trace_btrfs_space_reservation(fs_info, "pinned", | |
6607 | space_info->flags, len, 0); | |
4f4db217 | 6608 | space_info->max_extent_size = 0; |
dec59fa3 EL |
6609 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
6610 | -len, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
7b398f8e JB |
6611 | if (cache->ro) { |
6612 | space_info->bytes_readonly += len; | |
6613 | readonly = true; | |
6614 | } | |
11833d66 | 6615 | spin_unlock(&cache->lock); |
957780eb JB |
6616 | if (!readonly && return_free_space && |
6617 | global_rsv->space_info == space_info) { | |
6618 | u64 to_add = len; | |
92ac58ec | 6619 | |
7b398f8e JB |
6620 | spin_lock(&global_rsv->lock); |
6621 | if (!global_rsv->full) { | |
957780eb JB |
6622 | to_add = min(len, global_rsv->size - |
6623 | global_rsv->reserved); | |
6624 | global_rsv->reserved += to_add; | |
9f9b8e8d | 6625 | update_bytes_may_use(space_info, to_add); |
7b398f8e JB |
6626 | if (global_rsv->reserved >= global_rsv->size) |
6627 | global_rsv->full = 1; | |
957780eb JB |
6628 | trace_btrfs_space_reservation(fs_info, |
6629 | "space_info", | |
6630 | space_info->flags, | |
6631 | to_add, 1); | |
6632 | len -= to_add; | |
7b398f8e JB |
6633 | } |
6634 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
6635 | /* Add to any tickets we may have */ |
6636 | if (len) | |
6637 | space_info_add_new_bytes(fs_info, space_info, | |
6638 | len); | |
7b398f8e JB |
6639 | } |
6640 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6641 | } |
11833d66 YZ |
6642 | |
6643 | if (cache) | |
6644 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6645 | return 0; |
6646 | } | |
6647 | ||
5ead2dd0 | 6648 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 6649 | { |
5ead2dd0 | 6650 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e33e17ee JM |
6651 | struct btrfs_block_group_cache *block_group, *tmp; |
6652 | struct list_head *deleted_bgs; | |
11833d66 | 6653 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6654 | u64 start; |
6655 | u64 end; | |
a28ec197 | 6656 | int ret; |
a28ec197 | 6657 | |
11833d66 YZ |
6658 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6659 | unpin = &fs_info->freed_extents[1]; | |
6660 | else | |
6661 | unpin = &fs_info->freed_extents[0]; | |
6662 | ||
e33e17ee | 6663 | while (!trans->aborted) { |
d4b450cd | 6664 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6665 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 6666 | EXTENT_DIRTY, NULL); |
d4b450cd FM |
6667 | if (ret) { |
6668 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6669 | break; |
d4b450cd | 6670 | } |
1f3c79a2 | 6671 | |
0b246afa | 6672 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 6673 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 6674 | end + 1 - start, NULL); |
1f3c79a2 | 6675 | |
af6f8f60 | 6676 | clear_extent_dirty(unpin, start, end); |
2ff7e61e | 6677 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 6678 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
b9473439 | 6679 | cond_resched(); |
a28ec197 | 6680 | } |
817d52f8 | 6681 | |
e33e17ee JM |
6682 | /* |
6683 | * Transaction is finished. We don't need the lock anymore. We | |
6684 | * do need to clean up the block groups in case of a transaction | |
6685 | * abort. | |
6686 | */ | |
6687 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6688 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6689 | u64 trimmed = 0; | |
6690 | ||
6691 | ret = -EROFS; | |
6692 | if (!trans->aborted) | |
2ff7e61e | 6693 | ret = btrfs_discard_extent(fs_info, |
e33e17ee JM |
6694 | block_group->key.objectid, |
6695 | block_group->key.offset, | |
6696 | &trimmed); | |
6697 | ||
6698 | list_del_init(&block_group->bg_list); | |
6699 | btrfs_put_block_group_trimming(block_group); | |
6700 | btrfs_put_block_group(block_group); | |
6701 | ||
6702 | if (ret) { | |
6703 | const char *errstr = btrfs_decode_error(ret); | |
6704 | btrfs_warn(fs_info, | |
913e1535 | 6705 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
6706 | ret, errstr); |
6707 | } | |
6708 | } | |
6709 | ||
e20d96d6 CM |
6710 | return 0; |
6711 | } | |
6712 | ||
5d4f98a2 | 6713 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
6714 | struct btrfs_delayed_ref_node *node, u64 parent, |
6715 | u64 root_objectid, u64 owner_objectid, | |
6716 | u64 owner_offset, int refs_to_drop, | |
6717 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 6718 | { |
e72cb923 | 6719 | struct btrfs_fs_info *info = trans->fs_info; |
e2fa7227 | 6720 | struct btrfs_key key; |
5d4f98a2 | 6721 | struct btrfs_path *path; |
1261ec42 | 6722 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 6723 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6724 | struct btrfs_extent_item *ei; |
6725 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6726 | int ret; |
5d4f98a2 | 6727 | int is_data; |
952fccac CM |
6728 | int extent_slot = 0; |
6729 | int found_extent = 0; | |
6730 | int num_to_del = 1; | |
5d4f98a2 YZ |
6731 | u32 item_size; |
6732 | u64 refs; | |
c682f9b3 QW |
6733 | u64 bytenr = node->bytenr; |
6734 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6735 | int last_ref = 0; |
0b246afa | 6736 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 6737 | |
5caf2a00 | 6738 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6739 | if (!path) |
6740 | return -ENOMEM; | |
5f26f772 | 6741 | |
e4058b54 | 6742 | path->reada = READA_FORWARD; |
b9473439 | 6743 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6744 | |
6745 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6746 | BUG_ON(!is_data && refs_to_drop != 1); | |
6747 | ||
3173a18f | 6748 | if (is_data) |
897ca819 | 6749 | skinny_metadata = false; |
3173a18f | 6750 | |
fbe4801b NB |
6751 | ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes, |
6752 | parent, root_objectid, owner_objectid, | |
5d4f98a2 | 6753 | owner_offset); |
7bb86316 | 6754 | if (ret == 0) { |
952fccac | 6755 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6756 | while (extent_slot >= 0) { |
6757 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6758 | extent_slot); |
5d4f98a2 | 6759 | if (key.objectid != bytenr) |
952fccac | 6760 | break; |
5d4f98a2 YZ |
6761 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6762 | key.offset == num_bytes) { | |
952fccac CM |
6763 | found_extent = 1; |
6764 | break; | |
6765 | } | |
3173a18f JB |
6766 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
6767 | key.offset == owner_objectid) { | |
6768 | found_extent = 1; | |
6769 | break; | |
6770 | } | |
952fccac CM |
6771 | if (path->slots[0] - extent_slot > 5) |
6772 | break; | |
5d4f98a2 | 6773 | extent_slot--; |
952fccac | 6774 | } |
a79865c6 | 6775 | |
31840ae1 | 6776 | if (!found_extent) { |
5d4f98a2 | 6777 | BUG_ON(iref); |
87cc7a8a | 6778 | ret = remove_extent_backref(trans, path, NULL, |
87bde3cd | 6779 | refs_to_drop, |
fcebe456 | 6780 | is_data, &last_ref); |
005d6427 | 6781 | if (ret) { |
66642832 | 6782 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6783 | goto out; |
6784 | } | |
b3b4aa74 | 6785 | btrfs_release_path(path); |
b9473439 | 6786 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6787 | |
6788 | key.objectid = bytenr; | |
6789 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6790 | key.offset = num_bytes; | |
6791 | ||
3173a18f JB |
6792 | if (!is_data && skinny_metadata) { |
6793 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6794 | key.offset = owner_objectid; | |
6795 | } | |
6796 | ||
31840ae1 ZY |
6797 | ret = btrfs_search_slot(trans, extent_root, |
6798 | &key, path, -1, 1); | |
3173a18f JB |
6799 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
6800 | /* | |
6801 | * Couldn't find our skinny metadata item, | |
6802 | * see if we have ye olde extent item. | |
6803 | */ | |
6804 | path->slots[0]--; | |
6805 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6806 | path->slots[0]); | |
6807 | if (key.objectid == bytenr && | |
6808 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6809 | key.offset == num_bytes) | |
6810 | ret = 0; | |
6811 | } | |
6812 | ||
6813 | if (ret > 0 && skinny_metadata) { | |
6814 | skinny_metadata = false; | |
9ce49a0b | 6815 | key.objectid = bytenr; |
3173a18f JB |
6816 | key.type = BTRFS_EXTENT_ITEM_KEY; |
6817 | key.offset = num_bytes; | |
6818 | btrfs_release_path(path); | |
6819 | ret = btrfs_search_slot(trans, extent_root, | |
6820 | &key, path, -1, 1); | |
6821 | } | |
6822 | ||
f3465ca4 | 6823 | if (ret) { |
5d163e0e JM |
6824 | btrfs_err(info, |
6825 | "umm, got %d back from search, was looking for %llu", | |
6826 | ret, bytenr); | |
b783e62d | 6827 | if (ret > 0) |
a4f78750 | 6828 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 6829 | } |
005d6427 | 6830 | if (ret < 0) { |
66642832 | 6831 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6832 | goto out; |
6833 | } | |
31840ae1 ZY |
6834 | extent_slot = path->slots[0]; |
6835 | } | |
fae7f21c | 6836 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 6837 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
6838 | btrfs_err(info, |
6839 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
6840 | bytenr, parent, root_objectid, owner_objectid, |
6841 | owner_offset); | |
66642832 | 6842 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 6843 | goto out; |
79787eaa | 6844 | } else { |
66642832 | 6845 | btrfs_abort_transaction(trans, ret); |
005d6427 | 6846 | goto out; |
7bb86316 | 6847 | } |
5f39d397 CM |
6848 | |
6849 | leaf = path->nodes[0]; | |
5d4f98a2 | 6850 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6d8ff4e4 | 6851 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
6852 | ret = -EINVAL; |
6853 | btrfs_print_v0_err(info); | |
6854 | btrfs_abort_transaction(trans, ret); | |
6855 | goto out; | |
6856 | } | |
952fccac | 6857 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 6858 | struct btrfs_extent_item); |
3173a18f JB |
6859 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
6860 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
6861 | struct btrfs_tree_block_info *bi; |
6862 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
6863 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
6864 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
6865 | } | |
56bec294 | 6866 | |
5d4f98a2 | 6867 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 6868 | if (refs < refs_to_drop) { |
5d163e0e JM |
6869 | btrfs_err(info, |
6870 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
6871 | refs_to_drop, refs, bytenr); | |
32b02538 | 6872 | ret = -EINVAL; |
66642832 | 6873 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
6874 | goto out; |
6875 | } | |
56bec294 | 6876 | refs -= refs_to_drop; |
5f39d397 | 6877 | |
5d4f98a2 YZ |
6878 | if (refs > 0) { |
6879 | if (extent_op) | |
6880 | __run_delayed_extent_op(extent_op, leaf, ei); | |
6881 | /* | |
6882 | * In the case of inline back ref, reference count will | |
6883 | * be updated by remove_extent_backref | |
952fccac | 6884 | */ |
5d4f98a2 YZ |
6885 | if (iref) { |
6886 | BUG_ON(!found_extent); | |
6887 | } else { | |
6888 | btrfs_set_extent_refs(leaf, ei, refs); | |
6889 | btrfs_mark_buffer_dirty(leaf); | |
6890 | } | |
6891 | if (found_extent) { | |
87cc7a8a NB |
6892 | ret = remove_extent_backref(trans, path, iref, |
6893 | refs_to_drop, is_data, | |
6894 | &last_ref); | |
005d6427 | 6895 | if (ret) { |
66642832 | 6896 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6897 | goto out; |
6898 | } | |
952fccac | 6899 | } |
5d4f98a2 | 6900 | } else { |
5d4f98a2 YZ |
6901 | if (found_extent) { |
6902 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 6903 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
6904 | if (iref) { |
6905 | BUG_ON(path->slots[0] != extent_slot); | |
6906 | } else { | |
6907 | BUG_ON(path->slots[0] != extent_slot + 1); | |
6908 | path->slots[0] = extent_slot; | |
6909 | num_to_del = 2; | |
6910 | } | |
78fae27e | 6911 | } |
b9473439 | 6912 | |
fcebe456 | 6913 | last_ref = 1; |
952fccac CM |
6914 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
6915 | num_to_del); | |
005d6427 | 6916 | if (ret) { |
66642832 | 6917 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6918 | goto out; |
6919 | } | |
b3b4aa74 | 6920 | btrfs_release_path(path); |
21af804c | 6921 | |
5d4f98a2 | 6922 | if (is_data) { |
5b4aacef | 6923 | ret = btrfs_del_csums(trans, info, bytenr, num_bytes); |
005d6427 | 6924 | if (ret) { |
66642832 | 6925 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6926 | goto out; |
6927 | } | |
459931ec CM |
6928 | } |
6929 | ||
e7355e50 | 6930 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
1e144fb8 | 6931 | if (ret) { |
66642832 | 6932 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
6933 | goto out; |
6934 | } | |
6935 | ||
0b246afa | 6936 | ret = update_block_group(trans, info, bytenr, num_bytes, 0); |
005d6427 | 6937 | if (ret) { |
66642832 | 6938 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6939 | goto out; |
6940 | } | |
a28ec197 | 6941 | } |
fcebe456 JB |
6942 | btrfs_release_path(path); |
6943 | ||
79787eaa | 6944 | out: |
5caf2a00 | 6945 | btrfs_free_path(path); |
a28ec197 CM |
6946 | return ret; |
6947 | } | |
6948 | ||
1887be66 | 6949 | /* |
f0486c68 | 6950 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
6951 | * delayed ref for that extent as well. This searches the delayed ref tree for |
6952 | * a given extent, and if there are no other delayed refs to be processed, it | |
6953 | * removes it from the tree. | |
6954 | */ | |
6955 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 6956 | u64 bytenr) |
1887be66 CM |
6957 | { |
6958 | struct btrfs_delayed_ref_head *head; | |
6959 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 6960 | int ret = 0; |
1887be66 CM |
6961 | |
6962 | delayed_refs = &trans->transaction->delayed_refs; | |
6963 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 6964 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 6965 | if (!head) |
cf93da7b | 6966 | goto out_delayed_unlock; |
1887be66 | 6967 | |
d7df2c79 | 6968 | spin_lock(&head->lock); |
e3d03965 | 6969 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
1887be66 CM |
6970 | goto out; |
6971 | ||
5d4f98a2 YZ |
6972 | if (head->extent_op) { |
6973 | if (!head->must_insert_reserved) | |
6974 | goto out; | |
78a6184a | 6975 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
6976 | head->extent_op = NULL; |
6977 | } | |
6978 | ||
1887be66 CM |
6979 | /* |
6980 | * waiting for the lock here would deadlock. If someone else has it | |
6981 | * locked they are already in the process of dropping it anyway | |
6982 | */ | |
6983 | if (!mutex_trylock(&head->mutex)) | |
6984 | goto out; | |
6985 | ||
6986 | /* | |
6987 | * at this point we have a head with no other entries. Go | |
6988 | * ahead and process it. | |
6989 | */ | |
5c9d028b | 6990 | rb_erase_cached(&head->href_node, &delayed_refs->href_root); |
d278850e | 6991 | RB_CLEAR_NODE(&head->href_node); |
d7df2c79 | 6992 | atomic_dec(&delayed_refs->num_entries); |
1887be66 CM |
6993 | |
6994 | /* | |
6995 | * we don't take a ref on the node because we're removing it from the | |
6996 | * tree, so we just steal the ref the tree was holding. | |
6997 | */ | |
c3e69d58 | 6998 | delayed_refs->num_heads--; |
d7df2c79 | 6999 | if (head->processing == 0) |
c3e69d58 | 7000 | delayed_refs->num_heads_ready--; |
d7df2c79 JB |
7001 | head->processing = 0; |
7002 | spin_unlock(&head->lock); | |
1887be66 CM |
7003 | spin_unlock(&delayed_refs->lock); |
7004 | ||
f0486c68 YZ |
7005 | BUG_ON(head->extent_op); |
7006 | if (head->must_insert_reserved) | |
7007 | ret = 1; | |
7008 | ||
7009 | mutex_unlock(&head->mutex); | |
d278850e | 7010 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 7011 | return ret; |
1887be66 | 7012 | out: |
d7df2c79 | 7013 | spin_unlock(&head->lock); |
cf93da7b CM |
7014 | |
7015 | out_delayed_unlock: | |
1887be66 CM |
7016 | spin_unlock(&delayed_refs->lock); |
7017 | return 0; | |
7018 | } | |
7019 | ||
f0486c68 YZ |
7020 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
7021 | struct btrfs_root *root, | |
7022 | struct extent_buffer *buf, | |
5581a51a | 7023 | u64 parent, int last_ref) |
f0486c68 | 7024 | { |
0b246afa | 7025 | struct btrfs_fs_info *fs_info = root->fs_info; |
b150a4f1 | 7026 | int pin = 1; |
f0486c68 YZ |
7027 | int ret; |
7028 | ||
7029 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
d7eae340 OS |
7030 | int old_ref_mod, new_ref_mod; |
7031 | ||
fd708b81 JB |
7032 | btrfs_ref_tree_mod(root, buf->start, buf->len, parent, |
7033 | root->root_key.objectid, | |
7034 | btrfs_header_level(buf), 0, | |
7035 | BTRFS_DROP_DELAYED_REF); | |
44e1c47d | 7036 | ret = btrfs_add_delayed_tree_ref(trans, buf->start, |
7be07912 | 7037 | buf->len, parent, |
0b246afa JM |
7038 | root->root_key.objectid, |
7039 | btrfs_header_level(buf), | |
7be07912 | 7040 | BTRFS_DROP_DELAYED_REF, NULL, |
d7eae340 | 7041 | &old_ref_mod, &new_ref_mod); |
79787eaa | 7042 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 7043 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
7044 | } |
7045 | ||
0a16c7d7 | 7046 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
7047 | struct btrfs_block_group_cache *cache; |
7048 | ||
f0486c68 | 7049 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 7050 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 7051 | if (!ret) |
37be25bc | 7052 | goto out; |
f0486c68 YZ |
7053 | } |
7054 | ||
4da8b76d | 7055 | pin = 0; |
0b246afa | 7056 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 7057 | |
f0486c68 | 7058 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
2ff7e61e JM |
7059 | pin_down_extent(fs_info, cache, buf->start, |
7060 | buf->len, 1); | |
6219872d | 7061 | btrfs_put_block_group(cache); |
37be25bc | 7062 | goto out; |
f0486c68 YZ |
7063 | } |
7064 | ||
7065 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
7066 | ||
7067 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 7068 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 7069 | btrfs_put_block_group(cache); |
71ff6437 | 7070 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
7071 | } |
7072 | out: | |
b150a4f1 | 7073 | if (pin) |
29d2b84c | 7074 | add_pinned_bytes(fs_info, buf->len, true, |
b150a4f1 JB |
7075 | root->root_key.objectid); |
7076 | ||
0a16c7d7 OS |
7077 | if (last_ref) { |
7078 | /* | |
7079 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
7080 | * matter anymore. | |
7081 | */ | |
7082 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
7083 | } | |
f0486c68 YZ |
7084 | } |
7085 | ||
79787eaa | 7086 | /* Can return -ENOMEM */ |
2ff7e61e | 7087 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
84f7d8e6 | 7088 | struct btrfs_root *root, |
66d7e7f0 | 7089 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, |
b06c4bf5 | 7090 | u64 owner, u64 offset) |
925baedd | 7091 | { |
84f7d8e6 | 7092 | struct btrfs_fs_info *fs_info = root->fs_info; |
d7eae340 | 7093 | int old_ref_mod, new_ref_mod; |
925baedd CM |
7094 | int ret; |
7095 | ||
f5ee5c9a | 7096 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 7097 | return 0; |
fccb84c9 | 7098 | |
fd708b81 JB |
7099 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) |
7100 | btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, | |
7101 | root_objectid, owner, offset, | |
7102 | BTRFS_DROP_DELAYED_REF); | |
7103 | ||
56bec294 CM |
7104 | /* |
7105 | * tree log blocks never actually go into the extent allocation | |
7106 | * tree, just update pinning info and exit early. | |
56bec294 | 7107 | */ |
5d4f98a2 YZ |
7108 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
7109 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 7110 | /* unlocks the pinned mutex */ |
2ff7e61e | 7111 | btrfs_pin_extent(fs_info, bytenr, num_bytes, 1); |
d7eae340 | 7112 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 7113 | ret = 0; |
5d4f98a2 | 7114 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
44e1c47d | 7115 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, |
7be07912 OS |
7116 | num_bytes, parent, |
7117 | root_objectid, (int)owner, | |
7118 | BTRFS_DROP_DELAYED_REF, NULL, | |
d7eae340 | 7119 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 7120 | } else { |
88a979c6 | 7121 | ret = btrfs_add_delayed_data_ref(trans, bytenr, |
7be07912 OS |
7122 | num_bytes, parent, |
7123 | root_objectid, owner, offset, | |
7124 | 0, BTRFS_DROP_DELAYED_REF, | |
d7eae340 | 7125 | &old_ref_mod, &new_ref_mod); |
56bec294 | 7126 | } |
d7eae340 | 7127 | |
29d2b84c NB |
7128 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) { |
7129 | bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID; | |
7130 | ||
7131 | add_pinned_bytes(fs_info, num_bytes, metadata, root_objectid); | |
7132 | } | |
d7eae340 | 7133 | |
925baedd CM |
7134 | return ret; |
7135 | } | |
7136 | ||
817d52f8 JB |
7137 | /* |
7138 | * when we wait for progress in the block group caching, its because | |
7139 | * our allocation attempt failed at least once. So, we must sleep | |
7140 | * and let some progress happen before we try again. | |
7141 | * | |
7142 | * This function will sleep at least once waiting for new free space to | |
7143 | * show up, and then it will check the block group free space numbers | |
7144 | * for our min num_bytes. Another option is to have it go ahead | |
7145 | * and look in the rbtree for a free extent of a given size, but this | |
7146 | * is a good start. | |
36cce922 JB |
7147 | * |
7148 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
7149 | * any of the information in this block group. | |
817d52f8 | 7150 | */ |
36cce922 | 7151 | static noinline void |
817d52f8 JB |
7152 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
7153 | u64 num_bytes) | |
7154 | { | |
11833d66 | 7155 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 7156 | |
11833d66 YZ |
7157 | caching_ctl = get_caching_control(cache); |
7158 | if (!caching_ctl) | |
36cce922 | 7159 | return; |
817d52f8 | 7160 | |
11833d66 | 7161 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 7162 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
7163 | |
7164 | put_caching_control(caching_ctl); | |
11833d66 YZ |
7165 | } |
7166 | ||
7167 | static noinline int | |
7168 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
7169 | { | |
7170 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 7171 | int ret = 0; |
11833d66 YZ |
7172 | |
7173 | caching_ctl = get_caching_control(cache); | |
7174 | if (!caching_ctl) | |
36cce922 | 7175 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
7176 | |
7177 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
7178 | if (cache->cached == BTRFS_CACHE_ERROR) |
7179 | ret = -EIO; | |
11833d66 | 7180 | put_caching_control(caching_ctl); |
36cce922 | 7181 | return ret; |
817d52f8 JB |
7182 | } |
7183 | ||
7184 | enum btrfs_loop_type { | |
285ff5af JB |
7185 | LOOP_CACHING_NOWAIT = 0, |
7186 | LOOP_CACHING_WAIT = 1, | |
7187 | LOOP_ALLOC_CHUNK = 2, | |
7188 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
7189 | }; |
7190 | ||
e570fd27 MX |
7191 | static inline void |
7192 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
7193 | int delalloc) | |
7194 | { | |
7195 | if (delalloc) | |
7196 | down_read(&cache->data_rwsem); | |
7197 | } | |
7198 | ||
7199 | static inline void | |
7200 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
7201 | int delalloc) | |
7202 | { | |
7203 | btrfs_get_block_group(cache); | |
7204 | if (delalloc) | |
7205 | down_read(&cache->data_rwsem); | |
7206 | } | |
7207 | ||
7208 | static struct btrfs_block_group_cache * | |
7209 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
7210 | struct btrfs_free_cluster *cluster, | |
7211 | int delalloc) | |
7212 | { | |
89771cc9 | 7213 | struct btrfs_block_group_cache *used_bg = NULL; |
6719afdc | 7214 | |
e570fd27 | 7215 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
7216 | while (1) { |
7217 | used_bg = cluster->block_group; | |
7218 | if (!used_bg) | |
7219 | return NULL; | |
7220 | ||
7221 | if (used_bg == block_group) | |
e570fd27 MX |
7222 | return used_bg; |
7223 | ||
6719afdc | 7224 | btrfs_get_block_group(used_bg); |
e570fd27 | 7225 | |
6719afdc GU |
7226 | if (!delalloc) |
7227 | return used_bg; | |
e570fd27 | 7228 | |
6719afdc GU |
7229 | if (down_read_trylock(&used_bg->data_rwsem)) |
7230 | return used_bg; | |
e570fd27 | 7231 | |
6719afdc | 7232 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 7233 | |
e321f8a8 LB |
7234 | /* We should only have one-level nested. */ |
7235 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 7236 | |
6719afdc GU |
7237 | spin_lock(&cluster->refill_lock); |
7238 | if (used_bg == cluster->block_group) | |
7239 | return used_bg; | |
e570fd27 | 7240 | |
6719afdc GU |
7241 | up_read(&used_bg->data_rwsem); |
7242 | btrfs_put_block_group(used_bg); | |
7243 | } | |
e570fd27 MX |
7244 | } |
7245 | ||
7246 | static inline void | |
7247 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7248 | int delalloc) | |
7249 | { | |
7250 | if (delalloc) | |
7251 | up_read(&cache->data_rwsem); | |
7252 | btrfs_put_block_group(cache); | |
7253 | } | |
7254 | ||
fec577fb CM |
7255 | /* |
7256 | * walks the btree of allocated extents and find a hole of a given size. | |
7257 | * The key ins is changed to record the hole: | |
a4820398 | 7258 | * ins->objectid == start position |
62e2749e | 7259 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7260 | * ins->offset == the size of the hole. |
fec577fb | 7261 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7262 | * |
7263 | * If there is no suitable free space, we will record the max size of | |
7264 | * the free space extent currently. | |
fec577fb | 7265 | */ |
87bde3cd | 7266 | static noinline int find_free_extent(struct btrfs_fs_info *fs_info, |
18513091 WX |
7267 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
7268 | u64 hint_byte, struct btrfs_key *ins, | |
7269 | u64 flags, int delalloc) | |
fec577fb | 7270 | { |
80eb234a | 7271 | int ret = 0; |
0b246afa | 7272 | struct btrfs_root *root = fs_info->extent_root; |
fa9c0d79 | 7273 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7274 | struct btrfs_block_group_cache *block_group = NULL; |
81c9ad23 | 7275 | u64 search_start = 0; |
a4820398 | 7276 | u64 max_extent_size = 0; |
fb5c39d7 | 7277 | u64 max_free_space = 0; |
c759c4e1 | 7278 | u64 empty_cluster = 0; |
80eb234a | 7279 | struct btrfs_space_info *space_info; |
fa9c0d79 | 7280 | int loop = 0; |
3e72ee88 | 7281 | int index = btrfs_bg_flags_to_raid_index(flags); |
0a24325e | 7282 | bool failed_cluster_refill = false; |
1cdda9b8 | 7283 | bool failed_alloc = false; |
67377734 | 7284 | bool use_cluster = true; |
60d2adbb | 7285 | bool have_caching_bg = false; |
13a0db5a | 7286 | bool orig_have_caching_bg = false; |
a5e681d9 | 7287 | bool full_search = false; |
fec577fb | 7288 | |
0b246afa | 7289 | WARN_ON(num_bytes < fs_info->sectorsize); |
962a298f | 7290 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7291 | ins->objectid = 0; |
7292 | ins->offset = 0; | |
b1a4d965 | 7293 | |
71ff6437 | 7294 | trace_find_free_extent(fs_info, num_bytes, empty_size, flags); |
3f7de037 | 7295 | |
0b246afa | 7296 | space_info = __find_space_info(fs_info, flags); |
1b1d1f66 | 7297 | if (!space_info) { |
0b246afa | 7298 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7299 | return -ENOSPC; |
7300 | } | |
2552d17e | 7301 | |
67377734 | 7302 | /* |
4f4db217 JB |
7303 | * If our free space is heavily fragmented we may not be able to make |
7304 | * big contiguous allocations, so instead of doing the expensive search | |
7305 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7306 | * can go ahead and search for a more manageable chunk. | |
7307 | * | |
7308 | * If our max_extent_size is large enough for our allocation simply | |
7309 | * disable clustering since we will likely not be able to find enough | |
7310 | * space to create a cluster and induce latency trying. | |
67377734 | 7311 | */ |
4f4db217 JB |
7312 | if (unlikely(space_info->max_extent_size)) { |
7313 | spin_lock(&space_info->lock); | |
7314 | if (space_info->max_extent_size && | |
7315 | num_bytes > space_info->max_extent_size) { | |
7316 | ins->offset = space_info->max_extent_size; | |
7317 | spin_unlock(&space_info->lock); | |
7318 | return -ENOSPC; | |
7319 | } else if (space_info->max_extent_size) { | |
7320 | use_cluster = false; | |
7321 | } | |
7322 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7323 | } |
0f9dd46c | 7324 | |
2ff7e61e | 7325 | last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster); |
239b14b3 | 7326 | if (last_ptr) { |
fa9c0d79 CM |
7327 | spin_lock(&last_ptr->lock); |
7328 | if (last_ptr->block_group) | |
7329 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7330 | if (last_ptr->fragmented) { |
7331 | /* | |
7332 | * We still set window_start so we can keep track of the | |
7333 | * last place we found an allocation to try and save | |
7334 | * some time. | |
7335 | */ | |
7336 | hint_byte = last_ptr->window_start; | |
7337 | use_cluster = false; | |
7338 | } | |
fa9c0d79 | 7339 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7340 | } |
fa9c0d79 | 7341 | |
2ff7e61e | 7342 | search_start = max(search_start, first_logical_byte(fs_info, 0)); |
239b14b3 | 7343 | search_start = max(search_start, hint_byte); |
2552d17e | 7344 | if (search_start == hint_byte) { |
0b246afa | 7345 | block_group = btrfs_lookup_block_group(fs_info, search_start); |
817d52f8 JB |
7346 | /* |
7347 | * we don't want to use the block group if it doesn't match our | |
7348 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7349 | * |
7350 | * However if we are re-searching with an ideal block group | |
7351 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7352 | */ |
b6919a58 | 7353 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7354 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7355 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7356 | if (list_empty(&block_group->list) || |
7357 | block_group->ro) { | |
7358 | /* | |
7359 | * someone is removing this block group, | |
7360 | * we can't jump into the have_block_group | |
7361 | * target because our list pointers are not | |
7362 | * valid | |
7363 | */ | |
7364 | btrfs_put_block_group(block_group); | |
7365 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7366 | } else { |
3e72ee88 QW |
7367 | index = btrfs_bg_flags_to_raid_index( |
7368 | block_group->flags); | |
e570fd27 | 7369 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7370 | goto have_block_group; |
ccf0e725 | 7371 | } |
2552d17e | 7372 | } else if (block_group) { |
fa9c0d79 | 7373 | btrfs_put_block_group(block_group); |
2552d17e | 7374 | } |
42e70e7a | 7375 | } |
2552d17e | 7376 | search: |
60d2adbb | 7377 | have_caching_bg = false; |
3e72ee88 | 7378 | if (index == 0 || index == btrfs_bg_flags_to_raid_index(flags)) |
a5e681d9 | 7379 | full_search = true; |
80eb234a | 7380 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
7381 | list_for_each_entry(block_group, &space_info->block_groups[index], |
7382 | list) { | |
6226cb0a | 7383 | u64 offset; |
817d52f8 | 7384 | int cached; |
8a1413a2 | 7385 | |
14443937 JM |
7386 | /* If the block group is read-only, we can skip it entirely. */ |
7387 | if (unlikely(block_group->ro)) | |
7388 | continue; | |
7389 | ||
e570fd27 | 7390 | btrfs_grab_block_group(block_group, delalloc); |
2552d17e | 7391 | search_start = block_group->key.objectid; |
42e70e7a | 7392 | |
83a50de9 CM |
7393 | /* |
7394 | * this can happen if we end up cycling through all the | |
7395 | * raid types, but we want to make sure we only allocate | |
7396 | * for the proper type. | |
7397 | */ | |
b6919a58 | 7398 | if (!block_group_bits(block_group, flags)) { |
bece2e82 | 7399 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
83a50de9 | 7400 | BTRFS_BLOCK_GROUP_RAID1 | |
53b381b3 DW |
7401 | BTRFS_BLOCK_GROUP_RAID5 | |
7402 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
7403 | BTRFS_BLOCK_GROUP_RAID10; |
7404 | ||
7405 | /* | |
7406 | * if they asked for extra copies and this block group | |
7407 | * doesn't provide them, bail. This does allow us to | |
7408 | * fill raid0 from raid1. | |
7409 | */ | |
b6919a58 | 7410 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
7411 | goto loop; |
7412 | } | |
7413 | ||
2552d17e | 7414 | have_block_group: |
291c7d2f JB |
7415 | cached = block_group_cache_done(block_group); |
7416 | if (unlikely(!cached)) { | |
a5e681d9 | 7417 | have_caching_bg = true; |
f6373bf3 | 7418 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7419 | BUG_ON(ret < 0); |
7420 | ret = 0; | |
817d52f8 JB |
7421 | } |
7422 | ||
36cce922 JB |
7423 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7424 | goto loop; | |
0f9dd46c | 7425 | |
0a24325e | 7426 | /* |
062c05c4 AO |
7427 | * Ok we want to try and use the cluster allocator, so |
7428 | * lets look there | |
0a24325e | 7429 | */ |
c759c4e1 | 7430 | if (last_ptr && use_cluster) { |
215a63d1 | 7431 | struct btrfs_block_group_cache *used_block_group; |
8de972b4 | 7432 | unsigned long aligned_cluster; |
fa9c0d79 CM |
7433 | /* |
7434 | * the refill lock keeps out other | |
7435 | * people trying to start a new cluster | |
7436 | */ | |
e570fd27 MX |
7437 | used_block_group = btrfs_lock_cluster(block_group, |
7438 | last_ptr, | |
7439 | delalloc); | |
7440 | if (!used_block_group) | |
44fb5511 | 7441 | goto refill_cluster; |
274bd4fb | 7442 | |
e570fd27 MX |
7443 | if (used_block_group != block_group && |
7444 | (used_block_group->ro || | |
7445 | !block_group_bits(used_block_group, flags))) | |
7446 | goto release_cluster; | |
44fb5511 | 7447 | |
274bd4fb | 7448 | offset = btrfs_alloc_from_cluster(used_block_group, |
a4820398 MX |
7449 | last_ptr, |
7450 | num_bytes, | |
7451 | used_block_group->key.objectid, | |
7452 | &max_extent_size); | |
fa9c0d79 CM |
7453 | if (offset) { |
7454 | /* we have a block, we're done */ | |
7455 | spin_unlock(&last_ptr->refill_lock); | |
3dca5c94 | 7456 | trace_btrfs_reserve_extent_cluster( |
89d4346a MX |
7457 | used_block_group, |
7458 | search_start, num_bytes); | |
215a63d1 | 7459 | if (used_block_group != block_group) { |
e570fd27 MX |
7460 | btrfs_release_block_group(block_group, |
7461 | delalloc); | |
215a63d1 MX |
7462 | block_group = used_block_group; |
7463 | } | |
fa9c0d79 CM |
7464 | goto checks; |
7465 | } | |
7466 | ||
274bd4fb | 7467 | WARN_ON(last_ptr->block_group != used_block_group); |
e570fd27 | 7468 | release_cluster: |
062c05c4 AO |
7469 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
7470 | * set up a new clusters, so lets just skip it | |
7471 | * and let the allocator find whatever block | |
7472 | * it can find. If we reach this point, we | |
7473 | * will have tried the cluster allocator | |
7474 | * plenty of times and not have found | |
7475 | * anything, so we are likely way too | |
7476 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
7477 | * anything. |
7478 | * | |
7479 | * However, if the cluster is taken from the | |
7480 | * current block group, release the cluster | |
7481 | * first, so that we stand a better chance of | |
7482 | * succeeding in the unclustered | |
7483 | * allocation. */ | |
7484 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
e570fd27 | 7485 | used_block_group != block_group) { |
062c05c4 | 7486 | spin_unlock(&last_ptr->refill_lock); |
e570fd27 MX |
7487 | btrfs_release_block_group(used_block_group, |
7488 | delalloc); | |
062c05c4 AO |
7489 | goto unclustered_alloc; |
7490 | } | |
7491 | ||
fa9c0d79 CM |
7492 | /* |
7493 | * this cluster didn't work out, free it and | |
7494 | * start over | |
7495 | */ | |
7496 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7497 | ||
e570fd27 MX |
7498 | if (used_block_group != block_group) |
7499 | btrfs_release_block_group(used_block_group, | |
7500 | delalloc); | |
7501 | refill_cluster: | |
a5f6f719 AO |
7502 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
7503 | spin_unlock(&last_ptr->refill_lock); | |
7504 | goto unclustered_alloc; | |
7505 | } | |
7506 | ||
8de972b4 CM |
7507 | aligned_cluster = max_t(unsigned long, |
7508 | empty_cluster + empty_size, | |
7509 | block_group->full_stripe_len); | |
7510 | ||
fa9c0d79 | 7511 | /* allocate a cluster in this block group */ |
2ff7e61e | 7512 | ret = btrfs_find_space_cluster(fs_info, block_group, |
00361589 JB |
7513 | last_ptr, search_start, |
7514 | num_bytes, | |
7515 | aligned_cluster); | |
fa9c0d79 CM |
7516 | if (ret == 0) { |
7517 | /* | |
7518 | * now pull our allocation out of this | |
7519 | * cluster | |
7520 | */ | |
7521 | offset = btrfs_alloc_from_cluster(block_group, | |
a4820398 MX |
7522 | last_ptr, |
7523 | num_bytes, | |
7524 | search_start, | |
7525 | &max_extent_size); | |
fa9c0d79 CM |
7526 | if (offset) { |
7527 | /* we found one, proceed */ | |
7528 | spin_unlock(&last_ptr->refill_lock); | |
3dca5c94 | 7529 | trace_btrfs_reserve_extent_cluster( |
3f7de037 JB |
7530 | block_group, search_start, |
7531 | num_bytes); | |
fa9c0d79 CM |
7532 | goto checks; |
7533 | } | |
0a24325e JB |
7534 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
7535 | && !failed_cluster_refill) { | |
817d52f8 JB |
7536 | spin_unlock(&last_ptr->refill_lock); |
7537 | ||
0a24325e | 7538 | failed_cluster_refill = true; |
817d52f8 JB |
7539 | wait_block_group_cache_progress(block_group, |
7540 | num_bytes + empty_cluster + empty_size); | |
7541 | goto have_block_group; | |
fa9c0d79 | 7542 | } |
817d52f8 | 7543 | |
fa9c0d79 CM |
7544 | /* |
7545 | * at this point we either didn't find a cluster | |
7546 | * or we weren't able to allocate a block from our | |
7547 | * cluster. Free the cluster we've been trying | |
7548 | * to use, and go to the next block group | |
7549 | */ | |
0a24325e | 7550 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 7551 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 7552 | goto loop; |
fa9c0d79 CM |
7553 | } |
7554 | ||
062c05c4 | 7555 | unclustered_alloc: |
c759c4e1 JB |
7556 | /* |
7557 | * We are doing an unclustered alloc, set the fragmented flag so | |
7558 | * we don't bother trying to setup a cluster again until we get | |
7559 | * more space. | |
7560 | */ | |
7561 | if (unlikely(last_ptr)) { | |
7562 | spin_lock(&last_ptr->lock); | |
7563 | last_ptr->fragmented = 1; | |
7564 | spin_unlock(&last_ptr->lock); | |
7565 | } | |
0c9b36e0 LB |
7566 | if (cached) { |
7567 | struct btrfs_free_space_ctl *ctl = | |
7568 | block_group->free_space_ctl; | |
7569 | ||
7570 | spin_lock(&ctl->tree_lock); | |
7571 | if (ctl->free_space < | |
7572 | num_bytes + empty_cluster + empty_size) { | |
fb5c39d7 JB |
7573 | max_free_space = max(max_free_space, |
7574 | ctl->free_space); | |
0c9b36e0 LB |
7575 | spin_unlock(&ctl->tree_lock); |
7576 | goto loop; | |
7577 | } | |
7578 | spin_unlock(&ctl->tree_lock); | |
a5f6f719 | 7579 | } |
a5f6f719 | 7580 | |
6226cb0a | 7581 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
a4820398 MX |
7582 | num_bytes, empty_size, |
7583 | &max_extent_size); | |
1cdda9b8 JB |
7584 | /* |
7585 | * If we didn't find a chunk, and we haven't failed on this | |
7586 | * block group before, and this block group is in the middle of | |
7587 | * caching and we are ok with waiting, then go ahead and wait | |
7588 | * for progress to be made, and set failed_alloc to true. | |
7589 | * | |
7590 | * If failed_alloc is true then we've already waited on this | |
7591 | * block group once and should move on to the next block group. | |
7592 | */ | |
7593 | if (!offset && !failed_alloc && !cached && | |
7594 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 7595 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
7596 | num_bytes + empty_size); |
7597 | failed_alloc = true; | |
817d52f8 | 7598 | goto have_block_group; |
1cdda9b8 JB |
7599 | } else if (!offset) { |
7600 | goto loop; | |
817d52f8 | 7601 | } |
fa9c0d79 | 7602 | checks: |
5e23a6fe | 7603 | search_start = round_up(offset, fs_info->stripesize); |
25179201 | 7604 | |
2552d17e JB |
7605 | /* move on to the next group */ |
7606 | if (search_start + num_bytes > | |
215a63d1 MX |
7607 | block_group->key.objectid + block_group->key.offset) { |
7608 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 7609 | goto loop; |
6226cb0a | 7610 | } |
f5a31e16 | 7611 | |
f0486c68 | 7612 | if (offset < search_start) |
215a63d1 | 7613 | btrfs_add_free_space(block_group, offset, |
f0486c68 | 7614 | search_start - offset); |
2552d17e | 7615 | |
18513091 WX |
7616 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
7617 | num_bytes, delalloc); | |
f0486c68 | 7618 | if (ret == -EAGAIN) { |
215a63d1 | 7619 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 7620 | goto loop; |
0f9dd46c | 7621 | } |
9cfa3e34 | 7622 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 7623 | |
f0486c68 | 7624 | /* we are all good, lets return */ |
2552d17e JB |
7625 | ins->objectid = search_start; |
7626 | ins->offset = num_bytes; | |
d2fb3437 | 7627 | |
3dca5c94 | 7628 | trace_btrfs_reserve_extent(block_group, search_start, num_bytes); |
e570fd27 | 7629 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7630 | break; |
7631 | loop: | |
0a24325e | 7632 | failed_cluster_refill = false; |
1cdda9b8 | 7633 | failed_alloc = false; |
3e72ee88 QW |
7634 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != |
7635 | index); | |
e570fd27 | 7636 | btrfs_release_block_group(block_group, delalloc); |
14443937 | 7637 | cond_resched(); |
2552d17e JB |
7638 | } |
7639 | up_read(&space_info->groups_sem); | |
7640 | ||
13a0db5a | 7641 | if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg |
7642 | && !orig_have_caching_bg) | |
7643 | orig_have_caching_bg = true; | |
7644 | ||
60d2adbb MX |
7645 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
7646 | goto search; | |
7647 | ||
b742bb82 YZ |
7648 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
7649 | goto search; | |
7650 | ||
285ff5af | 7651 | /* |
ccf0e725 JB |
7652 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
7653 | * caching kthreads as we move along | |
817d52f8 JB |
7654 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
7655 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7656 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7657 | * again | |
fa9c0d79 | 7658 | */ |
723bda20 | 7659 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 7660 | index = 0; |
a5e681d9 JB |
7661 | if (loop == LOOP_CACHING_NOWAIT) { |
7662 | /* | |
7663 | * We want to skip the LOOP_CACHING_WAIT step if we | |
01327610 | 7664 | * don't have any uncached bgs and we've already done a |
a5e681d9 JB |
7665 | * full search through. |
7666 | */ | |
13a0db5a | 7667 | if (orig_have_caching_bg || !full_search) |
a5e681d9 JB |
7668 | loop = LOOP_CACHING_WAIT; |
7669 | else | |
7670 | loop = LOOP_ALLOC_CHUNK; | |
7671 | } else { | |
7672 | loop++; | |
7673 | } | |
7674 | ||
817d52f8 | 7675 | if (loop == LOOP_ALLOC_CHUNK) { |
00361589 | 7676 | struct btrfs_trans_handle *trans; |
f017f15f WS |
7677 | int exist = 0; |
7678 | ||
7679 | trans = current->journal_info; | |
7680 | if (trans) | |
7681 | exist = 1; | |
7682 | else | |
7683 | trans = btrfs_join_transaction(root); | |
00361589 | 7684 | |
00361589 JB |
7685 | if (IS_ERR(trans)) { |
7686 | ret = PTR_ERR(trans); | |
7687 | goto out; | |
7688 | } | |
7689 | ||
01458828 | 7690 | ret = do_chunk_alloc(trans, flags, CHUNK_ALLOC_FORCE); |
a5e681d9 JB |
7691 | |
7692 | /* | |
7693 | * If we can't allocate a new chunk we've already looped | |
7694 | * through at least once, move on to the NO_EMPTY_SIZE | |
7695 | * case. | |
7696 | */ | |
7697 | if (ret == -ENOSPC) | |
7698 | loop = LOOP_NO_EMPTY_SIZE; | |
7699 | ||
ea658bad JB |
7700 | /* |
7701 | * Do not bail out on ENOSPC since we | |
7702 | * can do more things. | |
7703 | */ | |
00361589 | 7704 | if (ret < 0 && ret != -ENOSPC) |
66642832 | 7705 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
7706 | else |
7707 | ret = 0; | |
f017f15f | 7708 | if (!exist) |
3a45bb20 | 7709 | btrfs_end_transaction(trans); |
00361589 | 7710 | if (ret) |
ea658bad | 7711 | goto out; |
2552d17e JB |
7712 | } |
7713 | ||
723bda20 | 7714 | if (loop == LOOP_NO_EMPTY_SIZE) { |
a5e681d9 JB |
7715 | /* |
7716 | * Don't loop again if we already have no empty_size and | |
7717 | * no empty_cluster. | |
7718 | */ | |
7719 | if (empty_size == 0 && | |
7720 | empty_cluster == 0) { | |
7721 | ret = -ENOSPC; | |
7722 | goto out; | |
7723 | } | |
723bda20 JB |
7724 | empty_size = 0; |
7725 | empty_cluster = 0; | |
fa9c0d79 | 7726 | } |
723bda20 JB |
7727 | |
7728 | goto search; | |
2552d17e JB |
7729 | } else if (!ins->objectid) { |
7730 | ret = -ENOSPC; | |
d82a6f1d | 7731 | } else if (ins->objectid) { |
c759c4e1 JB |
7732 | if (!use_cluster && last_ptr) { |
7733 | spin_lock(&last_ptr->lock); | |
7734 | last_ptr->window_start = ins->objectid; | |
7735 | spin_unlock(&last_ptr->lock); | |
7736 | } | |
80eb234a | 7737 | ret = 0; |
be744175 | 7738 | } |
79787eaa | 7739 | out: |
4f4db217 | 7740 | if (ret == -ENOSPC) { |
fb5c39d7 JB |
7741 | if (!max_extent_size) |
7742 | max_extent_size = max_free_space; | |
4f4db217 JB |
7743 | spin_lock(&space_info->lock); |
7744 | space_info->max_extent_size = max_extent_size; | |
7745 | spin_unlock(&space_info->lock); | |
a4820398 | 7746 | ins->offset = max_extent_size; |
4f4db217 | 7747 | } |
0f70abe2 | 7748 | return ret; |
fec577fb | 7749 | } |
ec44a35c | 7750 | |
ab8d0fc4 JM |
7751 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
7752 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 7753 | int dump_block_groups) |
0f9dd46c JB |
7754 | { |
7755 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 7756 | int index = 0; |
0f9dd46c | 7757 | |
9ed74f2d | 7758 | spin_lock(&info->lock); |
ab8d0fc4 JM |
7759 | btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull", |
7760 | info->flags, | |
4136135b LB |
7761 | info->total_bytes - btrfs_space_info_used(info, true), |
7762 | info->full ? "" : "not "); | |
ab8d0fc4 JM |
7763 | btrfs_info(fs_info, |
7764 | "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu", | |
7765 | info->total_bytes, info->bytes_used, info->bytes_pinned, | |
7766 | info->bytes_reserved, info->bytes_may_use, | |
7767 | info->bytes_readonly); | |
9ed74f2d JB |
7768 | spin_unlock(&info->lock); |
7769 | ||
7770 | if (!dump_block_groups) | |
7771 | return; | |
0f9dd46c | 7772 | |
80eb234a | 7773 | down_read(&info->groups_sem); |
b742bb82 YZ |
7774 | again: |
7775 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 7776 | spin_lock(&cache->lock); |
ab8d0fc4 JM |
7777 | btrfs_info(fs_info, |
7778 | "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s", | |
7779 | cache->key.objectid, cache->key.offset, | |
7780 | btrfs_block_group_used(&cache->item), cache->pinned, | |
7781 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
7782 | btrfs_dump_free_space(cache, bytes); |
7783 | spin_unlock(&cache->lock); | |
7784 | } | |
b742bb82 YZ |
7785 | if (++index < BTRFS_NR_RAID_TYPES) |
7786 | goto again; | |
80eb234a | 7787 | up_read(&info->groups_sem); |
0f9dd46c | 7788 | } |
e8569813 | 7789 | |
6f47c706 NB |
7790 | /* |
7791 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a | |
7792 | * hole that is at least as big as @num_bytes. | |
7793 | * | |
7794 | * @root - The root that will contain this extent | |
7795 | * | |
7796 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
7797 | * is used for accounting purposes. This value differs | |
7798 | * from @num_bytes only in the case of compressed extents. | |
7799 | * | |
7800 | * @num_bytes - Number of bytes to allocate on-disk. | |
7801 | * | |
7802 | * @min_alloc_size - Indicates the minimum amount of space that the | |
7803 | * allocator should try to satisfy. In some cases | |
7804 | * @num_bytes may be larger than what is required and if | |
7805 | * the filesystem is fragmented then allocation fails. | |
7806 | * However, the presence of @min_alloc_size gives a | |
7807 | * chance to try and satisfy the smaller allocation. | |
7808 | * | |
7809 | * @empty_size - A hint that you plan on doing more COW. This is the | |
7810 | * size in bytes the allocator should try to find free | |
7811 | * next to the block it returns. This is just a hint and | |
7812 | * may be ignored by the allocator. | |
7813 | * | |
7814 | * @hint_byte - Hint to the allocator to start searching above the byte | |
7815 | * address passed. It might be ignored. | |
7816 | * | |
7817 | * @ins - This key is modified to record the found hole. It will | |
7818 | * have the following values: | |
7819 | * ins->objectid == start position | |
7820 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
7821 | * ins->offset == the size of the hole. | |
7822 | * | |
7823 | * @is_data - Boolean flag indicating whether an extent is | |
7824 | * allocated for data (true) or metadata (false) | |
7825 | * | |
7826 | * @delalloc - Boolean flag indicating whether this allocation is for | |
7827 | * delalloc or not. If 'true' data_rwsem of block groups | |
7828 | * is going to be acquired. | |
7829 | * | |
7830 | * | |
7831 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
7832 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
7833 | * largest available hole the allocator managed to find. | |
7834 | */ | |
18513091 | 7835 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
7836 | u64 num_bytes, u64 min_alloc_size, |
7837 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 7838 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 7839 | { |
ab8d0fc4 | 7840 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 7841 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 7842 | u64 flags; |
fec577fb | 7843 | int ret; |
925baedd | 7844 | |
1b86826d | 7845 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 7846 | again: |
0b246afa | 7847 | WARN_ON(num_bytes < fs_info->sectorsize); |
87bde3cd | 7848 | ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size, |
18513091 | 7849 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 7850 | if (!ret && !is_data) { |
ab8d0fc4 | 7851 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 7852 | } else if (ret == -ENOSPC) { |
a4820398 MX |
7853 | if (!final_tried && ins->offset) { |
7854 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 7855 | num_bytes = round_down(num_bytes, |
0b246afa | 7856 | fs_info->sectorsize); |
9e622d6b | 7857 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 7858 | ram_bytes = num_bytes; |
9e622d6b MX |
7859 | if (num_bytes == min_alloc_size) |
7860 | final_tried = true; | |
7861 | goto again; | |
ab8d0fc4 | 7862 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
7863 | struct btrfs_space_info *sinfo; |
7864 | ||
ab8d0fc4 | 7865 | sinfo = __find_space_info(fs_info, flags); |
0b246afa | 7866 | btrfs_err(fs_info, |
5d163e0e JM |
7867 | "allocation failed flags %llu, wanted %llu", |
7868 | flags, num_bytes); | |
53804280 | 7869 | if (sinfo) |
ab8d0fc4 | 7870 | dump_space_info(fs_info, sinfo, num_bytes, 1); |
9e622d6b | 7871 | } |
925baedd | 7872 | } |
0f9dd46c JB |
7873 | |
7874 | return ret; | |
e6dcd2dc CM |
7875 | } |
7876 | ||
2ff7e61e | 7877 | static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
7878 | u64 start, u64 len, |
7879 | int pin, int delalloc) | |
65b51a00 | 7880 | { |
0f9dd46c | 7881 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 7882 | int ret = 0; |
0f9dd46c | 7883 | |
0b246afa | 7884 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 7885 | if (!cache) { |
0b246afa JM |
7886 | btrfs_err(fs_info, "Unable to find block group for %llu", |
7887 | start); | |
0f9dd46c JB |
7888 | return -ENOSPC; |
7889 | } | |
1f3c79a2 | 7890 | |
e688b725 | 7891 | if (pin) |
2ff7e61e | 7892 | pin_down_extent(fs_info, cache, start, len, 1); |
e688b725 | 7893 | else { |
0b246afa | 7894 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 7895 | ret = btrfs_discard_extent(fs_info, start, len, NULL); |
e688b725 | 7896 | btrfs_add_free_space(cache, start, len); |
4824f1f4 | 7897 | btrfs_free_reserved_bytes(cache, len, delalloc); |
71ff6437 | 7898 | trace_btrfs_reserved_extent_free(fs_info, start, len); |
e688b725 | 7899 | } |
31193213 | 7900 | |
fa9c0d79 | 7901 | btrfs_put_block_group(cache); |
e6dcd2dc CM |
7902 | return ret; |
7903 | } | |
7904 | ||
2ff7e61e | 7905 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 | 7906 | u64 start, u64 len, int delalloc) |
e688b725 | 7907 | { |
2ff7e61e | 7908 | return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc); |
e688b725 CM |
7909 | } |
7910 | ||
2ff7e61e | 7911 | int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info, |
e688b725 CM |
7912 | u64 start, u64 len) |
7913 | { | |
2ff7e61e | 7914 | return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0); |
e688b725 CM |
7915 | } |
7916 | ||
5d4f98a2 | 7917 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
7918 | u64 parent, u64 root_objectid, |
7919 | u64 flags, u64 owner, u64 offset, | |
7920 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc | 7921 | { |
ef89b824 | 7922 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 7923 | int ret; |
e6dcd2dc | 7924 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 7925 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 7926 | struct btrfs_path *path; |
5d4f98a2 YZ |
7927 | struct extent_buffer *leaf; |
7928 | int type; | |
7929 | u32 size; | |
26b8003f | 7930 | |
5d4f98a2 YZ |
7931 | if (parent > 0) |
7932 | type = BTRFS_SHARED_DATA_REF_KEY; | |
7933 | else | |
7934 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 7935 | |
5d4f98a2 | 7936 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
7937 | |
7938 | path = btrfs_alloc_path(); | |
db5b493a TI |
7939 | if (!path) |
7940 | return -ENOMEM; | |
47e4bb98 | 7941 | |
b9473439 | 7942 | path->leave_spinning = 1; |
5d4f98a2 YZ |
7943 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
7944 | ins, size); | |
79787eaa JM |
7945 | if (ret) { |
7946 | btrfs_free_path(path); | |
7947 | return ret; | |
7948 | } | |
0f9dd46c | 7949 | |
5d4f98a2 YZ |
7950 | leaf = path->nodes[0]; |
7951 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 7952 | struct btrfs_extent_item); |
5d4f98a2 YZ |
7953 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
7954 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
7955 | btrfs_set_extent_flags(leaf, extent_item, | |
7956 | flags | BTRFS_EXTENT_FLAG_DATA); | |
7957 | ||
7958 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
7959 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
7960 | if (parent > 0) { | |
7961 | struct btrfs_shared_data_ref *ref; | |
7962 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
7963 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
7964 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
7965 | } else { | |
7966 | struct btrfs_extent_data_ref *ref; | |
7967 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
7968 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
7969 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
7970 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
7971 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
7972 | } | |
47e4bb98 CM |
7973 | |
7974 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 7975 | btrfs_free_path(path); |
f510cfec | 7976 | |
25a356d3 | 7977 | ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset); |
1e144fb8 OS |
7978 | if (ret) |
7979 | return ret; | |
7980 | ||
6202df69 | 7981 | ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1); |
79787eaa | 7982 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 7983 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 7984 | ins->objectid, ins->offset); |
f5947066 CM |
7985 | BUG(); |
7986 | } | |
71ff6437 | 7987 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
7988 | return ret; |
7989 | } | |
7990 | ||
5d4f98a2 | 7991 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 7992 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 7993 | struct btrfs_delayed_extent_op *extent_op) |
e6dcd2dc | 7994 | { |
9dcdbe01 | 7995 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 7996 | int ret; |
5d4f98a2 | 7997 | struct btrfs_extent_item *extent_item; |
4e6bd4e0 | 7998 | struct btrfs_key extent_key; |
5d4f98a2 YZ |
7999 | struct btrfs_tree_block_info *block_info; |
8000 | struct btrfs_extent_inline_ref *iref; | |
8001 | struct btrfs_path *path; | |
8002 | struct extent_buffer *leaf; | |
4e6bd4e0 | 8003 | struct btrfs_delayed_tree_ref *ref; |
3173a18f | 8004 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
4e6bd4e0 | 8005 | u64 num_bytes; |
21ebfbe7 | 8006 | u64 flags = extent_op->flags_to_set; |
0b246afa | 8007 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f | 8008 | |
4e6bd4e0 NB |
8009 | ref = btrfs_delayed_node_to_tree_ref(node); |
8010 | ||
4e6bd4e0 NB |
8011 | extent_key.objectid = node->bytenr; |
8012 | if (skinny_metadata) { | |
8013 | extent_key.offset = ref->level; | |
8014 | extent_key.type = BTRFS_METADATA_ITEM_KEY; | |
8015 | num_bytes = fs_info->nodesize; | |
8016 | } else { | |
8017 | extent_key.offset = node->num_bytes; | |
8018 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
3173a18f | 8019 | size += sizeof(*block_info); |
4e6bd4e0 NB |
8020 | num_bytes = node->num_bytes; |
8021 | } | |
1c2308f8 | 8022 | |
5d4f98a2 | 8023 | path = btrfs_alloc_path(); |
80ee54bf | 8024 | if (!path) |
d8926bb3 | 8025 | return -ENOMEM; |
56bec294 | 8026 | |
5d4f98a2 YZ |
8027 | path->leave_spinning = 1; |
8028 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
4e6bd4e0 | 8029 | &extent_key, size); |
79787eaa | 8030 | if (ret) { |
dd825259 | 8031 | btrfs_free_path(path); |
79787eaa JM |
8032 | return ret; |
8033 | } | |
5d4f98a2 YZ |
8034 | |
8035 | leaf = path->nodes[0]; | |
8036 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
8037 | struct btrfs_extent_item); | |
8038 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
8039 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8040 | btrfs_set_extent_flags(leaf, extent_item, | |
8041 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 8042 | |
3173a18f JB |
8043 | if (skinny_metadata) { |
8044 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8045 | } else { | |
8046 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
21ebfbe7 | 8047 | btrfs_set_tree_block_key(leaf, block_info, &extent_op->key); |
4e6bd4e0 | 8048 | btrfs_set_tree_block_level(leaf, block_info, ref->level); |
3173a18f JB |
8049 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
8050 | } | |
5d4f98a2 | 8051 | |
d4b20733 | 8052 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) { |
5d4f98a2 YZ |
8053 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); |
8054 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8055 | BTRFS_SHARED_BLOCK_REF_KEY); | |
d4b20733 | 8056 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent); |
5d4f98a2 YZ |
8057 | } else { |
8058 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8059 | BTRFS_TREE_BLOCK_REF_KEY); | |
4e6bd4e0 | 8060 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root); |
5d4f98a2 YZ |
8061 | } |
8062 | ||
8063 | btrfs_mark_buffer_dirty(leaf); | |
8064 | btrfs_free_path(path); | |
8065 | ||
4e6bd4e0 NB |
8066 | ret = remove_from_free_space_tree(trans, extent_key.objectid, |
8067 | num_bytes); | |
1e144fb8 OS |
8068 | if (ret) |
8069 | return ret; | |
8070 | ||
4e6bd4e0 | 8071 | ret = update_block_group(trans, fs_info, extent_key.objectid, |
6202df69 | 8072 | fs_info->nodesize, 1); |
79787eaa | 8073 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8074 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
4e6bd4e0 | 8075 | extent_key.objectid, extent_key.offset); |
5d4f98a2 YZ |
8076 | BUG(); |
8077 | } | |
0be5dc67 | 8078 | |
4e6bd4e0 | 8079 | trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid, |
0b246afa | 8080 | fs_info->nodesize); |
5d4f98a2 YZ |
8081 | return ret; |
8082 | } | |
8083 | ||
8084 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 8085 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
8086 | u64 offset, u64 ram_bytes, |
8087 | struct btrfs_key *ins) | |
5d4f98a2 YZ |
8088 | { |
8089 | int ret; | |
8090 | ||
84f7d8e6 | 8091 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 8092 | |
fd708b81 JB |
8093 | btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0, |
8094 | root->root_key.objectid, owner, offset, | |
8095 | BTRFS_ADD_DELAYED_EXTENT); | |
8096 | ||
88a979c6 | 8097 | ret = btrfs_add_delayed_data_ref(trans, ins->objectid, |
84f7d8e6 JB |
8098 | ins->offset, 0, |
8099 | root->root_key.objectid, owner, | |
7be07912 OS |
8100 | offset, ram_bytes, |
8101 | BTRFS_ADD_DELAYED_EXTENT, NULL, NULL); | |
e6dcd2dc CM |
8102 | return ret; |
8103 | } | |
e02119d5 CM |
8104 | |
8105 | /* | |
8106 | * this is used by the tree logging recovery code. It records that | |
8107 | * an extent has been allocated and makes sure to clear the free | |
8108 | * space cache bits as well | |
8109 | */ | |
5d4f98a2 | 8110 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
8111 | u64 root_objectid, u64 owner, u64 offset, |
8112 | struct btrfs_key *ins) | |
e02119d5 | 8113 | { |
61da2abf | 8114 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e02119d5 CM |
8115 | int ret; |
8116 | struct btrfs_block_group_cache *block_group; | |
ed7a6948 | 8117 | struct btrfs_space_info *space_info; |
11833d66 | 8118 | |
8c2a1a30 JB |
8119 | /* |
8120 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 8121 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 8122 | */ |
0b246afa | 8123 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
8124 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
8125 | ins->offset); | |
b50c6e25 | 8126 | if (ret) |
8c2a1a30 | 8127 | return ret; |
11833d66 YZ |
8128 | } |
8129 | ||
0b246afa | 8130 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
8131 | if (!block_group) |
8132 | return -EINVAL; | |
8133 | ||
ed7a6948 WX |
8134 | space_info = block_group->space_info; |
8135 | spin_lock(&space_info->lock); | |
8136 | spin_lock(&block_group->lock); | |
8137 | space_info->bytes_reserved += ins->offset; | |
8138 | block_group->reserved += ins->offset; | |
8139 | spin_unlock(&block_group->lock); | |
8140 | spin_unlock(&space_info->lock); | |
8141 | ||
ef89b824 NB |
8142 | ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner, |
8143 | offset, ins, 1); | |
b50c6e25 | 8144 | btrfs_put_block_group(block_group); |
e02119d5 CM |
8145 | return ret; |
8146 | } | |
8147 | ||
48a3b636 ES |
8148 | static struct extent_buffer * |
8149 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
bc877d28 | 8150 | u64 bytenr, int level, u64 owner) |
65b51a00 | 8151 | { |
0b246afa | 8152 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
8153 | struct extent_buffer *buf; |
8154 | ||
2ff7e61e | 8155 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8156 | if (IS_ERR(buf)) |
8157 | return buf; | |
8158 | ||
b72c3aba QW |
8159 | /* |
8160 | * Extra safety check in case the extent tree is corrupted and extent | |
8161 | * allocator chooses to use a tree block which is already used and | |
8162 | * locked. | |
8163 | */ | |
8164 | if (buf->lock_owner == current->pid) { | |
8165 | btrfs_err_rl(fs_info, | |
8166 | "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected", | |
8167 | buf->start, btrfs_header_owner(buf), current->pid); | |
8168 | free_extent_buffer(buf); | |
8169 | return ERR_PTR(-EUCLEAN); | |
8170 | } | |
8171 | ||
85d4e461 | 8172 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 8173 | btrfs_tree_lock(buf); |
7c302b49 | 8174 | clean_tree_block(fs_info, buf); |
3083ee2e | 8175 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
8176 | |
8177 | btrfs_set_lock_blocking(buf); | |
4db8c528 | 8178 | set_extent_buffer_uptodate(buf); |
b4ce94de | 8179 | |
bc877d28 NB |
8180 | memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header)); |
8181 | btrfs_set_header_level(buf, level); | |
8182 | btrfs_set_header_bytenr(buf, buf->start); | |
8183 | btrfs_set_header_generation(buf, trans->transid); | |
8184 | btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV); | |
8185 | btrfs_set_header_owner(buf, owner); | |
8186 | write_extent_buffer_fsid(buf, fs_info->fsid); | |
8187 | write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid); | |
d0c803c4 | 8188 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 8189 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
8190 | /* |
8191 | * we allow two log transactions at a time, use different | |
8192 | * EXENT bit to differentiate dirty pages. | |
8193 | */ | |
656f30db | 8194 | if (buf->log_index == 0) |
8cef4e16 YZ |
8195 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
8196 | buf->start + buf->len - 1, GFP_NOFS); | |
8197 | else | |
8198 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 8199 | buf->start + buf->len - 1); |
d0c803c4 | 8200 | } else { |
656f30db | 8201 | buf->log_index = -1; |
d0c803c4 | 8202 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 8203 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 8204 | } |
64c12921 | 8205 | trans->dirty = true; |
b4ce94de | 8206 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
8207 | return buf; |
8208 | } | |
8209 | ||
f0486c68 YZ |
8210 | static struct btrfs_block_rsv * |
8211 | use_block_rsv(struct btrfs_trans_handle *trans, | |
8212 | struct btrfs_root *root, u32 blocksize) | |
8213 | { | |
0b246afa | 8214 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8215 | struct btrfs_block_rsv *block_rsv; |
0b246afa | 8216 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
f0486c68 | 8217 | int ret; |
d88033db | 8218 | bool global_updated = false; |
f0486c68 YZ |
8219 | |
8220 | block_rsv = get_block_rsv(trans, root); | |
8221 | ||
b586b323 MX |
8222 | if (unlikely(block_rsv->size == 0)) |
8223 | goto try_reserve; | |
d88033db | 8224 | again: |
f0486c68 YZ |
8225 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
8226 | if (!ret) | |
8227 | return block_rsv; | |
8228 | ||
b586b323 MX |
8229 | if (block_rsv->failfast) |
8230 | return ERR_PTR(ret); | |
8231 | ||
d88033db MX |
8232 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
8233 | global_updated = true; | |
0b246afa | 8234 | update_global_block_rsv(fs_info); |
d88033db MX |
8235 | goto again; |
8236 | } | |
8237 | ||
0b246afa | 8238 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
b586b323 MX |
8239 | static DEFINE_RATELIMIT_STATE(_rs, |
8240 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
8241 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
8242 | if (__ratelimit(&_rs)) | |
8243 | WARN(1, KERN_DEBUG | |
efe120a0 | 8244 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
8245 | } |
8246 | try_reserve: | |
8247 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
8248 | BTRFS_RESERVE_NO_FLUSH); | |
8249 | if (!ret) | |
8250 | return block_rsv; | |
8251 | /* | |
8252 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
8253 | * the global reserve if its space type is the same as the global |
8254 | * reservation. | |
b586b323 | 8255 | */ |
5881cfc9 MX |
8256 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
8257 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
8258 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
8259 | if (!ret) | |
8260 | return global_rsv; | |
8261 | } | |
8262 | return ERR_PTR(ret); | |
f0486c68 YZ |
8263 | } |
8264 | ||
8c2a3ca2 JB |
8265 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
8266 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 | 8267 | { |
3a584174 | 8268 | block_rsv_add_bytes(block_rsv, blocksize, false); |
ff6bc37e | 8269 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL); |
f0486c68 YZ |
8270 | } |
8271 | ||
fec577fb | 8272 | /* |
f0486c68 | 8273 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 8274 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 8275 | */ |
4d75f8a9 | 8276 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
8277 | struct btrfs_root *root, |
8278 | u64 parent, u64 root_objectid, | |
8279 | const struct btrfs_disk_key *key, | |
8280 | int level, u64 hint, | |
8281 | u64 empty_size) | |
fec577fb | 8282 | { |
0b246afa | 8283 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 8284 | struct btrfs_key ins; |
f0486c68 | 8285 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 8286 | struct extent_buffer *buf; |
67b7859e | 8287 | struct btrfs_delayed_extent_op *extent_op; |
f0486c68 YZ |
8288 | u64 flags = 0; |
8289 | int ret; | |
0b246afa JM |
8290 | u32 blocksize = fs_info->nodesize; |
8291 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 8292 | |
05653ef3 | 8293 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 8294 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 8295 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
bc877d28 | 8296 | level, root_objectid); |
faa2dbf0 JB |
8297 | if (!IS_ERR(buf)) |
8298 | root->alloc_bytenr += blocksize; | |
8299 | return buf; | |
8300 | } | |
05653ef3 | 8301 | #endif |
fccb84c9 | 8302 | |
f0486c68 YZ |
8303 | block_rsv = use_block_rsv(trans, root, blocksize); |
8304 | if (IS_ERR(block_rsv)) | |
8305 | return ERR_CAST(block_rsv); | |
8306 | ||
18513091 | 8307 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 8308 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
8309 | if (ret) |
8310 | goto out_unuse; | |
55c69072 | 8311 | |
bc877d28 NB |
8312 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level, |
8313 | root_objectid); | |
67b7859e OS |
8314 | if (IS_ERR(buf)) { |
8315 | ret = PTR_ERR(buf); | |
8316 | goto out_free_reserved; | |
8317 | } | |
f0486c68 YZ |
8318 | |
8319 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8320 | if (parent == 0) | |
8321 | parent = ins.objectid; | |
8322 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8323 | } else | |
8324 | BUG_ON(parent > 0); | |
8325 | ||
8326 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 8327 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
8328 | if (!extent_op) { |
8329 | ret = -ENOMEM; | |
8330 | goto out_free_buf; | |
8331 | } | |
f0486c68 YZ |
8332 | if (key) |
8333 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8334 | else | |
8335 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8336 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
8337 | extent_op->update_key = skinny_metadata ? false : true; |
8338 | extent_op->update_flags = true; | |
8339 | extent_op->is_data = false; | |
b1c79e09 | 8340 | extent_op->level = level; |
f0486c68 | 8341 | |
fd708b81 JB |
8342 | btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent, |
8343 | root_objectid, level, 0, | |
8344 | BTRFS_ADD_DELAYED_EXTENT); | |
44e1c47d | 8345 | ret = btrfs_add_delayed_tree_ref(trans, ins.objectid, |
7be07912 OS |
8346 | ins.offset, parent, |
8347 | root_objectid, level, | |
67b7859e | 8348 | BTRFS_ADD_DELAYED_EXTENT, |
7be07912 | 8349 | extent_op, NULL, NULL); |
67b7859e OS |
8350 | if (ret) |
8351 | goto out_free_delayed; | |
f0486c68 | 8352 | } |
fec577fb | 8353 | return buf; |
67b7859e OS |
8354 | |
8355 | out_free_delayed: | |
8356 | btrfs_free_delayed_extent_op(extent_op); | |
8357 | out_free_buf: | |
8358 | free_extent_buffer(buf); | |
8359 | out_free_reserved: | |
2ff7e61e | 8360 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 8361 | out_unuse: |
0b246afa | 8362 | unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 8363 | return ERR_PTR(ret); |
fec577fb | 8364 | } |
a28ec197 | 8365 | |
2c47e605 YZ |
8366 | struct walk_control { |
8367 | u64 refs[BTRFS_MAX_LEVEL]; | |
8368 | u64 flags[BTRFS_MAX_LEVEL]; | |
8369 | struct btrfs_key update_progress; | |
8370 | int stage; | |
8371 | int level; | |
8372 | int shared_level; | |
8373 | int update_ref; | |
8374 | int keep_locks; | |
1c4850e2 YZ |
8375 | int reada_slot; |
8376 | int reada_count; | |
2c47e605 YZ |
8377 | }; |
8378 | ||
8379 | #define DROP_REFERENCE 1 | |
8380 | #define UPDATE_BACKREF 2 | |
8381 | ||
1c4850e2 YZ |
8382 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8383 | struct btrfs_root *root, | |
8384 | struct walk_control *wc, | |
8385 | struct btrfs_path *path) | |
6407bf6d | 8386 | { |
0b246afa | 8387 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8388 | u64 bytenr; |
8389 | u64 generation; | |
8390 | u64 refs; | |
94fcca9f | 8391 | u64 flags; |
5d4f98a2 | 8392 | u32 nritems; |
1c4850e2 YZ |
8393 | struct btrfs_key key; |
8394 | struct extent_buffer *eb; | |
6407bf6d | 8395 | int ret; |
1c4850e2 YZ |
8396 | int slot; |
8397 | int nread = 0; | |
6407bf6d | 8398 | |
1c4850e2 YZ |
8399 | if (path->slots[wc->level] < wc->reada_slot) { |
8400 | wc->reada_count = wc->reada_count * 2 / 3; | |
8401 | wc->reada_count = max(wc->reada_count, 2); | |
8402 | } else { | |
8403 | wc->reada_count = wc->reada_count * 3 / 2; | |
8404 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 8405 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 8406 | } |
7bb86316 | 8407 | |
1c4850e2 YZ |
8408 | eb = path->nodes[wc->level]; |
8409 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 8410 | |
1c4850e2 YZ |
8411 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8412 | if (nread >= wc->reada_count) | |
8413 | break; | |
bd56b302 | 8414 | |
2dd3e67b | 8415 | cond_resched(); |
1c4850e2 YZ |
8416 | bytenr = btrfs_node_blockptr(eb, slot); |
8417 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8418 | |
1c4850e2 YZ |
8419 | if (slot == path->slots[wc->level]) |
8420 | goto reada; | |
5d4f98a2 | 8421 | |
1c4850e2 YZ |
8422 | if (wc->stage == UPDATE_BACKREF && |
8423 | generation <= root->root_key.offset) | |
bd56b302 CM |
8424 | continue; |
8425 | ||
94fcca9f | 8426 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 8427 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
8428 | wc->level - 1, 1, &refs, |
8429 | &flags); | |
79787eaa JM |
8430 | /* We don't care about errors in readahead. */ |
8431 | if (ret < 0) | |
8432 | continue; | |
94fcca9f YZ |
8433 | BUG_ON(refs == 0); |
8434 | ||
1c4850e2 | 8435 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8436 | if (refs == 1) |
8437 | goto reada; | |
bd56b302 | 8438 | |
94fcca9f YZ |
8439 | if (wc->level == 1 && |
8440 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8441 | continue; | |
1c4850e2 YZ |
8442 | if (!wc->update_ref || |
8443 | generation <= root->root_key.offset) | |
8444 | continue; | |
8445 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8446 | ret = btrfs_comp_cpu_keys(&key, | |
8447 | &wc->update_progress); | |
8448 | if (ret < 0) | |
8449 | continue; | |
94fcca9f YZ |
8450 | } else { |
8451 | if (wc->level == 1 && | |
8452 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8453 | continue; | |
6407bf6d | 8454 | } |
1c4850e2 | 8455 | reada: |
2ff7e61e | 8456 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 8457 | nread++; |
20524f02 | 8458 | } |
1c4850e2 | 8459 | wc->reada_slot = slot; |
20524f02 | 8460 | } |
2c47e605 | 8461 | |
f82d02d9 | 8462 | /* |
2c016dc2 | 8463 | * helper to process tree block while walking down the tree. |
2c47e605 | 8464 | * |
2c47e605 YZ |
8465 | * when wc->stage == UPDATE_BACKREF, this function updates |
8466 | * back refs for pointers in the block. | |
8467 | * | |
8468 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8469 | */ |
2c47e605 | 8470 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8471 | struct btrfs_root *root, |
2c47e605 | 8472 | struct btrfs_path *path, |
94fcca9f | 8473 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8474 | { |
2ff7e61e | 8475 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
8476 | int level = wc->level; |
8477 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 8478 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
8479 | int ret; |
8480 | ||
2c47e605 YZ |
8481 | if (wc->stage == UPDATE_BACKREF && |
8482 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8483 | return 1; | |
f82d02d9 | 8484 | |
2c47e605 YZ |
8485 | /* |
8486 | * when reference count of tree block is 1, it won't increase | |
8487 | * again. once full backref flag is set, we never clear it. | |
8488 | */ | |
94fcca9f YZ |
8489 | if (lookup_info && |
8490 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8491 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 8492 | BUG_ON(!path->locks[level]); |
2ff7e61e | 8493 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8494 | eb->start, level, 1, |
2c47e605 YZ |
8495 | &wc->refs[level], |
8496 | &wc->flags[level]); | |
79787eaa JM |
8497 | BUG_ON(ret == -ENOMEM); |
8498 | if (ret) | |
8499 | return ret; | |
2c47e605 YZ |
8500 | BUG_ON(wc->refs[level] == 0); |
8501 | } | |
5d4f98a2 | 8502 | |
2c47e605 YZ |
8503 | if (wc->stage == DROP_REFERENCE) { |
8504 | if (wc->refs[level] > 1) | |
8505 | return 1; | |
f82d02d9 | 8506 | |
2c47e605 | 8507 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 8508 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8509 | path->locks[level] = 0; |
8510 | } | |
8511 | return 0; | |
8512 | } | |
f82d02d9 | 8513 | |
2c47e605 YZ |
8514 | /* wc->stage == UPDATE_BACKREF */ |
8515 | if (!(wc->flags[level] & flag)) { | |
8516 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 8517 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 8518 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 8519 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8520 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8521 | ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start, |
b1c79e09 JB |
8522 | eb->len, flag, |
8523 | btrfs_header_level(eb), 0); | |
79787eaa | 8524 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
8525 | wc->flags[level] |= flag; |
8526 | } | |
8527 | ||
8528 | /* | |
8529 | * the block is shared by multiple trees, so it's not good to | |
8530 | * keep the tree lock | |
8531 | */ | |
8532 | if (path->locks[level] && level > 0) { | |
bd681513 | 8533 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8534 | path->locks[level] = 0; |
8535 | } | |
8536 | return 0; | |
8537 | } | |
8538 | ||
1c4850e2 | 8539 | /* |
2c016dc2 | 8540 | * helper to process tree block pointer. |
1c4850e2 YZ |
8541 | * |
8542 | * when wc->stage == DROP_REFERENCE, this function checks | |
8543 | * reference count of the block pointed to. if the block | |
8544 | * is shared and we need update back refs for the subtree | |
8545 | * rooted at the block, this function changes wc->stage to | |
8546 | * UPDATE_BACKREF. if the block is shared and there is no | |
8547 | * need to update back, this function drops the reference | |
8548 | * to the block. | |
8549 | * | |
8550 | * NOTE: return value 1 means we should stop walking down. | |
8551 | */ | |
8552 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8553 | struct btrfs_root *root, | |
8554 | struct btrfs_path *path, | |
94fcca9f | 8555 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 8556 | { |
0b246afa | 8557 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8558 | u64 bytenr; |
8559 | u64 generation; | |
8560 | u64 parent; | |
8561 | u32 blocksize; | |
8562 | struct btrfs_key key; | |
581c1760 | 8563 | struct btrfs_key first_key; |
1c4850e2 YZ |
8564 | struct extent_buffer *next; |
8565 | int level = wc->level; | |
8566 | int reada = 0; | |
8567 | int ret = 0; | |
1152651a | 8568 | bool need_account = false; |
1c4850e2 YZ |
8569 | |
8570 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8571 | path->slots[level]); | |
8572 | /* | |
8573 | * if the lower level block was created before the snapshot | |
8574 | * was created, we know there is no need to update back refs | |
8575 | * for the subtree | |
8576 | */ | |
8577 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8578 | generation <= root->root_key.offset) { |
8579 | *lookup_info = 1; | |
1c4850e2 | 8580 | return 1; |
94fcca9f | 8581 | } |
1c4850e2 YZ |
8582 | |
8583 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
581c1760 QW |
8584 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
8585 | path->slots[level]); | |
0b246afa | 8586 | blocksize = fs_info->nodesize; |
1c4850e2 | 8587 | |
0b246afa | 8588 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 8589 | if (!next) { |
2ff7e61e | 8590 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8591 | if (IS_ERR(next)) |
8592 | return PTR_ERR(next); | |
8593 | ||
b2aaaa3b JB |
8594 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8595 | level - 1); | |
1c4850e2 YZ |
8596 | reada = 1; |
8597 | } | |
8598 | btrfs_tree_lock(next); | |
8599 | btrfs_set_lock_blocking(next); | |
8600 | ||
2ff7e61e | 8601 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
8602 | &wc->refs[level - 1], |
8603 | &wc->flags[level - 1]); | |
4867268c JB |
8604 | if (ret < 0) |
8605 | goto out_unlock; | |
79787eaa | 8606 | |
c2cf52eb | 8607 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 8608 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
8609 | ret = -EIO; |
8610 | goto out_unlock; | |
c2cf52eb | 8611 | } |
94fcca9f | 8612 | *lookup_info = 0; |
1c4850e2 | 8613 | |
94fcca9f | 8614 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8615 | if (wc->refs[level - 1] > 1) { |
1152651a | 8616 | need_account = true; |
94fcca9f YZ |
8617 | if (level == 1 && |
8618 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8619 | goto skip; | |
8620 | ||
1c4850e2 YZ |
8621 | if (!wc->update_ref || |
8622 | generation <= root->root_key.offset) | |
8623 | goto skip; | |
8624 | ||
8625 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8626 | path->slots[level]); | |
8627 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8628 | if (ret < 0) | |
8629 | goto skip; | |
8630 | ||
8631 | wc->stage = UPDATE_BACKREF; | |
8632 | wc->shared_level = level - 1; | |
8633 | } | |
94fcca9f YZ |
8634 | } else { |
8635 | if (level == 1 && | |
8636 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8637 | goto skip; | |
1c4850e2 YZ |
8638 | } |
8639 | ||
b9fab919 | 8640 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8641 | btrfs_tree_unlock(next); |
8642 | free_extent_buffer(next); | |
8643 | next = NULL; | |
94fcca9f | 8644 | *lookup_info = 1; |
1c4850e2 YZ |
8645 | } |
8646 | ||
8647 | if (!next) { | |
8648 | if (reada && level == 1) | |
8649 | reada_walk_down(trans, root, wc, path); | |
581c1760 QW |
8650 | next = read_tree_block(fs_info, bytenr, generation, level - 1, |
8651 | &first_key); | |
64c043de LB |
8652 | if (IS_ERR(next)) { |
8653 | return PTR_ERR(next); | |
8654 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8655 | free_extent_buffer(next); |
97d9a8a4 | 8656 | return -EIO; |
416bc658 | 8657 | } |
1c4850e2 YZ |
8658 | btrfs_tree_lock(next); |
8659 | btrfs_set_lock_blocking(next); | |
8660 | } | |
8661 | ||
8662 | level--; | |
4867268c JB |
8663 | ASSERT(level == btrfs_header_level(next)); |
8664 | if (level != btrfs_header_level(next)) { | |
8665 | btrfs_err(root->fs_info, "mismatched level"); | |
8666 | ret = -EIO; | |
8667 | goto out_unlock; | |
8668 | } | |
1c4850e2 YZ |
8669 | path->nodes[level] = next; |
8670 | path->slots[level] = 0; | |
bd681513 | 8671 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8672 | wc->level = level; |
8673 | if (wc->level == 1) | |
8674 | wc->reada_slot = 0; | |
8675 | return 0; | |
8676 | skip: | |
8677 | wc->refs[level - 1] = 0; | |
8678 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
8679 | if (wc->stage == DROP_REFERENCE) { |
8680 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8681 | parent = path->nodes[level]->start; | |
8682 | } else { | |
4867268c | 8683 | ASSERT(root->root_key.objectid == |
94fcca9f | 8684 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
8685 | if (root->root_key.objectid != |
8686 | btrfs_header_owner(path->nodes[level])) { | |
8687 | btrfs_err(root->fs_info, | |
8688 | "mismatched block owner"); | |
8689 | ret = -EIO; | |
8690 | goto out_unlock; | |
8691 | } | |
94fcca9f YZ |
8692 | parent = 0; |
8693 | } | |
1c4850e2 | 8694 | |
2cd86d30 QW |
8695 | /* |
8696 | * Reloc tree doesn't contribute to qgroup numbers, and we have | |
8697 | * already accounted them at merge time (replace_path), | |
8698 | * thus we could skip expensive subtree trace here. | |
8699 | */ | |
8700 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
8701 | need_account) { | |
deb40627 | 8702 | ret = btrfs_qgroup_trace_subtree(trans, next, |
33d1f05c | 8703 | generation, level - 1); |
1152651a | 8704 | if (ret) { |
0b246afa | 8705 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8706 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
8707 | ret); | |
1152651a MF |
8708 | } |
8709 | } | |
84f7d8e6 | 8710 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, |
2ff7e61e JM |
8711 | parent, root->root_key.objectid, |
8712 | level - 1, 0); | |
4867268c JB |
8713 | if (ret) |
8714 | goto out_unlock; | |
1c4850e2 | 8715 | } |
4867268c JB |
8716 | |
8717 | *lookup_info = 1; | |
8718 | ret = 1; | |
8719 | ||
8720 | out_unlock: | |
1c4850e2 YZ |
8721 | btrfs_tree_unlock(next); |
8722 | free_extent_buffer(next); | |
4867268c JB |
8723 | |
8724 | return ret; | |
1c4850e2 YZ |
8725 | } |
8726 | ||
2c47e605 | 8727 | /* |
2c016dc2 | 8728 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
8729 | * |
8730 | * when wc->stage == DROP_REFERENCE, this function drops | |
8731 | * reference count on the block. | |
8732 | * | |
8733 | * when wc->stage == UPDATE_BACKREF, this function changes | |
8734 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
8735 | * to UPDATE_BACKREF previously while processing the block. | |
8736 | * | |
8737 | * NOTE: return value 1 means we should stop walking up. | |
8738 | */ | |
8739 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
8740 | struct btrfs_root *root, | |
8741 | struct btrfs_path *path, | |
8742 | struct walk_control *wc) | |
8743 | { | |
0b246afa | 8744 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8745 | int ret; |
2c47e605 YZ |
8746 | int level = wc->level; |
8747 | struct extent_buffer *eb = path->nodes[level]; | |
8748 | u64 parent = 0; | |
8749 | ||
8750 | if (wc->stage == UPDATE_BACKREF) { | |
8751 | BUG_ON(wc->shared_level < level); | |
8752 | if (level < wc->shared_level) | |
8753 | goto out; | |
8754 | ||
2c47e605 YZ |
8755 | ret = find_next_key(path, level + 1, &wc->update_progress); |
8756 | if (ret > 0) | |
8757 | wc->update_ref = 0; | |
8758 | ||
8759 | wc->stage = DROP_REFERENCE; | |
8760 | wc->shared_level = -1; | |
8761 | path->slots[level] = 0; | |
8762 | ||
8763 | /* | |
8764 | * check reference count again if the block isn't locked. | |
8765 | * we should start walking down the tree again if reference | |
8766 | * count is one. | |
8767 | */ | |
8768 | if (!path->locks[level]) { | |
8769 | BUG_ON(level == 0); | |
8770 | btrfs_tree_lock(eb); | |
8771 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8772 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8773 | |
2ff7e61e | 8774 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8775 | eb->start, level, 1, |
2c47e605 YZ |
8776 | &wc->refs[level], |
8777 | &wc->flags[level]); | |
79787eaa JM |
8778 | if (ret < 0) { |
8779 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 8780 | path->locks[level] = 0; |
79787eaa JM |
8781 | return ret; |
8782 | } | |
2c47e605 YZ |
8783 | BUG_ON(wc->refs[level] == 0); |
8784 | if (wc->refs[level] == 1) { | |
bd681513 | 8785 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 8786 | path->locks[level] = 0; |
2c47e605 YZ |
8787 | return 1; |
8788 | } | |
f82d02d9 | 8789 | } |
2c47e605 | 8790 | } |
f82d02d9 | 8791 | |
2c47e605 YZ |
8792 | /* wc->stage == DROP_REFERENCE */ |
8793 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 8794 | |
2c47e605 YZ |
8795 | if (wc->refs[level] == 1) { |
8796 | if (level == 0) { | |
8797 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 8798 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 8799 | else |
e339a6b0 | 8800 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8801 | BUG_ON(ret); /* -ENOMEM */ |
8d38d7eb | 8802 | ret = btrfs_qgroup_trace_leaf_items(trans, eb); |
1152651a | 8803 | if (ret) { |
0b246afa | 8804 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8805 | "error %d accounting leaf items. Quota is out of sync, rescan required.", |
8806 | ret); | |
1152651a | 8807 | } |
2c47e605 YZ |
8808 | } |
8809 | /* make block locked assertion in clean_tree_block happy */ | |
8810 | if (!path->locks[level] && | |
8811 | btrfs_header_generation(eb) == trans->transid) { | |
8812 | btrfs_tree_lock(eb); | |
8813 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8814 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8815 | } |
7c302b49 | 8816 | clean_tree_block(fs_info, eb); |
2c47e605 YZ |
8817 | } |
8818 | ||
8819 | if (eb == root->node) { | |
8820 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8821 | parent = eb->start; | |
65c6e82b QW |
8822 | else if (root->root_key.objectid != btrfs_header_owner(eb)) |
8823 | goto owner_mismatch; | |
2c47e605 YZ |
8824 | } else { |
8825 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8826 | parent = path->nodes[level + 1]->start; | |
65c6e82b QW |
8827 | else if (root->root_key.objectid != |
8828 | btrfs_header_owner(path->nodes[level + 1])) | |
8829 | goto owner_mismatch; | |
f82d02d9 | 8830 | } |
f82d02d9 | 8831 | |
5581a51a | 8832 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
8833 | out: |
8834 | wc->refs[level] = 0; | |
8835 | wc->flags[level] = 0; | |
f0486c68 | 8836 | return 0; |
65c6e82b QW |
8837 | |
8838 | owner_mismatch: | |
8839 | btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu", | |
8840 | btrfs_header_owner(eb), root->root_key.objectid); | |
8841 | return -EUCLEAN; | |
2c47e605 YZ |
8842 | } |
8843 | ||
8844 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
8845 | struct btrfs_root *root, | |
8846 | struct btrfs_path *path, | |
8847 | struct walk_control *wc) | |
8848 | { | |
2c47e605 | 8849 | int level = wc->level; |
94fcca9f | 8850 | int lookup_info = 1; |
2c47e605 YZ |
8851 | int ret; |
8852 | ||
8853 | while (level >= 0) { | |
94fcca9f | 8854 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
8855 | if (ret > 0) |
8856 | break; | |
8857 | ||
8858 | if (level == 0) | |
8859 | break; | |
8860 | ||
7a7965f8 YZ |
8861 | if (path->slots[level] >= |
8862 | btrfs_header_nritems(path->nodes[level])) | |
8863 | break; | |
8864 | ||
94fcca9f | 8865 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
8866 | if (ret > 0) { |
8867 | path->slots[level]++; | |
8868 | continue; | |
90d2c51d MX |
8869 | } else if (ret < 0) |
8870 | return ret; | |
1c4850e2 | 8871 | level = wc->level; |
f82d02d9 | 8872 | } |
f82d02d9 YZ |
8873 | return 0; |
8874 | } | |
8875 | ||
d397712b | 8876 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 8877 | struct btrfs_root *root, |
f82d02d9 | 8878 | struct btrfs_path *path, |
2c47e605 | 8879 | struct walk_control *wc, int max_level) |
20524f02 | 8880 | { |
2c47e605 | 8881 | int level = wc->level; |
20524f02 | 8882 | int ret; |
9f3a7427 | 8883 | |
2c47e605 YZ |
8884 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
8885 | while (level < max_level && path->nodes[level]) { | |
8886 | wc->level = level; | |
8887 | if (path->slots[level] + 1 < | |
8888 | btrfs_header_nritems(path->nodes[level])) { | |
8889 | path->slots[level]++; | |
20524f02 CM |
8890 | return 0; |
8891 | } else { | |
2c47e605 YZ |
8892 | ret = walk_up_proc(trans, root, path, wc); |
8893 | if (ret > 0) | |
8894 | return 0; | |
65c6e82b QW |
8895 | if (ret < 0) |
8896 | return ret; | |
bd56b302 | 8897 | |
2c47e605 | 8898 | if (path->locks[level]) { |
bd681513 CM |
8899 | btrfs_tree_unlock_rw(path->nodes[level], |
8900 | path->locks[level]); | |
2c47e605 | 8901 | path->locks[level] = 0; |
f82d02d9 | 8902 | } |
2c47e605 YZ |
8903 | free_extent_buffer(path->nodes[level]); |
8904 | path->nodes[level] = NULL; | |
8905 | level++; | |
20524f02 CM |
8906 | } |
8907 | } | |
8908 | return 1; | |
8909 | } | |
8910 | ||
9aca1d51 | 8911 | /* |
2c47e605 YZ |
8912 | * drop a subvolume tree. |
8913 | * | |
8914 | * this function traverses the tree freeing any blocks that only | |
8915 | * referenced by the tree. | |
8916 | * | |
8917 | * when a shared tree block is found. this function decreases its | |
8918 | * reference count by one. if update_ref is true, this function | |
8919 | * also make sure backrefs for the shared block and all lower level | |
8920 | * blocks are properly updated. | |
9d1a2a3a DS |
8921 | * |
8922 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 8923 | */ |
2c536799 | 8924 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
8925 | struct btrfs_block_rsv *block_rsv, int update_ref, |
8926 | int for_reloc) | |
20524f02 | 8927 | { |
ab8d0fc4 | 8928 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 8929 | struct btrfs_path *path; |
2c47e605 | 8930 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 8931 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 8932 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
8933 | struct walk_control *wc; |
8934 | struct btrfs_key key; | |
8935 | int err = 0; | |
8936 | int ret; | |
8937 | int level; | |
d29a9f62 | 8938 | bool root_dropped = false; |
20524f02 | 8939 | |
4fd786e6 | 8940 | btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid); |
1152651a | 8941 | |
5caf2a00 | 8942 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
8943 | if (!path) { |
8944 | err = -ENOMEM; | |
8945 | goto out; | |
8946 | } | |
20524f02 | 8947 | |
2c47e605 | 8948 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
8949 | if (!wc) { |
8950 | btrfs_free_path(path); | |
cb1b69f4 TI |
8951 | err = -ENOMEM; |
8952 | goto out; | |
38a1a919 | 8953 | } |
2c47e605 | 8954 | |
a22285a6 | 8955 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
8956 | if (IS_ERR(trans)) { |
8957 | err = PTR_ERR(trans); | |
8958 | goto out_free; | |
8959 | } | |
98d5dc13 | 8960 | |
3fd0a558 YZ |
8961 | if (block_rsv) |
8962 | trans->block_rsv = block_rsv; | |
2c47e605 | 8963 | |
9f3a7427 | 8964 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 8965 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
8966 | path->nodes[level] = btrfs_lock_root_node(root); |
8967 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 8968 | path->slots[level] = 0; |
bd681513 | 8969 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
8970 | memset(&wc->update_progress, 0, |
8971 | sizeof(wc->update_progress)); | |
9f3a7427 | 8972 | } else { |
9f3a7427 | 8973 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
8974 | memcpy(&wc->update_progress, &key, |
8975 | sizeof(wc->update_progress)); | |
8976 | ||
6702ed49 | 8977 | level = root_item->drop_level; |
2c47e605 | 8978 | BUG_ON(level == 0); |
6702ed49 | 8979 | path->lowest_level = level; |
2c47e605 YZ |
8980 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
8981 | path->lowest_level = 0; | |
8982 | if (ret < 0) { | |
8983 | err = ret; | |
79787eaa | 8984 | goto out_end_trans; |
9f3a7427 | 8985 | } |
1c4850e2 | 8986 | WARN_ON(ret > 0); |
2c47e605 | 8987 | |
7d9eb12c CM |
8988 | /* |
8989 | * unlock our path, this is safe because only this | |
8990 | * function is allowed to delete this snapshot | |
8991 | */ | |
5d4f98a2 | 8992 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
8993 | |
8994 | level = btrfs_header_level(root->node); | |
8995 | while (1) { | |
8996 | btrfs_tree_lock(path->nodes[level]); | |
8997 | btrfs_set_lock_blocking(path->nodes[level]); | |
fec386ac | 8998 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8999 | |
2ff7e61e | 9000 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 9001 | path->nodes[level]->start, |
3173a18f | 9002 | level, 1, &wc->refs[level], |
2c47e605 | 9003 | &wc->flags[level]); |
79787eaa JM |
9004 | if (ret < 0) { |
9005 | err = ret; | |
9006 | goto out_end_trans; | |
9007 | } | |
2c47e605 YZ |
9008 | BUG_ON(wc->refs[level] == 0); |
9009 | ||
9010 | if (level == root_item->drop_level) | |
9011 | break; | |
9012 | ||
9013 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 9014 | path->locks[level] = 0; |
2c47e605 YZ |
9015 | WARN_ON(wc->refs[level] != 1); |
9016 | level--; | |
9017 | } | |
9f3a7427 | 9018 | } |
2c47e605 YZ |
9019 | |
9020 | wc->level = level; | |
9021 | wc->shared_level = -1; | |
9022 | wc->stage = DROP_REFERENCE; | |
9023 | wc->update_ref = update_ref; | |
9024 | wc->keep_locks = 0; | |
0b246afa | 9025 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 9026 | |
d397712b | 9027 | while (1) { |
9d1a2a3a | 9028 | |
2c47e605 YZ |
9029 | ret = walk_down_tree(trans, root, path, wc); |
9030 | if (ret < 0) { | |
9031 | err = ret; | |
20524f02 | 9032 | break; |
2c47e605 | 9033 | } |
9aca1d51 | 9034 | |
2c47e605 YZ |
9035 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
9036 | if (ret < 0) { | |
9037 | err = ret; | |
20524f02 | 9038 | break; |
2c47e605 YZ |
9039 | } |
9040 | ||
9041 | if (ret > 0) { | |
9042 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
9043 | break; |
9044 | } | |
2c47e605 YZ |
9045 | |
9046 | if (wc->stage == DROP_REFERENCE) { | |
9047 | level = wc->level; | |
9048 | btrfs_node_key(path->nodes[level], | |
9049 | &root_item->drop_progress, | |
9050 | path->slots[level]); | |
9051 | root_item->drop_level = level; | |
9052 | } | |
9053 | ||
9054 | BUG_ON(wc->level == 0); | |
3a45bb20 | 9055 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 9056 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
9057 | ret = btrfs_update_root(trans, tree_root, |
9058 | &root->root_key, | |
9059 | root_item); | |
79787eaa | 9060 | if (ret) { |
66642832 | 9061 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9062 | err = ret; |
9063 | goto out_end_trans; | |
9064 | } | |
2c47e605 | 9065 | |
3a45bb20 | 9066 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 9067 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
9068 | btrfs_debug(fs_info, |
9069 | "drop snapshot early exit"); | |
3c8f2422 JB |
9070 | err = -EAGAIN; |
9071 | goto out_free; | |
9072 | } | |
9073 | ||
a22285a6 | 9074 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9075 | if (IS_ERR(trans)) { |
9076 | err = PTR_ERR(trans); | |
9077 | goto out_free; | |
9078 | } | |
3fd0a558 YZ |
9079 | if (block_rsv) |
9080 | trans->block_rsv = block_rsv; | |
c3e69d58 | 9081 | } |
20524f02 | 9082 | } |
b3b4aa74 | 9083 | btrfs_release_path(path); |
79787eaa JM |
9084 | if (err) |
9085 | goto out_end_trans; | |
2c47e605 | 9086 | |
ab9ce7d4 | 9087 | ret = btrfs_del_root(trans, &root->root_key); |
79787eaa | 9088 | if (ret) { |
66642832 | 9089 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 9090 | err = ret; |
79787eaa JM |
9091 | goto out_end_trans; |
9092 | } | |
2c47e605 | 9093 | |
76dda93c | 9094 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
9095 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
9096 | NULL, NULL); | |
79787eaa | 9097 | if (ret < 0) { |
66642832 | 9098 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9099 | err = ret; |
9100 | goto out_end_trans; | |
9101 | } else if (ret > 0) { | |
84cd948c JB |
9102 | /* if we fail to delete the orphan item this time |
9103 | * around, it'll get picked up the next time. | |
9104 | * | |
9105 | * The most common failure here is just -ENOENT. | |
9106 | */ | |
9107 | btrfs_del_orphan_item(trans, tree_root, | |
9108 | root->root_key.objectid); | |
76dda93c YZ |
9109 | } |
9110 | } | |
9111 | ||
27cdeb70 | 9112 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 9113 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
9114 | } else { |
9115 | free_extent_buffer(root->node); | |
9116 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 9117 | btrfs_put_fs_root(root); |
76dda93c | 9118 | } |
d29a9f62 | 9119 | root_dropped = true; |
79787eaa | 9120 | out_end_trans: |
3a45bb20 | 9121 | btrfs_end_transaction_throttle(trans); |
79787eaa | 9122 | out_free: |
2c47e605 | 9123 | kfree(wc); |
5caf2a00 | 9124 | btrfs_free_path(path); |
cb1b69f4 | 9125 | out: |
d29a9f62 JB |
9126 | /* |
9127 | * So if we need to stop dropping the snapshot for whatever reason we | |
9128 | * need to make sure to add it back to the dead root list so that we | |
9129 | * keep trying to do the work later. This also cleans up roots if we | |
9130 | * don't have it in the radix (like when we recover after a power fail | |
9131 | * or unmount) so we don't leak memory. | |
9132 | */ | |
897ca819 | 9133 | if (!for_reloc && !root_dropped) |
d29a9f62 | 9134 | btrfs_add_dead_root(root); |
90515e7f | 9135 | if (err && err != -EAGAIN) |
ab8d0fc4 | 9136 | btrfs_handle_fs_error(fs_info, err, NULL); |
2c536799 | 9137 | return err; |
20524f02 | 9138 | } |
9078a3e1 | 9139 | |
2c47e605 YZ |
9140 | /* |
9141 | * drop subtree rooted at tree block 'node'. | |
9142 | * | |
9143 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 9144 | * only used by relocation code |
2c47e605 | 9145 | */ |
f82d02d9 YZ |
9146 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
9147 | struct btrfs_root *root, | |
9148 | struct extent_buffer *node, | |
9149 | struct extent_buffer *parent) | |
9150 | { | |
0b246afa | 9151 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 9152 | struct btrfs_path *path; |
2c47e605 | 9153 | struct walk_control *wc; |
f82d02d9 YZ |
9154 | int level; |
9155 | int parent_level; | |
9156 | int ret = 0; | |
9157 | int wret; | |
9158 | ||
2c47e605 YZ |
9159 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
9160 | ||
f82d02d9 | 9161 | path = btrfs_alloc_path(); |
db5b493a TI |
9162 | if (!path) |
9163 | return -ENOMEM; | |
f82d02d9 | 9164 | |
2c47e605 | 9165 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
9166 | if (!wc) { |
9167 | btrfs_free_path(path); | |
9168 | return -ENOMEM; | |
9169 | } | |
2c47e605 | 9170 | |
b9447ef8 | 9171 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
9172 | parent_level = btrfs_header_level(parent); |
9173 | extent_buffer_get(parent); | |
9174 | path->nodes[parent_level] = parent; | |
9175 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9176 | ||
b9447ef8 | 9177 | btrfs_assert_tree_locked(node); |
f82d02d9 | 9178 | level = btrfs_header_level(node); |
f82d02d9 YZ |
9179 | path->nodes[level] = node; |
9180 | path->slots[level] = 0; | |
bd681513 | 9181 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9182 | |
9183 | wc->refs[parent_level] = 1; | |
9184 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9185 | wc->level = level; | |
9186 | wc->shared_level = -1; | |
9187 | wc->stage = DROP_REFERENCE; | |
9188 | wc->update_ref = 0; | |
9189 | wc->keep_locks = 1; | |
0b246afa | 9190 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
9191 | |
9192 | while (1) { | |
2c47e605 YZ |
9193 | wret = walk_down_tree(trans, root, path, wc); |
9194 | if (wret < 0) { | |
f82d02d9 | 9195 | ret = wret; |
f82d02d9 | 9196 | break; |
2c47e605 | 9197 | } |
f82d02d9 | 9198 | |
2c47e605 | 9199 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9200 | if (wret < 0) |
9201 | ret = wret; | |
9202 | if (wret != 0) | |
9203 | break; | |
9204 | } | |
9205 | ||
2c47e605 | 9206 | kfree(wc); |
f82d02d9 YZ |
9207 | btrfs_free_path(path); |
9208 | return ret; | |
9209 | } | |
9210 | ||
6202df69 | 9211 | static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c CM |
9212 | { |
9213 | u64 num_devices; | |
fc67c450 | 9214 | u64 stripped; |
e4d8ec0f | 9215 | |
fc67c450 ID |
9216 | /* |
9217 | * if restripe for this chunk_type is on pick target profile and | |
9218 | * return, otherwise do the usual balance | |
9219 | */ | |
6202df69 | 9220 | stripped = get_restripe_target(fs_info, flags); |
fc67c450 ID |
9221 | if (stripped) |
9222 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9223 | |
6202df69 | 9224 | num_devices = fs_info->fs_devices->rw_devices; |
cd02dca5 | 9225 | |
fc67c450 | 9226 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 9227 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
9228 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
9229 | ||
ec44a35c CM |
9230 | if (num_devices == 1) { |
9231 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9232 | stripped = flags & ~stripped; | |
9233 | ||
9234 | /* turn raid0 into single device chunks */ | |
9235 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9236 | return stripped; | |
9237 | ||
9238 | /* turn mirroring into duplication */ | |
9239 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9240 | BTRFS_BLOCK_GROUP_RAID10)) | |
9241 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9242 | } else { |
9243 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9244 | if (flags & stripped) |
9245 | return flags; | |
9246 | ||
9247 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9248 | stripped = flags & ~stripped; | |
9249 | ||
9250 | /* switch duplicated blocks with raid1 */ | |
9251 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9252 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9253 | ||
e3176ca2 | 9254 | /* this is drive concat, leave it alone */ |
ec44a35c | 9255 | } |
e3176ca2 | 9256 | |
ec44a35c CM |
9257 | return flags; |
9258 | } | |
9259 | ||
868f401a | 9260 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9261 | { |
f0486c68 YZ |
9262 | struct btrfs_space_info *sinfo = cache->space_info; |
9263 | u64 num_bytes; | |
199c36ea | 9264 | u64 min_allocable_bytes; |
f0486c68 | 9265 | int ret = -ENOSPC; |
0ef3e66b | 9266 | |
199c36ea MX |
9267 | /* |
9268 | * We need some metadata space and system metadata space for | |
9269 | * allocating chunks in some corner cases until we force to set | |
9270 | * it to be readonly. | |
9271 | */ | |
9272 | if ((sinfo->flags & | |
9273 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9274 | !force) | |
ee22184b | 9275 | min_allocable_bytes = SZ_1M; |
199c36ea MX |
9276 | else |
9277 | min_allocable_bytes = 0; | |
9278 | ||
f0486c68 YZ |
9279 | spin_lock(&sinfo->lock); |
9280 | spin_lock(&cache->lock); | |
61cfea9b W |
9281 | |
9282 | if (cache->ro) { | |
868f401a | 9283 | cache->ro++; |
61cfea9b W |
9284 | ret = 0; |
9285 | goto out; | |
9286 | } | |
9287 | ||
f0486c68 YZ |
9288 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9289 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
9290 | ||
4136135b | 9291 | if (btrfs_space_info_used(sinfo, true) + num_bytes + |
37be25bc | 9292 | min_allocable_bytes <= sinfo->total_bytes) { |
f0486c68 | 9293 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9294 | cache->ro++; |
633c0aad | 9295 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9296 | ret = 0; |
9297 | } | |
61cfea9b | 9298 | out: |
f0486c68 YZ |
9299 | spin_unlock(&cache->lock); |
9300 | spin_unlock(&sinfo->lock); | |
9301 | return ret; | |
9302 | } | |
7d9eb12c | 9303 | |
c83488af | 9304 | int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache) |
c286ac48 | 9305 | |
f0486c68 | 9306 | { |
c83488af | 9307 | struct btrfs_fs_info *fs_info = cache->fs_info; |
f0486c68 YZ |
9308 | struct btrfs_trans_handle *trans; |
9309 | u64 alloc_flags; | |
9310 | int ret; | |
7d9eb12c | 9311 | |
1bbc621e | 9312 | again: |
5e00f193 | 9313 | trans = btrfs_join_transaction(fs_info->extent_root); |
79787eaa JM |
9314 | if (IS_ERR(trans)) |
9315 | return PTR_ERR(trans); | |
5d4f98a2 | 9316 | |
1bbc621e CM |
9317 | /* |
9318 | * we're not allowed to set block groups readonly after the dirty | |
9319 | * block groups cache has started writing. If it already started, | |
9320 | * back off and let this transaction commit | |
9321 | */ | |
0b246afa | 9322 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c | 9323 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9324 | u64 transid = trans->transid; |
9325 | ||
0b246afa | 9326 | mutex_unlock(&fs_info->ro_block_group_mutex); |
3a45bb20 | 9327 | btrfs_end_transaction(trans); |
1bbc621e | 9328 | |
2ff7e61e | 9329 | ret = btrfs_wait_for_commit(fs_info, transid); |
1bbc621e CM |
9330 | if (ret) |
9331 | return ret; | |
9332 | goto again; | |
9333 | } | |
9334 | ||
153c35b6 CM |
9335 | /* |
9336 | * if we are changing raid levels, try to allocate a corresponding | |
9337 | * block group with the new raid level. | |
9338 | */ | |
0b246afa | 9339 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
153c35b6 | 9340 | if (alloc_flags != cache->flags) { |
01458828 | 9341 | ret = do_chunk_alloc(trans, alloc_flags, |
153c35b6 CM |
9342 | CHUNK_ALLOC_FORCE); |
9343 | /* | |
9344 | * ENOSPC is allowed here, we may have enough space | |
9345 | * already allocated at the new raid level to | |
9346 | * carry on | |
9347 | */ | |
9348 | if (ret == -ENOSPC) | |
9349 | ret = 0; | |
9350 | if (ret < 0) | |
9351 | goto out; | |
9352 | } | |
1bbc621e | 9353 | |
868f401a | 9354 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9355 | if (!ret) |
9356 | goto out; | |
2ff7e61e | 9357 | alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags); |
01458828 | 9358 | ret = do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9359 | if (ret < 0) |
9360 | goto out; | |
868f401a | 9361 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9362 | out: |
2f081088 | 9363 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 9364 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
34441361 | 9365 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 9366 | check_system_chunk(trans, alloc_flags); |
34441361 | 9367 | mutex_unlock(&fs_info->chunk_mutex); |
2f081088 | 9368 | } |
0b246afa | 9369 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2f081088 | 9370 | |
3a45bb20 | 9371 | btrfs_end_transaction(trans); |
f0486c68 YZ |
9372 | return ret; |
9373 | } | |
5d4f98a2 | 9374 | |
43a7e99d | 9375 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type) |
c87f08ca | 9376 | { |
43a7e99d | 9377 | u64 alloc_flags = get_alloc_profile(trans->fs_info, type); |
2ff7e61e | 9378 | |
01458828 | 9379 | return do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); |
c87f08ca CM |
9380 | } |
9381 | ||
6d07bcec MX |
9382 | /* |
9383 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9384 | * space_info. takes mirrors into account. |
6d07bcec | 9385 | */ |
633c0aad | 9386 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9387 | { |
9388 | struct btrfs_block_group_cache *block_group; | |
9389 | u64 free_bytes = 0; | |
9390 | int factor; | |
9391 | ||
01327610 | 9392 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
9393 | if (list_empty(&sinfo->ro_bgs)) |
9394 | return 0; | |
9395 | ||
9396 | spin_lock(&sinfo->lock); | |
9397 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9398 | spin_lock(&block_group->lock); |
9399 | ||
9400 | if (!block_group->ro) { | |
9401 | spin_unlock(&block_group->lock); | |
9402 | continue; | |
9403 | } | |
9404 | ||
46df06b8 | 9405 | factor = btrfs_bg_type_to_factor(block_group->flags); |
6d07bcec MX |
9406 | free_bytes += (block_group->key.offset - |
9407 | btrfs_block_group_used(&block_group->item)) * | |
9408 | factor; | |
9409 | ||
9410 | spin_unlock(&block_group->lock); | |
9411 | } | |
6d07bcec MX |
9412 | spin_unlock(&sinfo->lock); |
9413 | ||
9414 | return free_bytes; | |
9415 | } | |
9416 | ||
2ff7e61e | 9417 | void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9418 | { |
f0486c68 YZ |
9419 | struct btrfs_space_info *sinfo = cache->space_info; |
9420 | u64 num_bytes; | |
9421 | ||
9422 | BUG_ON(!cache->ro); | |
9423 | ||
9424 | spin_lock(&sinfo->lock); | |
9425 | spin_lock(&cache->lock); | |
868f401a Z |
9426 | if (!--cache->ro) { |
9427 | num_bytes = cache->key.offset - cache->reserved - | |
9428 | cache->pinned - cache->bytes_super - | |
9429 | btrfs_block_group_used(&cache->item); | |
9430 | sinfo->bytes_readonly -= num_bytes; | |
9431 | list_del_init(&cache->ro_list); | |
9432 | } | |
f0486c68 YZ |
9433 | spin_unlock(&cache->lock); |
9434 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9435 | } |
9436 | ||
ba1bf481 JB |
9437 | /* |
9438 | * checks to see if its even possible to relocate this block group. | |
9439 | * | |
9440 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9441 | * ok to go ahead and try. | |
9442 | */ | |
6bccf3ab | 9443 | int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr) |
1a40e23b | 9444 | { |
6bccf3ab | 9445 | struct btrfs_root *root = fs_info->extent_root; |
ba1bf481 JB |
9446 | struct btrfs_block_group_cache *block_group; |
9447 | struct btrfs_space_info *space_info; | |
0b246afa | 9448 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
ba1bf481 | 9449 | struct btrfs_device *device; |
6df9a95e | 9450 | struct btrfs_trans_handle *trans; |
cdcb725c | 9451 | u64 min_free; |
6719db6a JB |
9452 | u64 dev_min = 1; |
9453 | u64 dev_nr = 0; | |
4a5e98f5 | 9454 | u64 target; |
0305bc27 | 9455 | int debug; |
cdcb725c | 9456 | int index; |
ba1bf481 JB |
9457 | int full = 0; |
9458 | int ret = 0; | |
1a40e23b | 9459 | |
0b246afa | 9460 | debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG); |
0305bc27 | 9461 | |
0b246afa | 9462 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
1a40e23b | 9463 | |
ba1bf481 | 9464 | /* odd, couldn't find the block group, leave it alone */ |
0305bc27 QW |
9465 | if (!block_group) { |
9466 | if (debug) | |
0b246afa | 9467 | btrfs_warn(fs_info, |
0305bc27 QW |
9468 | "can't find block group for bytenr %llu", |
9469 | bytenr); | |
ba1bf481 | 9470 | return -1; |
0305bc27 | 9471 | } |
1a40e23b | 9472 | |
cdcb725c | 9473 | min_free = btrfs_block_group_used(&block_group->item); |
9474 | ||
ba1bf481 | 9475 | /* no bytes used, we're good */ |
cdcb725c | 9476 | if (!min_free) |
1a40e23b ZY |
9477 | goto out; |
9478 | ||
ba1bf481 JB |
9479 | space_info = block_group->space_info; |
9480 | spin_lock(&space_info->lock); | |
17d217fe | 9481 | |
ba1bf481 | 9482 | full = space_info->full; |
17d217fe | 9483 | |
ba1bf481 JB |
9484 | /* |
9485 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9486 | * relocate it unless we're able to allocate a new chunk below. |
9487 | * | |
9488 | * Otherwise, we need to make sure we have room in the space to handle | |
9489 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9490 | */ |
7ce618db | 9491 | if ((space_info->total_bytes != block_group->key.offset) && |
4136135b LB |
9492 | (btrfs_space_info_used(space_info, false) + min_free < |
9493 | space_info->total_bytes)) { | |
ba1bf481 JB |
9494 | spin_unlock(&space_info->lock); |
9495 | goto out; | |
17d217fe | 9496 | } |
ba1bf481 | 9497 | spin_unlock(&space_info->lock); |
ea8c2819 | 9498 | |
ba1bf481 JB |
9499 | /* |
9500 | * ok we don't have enough space, but maybe we have free space on our | |
9501 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9502 | * alloc devices and guess if we have enough space. if this block |
9503 | * group is going to be restriped, run checks against the target | |
9504 | * profile instead of the current one. | |
ba1bf481 JB |
9505 | */ |
9506 | ret = -1; | |
ea8c2819 | 9507 | |
cdcb725c | 9508 | /* |
9509 | * index: | |
9510 | * 0: raid10 | |
9511 | * 1: raid1 | |
9512 | * 2: dup | |
9513 | * 3: raid0 | |
9514 | * 4: single | |
9515 | */ | |
0b246afa | 9516 | target = get_restripe_target(fs_info, block_group->flags); |
4a5e98f5 | 9517 | if (target) { |
3e72ee88 | 9518 | index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9519 | } else { |
9520 | /* | |
9521 | * this is just a balance, so if we were marked as full | |
9522 | * we know there is no space for a new chunk | |
9523 | */ | |
0305bc27 QW |
9524 | if (full) { |
9525 | if (debug) | |
0b246afa JM |
9526 | btrfs_warn(fs_info, |
9527 | "no space to alloc new chunk for block group %llu", | |
9528 | block_group->key.objectid); | |
4a5e98f5 | 9529 | goto out; |
0305bc27 | 9530 | } |
4a5e98f5 | 9531 | |
3e72ee88 | 9532 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
4a5e98f5 ID |
9533 | } |
9534 | ||
e6ec716f | 9535 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9536 | dev_min = 4; |
6719db6a JB |
9537 | /* Divide by 2 */ |
9538 | min_free >>= 1; | |
e6ec716f | 9539 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9540 | dev_min = 2; |
e6ec716f | 9541 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9542 | /* Multiply by 2 */ |
9543 | min_free <<= 1; | |
e6ec716f | 9544 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9545 | dev_min = fs_devices->rw_devices; |
47c5713f | 9546 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9547 | } |
9548 | ||
6df9a95e JB |
9549 | /* We need to do this so that we can look at pending chunks */ |
9550 | trans = btrfs_join_transaction(root); | |
9551 | if (IS_ERR(trans)) { | |
9552 | ret = PTR_ERR(trans); | |
9553 | goto out; | |
9554 | } | |
9555 | ||
0b246afa | 9556 | mutex_lock(&fs_info->chunk_mutex); |
ba1bf481 | 9557 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
7bfc837d | 9558 | u64 dev_offset; |
56bec294 | 9559 | |
ba1bf481 JB |
9560 | /* |
9561 | * check to make sure we can actually find a chunk with enough | |
9562 | * space to fit our block group in. | |
9563 | */ | |
63a212ab | 9564 | if (device->total_bytes > device->bytes_used + min_free && |
401e29c1 | 9565 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
6df9a95e | 9566 | ret = find_free_dev_extent(trans, device, min_free, |
7bfc837d | 9567 | &dev_offset, NULL); |
ba1bf481 | 9568 | if (!ret) |
cdcb725c | 9569 | dev_nr++; |
9570 | ||
9571 | if (dev_nr >= dev_min) | |
73e48b27 | 9572 | break; |
cdcb725c | 9573 | |
ba1bf481 | 9574 | ret = -1; |
725c8463 | 9575 | } |
edbd8d4e | 9576 | } |
0305bc27 | 9577 | if (debug && ret == -1) |
0b246afa JM |
9578 | btrfs_warn(fs_info, |
9579 | "no space to allocate a new chunk for block group %llu", | |
9580 | block_group->key.objectid); | |
9581 | mutex_unlock(&fs_info->chunk_mutex); | |
3a45bb20 | 9582 | btrfs_end_transaction(trans); |
edbd8d4e | 9583 | out: |
ba1bf481 | 9584 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9585 | return ret; |
9586 | } | |
9587 | ||
6bccf3ab JM |
9588 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
9589 | struct btrfs_path *path, | |
9590 | struct btrfs_key *key) | |
0b86a832 | 9591 | { |
6bccf3ab | 9592 | struct btrfs_root *root = fs_info->extent_root; |
925baedd | 9593 | int ret = 0; |
0b86a832 CM |
9594 | struct btrfs_key found_key; |
9595 | struct extent_buffer *leaf; | |
514c7dca QW |
9596 | struct btrfs_block_group_item bg; |
9597 | u64 flags; | |
0b86a832 | 9598 | int slot; |
edbd8d4e | 9599 | |
0b86a832 CM |
9600 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9601 | if (ret < 0) | |
925baedd CM |
9602 | goto out; |
9603 | ||
d397712b | 9604 | while (1) { |
0b86a832 | 9605 | slot = path->slots[0]; |
edbd8d4e | 9606 | leaf = path->nodes[0]; |
0b86a832 CM |
9607 | if (slot >= btrfs_header_nritems(leaf)) { |
9608 | ret = btrfs_next_leaf(root, path); | |
9609 | if (ret == 0) | |
9610 | continue; | |
9611 | if (ret < 0) | |
925baedd | 9612 | goto out; |
0b86a832 | 9613 | break; |
edbd8d4e | 9614 | } |
0b86a832 | 9615 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9616 | |
0b86a832 | 9617 | if (found_key.objectid >= key->objectid && |
925baedd | 9618 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6fb37b75 LB |
9619 | struct extent_map_tree *em_tree; |
9620 | struct extent_map *em; | |
9621 | ||
9622 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
9623 | read_lock(&em_tree->lock); | |
9624 | em = lookup_extent_mapping(em_tree, found_key.objectid, | |
9625 | found_key.offset); | |
9626 | read_unlock(&em_tree->lock); | |
9627 | if (!em) { | |
0b246afa | 9628 | btrfs_err(fs_info, |
6fb37b75 LB |
9629 | "logical %llu len %llu found bg but no related chunk", |
9630 | found_key.objectid, found_key.offset); | |
9631 | ret = -ENOENT; | |
514c7dca QW |
9632 | } else if (em->start != found_key.objectid || |
9633 | em->len != found_key.offset) { | |
9634 | btrfs_err(fs_info, | |
9635 | "block group %llu len %llu mismatch with chunk %llu len %llu", | |
9636 | found_key.objectid, found_key.offset, | |
9637 | em->start, em->len); | |
9638 | ret = -EUCLEAN; | |
6fb37b75 | 9639 | } else { |
514c7dca QW |
9640 | read_extent_buffer(leaf, &bg, |
9641 | btrfs_item_ptr_offset(leaf, slot), | |
9642 | sizeof(bg)); | |
9643 | flags = btrfs_block_group_flags(&bg) & | |
9644 | BTRFS_BLOCK_GROUP_TYPE_MASK; | |
9645 | ||
9646 | if (flags != (em->map_lookup->type & | |
9647 | BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
9648 | btrfs_err(fs_info, | |
9649 | "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx", | |
9650 | found_key.objectid, | |
9651 | found_key.offset, flags, | |
9652 | (BTRFS_BLOCK_GROUP_TYPE_MASK & | |
9653 | em->map_lookup->type)); | |
9654 | ret = -EUCLEAN; | |
9655 | } else { | |
9656 | ret = 0; | |
9657 | } | |
6fb37b75 | 9658 | } |
187ee58c | 9659 | free_extent_map(em); |
925baedd CM |
9660 | goto out; |
9661 | } | |
0b86a832 | 9662 | path->slots[0]++; |
edbd8d4e | 9663 | } |
925baedd | 9664 | out: |
0b86a832 | 9665 | return ret; |
edbd8d4e CM |
9666 | } |
9667 | ||
0af3d00b JB |
9668 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
9669 | { | |
9670 | struct btrfs_block_group_cache *block_group; | |
9671 | u64 last = 0; | |
9672 | ||
9673 | while (1) { | |
9674 | struct inode *inode; | |
9675 | ||
9676 | block_group = btrfs_lookup_first_block_group(info, last); | |
9677 | while (block_group) { | |
3aa7c7a3 | 9678 | wait_block_group_cache_done(block_group); |
0af3d00b JB |
9679 | spin_lock(&block_group->lock); |
9680 | if (block_group->iref) | |
9681 | break; | |
9682 | spin_unlock(&block_group->lock); | |
2ff7e61e | 9683 | block_group = next_block_group(info, block_group); |
0af3d00b JB |
9684 | } |
9685 | if (!block_group) { | |
9686 | if (last == 0) | |
9687 | break; | |
9688 | last = 0; | |
9689 | continue; | |
9690 | } | |
9691 | ||
9692 | inode = block_group->inode; | |
9693 | block_group->iref = 0; | |
9694 | block_group->inode = NULL; | |
9695 | spin_unlock(&block_group->lock); | |
f3bca802 | 9696 | ASSERT(block_group->io_ctl.inode == NULL); |
0af3d00b JB |
9697 | iput(inode); |
9698 | last = block_group->key.objectid + block_group->key.offset; | |
9699 | btrfs_put_block_group(block_group); | |
9700 | } | |
9701 | } | |
9702 | ||
5cdd7db6 FM |
9703 | /* |
9704 | * Must be called only after stopping all workers, since we could have block | |
9705 | * group caching kthreads running, and therefore they could race with us if we | |
9706 | * freed the block groups before stopping them. | |
9707 | */ | |
1a40e23b ZY |
9708 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9709 | { | |
9710 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 9711 | struct btrfs_space_info *space_info; |
11833d66 | 9712 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
9713 | struct rb_node *n; |
9714 | ||
9e351cc8 | 9715 | down_write(&info->commit_root_sem); |
11833d66 YZ |
9716 | while (!list_empty(&info->caching_block_groups)) { |
9717 | caching_ctl = list_entry(info->caching_block_groups.next, | |
9718 | struct btrfs_caching_control, list); | |
9719 | list_del(&caching_ctl->list); | |
9720 | put_caching_control(caching_ctl); | |
9721 | } | |
9e351cc8 | 9722 | up_write(&info->commit_root_sem); |
11833d66 | 9723 | |
47ab2a6c JB |
9724 | spin_lock(&info->unused_bgs_lock); |
9725 | while (!list_empty(&info->unused_bgs)) { | |
9726 | block_group = list_first_entry(&info->unused_bgs, | |
9727 | struct btrfs_block_group_cache, | |
9728 | bg_list); | |
9729 | list_del_init(&block_group->bg_list); | |
9730 | btrfs_put_block_group(block_group); | |
9731 | } | |
9732 | spin_unlock(&info->unused_bgs_lock); | |
9733 | ||
1a40e23b ZY |
9734 | spin_lock(&info->block_group_cache_lock); |
9735 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
9736 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
9737 | cache_node); | |
1a40e23b ZY |
9738 | rb_erase(&block_group->cache_node, |
9739 | &info->block_group_cache_tree); | |
01eacb27 | 9740 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
9741 | spin_unlock(&info->block_group_cache_lock); |
9742 | ||
80eb234a | 9743 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 9744 | list_del(&block_group->list); |
80eb234a | 9745 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 9746 | |
3c14874a JB |
9747 | /* |
9748 | * We haven't cached this block group, which means we could | |
9749 | * possibly have excluded extents on this block group. | |
9750 | */ | |
36cce922 JB |
9751 | if (block_group->cached == BTRFS_CACHE_NO || |
9752 | block_group->cached == BTRFS_CACHE_ERROR) | |
9e715da8 | 9753 | free_excluded_extents(block_group); |
3c14874a | 9754 | |
817d52f8 | 9755 | btrfs_remove_free_space_cache(block_group); |
5cdd7db6 | 9756 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); |
f3bca802 LB |
9757 | ASSERT(list_empty(&block_group->dirty_list)); |
9758 | ASSERT(list_empty(&block_group->io_list)); | |
9759 | ASSERT(list_empty(&block_group->bg_list)); | |
9760 | ASSERT(atomic_read(&block_group->count) == 1); | |
11dfe35a | 9761 | btrfs_put_block_group(block_group); |
d899e052 YZ |
9762 | |
9763 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
9764 | } |
9765 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
9766 | |
9767 | /* now that all the block groups are freed, go through and | |
9768 | * free all the space_info structs. This is only called during | |
9769 | * the final stages of unmount, and so we know nobody is | |
9770 | * using them. We call synchronize_rcu() once before we start, | |
9771 | * just to be on the safe side. | |
9772 | */ | |
9773 | synchronize_rcu(); | |
9774 | ||
8929ecfa YZ |
9775 | release_global_block_rsv(info); |
9776 | ||
67871254 | 9777 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
9778 | int i; |
9779 | ||
4184ea7f CM |
9780 | space_info = list_entry(info->space_info.next, |
9781 | struct btrfs_space_info, | |
9782 | list); | |
d555b6c3 JB |
9783 | |
9784 | /* | |
9785 | * Do not hide this behind enospc_debug, this is actually | |
9786 | * important and indicates a real bug if this happens. | |
9787 | */ | |
9788 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
b069e0c3 | 9789 | space_info->bytes_reserved > 0 || |
d555b6c3 | 9790 | space_info->bytes_may_use > 0)) |
ab8d0fc4 | 9791 | dump_space_info(info, space_info, 0, 0); |
4184ea7f | 9792 | list_del(&space_info->list); |
6ab0a202 JM |
9793 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
9794 | struct kobject *kobj; | |
c1895442 JM |
9795 | kobj = space_info->block_group_kobjs[i]; |
9796 | space_info->block_group_kobjs[i] = NULL; | |
9797 | if (kobj) { | |
6ab0a202 JM |
9798 | kobject_del(kobj); |
9799 | kobject_put(kobj); | |
9800 | } | |
9801 | } | |
9802 | kobject_del(&space_info->kobj); | |
9803 | kobject_put(&space_info->kobj); | |
4184ea7f | 9804 | } |
1a40e23b ZY |
9805 | return 0; |
9806 | } | |
9807 | ||
75cb379d JM |
9808 | /* link_block_group will queue up kobjects to add when we're reclaim-safe */ |
9809 | void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info) | |
9810 | { | |
9811 | struct btrfs_space_info *space_info; | |
9812 | struct raid_kobject *rkobj; | |
9813 | LIST_HEAD(list); | |
9814 | int index; | |
9815 | int ret = 0; | |
9816 | ||
9817 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
9818 | list_splice_init(&fs_info->pending_raid_kobjs, &list); | |
9819 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
9820 | ||
9821 | list_for_each_entry(rkobj, &list, list) { | |
9822 | space_info = __find_space_info(fs_info, rkobj->flags); | |
9823 | index = btrfs_bg_flags_to_raid_index(rkobj->flags); | |
9824 | ||
9825 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
9826 | "%s", get_raid_name(index)); | |
9827 | if (ret) { | |
9828 | kobject_put(&rkobj->kobj); | |
9829 | break; | |
9830 | } | |
9831 | } | |
9832 | if (ret) | |
9833 | btrfs_warn(fs_info, | |
9834 | "failed to add kobject for block cache, ignoring"); | |
9835 | } | |
9836 | ||
c434d21c | 9837 | static void link_block_group(struct btrfs_block_group_cache *cache) |
b742bb82 | 9838 | { |
c434d21c | 9839 | struct btrfs_space_info *space_info = cache->space_info; |
75cb379d | 9840 | struct btrfs_fs_info *fs_info = cache->fs_info; |
3e72ee88 | 9841 | int index = btrfs_bg_flags_to_raid_index(cache->flags); |
ed55b6ac | 9842 | bool first = false; |
b742bb82 YZ |
9843 | |
9844 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
9845 | if (list_empty(&space_info->block_groups[index])) |
9846 | first = true; | |
9847 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
9848 | up_write(&space_info->groups_sem); | |
9849 | ||
9850 | if (first) { | |
75cb379d JM |
9851 | struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
9852 | if (!rkobj) { | |
9853 | btrfs_warn(cache->fs_info, | |
9854 | "couldn't alloc memory for raid level kobject"); | |
9855 | return; | |
6ab0a202 | 9856 | } |
75cb379d JM |
9857 | rkobj->flags = cache->flags; |
9858 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
9859 | ||
9860 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
9861 | list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs); | |
9862 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
c1895442 | 9863 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 9864 | } |
b742bb82 YZ |
9865 | } |
9866 | ||
920e4a58 | 9867 | static struct btrfs_block_group_cache * |
2ff7e61e JM |
9868 | btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info, |
9869 | u64 start, u64 size) | |
920e4a58 MX |
9870 | { |
9871 | struct btrfs_block_group_cache *cache; | |
9872 | ||
9873 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
9874 | if (!cache) | |
9875 | return NULL; | |
9876 | ||
9877 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
9878 | GFP_NOFS); | |
9879 | if (!cache->free_space_ctl) { | |
9880 | kfree(cache); | |
9881 | return NULL; | |
9882 | } | |
9883 | ||
9884 | cache->key.objectid = start; | |
9885 | cache->key.offset = size; | |
9886 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
9887 | ||
0b246afa | 9888 | cache->fs_info = fs_info; |
e4ff5fb5 | 9889 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); |
1e144fb8 OS |
9890 | set_free_space_tree_thresholds(cache); |
9891 | ||
920e4a58 MX |
9892 | atomic_set(&cache->count, 1); |
9893 | spin_lock_init(&cache->lock); | |
e570fd27 | 9894 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
9895 | INIT_LIST_HEAD(&cache->list); |
9896 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 9897 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 9898 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 9899 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 9900 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 9901 | btrfs_init_free_space_ctl(cache); |
04216820 | 9902 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 9903 | mutex_init(&cache->free_space_lock); |
0966a7b1 | 9904 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); |
920e4a58 MX |
9905 | |
9906 | return cache; | |
9907 | } | |
9908 | ||
7ef49515 QW |
9909 | |
9910 | /* | |
9911 | * Iterate all chunks and verify that each of them has the corresponding block | |
9912 | * group | |
9913 | */ | |
9914 | static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info) | |
9915 | { | |
9916 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
9917 | struct extent_map *em; | |
9918 | struct btrfs_block_group_cache *bg; | |
9919 | u64 start = 0; | |
9920 | int ret = 0; | |
9921 | ||
9922 | while (1) { | |
9923 | read_lock(&map_tree->map_tree.lock); | |
9924 | /* | |
9925 | * lookup_extent_mapping will return the first extent map | |
9926 | * intersecting the range, so setting @len to 1 is enough to | |
9927 | * get the first chunk. | |
9928 | */ | |
9929 | em = lookup_extent_mapping(&map_tree->map_tree, start, 1); | |
9930 | read_unlock(&map_tree->map_tree.lock); | |
9931 | if (!em) | |
9932 | break; | |
9933 | ||
9934 | bg = btrfs_lookup_block_group(fs_info, em->start); | |
9935 | if (!bg) { | |
9936 | btrfs_err(fs_info, | |
9937 | "chunk start=%llu len=%llu doesn't have corresponding block group", | |
9938 | em->start, em->len); | |
9939 | ret = -EUCLEAN; | |
9940 | free_extent_map(em); | |
9941 | break; | |
9942 | } | |
9943 | if (bg->key.objectid != em->start || | |
9944 | bg->key.offset != em->len || | |
9945 | (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) != | |
9946 | (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
9947 | btrfs_err(fs_info, | |
9948 | "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx", | |
9949 | em->start, em->len, | |
9950 | em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK, | |
9951 | bg->key.objectid, bg->key.offset, | |
9952 | bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK); | |
9953 | ret = -EUCLEAN; | |
9954 | free_extent_map(em); | |
9955 | btrfs_put_block_group(bg); | |
9956 | break; | |
9957 | } | |
9958 | start = em->start + em->len; | |
9959 | free_extent_map(em); | |
9960 | btrfs_put_block_group(bg); | |
9961 | } | |
9962 | return ret; | |
9963 | } | |
9964 | ||
5b4aacef | 9965 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
9078a3e1 CM |
9966 | { |
9967 | struct btrfs_path *path; | |
9968 | int ret; | |
9078a3e1 | 9969 | struct btrfs_block_group_cache *cache; |
6324fbf3 | 9970 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
9971 | struct btrfs_key key; |
9972 | struct btrfs_key found_key; | |
5f39d397 | 9973 | struct extent_buffer *leaf; |
0af3d00b JB |
9974 | int need_clear = 0; |
9975 | u64 cache_gen; | |
49303381 LB |
9976 | u64 feature; |
9977 | int mixed; | |
9978 | ||
9979 | feature = btrfs_super_incompat_flags(info->super_copy); | |
9980 | mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS); | |
96b5179d | 9981 | |
9078a3e1 | 9982 | key.objectid = 0; |
0b86a832 | 9983 | key.offset = 0; |
962a298f | 9984 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
9985 | path = btrfs_alloc_path(); |
9986 | if (!path) | |
9987 | return -ENOMEM; | |
e4058b54 | 9988 | path->reada = READA_FORWARD; |
9078a3e1 | 9989 | |
0b246afa JM |
9990 | cache_gen = btrfs_super_cache_generation(info->super_copy); |
9991 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
9992 | btrfs_super_generation(info->super_copy) != cache_gen) | |
0af3d00b | 9993 | need_clear = 1; |
0b246afa | 9994 | if (btrfs_test_opt(info, CLEAR_CACHE)) |
88c2ba3b | 9995 | need_clear = 1; |
0af3d00b | 9996 | |
d397712b | 9997 | while (1) { |
6bccf3ab | 9998 | ret = find_first_block_group(info, path, &key); |
b742bb82 YZ |
9999 | if (ret > 0) |
10000 | break; | |
0b86a832 CM |
10001 | if (ret != 0) |
10002 | goto error; | |
920e4a58 | 10003 | |
5f39d397 CM |
10004 | leaf = path->nodes[0]; |
10005 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 | 10006 | |
2ff7e61e | 10007 | cache = btrfs_create_block_group_cache(info, found_key.objectid, |
920e4a58 | 10008 | found_key.offset); |
9078a3e1 | 10009 | if (!cache) { |
0b86a832 | 10010 | ret = -ENOMEM; |
f0486c68 | 10011 | goto error; |
9078a3e1 | 10012 | } |
96303081 | 10013 | |
cf7c1ef6 LB |
10014 | if (need_clear) { |
10015 | /* | |
10016 | * When we mount with old space cache, we need to | |
10017 | * set BTRFS_DC_CLEAR and set dirty flag. | |
10018 | * | |
10019 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
10020 | * truncate the old free space cache inode and | |
10021 | * setup a new one. | |
10022 | * b) Setting 'dirty flag' makes sure that we flush | |
10023 | * the new space cache info onto disk. | |
10024 | */ | |
0b246afa | 10025 | if (btrfs_test_opt(info, SPACE_CACHE)) |
ce93ec54 | 10026 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 10027 | } |
0af3d00b | 10028 | |
5f39d397 CM |
10029 | read_extent_buffer(leaf, &cache->item, |
10030 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
10031 | sizeof(cache->item)); | |
920e4a58 | 10032 | cache->flags = btrfs_block_group_flags(&cache->item); |
49303381 LB |
10033 | if (!mixed && |
10034 | ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && | |
10035 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
10036 | btrfs_err(info, | |
10037 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
10038 | cache->key.objectid); | |
10039 | ret = -EINVAL; | |
10040 | goto error; | |
10041 | } | |
0b86a832 | 10042 | |
9078a3e1 | 10043 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 10044 | btrfs_release_path(path); |
34d52cb6 | 10045 | |
3c14874a JB |
10046 | /* |
10047 | * We need to exclude the super stripes now so that the space | |
10048 | * info has super bytes accounted for, otherwise we'll think | |
10049 | * we have more space than we actually do. | |
10050 | */ | |
3c4da657 | 10051 | ret = exclude_super_stripes(cache); |
835d974f JB |
10052 | if (ret) { |
10053 | /* | |
10054 | * We may have excluded something, so call this just in | |
10055 | * case. | |
10056 | */ | |
9e715da8 | 10057 | free_excluded_extents(cache); |
920e4a58 | 10058 | btrfs_put_block_group(cache); |
835d974f JB |
10059 | goto error; |
10060 | } | |
3c14874a | 10061 | |
817d52f8 JB |
10062 | /* |
10063 | * check for two cases, either we are full, and therefore | |
10064 | * don't need to bother with the caching work since we won't | |
10065 | * find any space, or we are empty, and we can just add all | |
10066 | * the space in and be done with it. This saves us _alot_ of | |
10067 | * time, particularly in the full case. | |
10068 | */ | |
10069 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 10070 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10071 | cache->cached = BTRFS_CACHE_FINISHED; |
9e715da8 | 10072 | free_excluded_extents(cache); |
817d52f8 | 10073 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 10074 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10075 | cache->cached = BTRFS_CACHE_FINISHED; |
4457c1c7 | 10076 | add_new_free_space(cache, found_key.objectid, |
817d52f8 JB |
10077 | found_key.objectid + |
10078 | found_key.offset); | |
9e715da8 | 10079 | free_excluded_extents(cache); |
817d52f8 | 10080 | } |
96b5179d | 10081 | |
0b246afa | 10082 | ret = btrfs_add_block_group_cache(info, cache); |
8c579fe7 JB |
10083 | if (ret) { |
10084 | btrfs_remove_free_space_cache(cache); | |
10085 | btrfs_put_block_group(cache); | |
10086 | goto error; | |
10087 | } | |
10088 | ||
0b246afa | 10089 | trace_btrfs_add_block_group(info, cache, 0); |
d2006e6d NB |
10090 | update_space_info(info, cache->flags, found_key.offset, |
10091 | btrfs_block_group_used(&cache->item), | |
10092 | cache->bytes_super, &space_info); | |
8c579fe7 | 10093 | |
6324fbf3 | 10094 | cache->space_info = space_info; |
1b2da372 | 10095 | |
c434d21c | 10096 | link_block_group(cache); |
0f9dd46c | 10097 | |
0b246afa | 10098 | set_avail_alloc_bits(info, cache->flags); |
2ff7e61e | 10099 | if (btrfs_chunk_readonly(info, cache->key.objectid)) { |
868f401a | 10100 | inc_block_group_ro(cache, 1); |
47ab2a6c | 10101 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
031f24da QW |
10102 | ASSERT(list_empty(&cache->bg_list)); |
10103 | btrfs_mark_bg_unused(cache); | |
47ab2a6c | 10104 | } |
9078a3e1 | 10105 | } |
b742bb82 | 10106 | |
0b246afa | 10107 | list_for_each_entry_rcu(space_info, &info->space_info, list) { |
2ff7e61e | 10108 | if (!(get_alloc_profile(info, space_info->flags) & |
b742bb82 YZ |
10109 | (BTRFS_BLOCK_GROUP_RAID10 | |
10110 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
10111 | BTRFS_BLOCK_GROUP_RAID5 | |
10112 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
10113 | BTRFS_BLOCK_GROUP_DUP))) |
10114 | continue; | |
10115 | /* | |
10116 | * avoid allocating from un-mirrored block group if there are | |
10117 | * mirrored block groups. | |
10118 | */ | |
1095cc0d | 10119 | list_for_each_entry(cache, |
10120 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
10121 | list) | |
868f401a | 10122 | inc_block_group_ro(cache, 1); |
1095cc0d | 10123 | list_for_each_entry(cache, |
10124 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
10125 | list) | |
868f401a | 10126 | inc_block_group_ro(cache, 1); |
9078a3e1 | 10127 | } |
f0486c68 | 10128 | |
75cb379d | 10129 | btrfs_add_raid_kobjects(info); |
f0486c68 | 10130 | init_global_block_rsv(info); |
7ef49515 | 10131 | ret = check_chunk_block_group_mappings(info); |
0b86a832 | 10132 | error: |
9078a3e1 | 10133 | btrfs_free_path(path); |
0b86a832 | 10134 | return ret; |
9078a3e1 | 10135 | } |
6324fbf3 | 10136 | |
6c686b35 | 10137 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) |
ea658bad | 10138 | { |
6c686b35 | 10139 | struct btrfs_fs_info *fs_info = trans->fs_info; |
545e3366 | 10140 | struct btrfs_block_group_cache *block_group; |
0b246afa | 10141 | struct btrfs_root *extent_root = fs_info->extent_root; |
ea658bad JB |
10142 | struct btrfs_block_group_item item; |
10143 | struct btrfs_key key; | |
10144 | int ret = 0; | |
10145 | ||
5ce55557 FM |
10146 | if (!trans->can_flush_pending_bgs) |
10147 | return; | |
10148 | ||
545e3366 JB |
10149 | while (!list_empty(&trans->new_bgs)) { |
10150 | block_group = list_first_entry(&trans->new_bgs, | |
10151 | struct btrfs_block_group_cache, | |
10152 | bg_list); | |
ea658bad | 10153 | if (ret) |
c92f6be3 | 10154 | goto next; |
ea658bad JB |
10155 | |
10156 | spin_lock(&block_group->lock); | |
10157 | memcpy(&item, &block_group->item, sizeof(item)); | |
10158 | memcpy(&key, &block_group->key, sizeof(key)); | |
10159 | spin_unlock(&block_group->lock); | |
10160 | ||
10161 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
10162 | sizeof(item)); | |
10163 | if (ret) | |
66642832 | 10164 | btrfs_abort_transaction(trans, ret); |
97aff912 | 10165 | ret = btrfs_finish_chunk_alloc(trans, key.objectid, key.offset); |
6df9a95e | 10166 | if (ret) |
66642832 | 10167 | btrfs_abort_transaction(trans, ret); |
e4e0711c | 10168 | add_block_group_free_space(trans, block_group); |
1e144fb8 | 10169 | /* already aborted the transaction if it failed. */ |
c92f6be3 FM |
10170 | next: |
10171 | list_del_init(&block_group->bg_list); | |
ea658bad | 10172 | } |
5ce55557 | 10173 | btrfs_trans_release_chunk_metadata(trans); |
ea658bad JB |
10174 | } |
10175 | ||
e7e02096 | 10176 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used, |
0174484d | 10177 | u64 type, u64 chunk_offset, u64 size) |
6324fbf3 | 10178 | { |
e7e02096 | 10179 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6324fbf3 | 10180 | struct btrfs_block_group_cache *cache; |
0b246afa | 10181 | int ret; |
6324fbf3 | 10182 | |
0b246afa | 10183 | btrfs_set_log_full_commit(fs_info, trans); |
e02119d5 | 10184 | |
2ff7e61e | 10185 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size); |
0f9dd46c JB |
10186 | if (!cache) |
10187 | return -ENOMEM; | |
34d52cb6 | 10188 | |
6324fbf3 | 10189 | btrfs_set_block_group_used(&cache->item, bytes_used); |
0174484d NB |
10190 | btrfs_set_block_group_chunk_objectid(&cache->item, |
10191 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
6324fbf3 CM |
10192 | btrfs_set_block_group_flags(&cache->item, type); |
10193 | ||
920e4a58 | 10194 | cache->flags = type; |
11833d66 | 10195 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10196 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 10197 | cache->needs_free_space = 1; |
3c4da657 | 10198 | ret = exclude_super_stripes(cache); |
835d974f JB |
10199 | if (ret) { |
10200 | /* | |
10201 | * We may have excluded something, so call this just in | |
10202 | * case. | |
10203 | */ | |
9e715da8 | 10204 | free_excluded_extents(cache); |
920e4a58 | 10205 | btrfs_put_block_group(cache); |
835d974f JB |
10206 | return ret; |
10207 | } | |
96303081 | 10208 | |
4457c1c7 | 10209 | add_new_free_space(cache, chunk_offset, chunk_offset + size); |
817d52f8 | 10210 | |
9e715da8 | 10211 | free_excluded_extents(cache); |
11833d66 | 10212 | |
d0bd4560 | 10213 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 10214 | if (btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
10215 | u64 new_bytes_used = size - bytes_used; |
10216 | ||
10217 | bytes_used += new_bytes_used >> 1; | |
2ff7e61e | 10218 | fragment_free_space(cache); |
d0bd4560 JB |
10219 | } |
10220 | #endif | |
2e6e5183 | 10221 | /* |
2be12ef7 NB |
10222 | * Ensure the corresponding space_info object is created and |
10223 | * assigned to our block group. We want our bg to be added to the rbtree | |
10224 | * with its ->space_info set. | |
2e6e5183 | 10225 | */ |
2be12ef7 | 10226 | cache->space_info = __find_space_info(fs_info, cache->flags); |
dc2d3005 | 10227 | ASSERT(cache->space_info); |
2e6e5183 | 10228 | |
0b246afa | 10229 | ret = btrfs_add_block_group_cache(fs_info, cache); |
8c579fe7 JB |
10230 | if (ret) { |
10231 | btrfs_remove_free_space_cache(cache); | |
10232 | btrfs_put_block_group(cache); | |
10233 | return ret; | |
10234 | } | |
10235 | ||
2e6e5183 FM |
10236 | /* |
10237 | * Now that our block group has its ->space_info set and is inserted in | |
10238 | * the rbtree, update the space info's counters. | |
10239 | */ | |
0b246afa | 10240 | trace_btrfs_add_block_group(fs_info, cache, 1); |
d2006e6d | 10241 | update_space_info(fs_info, cache->flags, size, bytes_used, |
e40edf2d | 10242 | cache->bytes_super, &cache->space_info); |
0b246afa | 10243 | update_global_block_rsv(fs_info); |
1b2da372 | 10244 | |
c434d21c | 10245 | link_block_group(cache); |
6324fbf3 | 10246 | |
47ab2a6c | 10247 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
6324fbf3 | 10248 | |
0b246afa | 10249 | set_avail_alloc_bits(fs_info, type); |
6324fbf3 CM |
10250 | return 0; |
10251 | } | |
1a40e23b | 10252 | |
10ea00f5 ID |
10253 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
10254 | { | |
899c81ea ID |
10255 | u64 extra_flags = chunk_to_extended(flags) & |
10256 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 10257 | |
de98ced9 | 10258 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10259 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
10260 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10261 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10262 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10263 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10264 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 10265 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10266 | } |
10267 | ||
1a40e23b | 10268 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
5a98ec01 | 10269 | u64 group_start, struct extent_map *em) |
1a40e23b | 10270 | { |
5a98ec01 | 10271 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 10272 | struct btrfs_root *root = fs_info->extent_root; |
1a40e23b ZY |
10273 | struct btrfs_path *path; |
10274 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 10275 | struct btrfs_free_cluster *cluster; |
0b246afa | 10276 | struct btrfs_root *tree_root = fs_info->tree_root; |
1a40e23b | 10277 | struct btrfs_key key; |
0af3d00b | 10278 | struct inode *inode; |
c1895442 | 10279 | struct kobject *kobj = NULL; |
1a40e23b | 10280 | int ret; |
10ea00f5 | 10281 | int index; |
89a55897 | 10282 | int factor; |
4f69cb98 | 10283 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 10284 | bool remove_em; |
1a40e23b | 10285 | |
6bccf3ab | 10286 | block_group = btrfs_lookup_block_group(fs_info, group_start); |
1a40e23b | 10287 | BUG_ON(!block_group); |
c146afad | 10288 | BUG_ON(!block_group->ro); |
1a40e23b | 10289 | |
4ed0a7a3 | 10290 | trace_btrfs_remove_block_group(block_group); |
9f7c43c9 | 10291 | /* |
10292 | * Free the reserved super bytes from this block group before | |
10293 | * remove it. | |
10294 | */ | |
9e715da8 | 10295 | free_excluded_extents(block_group); |
fd708b81 JB |
10296 | btrfs_free_ref_tree_range(fs_info, block_group->key.objectid, |
10297 | block_group->key.offset); | |
9f7c43c9 | 10298 | |
1a40e23b | 10299 | memcpy(&key, &block_group->key, sizeof(key)); |
3e72ee88 | 10300 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
46df06b8 | 10301 | factor = btrfs_bg_type_to_factor(block_group->flags); |
1a40e23b | 10302 | |
44fb5511 | 10303 | /* make sure this block group isn't part of an allocation cluster */ |
0b246afa | 10304 | cluster = &fs_info->data_alloc_cluster; |
44fb5511 CM |
10305 | spin_lock(&cluster->refill_lock); |
10306 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10307 | spin_unlock(&cluster->refill_lock); | |
10308 | ||
10309 | /* | |
10310 | * make sure this block group isn't part of a metadata | |
10311 | * allocation cluster | |
10312 | */ | |
0b246afa | 10313 | cluster = &fs_info->meta_alloc_cluster; |
44fb5511 CM |
10314 | spin_lock(&cluster->refill_lock); |
10315 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10316 | spin_unlock(&cluster->refill_lock); | |
10317 | ||
1a40e23b | 10318 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10319 | if (!path) { |
10320 | ret = -ENOMEM; | |
10321 | goto out; | |
10322 | } | |
1a40e23b | 10323 | |
1bbc621e CM |
10324 | /* |
10325 | * get the inode first so any iput calls done for the io_list | |
10326 | * aren't the final iput (no unlinks allowed now) | |
10327 | */ | |
77ab86bf | 10328 | inode = lookup_free_space_inode(fs_info, block_group, path); |
1bbc621e CM |
10329 | |
10330 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10331 | /* | |
10332 | * make sure our free spache cache IO is done before remove the | |
10333 | * free space inode | |
10334 | */ | |
10335 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10336 | if (!list_empty(&block_group->io_list)) { | |
10337 | list_del_init(&block_group->io_list); | |
10338 | ||
10339 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10340 | ||
10341 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
afdb5718 | 10342 | btrfs_wait_cache_io(trans, block_group, path); |
1bbc621e CM |
10343 | btrfs_put_block_group(block_group); |
10344 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10345 | } | |
10346 | ||
10347 | if (!list_empty(&block_group->dirty_list)) { | |
10348 | list_del_init(&block_group->dirty_list); | |
10349 | btrfs_put_block_group(block_group); | |
10350 | } | |
10351 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10352 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10353 | ||
0af3d00b | 10354 | if (!IS_ERR(inode)) { |
73f2e545 | 10355 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
79787eaa JM |
10356 | if (ret) { |
10357 | btrfs_add_delayed_iput(inode); | |
10358 | goto out; | |
10359 | } | |
0af3d00b JB |
10360 | clear_nlink(inode); |
10361 | /* One for the block groups ref */ | |
10362 | spin_lock(&block_group->lock); | |
10363 | if (block_group->iref) { | |
10364 | block_group->iref = 0; | |
10365 | block_group->inode = NULL; | |
10366 | spin_unlock(&block_group->lock); | |
10367 | iput(inode); | |
10368 | } else { | |
10369 | spin_unlock(&block_group->lock); | |
10370 | } | |
10371 | /* One for our lookup ref */ | |
455757c3 | 10372 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10373 | } |
10374 | ||
10375 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10376 | key.offset = block_group->key.objectid; | |
10377 | key.type = 0; | |
10378 | ||
10379 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10380 | if (ret < 0) | |
10381 | goto out; | |
10382 | if (ret > 0) | |
b3b4aa74 | 10383 | btrfs_release_path(path); |
0af3d00b JB |
10384 | if (ret == 0) { |
10385 | ret = btrfs_del_item(trans, tree_root, path); | |
10386 | if (ret) | |
10387 | goto out; | |
b3b4aa74 | 10388 | btrfs_release_path(path); |
0af3d00b JB |
10389 | } |
10390 | ||
0b246afa | 10391 | spin_lock(&fs_info->block_group_cache_lock); |
1a40e23b | 10392 | rb_erase(&block_group->cache_node, |
0b246afa | 10393 | &fs_info->block_group_cache_tree); |
292cbd51 | 10394 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd | 10395 | |
0b246afa JM |
10396 | if (fs_info->first_logical_byte == block_group->key.objectid) |
10397 | fs_info->first_logical_byte = (u64)-1; | |
10398 | spin_unlock(&fs_info->block_group_cache_lock); | |
817d52f8 | 10399 | |
80eb234a | 10400 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10401 | /* |
10402 | * we must use list_del_init so people can check to see if they | |
10403 | * are still on the list after taking the semaphore | |
10404 | */ | |
10405 | list_del_init(&block_group->list); | |
6ab0a202 | 10406 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10407 | kobj = block_group->space_info->block_group_kobjs[index]; |
10408 | block_group->space_info->block_group_kobjs[index] = NULL; | |
0b246afa | 10409 | clear_avail_alloc_bits(fs_info, block_group->flags); |
6ab0a202 | 10410 | } |
80eb234a | 10411 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
10412 | if (kobj) { |
10413 | kobject_del(kobj); | |
10414 | kobject_put(kobj); | |
10415 | } | |
1a40e23b | 10416 | |
4f69cb98 FM |
10417 | if (block_group->has_caching_ctl) |
10418 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10419 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10420 | wait_block_group_cache_done(block_group); |
4f69cb98 | 10421 | if (block_group->has_caching_ctl) { |
0b246afa | 10422 | down_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10423 | if (!caching_ctl) { |
10424 | struct btrfs_caching_control *ctl; | |
10425 | ||
10426 | list_for_each_entry(ctl, | |
0b246afa | 10427 | &fs_info->caching_block_groups, list) |
4f69cb98 FM |
10428 | if (ctl->block_group == block_group) { |
10429 | caching_ctl = ctl; | |
1e4f4714 | 10430 | refcount_inc(&caching_ctl->count); |
4f69cb98 FM |
10431 | break; |
10432 | } | |
10433 | } | |
10434 | if (caching_ctl) | |
10435 | list_del_init(&caching_ctl->list); | |
0b246afa | 10436 | up_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10437 | if (caching_ctl) { |
10438 | /* Once for the caching bgs list and once for us. */ | |
10439 | put_caching_control(caching_ctl); | |
10440 | put_caching_control(caching_ctl); | |
10441 | } | |
10442 | } | |
817d52f8 | 10443 | |
ce93ec54 JB |
10444 | spin_lock(&trans->transaction->dirty_bgs_lock); |
10445 | if (!list_empty(&block_group->dirty_list)) { | |
1bbc621e CM |
10446 | WARN_ON(1); |
10447 | } | |
10448 | if (!list_empty(&block_group->io_list)) { | |
10449 | WARN_ON(1); | |
ce93ec54 JB |
10450 | } |
10451 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
817d52f8 JB |
10452 | btrfs_remove_free_space_cache(block_group); |
10453 | ||
c146afad | 10454 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10455 | list_del_init(&block_group->ro_list); |
18d018ad | 10456 | |
0b246afa | 10457 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
18d018ad ZL |
10458 | WARN_ON(block_group->space_info->total_bytes |
10459 | < block_group->key.offset); | |
10460 | WARN_ON(block_group->space_info->bytes_readonly | |
10461 | < block_group->key.offset); | |
10462 | WARN_ON(block_group->space_info->disk_total | |
10463 | < block_group->key.offset * factor); | |
10464 | } | |
c146afad YZ |
10465 | block_group->space_info->total_bytes -= block_group->key.offset; |
10466 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10467 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10468 | |
c146afad | 10469 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10470 | |
0af3d00b JB |
10471 | memcpy(&key, &block_group->key, sizeof(key)); |
10472 | ||
34441361 | 10473 | mutex_lock(&fs_info->chunk_mutex); |
495e64f4 FM |
10474 | if (!list_empty(&em->list)) { |
10475 | /* We're in the transaction->pending_chunks list. */ | |
10476 | free_extent_map(em); | |
10477 | } | |
04216820 FM |
10478 | spin_lock(&block_group->lock); |
10479 | block_group->removed = 1; | |
10480 | /* | |
10481 | * At this point trimming can't start on this block group, because we | |
10482 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10483 | * so no one can't find it anymore and even if someone already got this | |
10484 | * block group before we removed it from the rbtree, they have already | |
10485 | * incremented block_group->trimming - if they didn't, they won't find | |
10486 | * any free space entries because we already removed them all when we | |
10487 | * called btrfs_remove_free_space_cache(). | |
10488 | * | |
10489 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10490 | * to prevent the same logical address range and physical device space | |
10491 | * ranges from being reused for a new block group. This is because our | |
10492 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10493 | * completely transactionless, so while it is trimming a range the | |
10494 | * currently running transaction might finish and a new one start, | |
10495 | * allowing for new block groups to be created that can reuse the same | |
10496 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10497 | * |
10498 | * There may also be an implicit trim operation if the file system | |
10499 | * is mounted with -odiscard. The same protections must remain | |
10500 | * in place until the extents have been discarded completely when | |
10501 | * the transaction commit has completed. | |
04216820 FM |
10502 | */ |
10503 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
10504 | /* | |
10505 | * Make sure a trimmer task always sees the em in the pinned_chunks list | |
10506 | * if it sees block_group->removed == 1 (needs to lock block_group->lock | |
10507 | * before checking block_group->removed). | |
10508 | */ | |
10509 | if (!remove_em) { | |
10510 | /* | |
10511 | * Our em might be in trans->transaction->pending_chunks which | |
10512 | * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks), | |
10513 | * and so is the fs_info->pinned_chunks list. | |
10514 | * | |
10515 | * So at this point we must be holding the chunk_mutex to avoid | |
10516 | * any races with chunk allocation (more specifically at | |
10517 | * volumes.c:contains_pending_extent()), to ensure it always | |
10518 | * sees the em, either in the pending_chunks list or in the | |
10519 | * pinned_chunks list. | |
10520 | */ | |
0b246afa | 10521 | list_move_tail(&em->list, &fs_info->pinned_chunks); |
04216820 FM |
10522 | } |
10523 | spin_unlock(&block_group->lock); | |
04216820 FM |
10524 | |
10525 | if (remove_em) { | |
10526 | struct extent_map_tree *em_tree; | |
10527 | ||
0b246afa | 10528 | em_tree = &fs_info->mapping_tree.map_tree; |
04216820 | 10529 | write_lock(&em_tree->lock); |
8dbcd10f FM |
10530 | /* |
10531 | * The em might be in the pending_chunks list, so make sure the | |
10532 | * chunk mutex is locked, since remove_extent_mapping() will | |
10533 | * delete us from that list. | |
10534 | */ | |
04216820 FM |
10535 | remove_extent_mapping(em_tree, em); |
10536 | write_unlock(&em_tree->lock); | |
10537 | /* once for the tree */ | |
10538 | free_extent_map(em); | |
10539 | } | |
10540 | ||
34441361 | 10541 | mutex_unlock(&fs_info->chunk_mutex); |
8dbcd10f | 10542 | |
f3f72779 | 10543 | ret = remove_block_group_free_space(trans, block_group); |
1e144fb8 OS |
10544 | if (ret) |
10545 | goto out; | |
10546 | ||
fa9c0d79 CM |
10547 | btrfs_put_block_group(block_group); |
10548 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10549 | |
10550 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10551 | if (ret > 0) | |
10552 | ret = -EIO; | |
10553 | if (ret < 0) | |
10554 | goto out; | |
10555 | ||
10556 | ret = btrfs_del_item(trans, root, path); | |
10557 | out: | |
10558 | btrfs_free_path(path); | |
10559 | return ret; | |
10560 | } | |
acce952b | 10561 | |
8eab77ff | 10562 | struct btrfs_trans_handle * |
7fd01182 FM |
10563 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10564 | const u64 chunk_offset) | |
8eab77ff | 10565 | { |
7fd01182 FM |
10566 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; |
10567 | struct extent_map *em; | |
10568 | struct map_lookup *map; | |
10569 | unsigned int num_items; | |
10570 | ||
10571 | read_lock(&em_tree->lock); | |
10572 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10573 | read_unlock(&em_tree->lock); | |
10574 | ASSERT(em && em->start == chunk_offset); | |
10575 | ||
8eab77ff | 10576 | /* |
7fd01182 FM |
10577 | * We need to reserve 3 + N units from the metadata space info in order |
10578 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10579 | * btrfs_remove_block_group()), which are used for: | |
10580 | * | |
8eab77ff FM |
10581 | * 1 unit for adding the free space inode's orphan (located in the tree |
10582 | * of tree roots). | |
7fd01182 FM |
10583 | * 1 unit for deleting the block group item (located in the extent |
10584 | * tree). | |
10585 | * 1 unit for deleting the free space item (located in tree of tree | |
10586 | * roots). | |
10587 | * N units for deleting N device extent items corresponding to each | |
10588 | * stripe (located in the device tree). | |
10589 | * | |
10590 | * In order to remove a block group we also need to reserve units in the | |
10591 | * system space info in order to update the chunk tree (update one or | |
10592 | * more device items and remove one chunk item), but this is done at | |
10593 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10594 | */ |
95617d69 | 10595 | map = em->map_lookup; |
7fd01182 FM |
10596 | num_items = 3 + map->num_stripes; |
10597 | free_extent_map(em); | |
10598 | ||
8eab77ff | 10599 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10600 | num_items, 1); |
8eab77ff FM |
10601 | } |
10602 | ||
47ab2a6c JB |
10603 | /* |
10604 | * Process the unused_bgs list and remove any that don't have any allocated | |
10605 | * space inside of them. | |
10606 | */ | |
10607 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10608 | { | |
10609 | struct btrfs_block_group_cache *block_group; | |
10610 | struct btrfs_space_info *space_info; | |
47ab2a6c JB |
10611 | struct btrfs_trans_handle *trans; |
10612 | int ret = 0; | |
10613 | ||
afcdd129 | 10614 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) |
47ab2a6c JB |
10615 | return; |
10616 | ||
10617 | spin_lock(&fs_info->unused_bgs_lock); | |
10618 | while (!list_empty(&fs_info->unused_bgs)) { | |
10619 | u64 start, end; | |
e33e17ee | 10620 | int trimming; |
47ab2a6c JB |
10621 | |
10622 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10623 | struct btrfs_block_group_cache, | |
10624 | bg_list); | |
47ab2a6c | 10625 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10626 | |
10627 | space_info = block_group->space_info; | |
10628 | ||
47ab2a6c JB |
10629 | if (ret || btrfs_mixed_space_info(space_info)) { |
10630 | btrfs_put_block_group(block_group); | |
10631 | continue; | |
10632 | } | |
10633 | spin_unlock(&fs_info->unused_bgs_lock); | |
10634 | ||
d5f2e33b | 10635 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10636 | |
47ab2a6c JB |
10637 | /* Don't want to race with allocators so take the groups_sem */ |
10638 | down_write(&space_info->groups_sem); | |
10639 | spin_lock(&block_group->lock); | |
43794446 | 10640 | if (block_group->reserved || block_group->pinned || |
47ab2a6c | 10641 | btrfs_block_group_used(&block_group->item) || |
19c4d2f9 | 10642 | block_group->ro || |
aefbe9a6 | 10643 | list_is_singular(&block_group->list)) { |
47ab2a6c JB |
10644 | /* |
10645 | * We want to bail if we made new allocations or have | |
10646 | * outstanding allocations in this block group. We do | |
10647 | * the ro check in case balance is currently acting on | |
10648 | * this block group. | |
10649 | */ | |
4ed0a7a3 | 10650 | trace_btrfs_skip_unused_block_group(block_group); |
47ab2a6c JB |
10651 | spin_unlock(&block_group->lock); |
10652 | up_write(&space_info->groups_sem); | |
10653 | goto next; | |
10654 | } | |
10655 | spin_unlock(&block_group->lock); | |
10656 | ||
10657 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 10658 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
10659 | up_write(&space_info->groups_sem); |
10660 | if (ret < 0) { | |
10661 | ret = 0; | |
10662 | goto next; | |
10663 | } | |
10664 | ||
10665 | /* | |
10666 | * Want to do this before we do anything else so we can recover | |
10667 | * properly if we fail to join the transaction. | |
10668 | */ | |
7fd01182 FM |
10669 | trans = btrfs_start_trans_remove_block_group(fs_info, |
10670 | block_group->key.objectid); | |
47ab2a6c | 10671 | if (IS_ERR(trans)) { |
2ff7e61e | 10672 | btrfs_dec_block_group_ro(block_group); |
47ab2a6c JB |
10673 | ret = PTR_ERR(trans); |
10674 | goto next; | |
10675 | } | |
10676 | ||
10677 | /* | |
10678 | * We could have pending pinned extents for this block group, | |
10679 | * just delete them, we don't care about them anymore. | |
10680 | */ | |
10681 | start = block_group->key.objectid; | |
10682 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
10683 | /* |
10684 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
10685 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
10686 | * another task might be running finish_extent_commit() for the | |
10687 | * previous transaction N - 1, and have seen a range belonging | |
10688 | * to the block group in freed_extents[] before we were able to | |
10689 | * clear the whole block group range from freed_extents[]. This | |
10690 | * means that task can lookup for the block group after we | |
10691 | * unpinned it from freed_extents[] and removed it, leading to | |
10692 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
10693 | */ | |
10694 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 10695 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
91166212 | 10696 | EXTENT_DIRTY); |
758eb51e | 10697 | if (ret) { |
d4b450cd | 10698 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10699 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10700 | goto end_trans; |
10701 | } | |
10702 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
91166212 | 10703 | EXTENT_DIRTY); |
758eb51e | 10704 | if (ret) { |
d4b450cd | 10705 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10706 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10707 | goto end_trans; |
10708 | } | |
d4b450cd | 10709 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
10710 | |
10711 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
10712 | spin_lock(&space_info->lock); |
10713 | spin_lock(&block_group->lock); | |
10714 | ||
10715 | space_info->bytes_pinned -= block_group->pinned; | |
10716 | space_info->bytes_readonly += block_group->pinned; | |
dec59fa3 EL |
10717 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
10718 | -block_group->pinned, | |
10719 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
47ab2a6c JB |
10720 | block_group->pinned = 0; |
10721 | ||
c30666d4 ZL |
10722 | spin_unlock(&block_group->lock); |
10723 | spin_unlock(&space_info->lock); | |
10724 | ||
e33e17ee | 10725 | /* DISCARD can flip during remount */ |
0b246afa | 10726 | trimming = btrfs_test_opt(fs_info, DISCARD); |
e33e17ee JM |
10727 | |
10728 | /* Implicit trim during transaction commit. */ | |
10729 | if (trimming) | |
10730 | btrfs_get_block_group_trimming(block_group); | |
10731 | ||
47ab2a6c JB |
10732 | /* |
10733 | * Btrfs_remove_chunk will abort the transaction if things go | |
10734 | * horribly wrong. | |
10735 | */ | |
97aff912 | 10736 | ret = btrfs_remove_chunk(trans, block_group->key.objectid); |
e33e17ee JM |
10737 | |
10738 | if (ret) { | |
10739 | if (trimming) | |
10740 | btrfs_put_block_group_trimming(block_group); | |
10741 | goto end_trans; | |
10742 | } | |
10743 | ||
10744 | /* | |
10745 | * If we're not mounted with -odiscard, we can just forget | |
10746 | * about this block group. Otherwise we'll need to wait | |
10747 | * until transaction commit to do the actual discard. | |
10748 | */ | |
10749 | if (trimming) { | |
348a0013 FM |
10750 | spin_lock(&fs_info->unused_bgs_lock); |
10751 | /* | |
10752 | * A concurrent scrub might have added us to the list | |
10753 | * fs_info->unused_bgs, so use a list_move operation | |
10754 | * to add the block group to the deleted_bgs list. | |
10755 | */ | |
e33e17ee JM |
10756 | list_move(&block_group->bg_list, |
10757 | &trans->transaction->deleted_bgs); | |
348a0013 | 10758 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
10759 | btrfs_get_block_group(block_group); |
10760 | } | |
758eb51e | 10761 | end_trans: |
3a45bb20 | 10762 | btrfs_end_transaction(trans); |
47ab2a6c | 10763 | next: |
d5f2e33b | 10764 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
10765 | btrfs_put_block_group(block_group); |
10766 | spin_lock(&fs_info->unused_bgs_lock); | |
10767 | } | |
10768 | spin_unlock(&fs_info->unused_bgs_lock); | |
10769 | } | |
10770 | ||
c59021f8 | 10771 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
10772 | { | |
1aba86d6 | 10773 | struct btrfs_super_block *disk_super; |
10774 | u64 features; | |
10775 | u64 flags; | |
10776 | int mixed = 0; | |
c59021f8 | 10777 | int ret; |
10778 | ||
6c41761f | 10779 | disk_super = fs_info->super_copy; |
1aba86d6 | 10780 | if (!btrfs_super_root(disk_super)) |
0dc924c5 | 10781 | return -EINVAL; |
c59021f8 | 10782 | |
1aba86d6 | 10783 | features = btrfs_super_incompat_flags(disk_super); |
10784 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
10785 | mixed = 1; | |
c59021f8 | 10786 | |
1aba86d6 | 10787 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
4ca61683 | 10788 | ret = create_space_info(fs_info, flags); |
c59021f8 | 10789 | if (ret) |
1aba86d6 | 10790 | goto out; |
c59021f8 | 10791 | |
1aba86d6 | 10792 | if (mixed) { |
10793 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
4ca61683 | 10794 | ret = create_space_info(fs_info, flags); |
1aba86d6 | 10795 | } else { |
10796 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
4ca61683 | 10797 | ret = create_space_info(fs_info, flags); |
1aba86d6 | 10798 | if (ret) |
10799 | goto out; | |
10800 | ||
10801 | flags = BTRFS_BLOCK_GROUP_DATA; | |
4ca61683 | 10802 | ret = create_space_info(fs_info, flags); |
1aba86d6 | 10803 | } |
10804 | out: | |
c59021f8 | 10805 | return ret; |
10806 | } | |
10807 | ||
2ff7e61e JM |
10808 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
10809 | u64 start, u64 end) | |
acce952b | 10810 | { |
2ff7e61e | 10811 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 10812 | } |
10813 | ||
499f377f JM |
10814 | /* |
10815 | * It used to be that old block groups would be left around forever. | |
10816 | * Iterating over them would be enough to trim unused space. Since we | |
10817 | * now automatically remove them, we also need to iterate over unallocated | |
10818 | * space. | |
10819 | * | |
10820 | * We don't want a transaction for this since the discard may take a | |
10821 | * substantial amount of time. We don't require that a transaction be | |
10822 | * running, but we do need to take a running transaction into account | |
fee7acc3 JM |
10823 | * to ensure that we're not discarding chunks that were released or |
10824 | * allocated in the current transaction. | |
499f377f JM |
10825 | * |
10826 | * Holding the chunks lock will prevent other threads from allocating | |
10827 | * or releasing chunks, but it won't prevent a running transaction | |
10828 | * from committing and releasing the memory that the pending chunks | |
10829 | * list head uses. For that, we need to take a reference to the | |
fee7acc3 JM |
10830 | * transaction and hold the commit root sem. We only need to hold |
10831 | * it while performing the free space search since we have already | |
10832 | * held back allocations. | |
499f377f JM |
10833 | */ |
10834 | static int btrfs_trim_free_extents(struct btrfs_device *device, | |
10835 | u64 minlen, u64 *trimmed) | |
10836 | { | |
10837 | u64 start = 0, len = 0; | |
10838 | int ret; | |
10839 | ||
10840 | *trimmed = 0; | |
10841 | ||
0be88e36 JM |
10842 | /* Discard not supported = nothing to do. */ |
10843 | if (!blk_queue_discard(bdev_get_queue(device->bdev))) | |
10844 | return 0; | |
10845 | ||
499f377f | 10846 | /* Not writeable = nothing to do. */ |
ebbede42 | 10847 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
10848 | return 0; |
10849 | ||
10850 | /* No free space = nothing to do. */ | |
10851 | if (device->total_bytes <= device->bytes_used) | |
10852 | return 0; | |
10853 | ||
10854 | ret = 0; | |
10855 | ||
10856 | while (1) { | |
fb456252 | 10857 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
10858 | struct btrfs_transaction *trans; |
10859 | u64 bytes; | |
10860 | ||
10861 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
10862 | if (ret) | |
fee7acc3 | 10863 | break; |
499f377f | 10864 | |
fee7acc3 JM |
10865 | ret = down_read_killable(&fs_info->commit_root_sem); |
10866 | if (ret) { | |
10867 | mutex_unlock(&fs_info->chunk_mutex); | |
10868 | break; | |
10869 | } | |
499f377f JM |
10870 | |
10871 | spin_lock(&fs_info->trans_lock); | |
10872 | trans = fs_info->running_transaction; | |
10873 | if (trans) | |
9b64f57d | 10874 | refcount_inc(&trans->use_count); |
499f377f JM |
10875 | spin_unlock(&fs_info->trans_lock); |
10876 | ||
fee7acc3 JM |
10877 | if (!trans) |
10878 | up_read(&fs_info->commit_root_sem); | |
10879 | ||
499f377f JM |
10880 | ret = find_free_dev_extent_start(trans, device, minlen, start, |
10881 | &start, &len); | |
fee7acc3 JM |
10882 | if (trans) { |
10883 | up_read(&fs_info->commit_root_sem); | |
499f377f | 10884 | btrfs_put_transaction(trans); |
fee7acc3 | 10885 | } |
499f377f JM |
10886 | |
10887 | if (ret) { | |
499f377f JM |
10888 | mutex_unlock(&fs_info->chunk_mutex); |
10889 | if (ret == -ENOSPC) | |
10890 | ret = 0; | |
10891 | break; | |
10892 | } | |
10893 | ||
10894 | ret = btrfs_issue_discard(device->bdev, start, len, &bytes); | |
499f377f JM |
10895 | mutex_unlock(&fs_info->chunk_mutex); |
10896 | ||
10897 | if (ret) | |
10898 | break; | |
10899 | ||
10900 | start += len; | |
10901 | *trimmed += bytes; | |
10902 | ||
10903 | if (fatal_signal_pending(current)) { | |
10904 | ret = -ERESTARTSYS; | |
10905 | break; | |
10906 | } | |
10907 | ||
10908 | cond_resched(); | |
10909 | } | |
10910 | ||
10911 | return ret; | |
10912 | } | |
10913 | ||
93bba24d QW |
10914 | /* |
10915 | * Trim the whole filesystem by: | |
10916 | * 1) trimming the free space in each block group | |
10917 | * 2) trimming the unallocated space on each device | |
10918 | * | |
10919 | * This will also continue trimming even if a block group or device encounters | |
10920 | * an error. The return value will be the last error, or 0 if nothing bad | |
10921 | * happens. | |
10922 | */ | |
2ff7e61e | 10923 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 10924 | { |
f7039b1d | 10925 | struct btrfs_block_group_cache *cache = NULL; |
499f377f JM |
10926 | struct btrfs_device *device; |
10927 | struct list_head *devices; | |
f7039b1d LD |
10928 | u64 group_trimmed; |
10929 | u64 start; | |
10930 | u64 end; | |
10931 | u64 trimmed = 0; | |
93bba24d QW |
10932 | u64 bg_failed = 0; |
10933 | u64 dev_failed = 0; | |
10934 | int bg_ret = 0; | |
10935 | int dev_ret = 0; | |
f7039b1d LD |
10936 | int ret = 0; |
10937 | ||
6ba9fc8e | 10938 | cache = btrfs_lookup_first_block_group(fs_info, range->start); |
93bba24d | 10939 | for (; cache; cache = next_block_group(fs_info, cache)) { |
f7039b1d LD |
10940 | if (cache->key.objectid >= (range->start + range->len)) { |
10941 | btrfs_put_block_group(cache); | |
10942 | break; | |
10943 | } | |
10944 | ||
10945 | start = max(range->start, cache->key.objectid); | |
10946 | end = min(range->start + range->len, | |
10947 | cache->key.objectid + cache->key.offset); | |
10948 | ||
10949 | if (end - start >= range->minlen) { | |
10950 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 10951 | ret = cache_block_group(cache, 0); |
1be41b78 | 10952 | if (ret) { |
93bba24d QW |
10953 | bg_failed++; |
10954 | bg_ret = ret; | |
10955 | continue; | |
1be41b78 JB |
10956 | } |
10957 | ret = wait_block_group_cache_done(cache); | |
10958 | if (ret) { | |
93bba24d QW |
10959 | bg_failed++; |
10960 | bg_ret = ret; | |
10961 | continue; | |
1be41b78 | 10962 | } |
f7039b1d LD |
10963 | } |
10964 | ret = btrfs_trim_block_group(cache, | |
10965 | &group_trimmed, | |
10966 | start, | |
10967 | end, | |
10968 | range->minlen); | |
10969 | ||
10970 | trimmed += group_trimmed; | |
10971 | if (ret) { | |
93bba24d QW |
10972 | bg_failed++; |
10973 | bg_ret = ret; | |
10974 | continue; | |
f7039b1d LD |
10975 | } |
10976 | } | |
f7039b1d LD |
10977 | } |
10978 | ||
93bba24d QW |
10979 | if (bg_failed) |
10980 | btrfs_warn(fs_info, | |
10981 | "failed to trim %llu block group(s), last error %d", | |
10982 | bg_failed, bg_ret); | |
0b246afa | 10983 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
d4e329de JM |
10984 | devices = &fs_info->fs_devices->devices; |
10985 | list_for_each_entry(device, devices, dev_list) { | |
499f377f JM |
10986 | ret = btrfs_trim_free_extents(device, range->minlen, |
10987 | &group_trimmed); | |
93bba24d QW |
10988 | if (ret) { |
10989 | dev_failed++; | |
10990 | dev_ret = ret; | |
499f377f | 10991 | break; |
93bba24d | 10992 | } |
499f377f JM |
10993 | |
10994 | trimmed += group_trimmed; | |
10995 | } | |
0b246afa | 10996 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 10997 | |
93bba24d QW |
10998 | if (dev_failed) |
10999 | btrfs_warn(fs_info, | |
11000 | "failed to trim %llu device(s), last error %d", | |
11001 | dev_failed, dev_ret); | |
f7039b1d | 11002 | range->len = trimmed; |
93bba24d QW |
11003 | if (bg_ret) |
11004 | return bg_ret; | |
11005 | return dev_ret; | |
f7039b1d | 11006 | } |
8257b2dc MX |
11007 | |
11008 | /* | |
ea14b57f | 11009 | * btrfs_{start,end}_write_no_snapshotting() are similar to |
9ea24bbe FM |
11010 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing |
11011 | * data into the page cache through nocow before the subvolume is snapshoted, | |
11012 | * but flush the data into disk after the snapshot creation, or to prevent | |
ea14b57f | 11013 | * operations while snapshotting is ongoing and that cause the snapshot to be |
9ea24bbe | 11014 | * inconsistent (writes followed by expanding truncates for example). |
8257b2dc | 11015 | */ |
ea14b57f | 11016 | void btrfs_end_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc MX |
11017 | { |
11018 | percpu_counter_dec(&root->subv_writers->counter); | |
093258e6 | 11019 | cond_wake_up(&root->subv_writers->wait); |
8257b2dc MX |
11020 | } |
11021 | ||
ea14b57f | 11022 | int btrfs_start_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc | 11023 | { |
ea14b57f | 11024 | if (atomic_read(&root->will_be_snapshotted)) |
8257b2dc MX |
11025 | return 0; |
11026 | ||
11027 | percpu_counter_inc(&root->subv_writers->counter); | |
11028 | /* | |
11029 | * Make sure counter is updated before we check for snapshot creation. | |
11030 | */ | |
11031 | smp_mb(); | |
ea14b57f DS |
11032 | if (atomic_read(&root->will_be_snapshotted)) { |
11033 | btrfs_end_write_no_snapshotting(root); | |
8257b2dc MX |
11034 | return 0; |
11035 | } | |
11036 | return 1; | |
11037 | } | |
0bc19f90 | 11038 | |
0bc19f90 ZL |
11039 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) |
11040 | { | |
11041 | while (true) { | |
11042 | int ret; | |
11043 | ||
ea14b57f | 11044 | ret = btrfs_start_write_no_snapshotting(root); |
0bc19f90 ZL |
11045 | if (ret) |
11046 | break; | |
4625956a PZ |
11047 | wait_var_event(&root->will_be_snapshotted, |
11048 | !atomic_read(&root->will_be_snapshotted)); | |
0bc19f90 ZL |
11049 | } |
11050 | } | |
031f24da QW |
11051 | |
11052 | void btrfs_mark_bg_unused(struct btrfs_block_group_cache *bg) | |
11053 | { | |
11054 | struct btrfs_fs_info *fs_info = bg->fs_info; | |
11055 | ||
11056 | spin_lock(&fs_info->unused_bgs_lock); | |
11057 | if (list_empty(&bg->bg_list)) { | |
11058 | btrfs_get_block_group(bg); | |
11059 | trace_btrfs_add_unused_block_group(bg); | |
11060 | list_add_tail(&bg->bg_list, &fs_info->unused_bgs); | |
11061 | } | |
11062 | spin_unlock(&fs_info->unused_bgs_lock); | |
11063 | } |