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" |
8719aaae | 31 | #include "space-info.h" |
fec577fb | 32 | |
709c0486 AJ |
33 | #undef SCRAMBLE_DELAYED_REFS |
34 | ||
9f9b8e8d | 35 | |
5d4f98a2 | 36 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
37 | struct btrfs_delayed_ref_node *node, u64 parent, |
38 | u64 root_objectid, u64 owner_objectid, | |
39 | u64 owner_offset, int refs_to_drop, | |
40 | struct btrfs_delayed_extent_op *extra_op); | |
5d4f98a2 YZ |
41 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
42 | struct extent_buffer *leaf, | |
43 | struct btrfs_extent_item *ei); | |
44 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
45 | u64 parent, u64 root_objectid, |
46 | u64 flags, u64 owner, u64 offset, | |
47 | struct btrfs_key *ins, int ref_mod); | |
48 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
4e6bd4e0 | 49 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 50 | struct btrfs_delayed_extent_op *extent_op); |
11833d66 YZ |
51 | static int find_next_key(struct btrfs_path *path, int level, |
52 | struct btrfs_key *key); | |
ab8d0fc4 JM |
53 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
54 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 55 | int dump_block_groups); |
6a63209f | 56 | |
817d52f8 JB |
57 | static noinline int |
58 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
59 | { | |
60 | smp_mb(); | |
36cce922 JB |
61 | return cache->cached == BTRFS_CACHE_FINISHED || |
62 | cache->cached == BTRFS_CACHE_ERROR; | |
817d52f8 JB |
63 | } |
64 | ||
0f9dd46c JB |
65 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
66 | { | |
67 | return (cache->flags & bits) == bits; | |
68 | } | |
69 | ||
758f2dfc | 70 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
71 | { |
72 | atomic_inc(&cache->count); | |
73 | } | |
74 | ||
75 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
76 | { | |
f0486c68 YZ |
77 | if (atomic_dec_and_test(&cache->count)) { |
78 | WARN_ON(cache->pinned > 0); | |
79 | WARN_ON(cache->reserved > 0); | |
0966a7b1 QW |
80 | |
81 | /* | |
82 | * If not empty, someone is still holding mutex of | |
83 | * full_stripe_lock, which can only be released by caller. | |
84 | * And it will definitely cause use-after-free when caller | |
85 | * tries to release full stripe lock. | |
86 | * | |
87 | * No better way to resolve, but only to warn. | |
88 | */ | |
89 | WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); | |
34d52cb6 | 90 | kfree(cache->free_space_ctl); |
11dfe35a | 91 | kfree(cache); |
f0486c68 | 92 | } |
11dfe35a JB |
93 | } |
94 | ||
0f9dd46c JB |
95 | /* |
96 | * this adds the block group to the fs_info rb tree for the block group | |
97 | * cache | |
98 | */ | |
b2950863 | 99 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
100 | struct btrfs_block_group_cache *block_group) |
101 | { | |
102 | struct rb_node **p; | |
103 | struct rb_node *parent = NULL; | |
104 | struct btrfs_block_group_cache *cache; | |
105 | ||
106 | spin_lock(&info->block_group_cache_lock); | |
107 | p = &info->block_group_cache_tree.rb_node; | |
108 | ||
109 | while (*p) { | |
110 | parent = *p; | |
111 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
112 | cache_node); | |
113 | if (block_group->key.objectid < cache->key.objectid) { | |
114 | p = &(*p)->rb_left; | |
115 | } else if (block_group->key.objectid > cache->key.objectid) { | |
116 | p = &(*p)->rb_right; | |
117 | } else { | |
118 | spin_unlock(&info->block_group_cache_lock); | |
119 | return -EEXIST; | |
120 | } | |
121 | } | |
122 | ||
123 | rb_link_node(&block_group->cache_node, parent, p); | |
124 | rb_insert_color(&block_group->cache_node, | |
125 | &info->block_group_cache_tree); | |
a1897fdd LB |
126 | |
127 | if (info->first_logical_byte > block_group->key.objectid) | |
128 | info->first_logical_byte = block_group->key.objectid; | |
129 | ||
0f9dd46c JB |
130 | spin_unlock(&info->block_group_cache_lock); |
131 | ||
132 | return 0; | |
133 | } | |
134 | ||
135 | /* | |
136 | * This will return the block group at or after bytenr if contains is 0, else | |
137 | * it will return the block group that contains the bytenr | |
138 | */ | |
139 | static struct btrfs_block_group_cache * | |
140 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
141 | int contains) | |
142 | { | |
143 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
144 | struct rb_node *n; | |
145 | u64 end, start; | |
146 | ||
147 | spin_lock(&info->block_group_cache_lock); | |
148 | n = info->block_group_cache_tree.rb_node; | |
149 | ||
150 | while (n) { | |
151 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
152 | cache_node); | |
153 | end = cache->key.objectid + cache->key.offset - 1; | |
154 | start = cache->key.objectid; | |
155 | ||
156 | if (bytenr < start) { | |
157 | if (!contains && (!ret || start < ret->key.objectid)) | |
158 | ret = cache; | |
159 | n = n->rb_left; | |
160 | } else if (bytenr > start) { | |
161 | if (contains && bytenr <= end) { | |
162 | ret = cache; | |
163 | break; | |
164 | } | |
165 | n = n->rb_right; | |
166 | } else { | |
167 | ret = cache; | |
168 | break; | |
169 | } | |
170 | } | |
a1897fdd | 171 | if (ret) { |
11dfe35a | 172 | btrfs_get_block_group(ret); |
a1897fdd LB |
173 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
174 | info->first_logical_byte = ret->key.objectid; | |
175 | } | |
0f9dd46c JB |
176 | spin_unlock(&info->block_group_cache_lock); |
177 | ||
178 | return ret; | |
179 | } | |
180 | ||
2ff7e61e | 181 | static int add_excluded_extent(struct btrfs_fs_info *fs_info, |
11833d66 | 182 | u64 start, u64 num_bytes) |
817d52f8 | 183 | { |
11833d66 | 184 | u64 end = start + num_bytes - 1; |
0b246afa | 185 | set_extent_bits(&fs_info->freed_extents[0], |
ceeb0ae7 | 186 | start, end, EXTENT_UPTODATE); |
0b246afa | 187 | set_extent_bits(&fs_info->freed_extents[1], |
ceeb0ae7 | 188 | start, end, EXTENT_UPTODATE); |
11833d66 YZ |
189 | return 0; |
190 | } | |
817d52f8 | 191 | |
9e715da8 | 192 | static void free_excluded_extents(struct btrfs_block_group_cache *cache) |
11833d66 | 193 | { |
9e715da8 | 194 | struct btrfs_fs_info *fs_info = cache->fs_info; |
11833d66 | 195 | u64 start, end; |
817d52f8 | 196 | |
11833d66 YZ |
197 | start = cache->key.objectid; |
198 | end = start + cache->key.offset - 1; | |
199 | ||
0b246afa | 200 | clear_extent_bits(&fs_info->freed_extents[0], |
91166212 | 201 | start, end, EXTENT_UPTODATE); |
0b246afa | 202 | clear_extent_bits(&fs_info->freed_extents[1], |
91166212 | 203 | start, end, EXTENT_UPTODATE); |
817d52f8 JB |
204 | } |
205 | ||
3c4da657 | 206 | static int exclude_super_stripes(struct btrfs_block_group_cache *cache) |
817d52f8 | 207 | { |
3c4da657 | 208 | struct btrfs_fs_info *fs_info = cache->fs_info; |
817d52f8 JB |
209 | u64 bytenr; |
210 | u64 *logical; | |
211 | int stripe_len; | |
212 | int i, nr, ret; | |
213 | ||
06b2331f YZ |
214 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
215 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
216 | cache->bytes_super += stripe_len; | |
2ff7e61e | 217 | ret = add_excluded_extent(fs_info, cache->key.objectid, |
06b2331f | 218 | stripe_len); |
835d974f JB |
219 | if (ret) |
220 | return ret; | |
06b2331f YZ |
221 | } |
222 | ||
817d52f8 JB |
223 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
224 | bytenr = btrfs_sb_offset(i); | |
0b246afa | 225 | ret = btrfs_rmap_block(fs_info, cache->key.objectid, |
63a9c7b9 | 226 | bytenr, &logical, &nr, &stripe_len); |
835d974f JB |
227 | if (ret) |
228 | return ret; | |
11833d66 | 229 | |
817d52f8 | 230 | while (nr--) { |
51bf5f0b JB |
231 | u64 start, len; |
232 | ||
233 | if (logical[nr] > cache->key.objectid + | |
234 | cache->key.offset) | |
235 | continue; | |
236 | ||
237 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
238 | continue; | |
239 | ||
240 | start = logical[nr]; | |
241 | if (start < cache->key.objectid) { | |
242 | start = cache->key.objectid; | |
243 | len = (logical[nr] + stripe_len) - start; | |
244 | } else { | |
245 | len = min_t(u64, stripe_len, | |
246 | cache->key.objectid + | |
247 | cache->key.offset - start); | |
248 | } | |
249 | ||
250 | cache->bytes_super += len; | |
2ff7e61e | 251 | ret = add_excluded_extent(fs_info, start, len); |
835d974f JB |
252 | if (ret) { |
253 | kfree(logical); | |
254 | return ret; | |
255 | } | |
817d52f8 | 256 | } |
11833d66 | 257 | |
817d52f8 JB |
258 | kfree(logical); |
259 | } | |
817d52f8 JB |
260 | return 0; |
261 | } | |
262 | ||
11833d66 YZ |
263 | static struct btrfs_caching_control * |
264 | get_caching_control(struct btrfs_block_group_cache *cache) | |
265 | { | |
266 | struct btrfs_caching_control *ctl; | |
267 | ||
268 | spin_lock(&cache->lock); | |
dde5abee JB |
269 | if (!cache->caching_ctl) { |
270 | spin_unlock(&cache->lock); | |
11833d66 YZ |
271 | return NULL; |
272 | } | |
273 | ||
274 | ctl = cache->caching_ctl; | |
1e4f4714 | 275 | refcount_inc(&ctl->count); |
11833d66 YZ |
276 | spin_unlock(&cache->lock); |
277 | return ctl; | |
278 | } | |
279 | ||
280 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
281 | { | |
1e4f4714 | 282 | if (refcount_dec_and_test(&ctl->count)) |
11833d66 YZ |
283 | kfree(ctl); |
284 | } | |
285 | ||
d0bd4560 | 286 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 287 | static void fragment_free_space(struct btrfs_block_group_cache *block_group) |
d0bd4560 | 288 | { |
2ff7e61e | 289 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
d0bd4560 JB |
290 | u64 start = block_group->key.objectid; |
291 | u64 len = block_group->key.offset; | |
292 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? | |
0b246afa | 293 | fs_info->nodesize : fs_info->sectorsize; |
d0bd4560 JB |
294 | u64 step = chunk << 1; |
295 | ||
296 | while (len > chunk) { | |
297 | btrfs_remove_free_space(block_group, start, chunk); | |
298 | start += step; | |
299 | if (len < step) | |
300 | len = 0; | |
301 | else | |
302 | len -= step; | |
303 | } | |
304 | } | |
305 | #endif | |
306 | ||
0f9dd46c JB |
307 | /* |
308 | * this is only called by cache_block_group, since we could have freed extents | |
309 | * we need to check the pinned_extents for any extents that can't be used yet | |
310 | * since their free space will be released as soon as the transaction commits. | |
311 | */ | |
a5ed9182 | 312 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
4457c1c7 | 313 | u64 start, u64 end) |
0f9dd46c | 314 | { |
4457c1c7 | 315 | struct btrfs_fs_info *info = block_group->fs_info; |
817d52f8 | 316 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
317 | int ret; |
318 | ||
319 | while (start < end) { | |
11833d66 | 320 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 321 | &extent_start, &extent_end, |
e6138876 JB |
322 | EXTENT_DIRTY | EXTENT_UPTODATE, |
323 | NULL); | |
0f9dd46c JB |
324 | if (ret) |
325 | break; | |
326 | ||
06b2331f | 327 | if (extent_start <= start) { |
0f9dd46c JB |
328 | start = extent_end + 1; |
329 | } else if (extent_start > start && extent_start < end) { | |
330 | size = extent_start - start; | |
817d52f8 | 331 | total_added += size; |
ea6a478e JB |
332 | ret = btrfs_add_free_space(block_group, start, |
333 | size); | |
79787eaa | 334 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
335 | start = extent_end + 1; |
336 | } else { | |
337 | break; | |
338 | } | |
339 | } | |
340 | ||
341 | if (start < end) { | |
342 | size = end - start; | |
817d52f8 | 343 | total_added += size; |
ea6a478e | 344 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 345 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
346 | } |
347 | ||
817d52f8 | 348 | return total_added; |
0f9dd46c JB |
349 | } |
350 | ||
73fa48b6 | 351 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) |
e37c9e69 | 352 | { |
0b246afa JM |
353 | struct btrfs_block_group_cache *block_group = caching_ctl->block_group; |
354 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
355 | struct btrfs_root *extent_root = fs_info->extent_root; | |
e37c9e69 | 356 | struct btrfs_path *path; |
5f39d397 | 357 | struct extent_buffer *leaf; |
11833d66 | 358 | struct btrfs_key key; |
817d52f8 | 359 | u64 total_found = 0; |
11833d66 YZ |
360 | u64 last = 0; |
361 | u32 nritems; | |
73fa48b6 | 362 | int ret; |
d0bd4560 | 363 | bool wakeup = true; |
f510cfec | 364 | |
e37c9e69 CM |
365 | path = btrfs_alloc_path(); |
366 | if (!path) | |
73fa48b6 | 367 | return -ENOMEM; |
7d7d6068 | 368 | |
817d52f8 | 369 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 370 | |
d0bd4560 JB |
371 | #ifdef CONFIG_BTRFS_DEBUG |
372 | /* | |
373 | * If we're fragmenting we don't want to make anybody think we can | |
374 | * allocate from this block group until we've had a chance to fragment | |
375 | * the free space. | |
376 | */ | |
2ff7e61e | 377 | if (btrfs_should_fragment_free_space(block_group)) |
d0bd4560 JB |
378 | wakeup = false; |
379 | #endif | |
5cd57b2c | 380 | /* |
817d52f8 JB |
381 | * We don't want to deadlock with somebody trying to allocate a new |
382 | * extent for the extent root while also trying to search the extent | |
383 | * root to add free space. So we skip locking and search the commit | |
384 | * root, since its read-only | |
5cd57b2c CM |
385 | */ |
386 | path->skip_locking = 1; | |
817d52f8 | 387 | path->search_commit_root = 1; |
e4058b54 | 388 | path->reada = READA_FORWARD; |
817d52f8 | 389 | |
e4404d6e | 390 | key.objectid = last; |
e37c9e69 | 391 | key.offset = 0; |
11833d66 | 392 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 393 | |
52ee28d2 | 394 | next: |
11833d66 | 395 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 396 | if (ret < 0) |
73fa48b6 | 397 | goto out; |
a512bbf8 | 398 | |
11833d66 YZ |
399 | leaf = path->nodes[0]; |
400 | nritems = btrfs_header_nritems(leaf); | |
401 | ||
d397712b | 402 | while (1) { |
7841cb28 | 403 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 404 | last = (u64)-1; |
817d52f8 | 405 | break; |
f25784b3 | 406 | } |
817d52f8 | 407 | |
11833d66 YZ |
408 | if (path->slots[0] < nritems) { |
409 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
410 | } else { | |
411 | ret = find_next_key(path, 0, &key); | |
412 | if (ret) | |
e37c9e69 | 413 | break; |
817d52f8 | 414 | |
c9ea7b24 | 415 | if (need_resched() || |
9e351cc8 | 416 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
d0bd4560 JB |
417 | if (wakeup) |
418 | caching_ctl->progress = last; | |
ff5714cc | 419 | btrfs_release_path(path); |
9e351cc8 | 420 | up_read(&fs_info->commit_root_sem); |
589d8ade | 421 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 422 | cond_resched(); |
73fa48b6 OS |
423 | mutex_lock(&caching_ctl->mutex); |
424 | down_read(&fs_info->commit_root_sem); | |
425 | goto next; | |
589d8ade | 426 | } |
0a3896d0 JB |
427 | |
428 | ret = btrfs_next_leaf(extent_root, path); | |
429 | if (ret < 0) | |
73fa48b6 | 430 | goto out; |
0a3896d0 JB |
431 | if (ret) |
432 | break; | |
589d8ade JB |
433 | leaf = path->nodes[0]; |
434 | nritems = btrfs_header_nritems(leaf); | |
435 | continue; | |
11833d66 | 436 | } |
817d52f8 | 437 | |
52ee28d2 LB |
438 | if (key.objectid < last) { |
439 | key.objectid = last; | |
440 | key.offset = 0; | |
441 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
442 | ||
d0bd4560 JB |
443 | if (wakeup) |
444 | caching_ctl->progress = last; | |
52ee28d2 LB |
445 | btrfs_release_path(path); |
446 | goto next; | |
447 | } | |
448 | ||
11833d66 YZ |
449 | if (key.objectid < block_group->key.objectid) { |
450 | path->slots[0]++; | |
817d52f8 | 451 | continue; |
e37c9e69 | 452 | } |
0f9dd46c | 453 | |
e37c9e69 | 454 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 455 | block_group->key.offset) |
e37c9e69 | 456 | break; |
7d7d6068 | 457 | |
3173a18f JB |
458 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
459 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
4457c1c7 | 460 | total_found += add_new_free_space(block_group, last, |
817d52f8 | 461 | key.objectid); |
3173a18f JB |
462 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
463 | last = key.objectid + | |
da17066c | 464 | fs_info->nodesize; |
3173a18f JB |
465 | else |
466 | last = key.objectid + key.offset; | |
817d52f8 | 467 | |
73fa48b6 | 468 | if (total_found > CACHING_CTL_WAKE_UP) { |
11833d66 | 469 | total_found = 0; |
d0bd4560 JB |
470 | if (wakeup) |
471 | wake_up(&caching_ctl->wait); | |
11833d66 | 472 | } |
817d52f8 | 473 | } |
e37c9e69 CM |
474 | path->slots[0]++; |
475 | } | |
817d52f8 | 476 | ret = 0; |
e37c9e69 | 477 | |
4457c1c7 | 478 | total_found += add_new_free_space(block_group, last, |
817d52f8 JB |
479 | block_group->key.objectid + |
480 | block_group->key.offset); | |
11833d66 | 481 | caching_ctl->progress = (u64)-1; |
817d52f8 | 482 | |
73fa48b6 OS |
483 | out: |
484 | btrfs_free_path(path); | |
485 | return ret; | |
486 | } | |
487 | ||
488 | static noinline void caching_thread(struct btrfs_work *work) | |
489 | { | |
490 | struct btrfs_block_group_cache *block_group; | |
491 | struct btrfs_fs_info *fs_info; | |
492 | struct btrfs_caching_control *caching_ctl; | |
493 | int ret; | |
494 | ||
495 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
496 | block_group = caching_ctl->block_group; | |
497 | fs_info = block_group->fs_info; | |
498 | ||
499 | mutex_lock(&caching_ctl->mutex); | |
500 | down_read(&fs_info->commit_root_sem); | |
501 | ||
1e144fb8 OS |
502 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
503 | ret = load_free_space_tree(caching_ctl); | |
504 | else | |
505 | ret = load_extent_tree_free(caching_ctl); | |
73fa48b6 | 506 | |
817d52f8 | 507 | spin_lock(&block_group->lock); |
11833d66 | 508 | block_group->caching_ctl = NULL; |
73fa48b6 | 509 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; |
817d52f8 | 510 | spin_unlock(&block_group->lock); |
0f9dd46c | 511 | |
d0bd4560 | 512 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 513 | if (btrfs_should_fragment_free_space(block_group)) { |
d0bd4560 JB |
514 | u64 bytes_used; |
515 | ||
516 | spin_lock(&block_group->space_info->lock); | |
517 | spin_lock(&block_group->lock); | |
518 | bytes_used = block_group->key.offset - | |
519 | btrfs_block_group_used(&block_group->item); | |
520 | block_group->space_info->bytes_used += bytes_used >> 1; | |
521 | spin_unlock(&block_group->lock); | |
522 | spin_unlock(&block_group->space_info->lock); | |
2ff7e61e | 523 | fragment_free_space(block_group); |
d0bd4560 JB |
524 | } |
525 | #endif | |
526 | ||
527 | caching_ctl->progress = (u64)-1; | |
11833d66 | 528 | |
9e351cc8 | 529 | up_read(&fs_info->commit_root_sem); |
9e715da8 | 530 | free_excluded_extents(block_group); |
11833d66 | 531 | mutex_unlock(&caching_ctl->mutex); |
73fa48b6 | 532 | |
11833d66 YZ |
533 | wake_up(&caching_ctl->wait); |
534 | ||
535 | put_caching_control(caching_ctl); | |
11dfe35a | 536 | btrfs_put_block_group(block_group); |
817d52f8 JB |
537 | } |
538 | ||
9d66e233 | 539 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 540 | int load_cache_only) |
817d52f8 | 541 | { |
291c7d2f | 542 | DEFINE_WAIT(wait); |
11833d66 YZ |
543 | struct btrfs_fs_info *fs_info = cache->fs_info; |
544 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
545 | int ret = 0; |
546 | ||
291c7d2f | 547 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
548 | if (!caching_ctl) |
549 | return -ENOMEM; | |
291c7d2f JB |
550 | |
551 | INIT_LIST_HEAD(&caching_ctl->list); | |
552 | mutex_init(&caching_ctl->mutex); | |
553 | init_waitqueue_head(&caching_ctl->wait); | |
554 | caching_ctl->block_group = cache; | |
555 | caching_ctl->progress = cache->key.objectid; | |
1e4f4714 | 556 | refcount_set(&caching_ctl->count, 1); |
9e0af237 LB |
557 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, |
558 | caching_thread, NULL, NULL); | |
291c7d2f JB |
559 | |
560 | spin_lock(&cache->lock); | |
561 | /* | |
562 | * This should be a rare occasion, but this could happen I think in the | |
563 | * case where one thread starts to load the space cache info, and then | |
564 | * some other thread starts a transaction commit which tries to do an | |
565 | * allocation while the other thread is still loading the space cache | |
566 | * info. The previous loop should have kept us from choosing this block | |
567 | * group, but if we've moved to the state where we will wait on caching | |
568 | * block groups we need to first check if we're doing a fast load here, | |
569 | * so we can wait for it to finish, otherwise we could end up allocating | |
570 | * from a block group who's cache gets evicted for one reason or | |
571 | * another. | |
572 | */ | |
573 | while (cache->cached == BTRFS_CACHE_FAST) { | |
574 | struct btrfs_caching_control *ctl; | |
575 | ||
576 | ctl = cache->caching_ctl; | |
1e4f4714 | 577 | refcount_inc(&ctl->count); |
291c7d2f JB |
578 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); |
579 | spin_unlock(&cache->lock); | |
580 | ||
581 | schedule(); | |
582 | ||
583 | finish_wait(&ctl->wait, &wait); | |
584 | put_caching_control(ctl); | |
585 | spin_lock(&cache->lock); | |
586 | } | |
587 | ||
588 | if (cache->cached != BTRFS_CACHE_NO) { | |
589 | spin_unlock(&cache->lock); | |
590 | kfree(caching_ctl); | |
11833d66 | 591 | return 0; |
291c7d2f JB |
592 | } |
593 | WARN_ON(cache->caching_ctl); | |
594 | cache->caching_ctl = caching_ctl; | |
595 | cache->cached = BTRFS_CACHE_FAST; | |
596 | spin_unlock(&cache->lock); | |
11833d66 | 597 | |
d8953d69 | 598 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cb83b7b8 | 599 | mutex_lock(&caching_ctl->mutex); |
bb6cb1c5 | 600 | ret = load_free_space_cache(cache); |
9d66e233 JB |
601 | |
602 | spin_lock(&cache->lock); | |
603 | if (ret == 1) { | |
291c7d2f | 604 | cache->caching_ctl = NULL; |
9d66e233 JB |
605 | cache->cached = BTRFS_CACHE_FINISHED; |
606 | cache->last_byte_to_unpin = (u64)-1; | |
cb83b7b8 | 607 | caching_ctl->progress = (u64)-1; |
9d66e233 | 608 | } else { |
291c7d2f JB |
609 | if (load_cache_only) { |
610 | cache->caching_ctl = NULL; | |
611 | cache->cached = BTRFS_CACHE_NO; | |
612 | } else { | |
613 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 614 | cache->has_caching_ctl = 1; |
291c7d2f | 615 | } |
9d66e233 JB |
616 | } |
617 | spin_unlock(&cache->lock); | |
d0bd4560 JB |
618 | #ifdef CONFIG_BTRFS_DEBUG |
619 | if (ret == 1 && | |
2ff7e61e | 620 | btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
621 | u64 bytes_used; |
622 | ||
623 | spin_lock(&cache->space_info->lock); | |
624 | spin_lock(&cache->lock); | |
625 | bytes_used = cache->key.offset - | |
626 | btrfs_block_group_used(&cache->item); | |
627 | cache->space_info->bytes_used += bytes_used >> 1; | |
628 | spin_unlock(&cache->lock); | |
629 | spin_unlock(&cache->space_info->lock); | |
2ff7e61e | 630 | fragment_free_space(cache); |
d0bd4560 JB |
631 | } |
632 | #endif | |
cb83b7b8 JB |
633 | mutex_unlock(&caching_ctl->mutex); |
634 | ||
291c7d2f | 635 | wake_up(&caching_ctl->wait); |
3c14874a | 636 | if (ret == 1) { |
291c7d2f | 637 | put_caching_control(caching_ctl); |
9e715da8 | 638 | free_excluded_extents(cache); |
9d66e233 | 639 | return 0; |
3c14874a | 640 | } |
291c7d2f JB |
641 | } else { |
642 | /* | |
1e144fb8 OS |
643 | * We're either using the free space tree or no caching at all. |
644 | * Set cached to the appropriate value and wakeup any waiters. | |
291c7d2f JB |
645 | */ |
646 | spin_lock(&cache->lock); | |
647 | if (load_cache_only) { | |
648 | cache->caching_ctl = NULL; | |
649 | cache->cached = BTRFS_CACHE_NO; | |
650 | } else { | |
651 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 652 | cache->has_caching_ctl = 1; |
291c7d2f JB |
653 | } |
654 | spin_unlock(&cache->lock); | |
655 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
656 | } |
657 | ||
291c7d2f JB |
658 | if (load_cache_only) { |
659 | put_caching_control(caching_ctl); | |
11833d66 | 660 | return 0; |
817d52f8 | 661 | } |
817d52f8 | 662 | |
9e351cc8 | 663 | down_write(&fs_info->commit_root_sem); |
1e4f4714 | 664 | refcount_inc(&caching_ctl->count); |
11833d66 | 665 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
9e351cc8 | 666 | up_write(&fs_info->commit_root_sem); |
11833d66 | 667 | |
11dfe35a | 668 | btrfs_get_block_group(cache); |
11833d66 | 669 | |
e66f0bb1 | 670 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 671 | |
ef8bbdfe | 672 | return ret; |
e37c9e69 CM |
673 | } |
674 | ||
0f9dd46c JB |
675 | /* |
676 | * return the block group that starts at or after bytenr | |
677 | */ | |
d397712b CM |
678 | static struct btrfs_block_group_cache * |
679 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 680 | { |
e2c89907 | 681 | return block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b CM |
682 | } |
683 | ||
0f9dd46c | 684 | /* |
9f55684c | 685 | * return the block group that contains the given bytenr |
0f9dd46c | 686 | */ |
d397712b CM |
687 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
688 | struct btrfs_fs_info *info, | |
689 | u64 bytenr) | |
be744175 | 690 | { |
e2c89907 | 691 | return block_group_cache_tree_search(info, bytenr, 1); |
be744175 | 692 | } |
0b86a832 | 693 | |
78192442 | 694 | static u64 generic_ref_to_space_flags(struct btrfs_ref *ref) |
0d9f824d | 695 | { |
ddf30cf0 QW |
696 | if (ref->type == BTRFS_REF_METADATA) { |
697 | if (ref->tree_ref.root == BTRFS_CHUNK_TREE_OBJECTID) | |
78192442 | 698 | return BTRFS_BLOCK_GROUP_SYSTEM; |
0d9f824d | 699 | else |
78192442 | 700 | return BTRFS_BLOCK_GROUP_METADATA; |
0d9f824d | 701 | } |
78192442 QW |
702 | return BTRFS_BLOCK_GROUP_DATA; |
703 | } | |
704 | ||
705 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, | |
706 | struct btrfs_ref *ref) | |
707 | { | |
708 | struct btrfs_space_info *space_info; | |
709 | u64 flags = generic_ref_to_space_flags(ref); | |
710 | ||
280c2908 | 711 | space_info = btrfs_find_space_info(fs_info, flags); |
78192442 QW |
712 | ASSERT(space_info); |
713 | percpu_counter_add_batch(&space_info->total_bytes_pinned, ref->len, | |
714 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
715 | } | |
716 | ||
717 | static void sub_pinned_bytes(struct btrfs_fs_info *fs_info, | |
718 | struct btrfs_ref *ref) | |
719 | { | |
720 | struct btrfs_space_info *space_info; | |
721 | u64 flags = generic_ref_to_space_flags(ref); | |
0d9f824d | 722 | |
280c2908 | 723 | space_info = btrfs_find_space_info(fs_info, flags); |
55e8196a | 724 | ASSERT(space_info); |
78192442 | 725 | percpu_counter_add_batch(&space_info->total_bytes_pinned, -ref->len, |
dec59fa3 | 726 | BTRFS_TOTAL_BYTES_PINNED_BATCH); |
0d9f824d OS |
727 | } |
728 | ||
1a4ed8fd | 729 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 730 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
731 | { |
732 | int ret; | |
733 | struct btrfs_key key; | |
31840ae1 | 734 | struct btrfs_path *path; |
e02119d5 | 735 | |
31840ae1 | 736 | path = btrfs_alloc_path(); |
d8926bb3 MF |
737 | if (!path) |
738 | return -ENOMEM; | |
739 | ||
e02119d5 CM |
740 | key.objectid = start; |
741 | key.offset = len; | |
3173a18f | 742 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 743 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 744 | btrfs_free_path(path); |
7bb86316 CM |
745 | return ret; |
746 | } | |
747 | ||
a22285a6 | 748 | /* |
3173a18f | 749 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
750 | * |
751 | * the head node for delayed ref is used to store the sum of all the | |
752 | * reference count modifications queued up in the rbtree. the head | |
753 | * node may also store the extent flags to set. This way you can check | |
754 | * to see what the reference count and extent flags would be if all of | |
755 | * the delayed refs are not processed. | |
756 | */ | |
757 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 758 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 759 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
760 | { |
761 | struct btrfs_delayed_ref_head *head; | |
762 | struct btrfs_delayed_ref_root *delayed_refs; | |
763 | struct btrfs_path *path; | |
764 | struct btrfs_extent_item *ei; | |
765 | struct extent_buffer *leaf; | |
766 | struct btrfs_key key; | |
767 | u32 item_size; | |
768 | u64 num_refs; | |
769 | u64 extent_flags; | |
770 | int ret; | |
771 | ||
3173a18f JB |
772 | /* |
773 | * If we don't have skinny metadata, don't bother doing anything | |
774 | * different | |
775 | */ | |
0b246afa JM |
776 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
777 | offset = fs_info->nodesize; | |
3173a18f JB |
778 | metadata = 0; |
779 | } | |
780 | ||
a22285a6 YZ |
781 | path = btrfs_alloc_path(); |
782 | if (!path) | |
783 | return -ENOMEM; | |
784 | ||
a22285a6 YZ |
785 | if (!trans) { |
786 | path->skip_locking = 1; | |
787 | path->search_commit_root = 1; | |
788 | } | |
639eefc8 FDBM |
789 | |
790 | search_again: | |
791 | key.objectid = bytenr; | |
792 | key.offset = offset; | |
793 | if (metadata) | |
794 | key.type = BTRFS_METADATA_ITEM_KEY; | |
795 | else | |
796 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
797 | ||
0b246afa | 798 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
799 | if (ret < 0) |
800 | goto out_free; | |
801 | ||
3173a18f | 802 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
803 | if (path->slots[0]) { |
804 | path->slots[0]--; | |
805 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
806 | path->slots[0]); | |
807 | if (key.objectid == bytenr && | |
808 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 809 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
810 | ret = 0; |
811 | } | |
3173a18f JB |
812 | } |
813 | ||
a22285a6 YZ |
814 | if (ret == 0) { |
815 | leaf = path->nodes[0]; | |
816 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
817 | if (item_size >= sizeof(*ei)) { | |
818 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
819 | struct btrfs_extent_item); | |
820 | num_refs = btrfs_extent_refs(leaf, ei); | |
821 | extent_flags = btrfs_extent_flags(leaf, ei); | |
822 | } else { | |
ba3c2b19 NB |
823 | ret = -EINVAL; |
824 | btrfs_print_v0_err(fs_info); | |
825 | if (trans) | |
826 | btrfs_abort_transaction(trans, ret); | |
827 | else | |
828 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
829 | ||
830 | goto out_free; | |
a22285a6 | 831 | } |
ba3c2b19 | 832 | |
a22285a6 YZ |
833 | BUG_ON(num_refs == 0); |
834 | } else { | |
835 | num_refs = 0; | |
836 | extent_flags = 0; | |
837 | ret = 0; | |
838 | } | |
839 | ||
840 | if (!trans) | |
841 | goto out; | |
842 | ||
843 | delayed_refs = &trans->transaction->delayed_refs; | |
844 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 845 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
846 | if (head) { |
847 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 848 | refcount_inc(&head->refs); |
a22285a6 YZ |
849 | spin_unlock(&delayed_refs->lock); |
850 | ||
b3b4aa74 | 851 | btrfs_release_path(path); |
a22285a6 | 852 | |
8cc33e5c DS |
853 | /* |
854 | * Mutex was contended, block until it's released and try | |
855 | * again | |
856 | */ | |
a22285a6 YZ |
857 | mutex_lock(&head->mutex); |
858 | mutex_unlock(&head->mutex); | |
d278850e | 859 | btrfs_put_delayed_ref_head(head); |
639eefc8 | 860 | goto search_again; |
a22285a6 | 861 | } |
d7df2c79 | 862 | spin_lock(&head->lock); |
a22285a6 YZ |
863 | if (head->extent_op && head->extent_op->update_flags) |
864 | extent_flags |= head->extent_op->flags_to_set; | |
865 | else | |
866 | BUG_ON(num_refs == 0); | |
867 | ||
d278850e | 868 | num_refs += head->ref_mod; |
d7df2c79 | 869 | spin_unlock(&head->lock); |
a22285a6 YZ |
870 | mutex_unlock(&head->mutex); |
871 | } | |
872 | spin_unlock(&delayed_refs->lock); | |
873 | out: | |
874 | WARN_ON(num_refs == 0); | |
875 | if (refs) | |
876 | *refs = num_refs; | |
877 | if (flags) | |
878 | *flags = extent_flags; | |
879 | out_free: | |
880 | btrfs_free_path(path); | |
881 | return ret; | |
882 | } | |
883 | ||
d8d5f3e1 CM |
884 | /* |
885 | * Back reference rules. Back refs have three main goals: | |
886 | * | |
887 | * 1) differentiate between all holders of references to an extent so that | |
888 | * when a reference is dropped we can make sure it was a valid reference | |
889 | * before freeing the extent. | |
890 | * | |
891 | * 2) Provide enough information to quickly find the holders of an extent | |
892 | * if we notice a given block is corrupted or bad. | |
893 | * | |
894 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
895 | * maintenance. This is actually the same as #2, but with a slightly | |
896 | * different use case. | |
897 | * | |
5d4f98a2 YZ |
898 | * There are two kinds of back refs. The implicit back refs is optimized |
899 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
900 | * back refs of this kind provide information about the block's owner tree | |
901 | * and the pointer's key. These information allow us to find the block by | |
902 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
903 | * referenced by their owner trees. The location of tree block is recorded | |
904 | * in the back refs. Actually the full back refs is generic, and can be | |
905 | * used in all cases the implicit back refs is used. The major shortcoming | |
906 | * of the full back refs is its overhead. Every time a tree block gets | |
907 | * COWed, we have to update back refs entry for all pointers in it. | |
908 | * | |
909 | * For a newly allocated tree block, we use implicit back refs for | |
910 | * pointers in it. This means most tree related operations only involve | |
911 | * implicit back refs. For a tree block created in old transaction, the | |
912 | * only way to drop a reference to it is COW it. So we can detect the | |
913 | * event that tree block loses its owner tree's reference and do the | |
914 | * back refs conversion. | |
915 | * | |
01327610 | 916 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
917 | * |
918 | * The reference count of the block is one and the tree is the block's | |
919 | * owner tree. Nothing to do in this case. | |
920 | * | |
921 | * The reference count of the block is one and the tree is not the | |
922 | * block's owner tree. In this case, full back refs is used for pointers | |
923 | * in the block. Remove these full back refs, add implicit back refs for | |
924 | * every pointers in the new block. | |
925 | * | |
926 | * The reference count of the block is greater than one and the tree is | |
927 | * the block's owner tree. In this case, implicit back refs is used for | |
928 | * pointers in the block. Add full back refs for every pointers in the | |
929 | * block, increase lower level extents' reference counts. The original | |
930 | * implicit back refs are entailed to the new block. | |
931 | * | |
932 | * The reference count of the block is greater than one and the tree is | |
933 | * not the block's owner tree. Add implicit back refs for every pointer in | |
934 | * the new block, increase lower level extents' reference count. | |
935 | * | |
936 | * Back Reference Key composing: | |
937 | * | |
938 | * The key objectid corresponds to the first byte in the extent, | |
939 | * The key type is used to differentiate between types of back refs. | |
940 | * There are different meanings of the key offset for different types | |
941 | * of back refs. | |
942 | * | |
d8d5f3e1 CM |
943 | * File extents can be referenced by: |
944 | * | |
945 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 946 | * - different files inside a single subvolume |
d8d5f3e1 CM |
947 | * - different offsets inside a file (bookend extents in file.c) |
948 | * | |
5d4f98a2 | 949 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
950 | * |
951 | * - Objectid of the subvolume root | |
d8d5f3e1 | 952 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
953 | * - original offset in the file |
954 | * - how many bookend extents | |
d8d5f3e1 | 955 | * |
5d4f98a2 YZ |
956 | * The key offset for the implicit back refs is hash of the first |
957 | * three fields. | |
d8d5f3e1 | 958 | * |
5d4f98a2 | 959 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 960 | * |
5d4f98a2 | 961 | * - number of pointers in the tree leaf |
d8d5f3e1 | 962 | * |
5d4f98a2 YZ |
963 | * The key offset for the implicit back refs is the first byte of |
964 | * the tree leaf | |
d8d5f3e1 | 965 | * |
5d4f98a2 YZ |
966 | * When a file extent is allocated, The implicit back refs is used. |
967 | * the fields are filled in: | |
d8d5f3e1 | 968 | * |
5d4f98a2 | 969 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 970 | * |
5d4f98a2 YZ |
971 | * When a file extent is removed file truncation, we find the |
972 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 973 | * |
5d4f98a2 | 974 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 975 | * |
5d4f98a2 | 976 | * Btree extents can be referenced by: |
d8d5f3e1 | 977 | * |
5d4f98a2 | 978 | * - Different subvolumes |
d8d5f3e1 | 979 | * |
5d4f98a2 YZ |
980 | * Both the implicit back refs and the full back refs for tree blocks |
981 | * only consist of key. The key offset for the implicit back refs is | |
982 | * objectid of block's owner tree. The key offset for the full back refs | |
983 | * is the first byte of parent block. | |
d8d5f3e1 | 984 | * |
5d4f98a2 YZ |
985 | * When implicit back refs is used, information about the lowest key and |
986 | * level of the tree block are required. These information are stored in | |
987 | * tree block info structure. | |
d8d5f3e1 | 988 | */ |
31840ae1 | 989 | |
167ce953 LB |
990 | /* |
991 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
52042d8e | 992 | * is_data == BTRFS_REF_TYPE_DATA, data type is requiried, |
167ce953 LB |
993 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. |
994 | */ | |
995 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
996 | struct btrfs_extent_inline_ref *iref, | |
997 | enum btrfs_inline_ref_type is_data) | |
998 | { | |
999 | int type = btrfs_extent_inline_ref_type(eb, iref); | |
64ecdb64 | 1000 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 LB |
1001 | |
1002 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
1003 | type == BTRFS_SHARED_BLOCK_REF_KEY || | |
1004 | type == BTRFS_SHARED_DATA_REF_KEY || | |
1005 | type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1006 | if (is_data == BTRFS_REF_TYPE_BLOCK) { | |
64ecdb64 | 1007 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
167ce953 | 1008 | return type; |
64ecdb64 LB |
1009 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
1010 | ASSERT(eb->fs_info); | |
1011 | /* | |
1012 | * Every shared one has parent tree | |
1013 | * block, which must be aligned to | |
1014 | * nodesize. | |
1015 | */ | |
1016 | if (offset && | |
1017 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1018 | return type; | |
1019 | } | |
167ce953 | 1020 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 1021 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 1022 | return type; |
64ecdb64 LB |
1023 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
1024 | ASSERT(eb->fs_info); | |
1025 | /* | |
1026 | * Every shared one has parent tree | |
1027 | * block, which must be aligned to | |
1028 | * nodesize. | |
1029 | */ | |
1030 | if (offset && | |
1031 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1032 | return type; | |
1033 | } | |
167ce953 LB |
1034 | } else { |
1035 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
1036 | return type; | |
1037 | } | |
1038 | } | |
1039 | ||
1040 | btrfs_print_leaf((struct extent_buffer *)eb); | |
1041 | btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d", | |
1042 | eb->start, type); | |
1043 | WARN_ON(1); | |
1044 | ||
1045 | return BTRFS_REF_TYPE_INVALID; | |
1046 | } | |
1047 | ||
5d4f98a2 YZ |
1048 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
1049 | { | |
1050 | u32 high_crc = ~(u32)0; | |
1051 | u32 low_crc = ~(u32)0; | |
1052 | __le64 lenum; | |
1053 | ||
1054 | lenum = cpu_to_le64(root_objectid); | |
65019df8 | 1055 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1056 | lenum = cpu_to_le64(owner); |
65019df8 | 1057 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1058 | lenum = cpu_to_le64(offset); |
65019df8 | 1059 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1060 | |
1061 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1062 | } | |
1063 | ||
1064 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1065 | struct btrfs_extent_data_ref *ref) | |
1066 | { | |
1067 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1068 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1069 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1070 | } | |
1071 | ||
1072 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1073 | struct btrfs_extent_data_ref *ref, | |
1074 | u64 root_objectid, u64 owner, u64 offset) | |
1075 | { | |
1076 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1077 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1078 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1079 | return 0; | |
1080 | return 1; | |
1081 | } | |
1082 | ||
1083 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1084 | struct btrfs_path *path, |
1085 | u64 bytenr, u64 parent, | |
1086 | u64 root_objectid, | |
1087 | u64 owner, u64 offset) | |
1088 | { | |
bd1d53ef | 1089 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 YZ |
1090 | struct btrfs_key key; |
1091 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1092 | struct extent_buffer *leaf; |
5d4f98a2 | 1093 | u32 nritems; |
74493f7a | 1094 | int ret; |
5d4f98a2 YZ |
1095 | int recow; |
1096 | int err = -ENOENT; | |
74493f7a | 1097 | |
31840ae1 | 1098 | key.objectid = bytenr; |
5d4f98a2 YZ |
1099 | if (parent) { |
1100 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1101 | key.offset = parent; | |
1102 | } else { | |
1103 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1104 | key.offset = hash_extent_data_ref(root_objectid, | |
1105 | owner, offset); | |
1106 | } | |
1107 | again: | |
1108 | recow = 0; | |
1109 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1110 | if (ret < 0) { | |
1111 | err = ret; | |
1112 | goto fail; | |
1113 | } | |
31840ae1 | 1114 | |
5d4f98a2 YZ |
1115 | if (parent) { |
1116 | if (!ret) | |
1117 | return 0; | |
5d4f98a2 | 1118 | goto fail; |
31840ae1 ZY |
1119 | } |
1120 | ||
1121 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1122 | nritems = btrfs_header_nritems(leaf); |
1123 | while (1) { | |
1124 | if (path->slots[0] >= nritems) { | |
1125 | ret = btrfs_next_leaf(root, path); | |
1126 | if (ret < 0) | |
1127 | err = ret; | |
1128 | if (ret) | |
1129 | goto fail; | |
1130 | ||
1131 | leaf = path->nodes[0]; | |
1132 | nritems = btrfs_header_nritems(leaf); | |
1133 | recow = 1; | |
1134 | } | |
1135 | ||
1136 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1137 | if (key.objectid != bytenr || | |
1138 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1139 | goto fail; | |
1140 | ||
1141 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1142 | struct btrfs_extent_data_ref); | |
1143 | ||
1144 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1145 | owner, offset)) { | |
1146 | if (recow) { | |
b3b4aa74 | 1147 | btrfs_release_path(path); |
5d4f98a2 YZ |
1148 | goto again; |
1149 | } | |
1150 | err = 0; | |
1151 | break; | |
1152 | } | |
1153 | path->slots[0]++; | |
31840ae1 | 1154 | } |
5d4f98a2 YZ |
1155 | fail: |
1156 | return err; | |
31840ae1 ZY |
1157 | } |
1158 | ||
5d4f98a2 | 1159 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1160 | struct btrfs_path *path, |
1161 | u64 bytenr, u64 parent, | |
1162 | u64 root_objectid, u64 owner, | |
1163 | u64 offset, int refs_to_add) | |
31840ae1 | 1164 | { |
62b895af | 1165 | struct btrfs_root *root = trans->fs_info->extent_root; |
31840ae1 ZY |
1166 | struct btrfs_key key; |
1167 | struct extent_buffer *leaf; | |
5d4f98a2 | 1168 | u32 size; |
31840ae1 ZY |
1169 | u32 num_refs; |
1170 | int ret; | |
74493f7a | 1171 | |
74493f7a | 1172 | key.objectid = bytenr; |
5d4f98a2 YZ |
1173 | if (parent) { |
1174 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1175 | key.offset = parent; | |
1176 | size = sizeof(struct btrfs_shared_data_ref); | |
1177 | } else { | |
1178 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1179 | key.offset = hash_extent_data_ref(root_objectid, | |
1180 | owner, offset); | |
1181 | size = sizeof(struct btrfs_extent_data_ref); | |
1182 | } | |
74493f7a | 1183 | |
5d4f98a2 YZ |
1184 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1185 | if (ret && ret != -EEXIST) | |
1186 | goto fail; | |
1187 | ||
1188 | leaf = path->nodes[0]; | |
1189 | if (parent) { | |
1190 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1191 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1192 | struct btrfs_shared_data_ref); |
1193 | if (ret == 0) { | |
1194 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1195 | } else { | |
1196 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1197 | num_refs += refs_to_add; | |
1198 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1199 | } |
5d4f98a2 YZ |
1200 | } else { |
1201 | struct btrfs_extent_data_ref *ref; | |
1202 | while (ret == -EEXIST) { | |
1203 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1204 | struct btrfs_extent_data_ref); | |
1205 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1206 | owner, offset)) | |
1207 | break; | |
b3b4aa74 | 1208 | btrfs_release_path(path); |
5d4f98a2 YZ |
1209 | key.offset++; |
1210 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1211 | size); | |
1212 | if (ret && ret != -EEXIST) | |
1213 | goto fail; | |
31840ae1 | 1214 | |
5d4f98a2 YZ |
1215 | leaf = path->nodes[0]; |
1216 | } | |
1217 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1218 | struct btrfs_extent_data_ref); | |
1219 | if (ret == 0) { | |
1220 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1221 | root_objectid); | |
1222 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1223 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1224 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1225 | } else { | |
1226 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1227 | num_refs += refs_to_add; | |
1228 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1229 | } |
31840ae1 | 1230 | } |
5d4f98a2 YZ |
1231 | btrfs_mark_buffer_dirty(leaf); |
1232 | ret = 0; | |
1233 | fail: | |
b3b4aa74 | 1234 | btrfs_release_path(path); |
7bb86316 | 1235 | return ret; |
74493f7a CM |
1236 | } |
1237 | ||
5d4f98a2 | 1238 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 | 1239 | struct btrfs_path *path, |
fcebe456 | 1240 | int refs_to_drop, int *last_ref) |
31840ae1 | 1241 | { |
5d4f98a2 YZ |
1242 | struct btrfs_key key; |
1243 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1244 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1245 | struct extent_buffer *leaf; |
5d4f98a2 | 1246 | u32 num_refs = 0; |
31840ae1 ZY |
1247 | int ret = 0; |
1248 | ||
1249 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1250 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1251 | ||
1252 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1253 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1254 | struct btrfs_extent_data_ref); | |
1255 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1256 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1257 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1258 | struct btrfs_shared_data_ref); | |
1259 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
6d8ff4e4 | 1260 | } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { |
ba3c2b19 NB |
1261 | btrfs_print_v0_err(trans->fs_info); |
1262 | btrfs_abort_transaction(trans, -EINVAL); | |
1263 | return -EINVAL; | |
5d4f98a2 YZ |
1264 | } else { |
1265 | BUG(); | |
1266 | } | |
1267 | ||
56bec294 CM |
1268 | BUG_ON(num_refs < refs_to_drop); |
1269 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1270 | |
31840ae1 | 1271 | if (num_refs == 0) { |
e9f6290d | 1272 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
fcebe456 | 1273 | *last_ref = 1; |
31840ae1 | 1274 | } else { |
5d4f98a2 YZ |
1275 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1276 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1277 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1278 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
31840ae1 ZY |
1279 | btrfs_mark_buffer_dirty(leaf); |
1280 | } | |
31840ae1 ZY |
1281 | return ret; |
1282 | } | |
1283 | ||
9ed0dea0 | 1284 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1285 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1286 | { |
5d4f98a2 YZ |
1287 | struct btrfs_key key; |
1288 | struct extent_buffer *leaf; | |
1289 | struct btrfs_extent_data_ref *ref1; | |
1290 | struct btrfs_shared_data_ref *ref2; | |
1291 | u32 num_refs = 0; | |
3de28d57 | 1292 | int type; |
5d4f98a2 YZ |
1293 | |
1294 | leaf = path->nodes[0]; | |
1295 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
ba3c2b19 NB |
1296 | |
1297 | BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); | |
5d4f98a2 | 1298 | if (iref) { |
3de28d57 LB |
1299 | /* |
1300 | * If type is invalid, we should have bailed out earlier than | |
1301 | * this call. | |
1302 | */ | |
1303 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
1304 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
1305 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
1306 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
1307 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1308 | } else { | |
1309 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1310 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1311 | } | |
1312 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1313 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1314 | struct btrfs_extent_data_ref); | |
1315 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1316 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1317 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1318 | struct btrfs_shared_data_ref); | |
1319 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
5d4f98a2 YZ |
1320 | } else { |
1321 | WARN_ON(1); | |
1322 | } | |
1323 | return num_refs; | |
1324 | } | |
15916de8 | 1325 | |
5d4f98a2 | 1326 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1327 | struct btrfs_path *path, |
1328 | u64 bytenr, u64 parent, | |
1329 | u64 root_objectid) | |
1f3c79a2 | 1330 | { |
b8582eea | 1331 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 | 1332 | struct btrfs_key key; |
1f3c79a2 | 1333 | int ret; |
1f3c79a2 | 1334 | |
5d4f98a2 YZ |
1335 | key.objectid = bytenr; |
1336 | if (parent) { | |
1337 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1338 | key.offset = parent; | |
1339 | } else { | |
1340 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1341 | key.offset = root_objectid; | |
1f3c79a2 LH |
1342 | } |
1343 | ||
5d4f98a2 YZ |
1344 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1345 | if (ret > 0) | |
1346 | ret = -ENOENT; | |
5d4f98a2 | 1347 | return ret; |
1f3c79a2 LH |
1348 | } |
1349 | ||
5d4f98a2 | 1350 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1351 | struct btrfs_path *path, |
1352 | u64 bytenr, u64 parent, | |
1353 | u64 root_objectid) | |
31840ae1 | 1354 | { |
5d4f98a2 | 1355 | struct btrfs_key key; |
31840ae1 | 1356 | int ret; |
31840ae1 | 1357 | |
5d4f98a2 YZ |
1358 | key.objectid = bytenr; |
1359 | if (parent) { | |
1360 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1361 | key.offset = parent; | |
1362 | } else { | |
1363 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1364 | key.offset = root_objectid; | |
1365 | } | |
1366 | ||
10728404 | 1367 | ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root, |
87bde3cd | 1368 | path, &key, 0); |
b3b4aa74 | 1369 | btrfs_release_path(path); |
31840ae1 ZY |
1370 | return ret; |
1371 | } | |
1372 | ||
5d4f98a2 | 1373 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1374 | { |
5d4f98a2 YZ |
1375 | int type; |
1376 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1377 | if (parent > 0) | |
1378 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1379 | else | |
1380 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1381 | } else { | |
1382 | if (parent > 0) | |
1383 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1384 | else | |
1385 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1386 | } | |
1387 | return type; | |
31840ae1 | 1388 | } |
56bec294 | 1389 | |
2c47e605 YZ |
1390 | static int find_next_key(struct btrfs_path *path, int level, |
1391 | struct btrfs_key *key) | |
56bec294 | 1392 | |
02217ed2 | 1393 | { |
2c47e605 | 1394 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1395 | if (!path->nodes[level]) |
1396 | break; | |
5d4f98a2 YZ |
1397 | if (path->slots[level] + 1 >= |
1398 | btrfs_header_nritems(path->nodes[level])) | |
1399 | continue; | |
1400 | if (level == 0) | |
1401 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1402 | path->slots[level] + 1); | |
1403 | else | |
1404 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1405 | path->slots[level] + 1); | |
1406 | return 0; | |
1407 | } | |
1408 | return 1; | |
1409 | } | |
037e6390 | 1410 | |
5d4f98a2 YZ |
1411 | /* |
1412 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1413 | * to the address of inline back ref, and 0 is returned. | |
1414 | * | |
1415 | * if back ref isn't found, *ref_ret is set to the address where it | |
1416 | * should be inserted, and -ENOENT is returned. | |
1417 | * | |
1418 | * if insert is true and there are too many inline back refs, the path | |
1419 | * points to the extent item, and -EAGAIN is returned. | |
1420 | * | |
1421 | * NOTE: inline back refs are ordered in the same way that back ref | |
1422 | * items in the tree are ordered. | |
1423 | */ | |
1424 | static noinline_for_stack | |
1425 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1426 | struct btrfs_path *path, |
1427 | struct btrfs_extent_inline_ref **ref_ret, | |
1428 | u64 bytenr, u64 num_bytes, | |
1429 | u64 parent, u64 root_objectid, | |
1430 | u64 owner, u64 offset, int insert) | |
1431 | { | |
867cc1fb | 1432 | struct btrfs_fs_info *fs_info = trans->fs_info; |
87bde3cd | 1433 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1434 | struct btrfs_key key; |
1435 | struct extent_buffer *leaf; | |
1436 | struct btrfs_extent_item *ei; | |
1437 | struct btrfs_extent_inline_ref *iref; | |
1438 | u64 flags; | |
1439 | u64 item_size; | |
1440 | unsigned long ptr; | |
1441 | unsigned long end; | |
1442 | int extra_size; | |
1443 | int type; | |
1444 | int want; | |
1445 | int ret; | |
1446 | int err = 0; | |
0b246afa | 1447 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 1448 | int needed; |
26b8003f | 1449 | |
db94535d | 1450 | key.objectid = bytenr; |
31840ae1 | 1451 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1452 | key.offset = num_bytes; |
31840ae1 | 1453 | |
5d4f98a2 YZ |
1454 | want = extent_ref_type(parent, owner); |
1455 | if (insert) { | |
1456 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1457 | path->keep_locks = 1; |
5d4f98a2 YZ |
1458 | } else |
1459 | extra_size = -1; | |
3173a18f JB |
1460 | |
1461 | /* | |
16d1c062 NB |
1462 | * Owner is our level, so we can just add one to get the level for the |
1463 | * block we are interested in. | |
3173a18f JB |
1464 | */ |
1465 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1466 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1467 | key.offset = owner; | |
1468 | } | |
1469 | ||
1470 | again: | |
5d4f98a2 | 1471 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1472 | if (ret < 0) { |
5d4f98a2 YZ |
1473 | err = ret; |
1474 | goto out; | |
1475 | } | |
3173a18f JB |
1476 | |
1477 | /* | |
1478 | * We may be a newly converted file system which still has the old fat | |
1479 | * extent entries for metadata, so try and see if we have one of those. | |
1480 | */ | |
1481 | if (ret > 0 && skinny_metadata) { | |
1482 | skinny_metadata = false; | |
1483 | if (path->slots[0]) { | |
1484 | path->slots[0]--; | |
1485 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1486 | path->slots[0]); | |
1487 | if (key.objectid == bytenr && | |
1488 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1489 | key.offset == num_bytes) | |
1490 | ret = 0; | |
1491 | } | |
1492 | if (ret) { | |
9ce49a0b | 1493 | key.objectid = bytenr; |
3173a18f JB |
1494 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1495 | key.offset = num_bytes; | |
1496 | btrfs_release_path(path); | |
1497 | goto again; | |
1498 | } | |
1499 | } | |
1500 | ||
79787eaa JM |
1501 | if (ret && !insert) { |
1502 | err = -ENOENT; | |
1503 | goto out; | |
fae7f21c | 1504 | } else if (WARN_ON(ret)) { |
492104c8 | 1505 | err = -EIO; |
492104c8 | 1506 | goto out; |
79787eaa | 1507 | } |
5d4f98a2 YZ |
1508 | |
1509 | leaf = path->nodes[0]; | |
1510 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
6d8ff4e4 | 1511 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
1512 | err = -EINVAL; |
1513 | btrfs_print_v0_err(fs_info); | |
1514 | btrfs_abort_transaction(trans, err); | |
1515 | goto out; | |
1516 | } | |
5d4f98a2 | 1517 | |
5d4f98a2 YZ |
1518 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1519 | flags = btrfs_extent_flags(leaf, ei); | |
1520 | ||
1521 | ptr = (unsigned long)(ei + 1); | |
1522 | end = (unsigned long)ei + item_size; | |
1523 | ||
3173a18f | 1524 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1525 | ptr += sizeof(struct btrfs_tree_block_info); |
1526 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1527 | } |
1528 | ||
3de28d57 LB |
1529 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
1530 | needed = BTRFS_REF_TYPE_DATA; | |
1531 | else | |
1532 | needed = BTRFS_REF_TYPE_BLOCK; | |
1533 | ||
5d4f98a2 YZ |
1534 | err = -ENOENT; |
1535 | while (1) { | |
1536 | if (ptr >= end) { | |
1537 | WARN_ON(ptr > end); | |
1538 | break; | |
1539 | } | |
1540 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
1541 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
1542 | if (type == BTRFS_REF_TYPE_INVALID) { | |
af431dcb | 1543 | err = -EUCLEAN; |
3de28d57 LB |
1544 | goto out; |
1545 | } | |
1546 | ||
5d4f98a2 YZ |
1547 | if (want < type) |
1548 | break; | |
1549 | if (want > type) { | |
1550 | ptr += btrfs_extent_inline_ref_size(type); | |
1551 | continue; | |
1552 | } | |
1553 | ||
1554 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1555 | struct btrfs_extent_data_ref *dref; | |
1556 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1557 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1558 | owner, offset)) { | |
1559 | err = 0; | |
1560 | break; | |
1561 | } | |
1562 | if (hash_extent_data_ref_item(leaf, dref) < | |
1563 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1564 | break; | |
1565 | } else { | |
1566 | u64 ref_offset; | |
1567 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1568 | if (parent > 0) { | |
1569 | if (parent == ref_offset) { | |
1570 | err = 0; | |
1571 | break; | |
1572 | } | |
1573 | if (ref_offset < parent) | |
1574 | break; | |
1575 | } else { | |
1576 | if (root_objectid == ref_offset) { | |
1577 | err = 0; | |
1578 | break; | |
1579 | } | |
1580 | if (ref_offset < root_objectid) | |
1581 | break; | |
1582 | } | |
1583 | } | |
1584 | ptr += btrfs_extent_inline_ref_size(type); | |
1585 | } | |
1586 | if (err == -ENOENT && insert) { | |
1587 | if (item_size + extra_size >= | |
1588 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1589 | err = -EAGAIN; | |
1590 | goto out; | |
1591 | } | |
1592 | /* | |
1593 | * To add new inline back ref, we have to make sure | |
1594 | * there is no corresponding back ref item. | |
1595 | * For simplicity, we just do not add new inline back | |
1596 | * ref if there is any kind of item for this block | |
1597 | */ | |
2c47e605 YZ |
1598 | if (find_next_key(path, 0, &key) == 0 && |
1599 | key.objectid == bytenr && | |
85d4198e | 1600 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1601 | err = -EAGAIN; |
1602 | goto out; | |
1603 | } | |
1604 | } | |
1605 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1606 | out: | |
85d4198e | 1607 | if (insert) { |
5d4f98a2 YZ |
1608 | path->keep_locks = 0; |
1609 | btrfs_unlock_up_safe(path, 1); | |
1610 | } | |
1611 | return err; | |
1612 | } | |
1613 | ||
1614 | /* | |
1615 | * helper to add new inline back ref | |
1616 | */ | |
1617 | static noinline_for_stack | |
87bde3cd | 1618 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1619 | struct btrfs_path *path, |
1620 | struct btrfs_extent_inline_ref *iref, | |
1621 | u64 parent, u64 root_objectid, | |
1622 | u64 owner, u64 offset, int refs_to_add, | |
1623 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1624 | { |
1625 | struct extent_buffer *leaf; | |
1626 | struct btrfs_extent_item *ei; | |
1627 | unsigned long ptr; | |
1628 | unsigned long end; | |
1629 | unsigned long item_offset; | |
1630 | u64 refs; | |
1631 | int size; | |
1632 | int type; | |
5d4f98a2 YZ |
1633 | |
1634 | leaf = path->nodes[0]; | |
1635 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1636 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1637 | ||
1638 | type = extent_ref_type(parent, owner); | |
1639 | size = btrfs_extent_inline_ref_size(type); | |
1640 | ||
c71dd880 | 1641 | btrfs_extend_item(path, size); |
5d4f98a2 YZ |
1642 | |
1643 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1644 | refs = btrfs_extent_refs(leaf, ei); | |
1645 | refs += refs_to_add; | |
1646 | btrfs_set_extent_refs(leaf, ei, refs); | |
1647 | if (extent_op) | |
1648 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1649 | ||
1650 | ptr = (unsigned long)ei + item_offset; | |
1651 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1652 | if (ptr < end - size) | |
1653 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1654 | end - size - ptr); | |
1655 | ||
1656 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1657 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1658 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1659 | struct btrfs_extent_data_ref *dref; | |
1660 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1661 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1662 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1663 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1664 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1665 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1666 | struct btrfs_shared_data_ref *sref; | |
1667 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1668 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1669 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1670 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1671 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1672 | } else { | |
1673 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1674 | } | |
1675 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1676 | } |
1677 | ||
1678 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1679 | struct btrfs_path *path, |
1680 | struct btrfs_extent_inline_ref **ref_ret, | |
1681 | u64 bytenr, u64 num_bytes, u64 parent, | |
1682 | u64 root_objectid, u64 owner, u64 offset) | |
1683 | { | |
1684 | int ret; | |
1685 | ||
867cc1fb NB |
1686 | ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr, |
1687 | num_bytes, parent, root_objectid, | |
1688 | owner, offset, 0); | |
5d4f98a2 | 1689 | if (ret != -ENOENT) |
54aa1f4d | 1690 | return ret; |
5d4f98a2 | 1691 | |
b3b4aa74 | 1692 | btrfs_release_path(path); |
5d4f98a2 YZ |
1693 | *ref_ret = NULL; |
1694 | ||
1695 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
b8582eea NB |
1696 | ret = lookup_tree_block_ref(trans, path, bytenr, parent, |
1697 | root_objectid); | |
5d4f98a2 | 1698 | } else { |
bd1d53ef NB |
1699 | ret = lookup_extent_data_ref(trans, path, bytenr, parent, |
1700 | root_objectid, owner, offset); | |
b9473439 | 1701 | } |
5d4f98a2 YZ |
1702 | return ret; |
1703 | } | |
31840ae1 | 1704 | |
5d4f98a2 YZ |
1705 | /* |
1706 | * helper to update/remove inline back ref | |
1707 | */ | |
1708 | static noinline_for_stack | |
61a18f1c | 1709 | void update_inline_extent_backref(struct btrfs_path *path, |
143bede5 JM |
1710 | struct btrfs_extent_inline_ref *iref, |
1711 | int refs_to_mod, | |
fcebe456 JB |
1712 | struct btrfs_delayed_extent_op *extent_op, |
1713 | int *last_ref) | |
5d4f98a2 | 1714 | { |
61a18f1c | 1715 | struct extent_buffer *leaf = path->nodes[0]; |
5d4f98a2 YZ |
1716 | struct btrfs_extent_item *ei; |
1717 | struct btrfs_extent_data_ref *dref = NULL; | |
1718 | struct btrfs_shared_data_ref *sref = NULL; | |
1719 | unsigned long ptr; | |
1720 | unsigned long end; | |
1721 | u32 item_size; | |
1722 | int size; | |
1723 | int type; | |
5d4f98a2 YZ |
1724 | u64 refs; |
1725 | ||
5d4f98a2 YZ |
1726 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1727 | refs = btrfs_extent_refs(leaf, ei); | |
1728 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1729 | refs += refs_to_mod; | |
1730 | btrfs_set_extent_refs(leaf, ei, refs); | |
1731 | if (extent_op) | |
1732 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1733 | ||
3de28d57 LB |
1734 | /* |
1735 | * If type is invalid, we should have bailed out after | |
1736 | * lookup_inline_extent_backref(). | |
1737 | */ | |
1738 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
1739 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
5d4f98a2 YZ |
1740 | |
1741 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1742 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1743 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1744 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1745 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1746 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1747 | } else { | |
1748 | refs = 1; | |
1749 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1750 | } |
31840ae1 | 1751 | |
5d4f98a2 YZ |
1752 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1753 | refs += refs_to_mod; | |
1754 | ||
1755 | if (refs > 0) { | |
1756 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1757 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1758 | else | |
1759 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1760 | } else { | |
fcebe456 | 1761 | *last_ref = 1; |
5d4f98a2 YZ |
1762 | size = btrfs_extent_inline_ref_size(type); |
1763 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1764 | ptr = (unsigned long)iref; | |
1765 | end = (unsigned long)ei + item_size; | |
1766 | if (ptr + size < end) | |
1767 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1768 | end - ptr - size); | |
1769 | item_size -= size; | |
78ac4f9e | 1770 | btrfs_truncate_item(path, item_size, 1); |
5d4f98a2 YZ |
1771 | } |
1772 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1773 | } |
1774 | ||
1775 | static noinline_for_stack | |
1776 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1777 | struct btrfs_path *path, |
1778 | u64 bytenr, u64 num_bytes, u64 parent, | |
1779 | u64 root_objectid, u64 owner, | |
1780 | u64 offset, int refs_to_add, | |
1781 | struct btrfs_delayed_extent_op *extent_op) | |
1782 | { | |
1783 | struct btrfs_extent_inline_ref *iref; | |
1784 | int ret; | |
1785 | ||
867cc1fb NB |
1786 | ret = lookup_inline_extent_backref(trans, path, &iref, bytenr, |
1787 | num_bytes, parent, root_objectid, | |
1788 | owner, offset, 1); | |
5d4f98a2 YZ |
1789 | if (ret == 0) { |
1790 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
61a18f1c NB |
1791 | update_inline_extent_backref(path, iref, refs_to_add, |
1792 | extent_op, NULL); | |
5d4f98a2 | 1793 | } else if (ret == -ENOENT) { |
a639cdeb | 1794 | setup_inline_extent_backref(trans->fs_info, path, iref, parent, |
143bede5 JM |
1795 | root_objectid, owner, offset, |
1796 | refs_to_add, extent_op); | |
1797 | ret = 0; | |
771ed689 | 1798 | } |
5d4f98a2 YZ |
1799 | return ret; |
1800 | } | |
31840ae1 | 1801 | |
5d4f98a2 | 1802 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1803 | struct btrfs_path *path, |
1804 | u64 bytenr, u64 parent, u64 root_objectid, | |
1805 | u64 owner, u64 offset, int refs_to_add) | |
1806 | { | |
1807 | int ret; | |
1808 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1809 | BUG_ON(refs_to_add != 1); | |
10728404 NB |
1810 | ret = insert_tree_block_ref(trans, path, bytenr, parent, |
1811 | root_objectid); | |
5d4f98a2 | 1812 | } else { |
62b895af NB |
1813 | ret = insert_extent_data_ref(trans, path, bytenr, parent, |
1814 | root_objectid, owner, offset, | |
1815 | refs_to_add); | |
5d4f98a2 YZ |
1816 | } |
1817 | return ret; | |
1818 | } | |
56bec294 | 1819 | |
5d4f98a2 | 1820 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1821 | struct btrfs_path *path, |
1822 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1823 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1824 | { |
143bede5 | 1825 | int ret = 0; |
b9473439 | 1826 | |
5d4f98a2 YZ |
1827 | BUG_ON(!is_data && refs_to_drop != 1); |
1828 | if (iref) { | |
61a18f1c NB |
1829 | update_inline_extent_backref(path, iref, -refs_to_drop, NULL, |
1830 | last_ref); | |
5d4f98a2 | 1831 | } else if (is_data) { |
e9f6290d | 1832 | ret = remove_extent_data_ref(trans, path, refs_to_drop, |
fcebe456 | 1833 | last_ref); |
5d4f98a2 | 1834 | } else { |
fcebe456 | 1835 | *last_ref = 1; |
87cc7a8a | 1836 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
5d4f98a2 YZ |
1837 | } |
1838 | return ret; | |
1839 | } | |
1840 | ||
d04c6b88 JM |
1841 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1842 | u64 *discarded_bytes) | |
5d4f98a2 | 1843 | { |
86557861 JM |
1844 | int j, ret = 0; |
1845 | u64 bytes_left, end; | |
4d89d377 | 1846 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1847 | |
4d89d377 JM |
1848 | if (WARN_ON(start != aligned_start)) { |
1849 | len -= aligned_start - start; | |
1850 | len = round_down(len, 1 << 9); | |
1851 | start = aligned_start; | |
1852 | } | |
d04c6b88 | 1853 | |
4d89d377 | 1854 | *discarded_bytes = 0; |
86557861 JM |
1855 | |
1856 | if (!len) | |
1857 | return 0; | |
1858 | ||
1859 | end = start + len; | |
1860 | bytes_left = len; | |
1861 | ||
1862 | /* Skip any superblocks on this device. */ | |
1863 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1864 | u64 sb_start = btrfs_sb_offset(j); | |
1865 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1866 | u64 size = sb_start - start; | |
1867 | ||
1868 | if (!in_range(sb_start, start, bytes_left) && | |
1869 | !in_range(sb_end, start, bytes_left) && | |
1870 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1871 | continue; | |
1872 | ||
1873 | /* | |
1874 | * Superblock spans beginning of range. Adjust start and | |
1875 | * try again. | |
1876 | */ | |
1877 | if (sb_start <= start) { | |
1878 | start += sb_end - start; | |
1879 | if (start > end) { | |
1880 | bytes_left = 0; | |
1881 | break; | |
1882 | } | |
1883 | bytes_left = end - start; | |
1884 | continue; | |
1885 | } | |
1886 | ||
1887 | if (size) { | |
1888 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
1889 | GFP_NOFS, 0); | |
1890 | if (!ret) | |
1891 | *discarded_bytes += size; | |
1892 | else if (ret != -EOPNOTSUPP) | |
1893 | return ret; | |
1894 | } | |
1895 | ||
1896 | start = sb_end; | |
1897 | if (start > end) { | |
1898 | bytes_left = 0; | |
1899 | break; | |
1900 | } | |
1901 | bytes_left = end - start; | |
1902 | } | |
1903 | ||
1904 | if (bytes_left) { | |
1905 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
1906 | GFP_NOFS, 0); |
1907 | if (!ret) | |
86557861 | 1908 | *discarded_bytes += bytes_left; |
4d89d377 | 1909 | } |
d04c6b88 | 1910 | return ret; |
5d4f98a2 | 1911 | } |
5d4f98a2 | 1912 | |
2ff7e61e | 1913 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 1914 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1915 | { |
5d4f98a2 | 1916 | int ret; |
5378e607 | 1917 | u64 discarded_bytes = 0; |
a1d3c478 | 1918 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1919 | |
e244a0ae | 1920 | |
2999241d FM |
1921 | /* |
1922 | * Avoid races with device replace and make sure our bbio has devices | |
1923 | * associated to its stripes that don't go away while we are discarding. | |
1924 | */ | |
0b246afa | 1925 | btrfs_bio_counter_inc_blocked(fs_info); |
5d4f98a2 | 1926 | /* Tell the block device(s) that the sectors can be discarded */ |
0b246afa JM |
1927 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes, |
1928 | &bbio, 0); | |
79787eaa | 1929 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 1930 | if (!ret) { |
a1d3c478 | 1931 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
1932 | int i; |
1933 | ||
5d4f98a2 | 1934 | |
a1d3c478 | 1935 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 1936 | u64 bytes; |
38b5f68e AJ |
1937 | struct request_queue *req_q; |
1938 | ||
627e0873 FM |
1939 | if (!stripe->dev->bdev) { |
1940 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); | |
1941 | continue; | |
1942 | } | |
38b5f68e AJ |
1943 | req_q = bdev_get_queue(stripe->dev->bdev); |
1944 | if (!blk_queue_discard(req_q)) | |
d5e2003c JB |
1945 | continue; |
1946 | ||
5378e607 LD |
1947 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1948 | stripe->physical, | |
d04c6b88 JM |
1949 | stripe->length, |
1950 | &bytes); | |
5378e607 | 1951 | if (!ret) |
d04c6b88 | 1952 | discarded_bytes += bytes; |
5378e607 | 1953 | else if (ret != -EOPNOTSUPP) |
79787eaa | 1954 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
1955 | |
1956 | /* | |
1957 | * Just in case we get back EOPNOTSUPP for some reason, | |
1958 | * just ignore the return value so we don't screw up | |
1959 | * people calling discard_extent. | |
1960 | */ | |
1961 | ret = 0; | |
5d4f98a2 | 1962 | } |
6e9606d2 | 1963 | btrfs_put_bbio(bbio); |
5d4f98a2 | 1964 | } |
0b246afa | 1965 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
1966 | |
1967 | if (actual_bytes) | |
1968 | *actual_bytes = discarded_bytes; | |
1969 | ||
5d4f98a2 | 1970 | |
53b381b3 DW |
1971 | if (ret == -EOPNOTSUPP) |
1972 | ret = 0; | |
5d4f98a2 | 1973 | return ret; |
5d4f98a2 YZ |
1974 | } |
1975 | ||
79787eaa | 1976 | /* Can return -ENOMEM */ |
5d4f98a2 | 1977 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
82fa113f | 1978 | struct btrfs_ref *generic_ref) |
5d4f98a2 | 1979 | { |
82fa113f | 1980 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d7eae340 | 1981 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 1982 | int ret; |
66d7e7f0 | 1983 | |
82fa113f QW |
1984 | ASSERT(generic_ref->type != BTRFS_REF_NOT_SET && |
1985 | generic_ref->action); | |
1986 | BUG_ON(generic_ref->type == BTRFS_REF_METADATA && | |
1987 | generic_ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID); | |
5d4f98a2 | 1988 | |
82fa113f QW |
1989 | if (generic_ref->type == BTRFS_REF_METADATA) |
1990 | ret = btrfs_add_delayed_tree_ref(trans, generic_ref, | |
ed4f255b | 1991 | NULL, &old_ref_mod, &new_ref_mod); |
82fa113f QW |
1992 | else |
1993 | ret = btrfs_add_delayed_data_ref(trans, generic_ref, 0, | |
d7eae340 | 1994 | &old_ref_mod, &new_ref_mod); |
d7eae340 | 1995 | |
82fa113f | 1996 | btrfs_ref_tree_mod(fs_info, generic_ref); |
8a5040f7 | 1997 | |
ddf30cf0 | 1998 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) |
78192442 | 1999 | sub_pinned_bytes(fs_info, generic_ref); |
d7eae340 | 2000 | |
5d4f98a2 YZ |
2001 | return ret; |
2002 | } | |
2003 | ||
bd3c685e NB |
2004 | /* |
2005 | * __btrfs_inc_extent_ref - insert backreference for a given extent | |
2006 | * | |
2007 | * @trans: Handle of transaction | |
2008 | * | |
2009 | * @node: The delayed ref node used to get the bytenr/length for | |
2010 | * extent whose references are incremented. | |
2011 | * | |
2012 | * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/ | |
2013 | * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical | |
2014 | * bytenr of the parent block. Since new extents are always | |
2015 | * created with indirect references, this will only be the case | |
2016 | * when relocating a shared extent. In that case, root_objectid | |
2017 | * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must | |
2018 | * be 0 | |
2019 | * | |
2020 | * @root_objectid: The id of the root where this modification has originated, | |
2021 | * this can be either one of the well-known metadata trees or | |
2022 | * the subvolume id which references this extent. | |
2023 | * | |
2024 | * @owner: For data extents it is the inode number of the owning file. | |
2025 | * For metadata extents this parameter holds the level in the | |
2026 | * tree of the extent. | |
2027 | * | |
2028 | * @offset: For metadata extents the offset is ignored and is currently | |
2029 | * always passed as 0. For data extents it is the fileoffset | |
2030 | * this extent belongs to. | |
2031 | * | |
2032 | * @refs_to_add Number of references to add | |
2033 | * | |
2034 | * @extent_op Pointer to a structure, holding information necessary when | |
2035 | * updating a tree block's flags | |
2036 | * | |
2037 | */ | |
5d4f98a2 | 2038 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
c682f9b3 | 2039 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2040 | u64 parent, u64 root_objectid, |
2041 | u64 owner, u64 offset, int refs_to_add, | |
2042 | struct btrfs_delayed_extent_op *extent_op) | |
2043 | { | |
2044 | struct btrfs_path *path; | |
2045 | struct extent_buffer *leaf; | |
2046 | struct btrfs_extent_item *item; | |
fcebe456 | 2047 | struct btrfs_key key; |
c682f9b3 QW |
2048 | u64 bytenr = node->bytenr; |
2049 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2050 | u64 refs; |
2051 | int ret; | |
5d4f98a2 YZ |
2052 | |
2053 | path = btrfs_alloc_path(); | |
2054 | if (!path) | |
2055 | return -ENOMEM; | |
2056 | ||
e4058b54 | 2057 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2058 | path->leave_spinning = 1; |
2059 | /* this will setup the path even if it fails to insert the back ref */ | |
a639cdeb NB |
2060 | ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes, |
2061 | parent, root_objectid, owner, | |
2062 | offset, refs_to_add, extent_op); | |
0ed4792a | 2063 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2064 | goto out; |
fcebe456 JB |
2065 | |
2066 | /* | |
2067 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2068 | * inline extent ref, so just update the reference count and add a | |
2069 | * normal backref. | |
2070 | */ | |
5d4f98a2 | 2071 | leaf = path->nodes[0]; |
fcebe456 | 2072 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2073 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2074 | refs = btrfs_extent_refs(leaf, item); | |
2075 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2076 | if (extent_op) | |
2077 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2078 | |
5d4f98a2 | 2079 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2080 | btrfs_release_path(path); |
56bec294 | 2081 | |
e4058b54 | 2082 | path->reada = READA_FORWARD; |
b9473439 | 2083 | path->leave_spinning = 1; |
56bec294 | 2084 | /* now insert the actual backref */ |
37593410 NB |
2085 | ret = insert_extent_backref(trans, path, bytenr, parent, root_objectid, |
2086 | owner, offset, refs_to_add); | |
79787eaa | 2087 | if (ret) |
66642832 | 2088 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 2089 | out: |
56bec294 | 2090 | btrfs_free_path(path); |
30d133fc | 2091 | return ret; |
56bec294 CM |
2092 | } |
2093 | ||
5d4f98a2 | 2094 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2095 | struct btrfs_delayed_ref_node *node, |
2096 | struct btrfs_delayed_extent_op *extent_op, | |
2097 | int insert_reserved) | |
56bec294 | 2098 | { |
5d4f98a2 YZ |
2099 | int ret = 0; |
2100 | struct btrfs_delayed_data_ref *ref; | |
2101 | struct btrfs_key ins; | |
2102 | u64 parent = 0; | |
2103 | u64 ref_root = 0; | |
2104 | u64 flags = 0; | |
2105 | ||
2106 | ins.objectid = node->bytenr; | |
2107 | ins.offset = node->num_bytes; | |
2108 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2109 | ||
2110 | ref = btrfs_delayed_node_to_data_ref(node); | |
2bf98ef3 | 2111 | trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 2112 | |
5d4f98a2 YZ |
2113 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2114 | parent = ref->parent; | |
fcebe456 | 2115 | ref_root = ref->root; |
5d4f98a2 YZ |
2116 | |
2117 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2118 | if (extent_op) |
5d4f98a2 | 2119 | flags |= extent_op->flags_to_set; |
ef89b824 NB |
2120 | ret = alloc_reserved_file_extent(trans, parent, ref_root, |
2121 | flags, ref->objectid, | |
2122 | ref->offset, &ins, | |
2123 | node->ref_mod); | |
5d4f98a2 | 2124 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
2125 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
2126 | ref->objectid, ref->offset, | |
2127 | node->ref_mod, extent_op); | |
5d4f98a2 | 2128 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 2129 | ret = __btrfs_free_extent(trans, node, parent, |
5d4f98a2 YZ |
2130 | ref_root, ref->objectid, |
2131 | ref->offset, node->ref_mod, | |
c682f9b3 | 2132 | extent_op); |
5d4f98a2 YZ |
2133 | } else { |
2134 | BUG(); | |
2135 | } | |
2136 | return ret; | |
2137 | } | |
2138 | ||
2139 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2140 | struct extent_buffer *leaf, | |
2141 | struct btrfs_extent_item *ei) | |
2142 | { | |
2143 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2144 | if (extent_op->update_flags) { | |
2145 | flags |= extent_op->flags_to_set; | |
2146 | btrfs_set_extent_flags(leaf, ei, flags); | |
2147 | } | |
2148 | ||
2149 | if (extent_op->update_key) { | |
2150 | struct btrfs_tree_block_info *bi; | |
2151 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2152 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2153 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2154 | } | |
2155 | } | |
2156 | ||
2157 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
d278850e | 2158 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
2159 | struct btrfs_delayed_extent_op *extent_op) |
2160 | { | |
20b9a2d6 | 2161 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
2162 | struct btrfs_key key; |
2163 | struct btrfs_path *path; | |
2164 | struct btrfs_extent_item *ei; | |
2165 | struct extent_buffer *leaf; | |
2166 | u32 item_size; | |
56bec294 | 2167 | int ret; |
5d4f98a2 | 2168 | int err = 0; |
b1c79e09 | 2169 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2170 | |
79787eaa JM |
2171 | if (trans->aborted) |
2172 | return 0; | |
2173 | ||
0b246afa | 2174 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
2175 | metadata = 0; |
2176 | ||
5d4f98a2 YZ |
2177 | path = btrfs_alloc_path(); |
2178 | if (!path) | |
2179 | return -ENOMEM; | |
2180 | ||
d278850e | 2181 | key.objectid = head->bytenr; |
5d4f98a2 | 2182 | |
3173a18f | 2183 | if (metadata) { |
3173a18f | 2184 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2185 | key.offset = extent_op->level; |
3173a18f JB |
2186 | } else { |
2187 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 2188 | key.offset = head->num_bytes; |
3173a18f JB |
2189 | } |
2190 | ||
2191 | again: | |
e4058b54 | 2192 | path->reada = READA_FORWARD; |
5d4f98a2 | 2193 | path->leave_spinning = 1; |
0b246afa | 2194 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
2195 | if (ret < 0) { |
2196 | err = ret; | |
2197 | goto out; | |
2198 | } | |
2199 | if (ret > 0) { | |
3173a18f | 2200 | if (metadata) { |
55994887 FDBM |
2201 | if (path->slots[0] > 0) { |
2202 | path->slots[0]--; | |
2203 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2204 | path->slots[0]); | |
d278850e | 2205 | if (key.objectid == head->bytenr && |
55994887 | 2206 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 2207 | key.offset == head->num_bytes) |
55994887 FDBM |
2208 | ret = 0; |
2209 | } | |
2210 | if (ret > 0) { | |
2211 | btrfs_release_path(path); | |
2212 | metadata = 0; | |
3173a18f | 2213 | |
d278850e JB |
2214 | key.objectid = head->bytenr; |
2215 | key.offset = head->num_bytes; | |
55994887 FDBM |
2216 | key.type = BTRFS_EXTENT_ITEM_KEY; |
2217 | goto again; | |
2218 | } | |
2219 | } else { | |
2220 | err = -EIO; | |
2221 | goto out; | |
3173a18f | 2222 | } |
5d4f98a2 YZ |
2223 | } |
2224 | ||
2225 | leaf = path->nodes[0]; | |
2226 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ba3c2b19 | 2227 | |
6d8ff4e4 | 2228 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
2229 | err = -EINVAL; |
2230 | btrfs_print_v0_err(fs_info); | |
2231 | btrfs_abort_transaction(trans, err); | |
2232 | goto out; | |
2233 | } | |
2234 | ||
5d4f98a2 YZ |
2235 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2236 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2237 | |
5d4f98a2 YZ |
2238 | btrfs_mark_buffer_dirty(leaf); |
2239 | out: | |
2240 | btrfs_free_path(path); | |
2241 | return err; | |
56bec294 CM |
2242 | } |
2243 | ||
5d4f98a2 | 2244 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2245 | struct btrfs_delayed_ref_node *node, |
2246 | struct btrfs_delayed_extent_op *extent_op, | |
2247 | int insert_reserved) | |
56bec294 CM |
2248 | { |
2249 | int ret = 0; | |
5d4f98a2 | 2250 | struct btrfs_delayed_tree_ref *ref; |
5d4f98a2 YZ |
2251 | u64 parent = 0; |
2252 | u64 ref_root = 0; | |
56bec294 | 2253 | |
5d4f98a2 | 2254 | ref = btrfs_delayed_node_to_tree_ref(node); |
f97806f2 | 2255 | trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 2256 | |
5d4f98a2 YZ |
2257 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2258 | parent = ref->parent; | |
fcebe456 | 2259 | ref_root = ref->root; |
5d4f98a2 | 2260 | |
02794222 | 2261 | if (node->ref_mod != 1) { |
f97806f2 | 2262 | btrfs_err(trans->fs_info, |
02794222 LB |
2263 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
2264 | node->bytenr, node->ref_mod, node->action, ref_root, | |
2265 | parent); | |
2266 | return -EIO; | |
2267 | } | |
5d4f98a2 | 2268 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 2269 | BUG_ON(!extent_op || !extent_op->update_flags); |
21ebfbe7 | 2270 | ret = alloc_reserved_tree_block(trans, node, extent_op); |
5d4f98a2 | 2271 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
2272 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
2273 | ref->level, 0, 1, extent_op); | |
5d4f98a2 | 2274 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 2275 | ret = __btrfs_free_extent(trans, node, parent, ref_root, |
c682f9b3 | 2276 | ref->level, 0, 1, extent_op); |
5d4f98a2 YZ |
2277 | } else { |
2278 | BUG(); | |
2279 | } | |
56bec294 CM |
2280 | return ret; |
2281 | } | |
2282 | ||
2283 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 2284 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2285 | struct btrfs_delayed_ref_node *node, |
2286 | struct btrfs_delayed_extent_op *extent_op, | |
2287 | int insert_reserved) | |
56bec294 | 2288 | { |
79787eaa JM |
2289 | int ret = 0; |
2290 | ||
857cc2fc JB |
2291 | if (trans->aborted) { |
2292 | if (insert_reserved) | |
5fac7f9e | 2293 | btrfs_pin_extent(trans->fs_info, node->bytenr, |
857cc2fc | 2294 | node->num_bytes, 1); |
79787eaa | 2295 | return 0; |
857cc2fc | 2296 | } |
79787eaa | 2297 | |
5d4f98a2 YZ |
2298 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2299 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
f97806f2 | 2300 | ret = run_delayed_tree_ref(trans, node, extent_op, |
5d4f98a2 YZ |
2301 | insert_reserved); |
2302 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2303 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2bf98ef3 | 2304 | ret = run_delayed_data_ref(trans, node, extent_op, |
5d4f98a2 YZ |
2305 | insert_reserved); |
2306 | else | |
2307 | BUG(); | |
80ee54bf JB |
2308 | if (ret && insert_reserved) |
2309 | btrfs_pin_extent(trans->fs_info, node->bytenr, | |
2310 | node->num_bytes, 1); | |
5d4f98a2 | 2311 | return ret; |
56bec294 CM |
2312 | } |
2313 | ||
c6fc2454 | 2314 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2315 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2316 | { | |
cffc3374 FM |
2317 | struct btrfs_delayed_ref_node *ref; |
2318 | ||
e3d03965 | 2319 | if (RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
c6fc2454 | 2320 | return NULL; |
d7df2c79 | 2321 | |
cffc3374 FM |
2322 | /* |
2323 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2324 | * This is to prevent a ref count from going down to zero, which deletes | |
2325 | * the extent item from the extent tree, when there still are references | |
2326 | * to add, which would fail because they would not find the extent item. | |
2327 | */ | |
1d57ee94 WX |
2328 | if (!list_empty(&head->ref_add_list)) |
2329 | return list_first_entry(&head->ref_add_list, | |
2330 | struct btrfs_delayed_ref_node, add_list); | |
2331 | ||
e3d03965 | 2332 | ref = rb_entry(rb_first_cached(&head->ref_tree), |
0e0adbcf | 2333 | struct btrfs_delayed_ref_node, ref_node); |
1d57ee94 WX |
2334 | ASSERT(list_empty(&ref->add_list)); |
2335 | return ref; | |
56bec294 CM |
2336 | } |
2337 | ||
2eadaa22 JB |
2338 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
2339 | struct btrfs_delayed_ref_head *head) | |
2340 | { | |
2341 | spin_lock(&delayed_refs->lock); | |
2342 | head->processing = 0; | |
2343 | delayed_refs->num_heads_ready++; | |
2344 | spin_unlock(&delayed_refs->lock); | |
2345 | btrfs_delayed_ref_unlock(head); | |
2346 | } | |
2347 | ||
bedc6617 JB |
2348 | static struct btrfs_delayed_extent_op *cleanup_extent_op( |
2349 | struct btrfs_delayed_ref_head *head) | |
b00e6250 JB |
2350 | { |
2351 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
b00e6250 JB |
2352 | |
2353 | if (!extent_op) | |
bedc6617 JB |
2354 | return NULL; |
2355 | ||
b00e6250 | 2356 | if (head->must_insert_reserved) { |
bedc6617 | 2357 | head->extent_op = NULL; |
b00e6250 | 2358 | btrfs_free_delayed_extent_op(extent_op); |
bedc6617 | 2359 | return NULL; |
b00e6250 | 2360 | } |
bedc6617 JB |
2361 | return extent_op; |
2362 | } | |
2363 | ||
2364 | static int run_and_cleanup_extent_op(struct btrfs_trans_handle *trans, | |
2365 | struct btrfs_delayed_ref_head *head) | |
2366 | { | |
2367 | struct btrfs_delayed_extent_op *extent_op; | |
2368 | int ret; | |
2369 | ||
2370 | extent_op = cleanup_extent_op(head); | |
2371 | if (!extent_op) | |
2372 | return 0; | |
2373 | head->extent_op = NULL; | |
b00e6250 | 2374 | spin_unlock(&head->lock); |
20b9a2d6 | 2375 | ret = run_delayed_extent_op(trans, head, extent_op); |
b00e6250 JB |
2376 | btrfs_free_delayed_extent_op(extent_op); |
2377 | return ret ? ret : 1; | |
2378 | } | |
2379 | ||
31890da0 JB |
2380 | void btrfs_cleanup_ref_head_accounting(struct btrfs_fs_info *fs_info, |
2381 | struct btrfs_delayed_ref_root *delayed_refs, | |
2382 | struct btrfs_delayed_ref_head *head) | |
07c47775 | 2383 | { |
ba2c4d4e | 2384 | int nr_items = 1; /* Dropping this ref head update. */ |
07c47775 JB |
2385 | |
2386 | if (head->total_ref_mod < 0) { | |
2387 | struct btrfs_space_info *space_info; | |
2388 | u64 flags; | |
2389 | ||
2390 | if (head->is_data) | |
2391 | flags = BTRFS_BLOCK_GROUP_DATA; | |
2392 | else if (head->is_system) | |
2393 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
2394 | else | |
2395 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
280c2908 | 2396 | space_info = btrfs_find_space_info(fs_info, flags); |
07c47775 JB |
2397 | ASSERT(space_info); |
2398 | percpu_counter_add_batch(&space_info->total_bytes_pinned, | |
2399 | -head->num_bytes, | |
2400 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
2401 | ||
ba2c4d4e JB |
2402 | /* |
2403 | * We had csum deletions accounted for in our delayed refs rsv, | |
2404 | * we need to drop the csum leaves for this update from our | |
2405 | * delayed_refs_rsv. | |
2406 | */ | |
07c47775 JB |
2407 | if (head->is_data) { |
2408 | spin_lock(&delayed_refs->lock); | |
2409 | delayed_refs->pending_csums -= head->num_bytes; | |
2410 | spin_unlock(&delayed_refs->lock); | |
ba2c4d4e JB |
2411 | nr_items += btrfs_csum_bytes_to_leaves(fs_info, |
2412 | head->num_bytes); | |
07c47775 JB |
2413 | } |
2414 | } | |
2415 | ||
ba2c4d4e | 2416 | btrfs_delayed_refs_rsv_release(fs_info, nr_items); |
07c47775 JB |
2417 | } |
2418 | ||
194ab0bc | 2419 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
194ab0bc JB |
2420 | struct btrfs_delayed_ref_head *head) |
2421 | { | |
f9871edd NB |
2422 | |
2423 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
194ab0bc JB |
2424 | struct btrfs_delayed_ref_root *delayed_refs; |
2425 | int ret; | |
2426 | ||
2427 | delayed_refs = &trans->transaction->delayed_refs; | |
2428 | ||
bedc6617 | 2429 | ret = run_and_cleanup_extent_op(trans, head); |
194ab0bc JB |
2430 | if (ret < 0) { |
2431 | unselect_delayed_ref_head(delayed_refs, head); | |
2432 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
2433 | return ret; | |
2434 | } else if (ret) { | |
2435 | return ret; | |
2436 | } | |
2437 | ||
2438 | /* | |
2439 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
2440 | * and then re-check to make sure nobody got added. | |
2441 | */ | |
2442 | spin_unlock(&head->lock); | |
2443 | spin_lock(&delayed_refs->lock); | |
2444 | spin_lock(&head->lock); | |
e3d03965 | 2445 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root) || head->extent_op) { |
194ab0bc JB |
2446 | spin_unlock(&head->lock); |
2447 | spin_unlock(&delayed_refs->lock); | |
2448 | return 1; | |
2449 | } | |
d7baffda | 2450 | btrfs_delete_ref_head(delayed_refs, head); |
c1103f7a | 2451 | spin_unlock(&head->lock); |
1e7a1421 | 2452 | spin_unlock(&delayed_refs->lock); |
c1103f7a | 2453 | |
c1103f7a | 2454 | if (head->must_insert_reserved) { |
d278850e JB |
2455 | btrfs_pin_extent(fs_info, head->bytenr, |
2456 | head->num_bytes, 1); | |
c1103f7a | 2457 | if (head->is_data) { |
d278850e JB |
2458 | ret = btrfs_del_csums(trans, fs_info, head->bytenr, |
2459 | head->num_bytes); | |
c1103f7a JB |
2460 | } |
2461 | } | |
2462 | ||
31890da0 | 2463 | btrfs_cleanup_ref_head_accounting(fs_info, delayed_refs, head); |
07c47775 JB |
2464 | |
2465 | trace_run_delayed_ref_head(fs_info, head, 0); | |
c1103f7a | 2466 | btrfs_delayed_ref_unlock(head); |
d278850e | 2467 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
2468 | return 0; |
2469 | } | |
2470 | ||
b1cdbcb5 NB |
2471 | static struct btrfs_delayed_ref_head *btrfs_obtain_ref_head( |
2472 | struct btrfs_trans_handle *trans) | |
2473 | { | |
2474 | struct btrfs_delayed_ref_root *delayed_refs = | |
2475 | &trans->transaction->delayed_refs; | |
2476 | struct btrfs_delayed_ref_head *head = NULL; | |
2477 | int ret; | |
2478 | ||
2479 | spin_lock(&delayed_refs->lock); | |
5637c74b | 2480 | head = btrfs_select_ref_head(delayed_refs); |
b1cdbcb5 NB |
2481 | if (!head) { |
2482 | spin_unlock(&delayed_refs->lock); | |
2483 | return head; | |
2484 | } | |
2485 | ||
2486 | /* | |
2487 | * Grab the lock that says we are going to process all the refs for | |
2488 | * this head | |
2489 | */ | |
9e920a6f | 2490 | ret = btrfs_delayed_ref_lock(delayed_refs, head); |
b1cdbcb5 NB |
2491 | spin_unlock(&delayed_refs->lock); |
2492 | ||
2493 | /* | |
2494 | * We may have dropped the spin lock to get the head mutex lock, and | |
2495 | * that might have given someone else time to free the head. If that's | |
2496 | * true, it has been removed from our list and we can move on. | |
2497 | */ | |
2498 | if (ret == -EAGAIN) | |
2499 | head = ERR_PTR(-EAGAIN); | |
2500 | ||
2501 | return head; | |
2502 | } | |
2503 | ||
e7261386 NB |
2504 | static int btrfs_run_delayed_refs_for_head(struct btrfs_trans_handle *trans, |
2505 | struct btrfs_delayed_ref_head *locked_ref, | |
2506 | unsigned long *run_refs) | |
2507 | { | |
2508 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
2509 | struct btrfs_delayed_ref_root *delayed_refs; | |
2510 | struct btrfs_delayed_extent_op *extent_op; | |
2511 | struct btrfs_delayed_ref_node *ref; | |
2512 | int must_insert_reserved = 0; | |
2513 | int ret; | |
2514 | ||
2515 | delayed_refs = &trans->transaction->delayed_refs; | |
2516 | ||
0110a4c4 NB |
2517 | lockdep_assert_held(&locked_ref->mutex); |
2518 | lockdep_assert_held(&locked_ref->lock); | |
2519 | ||
e7261386 NB |
2520 | while ((ref = select_delayed_ref(locked_ref))) { |
2521 | if (ref->seq && | |
2522 | btrfs_check_delayed_seq(fs_info, ref->seq)) { | |
2523 | spin_unlock(&locked_ref->lock); | |
2524 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
2525 | return -EAGAIN; | |
2526 | } | |
2527 | ||
2528 | (*run_refs)++; | |
2529 | ref->in_tree = 0; | |
2530 | rb_erase_cached(&ref->ref_node, &locked_ref->ref_tree); | |
2531 | RB_CLEAR_NODE(&ref->ref_node); | |
2532 | if (!list_empty(&ref->add_list)) | |
2533 | list_del(&ref->add_list); | |
2534 | /* | |
2535 | * When we play the delayed ref, also correct the ref_mod on | |
2536 | * head | |
2537 | */ | |
2538 | switch (ref->action) { | |
2539 | case BTRFS_ADD_DELAYED_REF: | |
2540 | case BTRFS_ADD_DELAYED_EXTENT: | |
2541 | locked_ref->ref_mod -= ref->ref_mod; | |
2542 | break; | |
2543 | case BTRFS_DROP_DELAYED_REF: | |
2544 | locked_ref->ref_mod += ref->ref_mod; | |
2545 | break; | |
2546 | default: | |
2547 | WARN_ON(1); | |
2548 | } | |
2549 | atomic_dec(&delayed_refs->num_entries); | |
2550 | ||
2551 | /* | |
2552 | * Record the must_insert_reserved flag before we drop the | |
2553 | * spin lock. | |
2554 | */ | |
2555 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2556 | locked_ref->must_insert_reserved = 0; | |
2557 | ||
2558 | extent_op = locked_ref->extent_op; | |
2559 | locked_ref->extent_op = NULL; | |
2560 | spin_unlock(&locked_ref->lock); | |
2561 | ||
2562 | ret = run_one_delayed_ref(trans, ref, extent_op, | |
2563 | must_insert_reserved); | |
2564 | ||
2565 | btrfs_free_delayed_extent_op(extent_op); | |
2566 | if (ret) { | |
2567 | unselect_delayed_ref_head(delayed_refs, locked_ref); | |
2568 | btrfs_put_delayed_ref(ref); | |
2569 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", | |
2570 | ret); | |
2571 | return ret; | |
2572 | } | |
2573 | ||
2574 | btrfs_put_delayed_ref(ref); | |
2575 | cond_resched(); | |
2576 | ||
2577 | spin_lock(&locked_ref->lock); | |
2578 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); | |
2579 | } | |
2580 | ||
2581 | return 0; | |
2582 | } | |
2583 | ||
79787eaa JM |
2584 | /* |
2585 | * Returns 0 on success or if called with an already aborted transaction. | |
2586 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2587 | */ | |
d7df2c79 | 2588 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
d7df2c79 | 2589 | unsigned long nr) |
56bec294 | 2590 | { |
0a1e458a | 2591 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 | 2592 | struct btrfs_delayed_ref_root *delayed_refs; |
56bec294 | 2593 | struct btrfs_delayed_ref_head *locked_ref = NULL; |
0a2b2a84 | 2594 | ktime_t start = ktime_get(); |
56bec294 | 2595 | int ret; |
d7df2c79 | 2596 | unsigned long count = 0; |
0a2b2a84 | 2597 | unsigned long actual_count = 0; |
56bec294 CM |
2598 | |
2599 | delayed_refs = &trans->transaction->delayed_refs; | |
0110a4c4 | 2600 | do { |
56bec294 | 2601 | if (!locked_ref) { |
b1cdbcb5 | 2602 | locked_ref = btrfs_obtain_ref_head(trans); |
0110a4c4 NB |
2603 | if (IS_ERR_OR_NULL(locked_ref)) { |
2604 | if (PTR_ERR(locked_ref) == -EAGAIN) { | |
2605 | continue; | |
2606 | } else { | |
2607 | break; | |
2608 | } | |
56bec294 | 2609 | } |
0110a4c4 | 2610 | count++; |
56bec294 | 2611 | } |
2c3cf7d5 FM |
2612 | /* |
2613 | * We need to try and merge add/drops of the same ref since we | |
2614 | * can run into issues with relocate dropping the implicit ref | |
2615 | * and then it being added back again before the drop can | |
2616 | * finish. If we merged anything we need to re-loop so we can | |
2617 | * get a good ref. | |
2618 | * Or we can get node references of the same type that weren't | |
2619 | * merged when created due to bumps in the tree mod seq, and | |
2620 | * we need to merge them to prevent adding an inline extent | |
2621 | * backref before dropping it (triggering a BUG_ON at | |
2622 | * insert_inline_extent_backref()). | |
2623 | */ | |
d7df2c79 | 2624 | spin_lock(&locked_ref->lock); |
be97f133 | 2625 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); |
ae1e206b | 2626 | |
0110a4c4 NB |
2627 | ret = btrfs_run_delayed_refs_for_head(trans, locked_ref, |
2628 | &actual_count); | |
2629 | if (ret < 0 && ret != -EAGAIN) { | |
2630 | /* | |
2631 | * Error, btrfs_run_delayed_refs_for_head already | |
2632 | * unlocked everything so just bail out | |
2633 | */ | |
2634 | return ret; | |
2635 | } else if (!ret) { | |
2636 | /* | |
2637 | * Success, perform the usual cleanup of a processed | |
2638 | * head | |
2639 | */ | |
f9871edd | 2640 | ret = cleanup_ref_head(trans, locked_ref); |
194ab0bc | 2641 | if (ret > 0 ) { |
b00e6250 JB |
2642 | /* We dropped our lock, we need to loop. */ |
2643 | ret = 0; | |
d7df2c79 | 2644 | continue; |
194ab0bc JB |
2645 | } else if (ret) { |
2646 | return ret; | |
5d4f98a2 | 2647 | } |
22cd2e7d | 2648 | } |
1ce7a5ec | 2649 | |
b00e6250 | 2650 | /* |
0110a4c4 NB |
2651 | * Either success case or btrfs_run_delayed_refs_for_head |
2652 | * returned -EAGAIN, meaning we need to select another head | |
b00e6250 | 2653 | */ |
b00e6250 | 2654 | |
0110a4c4 | 2655 | locked_ref = NULL; |
c3e69d58 | 2656 | cond_resched(); |
0110a4c4 | 2657 | } while ((nr != -1 && count < nr) || locked_ref); |
0a2b2a84 JB |
2658 | |
2659 | /* | |
2660 | * We don't want to include ref heads since we can have empty ref heads | |
2661 | * and those will drastically skew our runtime down since we just do | |
2662 | * accounting, no actual extent tree updates. | |
2663 | */ | |
2664 | if (actual_count > 0) { | |
2665 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2666 | u64 avg; | |
2667 | ||
2668 | /* | |
2669 | * We weigh the current average higher than our current runtime | |
2670 | * to avoid large swings in the average. | |
2671 | */ | |
2672 | spin_lock(&delayed_refs->lock); | |
2673 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2674 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2675 | spin_unlock(&delayed_refs->lock); |
2676 | } | |
d7df2c79 | 2677 | return 0; |
c3e69d58 CM |
2678 | } |
2679 | ||
709c0486 AJ |
2680 | #ifdef SCRAMBLE_DELAYED_REFS |
2681 | /* | |
2682 | * Normally delayed refs get processed in ascending bytenr order. This | |
2683 | * correlates in most cases to the order added. To expose dependencies on this | |
2684 | * order, we start to process the tree in the middle instead of the beginning | |
2685 | */ | |
2686 | static u64 find_middle(struct rb_root *root) | |
2687 | { | |
2688 | struct rb_node *n = root->rb_node; | |
2689 | struct btrfs_delayed_ref_node *entry; | |
2690 | int alt = 1; | |
2691 | u64 middle; | |
2692 | u64 first = 0, last = 0; | |
2693 | ||
2694 | n = rb_first(root); | |
2695 | if (n) { | |
2696 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2697 | first = entry->bytenr; | |
2698 | } | |
2699 | n = rb_last(root); | |
2700 | if (n) { | |
2701 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2702 | last = entry->bytenr; | |
2703 | } | |
2704 | n = root->rb_node; | |
2705 | ||
2706 | while (n) { | |
2707 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2708 | WARN_ON(!entry->in_tree); | |
2709 | ||
2710 | middle = entry->bytenr; | |
2711 | ||
2712 | if (alt) | |
2713 | n = n->rb_left; | |
2714 | else | |
2715 | n = n->rb_right; | |
2716 | ||
2717 | alt = 1 - alt; | |
2718 | } | |
2719 | return middle; | |
2720 | } | |
2721 | #endif | |
2722 | ||
2ff7e61e | 2723 | static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads) |
1be41b78 JB |
2724 | { |
2725 | u64 num_bytes; | |
2726 | ||
2727 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2728 | sizeof(struct btrfs_extent_inline_ref)); | |
0b246afa | 2729 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1be41b78 JB |
2730 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
2731 | ||
2732 | /* | |
2733 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2734 | * closer to what we're really going to want to use. |
1be41b78 | 2735 | */ |
0b246afa | 2736 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info)); |
1be41b78 JB |
2737 | } |
2738 | ||
1262133b JB |
2739 | /* |
2740 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2741 | * would require to store the csums for that many bytes. | |
2742 | */ | |
2ff7e61e | 2743 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2744 | { |
2745 | u64 csum_size; | |
2746 | u64 num_csums_per_leaf; | |
2747 | u64 num_csums; | |
2748 | ||
0b246afa | 2749 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2750 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2751 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2752 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2753 | num_csums += num_csums_per_leaf - 1; |
2754 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2755 | return num_csums; | |
2756 | } | |
2757 | ||
64403612 | 2758 | bool btrfs_check_space_for_delayed_refs(struct btrfs_fs_info *fs_info) |
1be41b78 | 2759 | { |
64403612 JB |
2760 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; |
2761 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
2762 | bool ret = false; | |
2763 | u64 reserved; | |
1be41b78 | 2764 | |
64403612 JB |
2765 | spin_lock(&global_rsv->lock); |
2766 | reserved = global_rsv->reserved; | |
2767 | spin_unlock(&global_rsv->lock); | |
1be41b78 JB |
2768 | |
2769 | /* | |
64403612 JB |
2770 | * Since the global reserve is just kind of magic we don't really want |
2771 | * to rely on it to save our bacon, so if our size is more than the | |
2772 | * delayed_refs_rsv and the global rsv then it's time to think about | |
2773 | * bailing. | |
1be41b78 | 2774 | */ |
64403612 JB |
2775 | spin_lock(&delayed_refs_rsv->lock); |
2776 | reserved += delayed_refs_rsv->reserved; | |
2777 | if (delayed_refs_rsv->size >= reserved) | |
2778 | ret = true; | |
2779 | spin_unlock(&delayed_refs_rsv->lock); | |
1be41b78 JB |
2780 | return ret; |
2781 | } | |
2782 | ||
7c861627 | 2783 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans) |
0a2b2a84 | 2784 | { |
0a2b2a84 JB |
2785 | u64 num_entries = |
2786 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2787 | u64 avg_runtime; | |
a79b7d4b | 2788 | u64 val; |
0a2b2a84 JB |
2789 | |
2790 | smp_mb(); | |
7c861627 | 2791 | avg_runtime = trans->fs_info->avg_delayed_ref_runtime; |
a79b7d4b | 2792 | val = num_entries * avg_runtime; |
dc1a90c6 | 2793 | if (val >= NSEC_PER_SEC) |
0a2b2a84 | 2794 | return 1; |
a79b7d4b CM |
2795 | if (val >= NSEC_PER_SEC / 2) |
2796 | return 2; | |
0a2b2a84 | 2797 | |
64403612 | 2798 | return btrfs_check_space_for_delayed_refs(trans->fs_info); |
0a2b2a84 JB |
2799 | } |
2800 | ||
c3e69d58 CM |
2801 | /* |
2802 | * this starts processing the delayed reference count updates and | |
2803 | * extent insertions we have queued up so far. count can be | |
2804 | * 0, which means to process everything in the tree at the start | |
2805 | * of the run (but not newly added entries), or it can be some target | |
2806 | * number you'd like to process. | |
79787eaa JM |
2807 | * |
2808 | * Returns 0 on success or if called with an aborted transaction | |
2809 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2810 | */ |
2811 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
c79a70b1 | 2812 | unsigned long count) |
c3e69d58 | 2813 | { |
c79a70b1 | 2814 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 CM |
2815 | struct rb_node *node; |
2816 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2817 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2818 | int ret; |
2819 | int run_all = count == (unsigned long)-1; | |
c3e69d58 | 2820 | |
79787eaa JM |
2821 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2822 | if (trans->aborted) | |
2823 | return 0; | |
2824 | ||
0b246afa | 2825 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2826 | return 0; |
2827 | ||
c3e69d58 | 2828 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 2829 | if (count == 0) |
d7df2c79 | 2830 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 2831 | |
c3e69d58 | 2832 | again: |
709c0486 AJ |
2833 | #ifdef SCRAMBLE_DELAYED_REFS |
2834 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2835 | #endif | |
0a1e458a | 2836 | ret = __btrfs_run_delayed_refs(trans, count); |
d7df2c79 | 2837 | if (ret < 0) { |
66642832 | 2838 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 2839 | return ret; |
eb099670 | 2840 | } |
c3e69d58 | 2841 | |
56bec294 | 2842 | if (run_all) { |
119e80df | 2843 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2844 | |
d7df2c79 | 2845 | spin_lock(&delayed_refs->lock); |
5c9d028b | 2846 | node = rb_first_cached(&delayed_refs->href_root); |
d7df2c79 JB |
2847 | if (!node) { |
2848 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2849 | goto out; |
d7df2c79 | 2850 | } |
d278850e JB |
2851 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2852 | href_node); | |
2853 | refcount_inc(&head->refs); | |
2854 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 2855 | |
d278850e JB |
2856 | /* Mutex was contended, block until it's released and retry. */ |
2857 | mutex_lock(&head->mutex); | |
2858 | mutex_unlock(&head->mutex); | |
56bec294 | 2859 | |
d278850e | 2860 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 2861 | cond_resched(); |
56bec294 | 2862 | goto again; |
5f39d397 | 2863 | } |
54aa1f4d | 2864 | out: |
a28ec197 CM |
2865 | return 0; |
2866 | } | |
2867 | ||
5d4f98a2 | 2868 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
5d4f98a2 | 2869 | u64 bytenr, u64 num_bytes, u64 flags, |
b1c79e09 | 2870 | int level, int is_data) |
5d4f98a2 YZ |
2871 | { |
2872 | struct btrfs_delayed_extent_op *extent_op; | |
2873 | int ret; | |
2874 | ||
78a6184a | 2875 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2876 | if (!extent_op) |
2877 | return -ENOMEM; | |
2878 | ||
2879 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
2880 | extent_op->update_flags = true; |
2881 | extent_op->update_key = false; | |
2882 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 2883 | extent_op->level = level; |
5d4f98a2 | 2884 | |
c6e340bc | 2885 | ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op); |
5d4f98a2 | 2886 | if (ret) |
78a6184a | 2887 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2888 | return ret; |
2889 | } | |
2890 | ||
e4c3b2dc | 2891 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2892 | struct btrfs_path *path, |
2893 | u64 objectid, u64 offset, u64 bytenr) | |
2894 | { | |
2895 | struct btrfs_delayed_ref_head *head; | |
2896 | struct btrfs_delayed_ref_node *ref; | |
2897 | struct btrfs_delayed_data_ref *data_ref; | |
2898 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 2899 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 2900 | struct rb_node *node; |
5d4f98a2 YZ |
2901 | int ret = 0; |
2902 | ||
998ac6d2 | 2903 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 2904 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 2905 | if (cur_trans) |
2906 | refcount_inc(&cur_trans->use_count); | |
2907 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
2908 | if (!cur_trans) |
2909 | return 0; | |
2910 | ||
2911 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 2912 | spin_lock(&delayed_refs->lock); |
f72ad18e | 2913 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
2914 | if (!head) { |
2915 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 2916 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
2917 | return 0; |
2918 | } | |
5d4f98a2 YZ |
2919 | |
2920 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 2921 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
2922 | spin_unlock(&delayed_refs->lock); |
2923 | ||
b3b4aa74 | 2924 | btrfs_release_path(path); |
5d4f98a2 | 2925 | |
8cc33e5c DS |
2926 | /* |
2927 | * Mutex was contended, block until it's released and let | |
2928 | * caller try again | |
2929 | */ | |
5d4f98a2 YZ |
2930 | mutex_lock(&head->mutex); |
2931 | mutex_unlock(&head->mutex); | |
d278850e | 2932 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 2933 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2934 | return -EAGAIN; |
2935 | } | |
d7df2c79 | 2936 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 2937 | |
d7df2c79 | 2938 | spin_lock(&head->lock); |
0e0adbcf JB |
2939 | /* |
2940 | * XXX: We should replace this with a proper search function in the | |
2941 | * future. | |
2942 | */ | |
e3d03965 LB |
2943 | for (node = rb_first_cached(&head->ref_tree); node; |
2944 | node = rb_next(node)) { | |
0e0adbcf | 2945 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
d7df2c79 JB |
2946 | /* If it's a shared ref we know a cross reference exists */ |
2947 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
2948 | ret = 1; | |
2949 | break; | |
2950 | } | |
5d4f98a2 | 2951 | |
d7df2c79 | 2952 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 2953 | |
d7df2c79 JB |
2954 | /* |
2955 | * If our ref doesn't match the one we're currently looking at | |
2956 | * then we have a cross reference. | |
2957 | */ | |
2958 | if (data_ref->root != root->root_key.objectid || | |
2959 | data_ref->objectid != objectid || | |
2960 | data_ref->offset != offset) { | |
2961 | ret = 1; | |
2962 | break; | |
2963 | } | |
5d4f98a2 | 2964 | } |
d7df2c79 | 2965 | spin_unlock(&head->lock); |
5d4f98a2 | 2966 | mutex_unlock(&head->mutex); |
998ac6d2 | 2967 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
2968 | return ret; |
2969 | } | |
2970 | ||
e4c3b2dc | 2971 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2972 | struct btrfs_path *path, |
2973 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d | 2974 | { |
0b246afa JM |
2975 | struct btrfs_fs_info *fs_info = root->fs_info; |
2976 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 2977 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
2978 | struct btrfs_extent_data_ref *ref; |
2979 | struct btrfs_extent_inline_ref *iref; | |
2980 | struct btrfs_extent_item *ei; | |
f321e491 | 2981 | struct btrfs_key key; |
5d4f98a2 | 2982 | u32 item_size; |
3de28d57 | 2983 | int type; |
be20aa9d | 2984 | int ret; |
925baedd | 2985 | |
be20aa9d | 2986 | key.objectid = bytenr; |
31840ae1 | 2987 | key.offset = (u64)-1; |
f321e491 | 2988 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 2989 | |
be20aa9d CM |
2990 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
2991 | if (ret < 0) | |
2992 | goto out; | |
79787eaa | 2993 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
2994 | |
2995 | ret = -ENOENT; | |
2996 | if (path->slots[0] == 0) | |
31840ae1 | 2997 | goto out; |
be20aa9d | 2998 | |
31840ae1 | 2999 | path->slots[0]--; |
f321e491 | 3000 | leaf = path->nodes[0]; |
5d4f98a2 | 3001 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3002 | |
5d4f98a2 | 3003 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3004 | goto out; |
f321e491 | 3005 | |
5d4f98a2 YZ |
3006 | ret = 1; |
3007 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
5d4f98a2 | 3008 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
bd09835d | 3009 | |
5d4f98a2 YZ |
3010 | if (item_size != sizeof(*ei) + |
3011 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3012 | goto out; | |
be20aa9d | 3013 | |
5d4f98a2 YZ |
3014 | if (btrfs_extent_generation(leaf, ei) <= |
3015 | btrfs_root_last_snapshot(&root->root_item)) | |
3016 | goto out; | |
3017 | ||
3018 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 LB |
3019 | |
3020 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
3021 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
3022 | goto out; |
3023 | ||
3024 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3025 | if (btrfs_extent_refs(leaf, ei) != | |
3026 | btrfs_extent_data_ref_count(leaf, ref) || | |
3027 | btrfs_extent_data_ref_root(leaf, ref) != | |
3028 | root->root_key.objectid || | |
3029 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3030 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3031 | goto out; | |
3032 | ||
3033 | ret = 0; | |
3034 | out: | |
3035 | return ret; | |
3036 | } | |
3037 | ||
e4c3b2dc LB |
3038 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
3039 | u64 bytenr) | |
5d4f98a2 YZ |
3040 | { |
3041 | struct btrfs_path *path; | |
3042 | int ret; | |
5d4f98a2 YZ |
3043 | |
3044 | path = btrfs_alloc_path(); | |
3045 | if (!path) | |
9132c4ff | 3046 | return -ENOMEM; |
5d4f98a2 YZ |
3047 | |
3048 | do { | |
e4c3b2dc | 3049 | ret = check_committed_ref(root, path, objectid, |
5d4f98a2 YZ |
3050 | offset, bytenr); |
3051 | if (ret && ret != -ENOENT) | |
f321e491 | 3052 | goto out; |
80ff3856 | 3053 | |
380fd066 MT |
3054 | ret = check_delayed_ref(root, path, objectid, offset, bytenr); |
3055 | } while (ret == -EAGAIN); | |
5d4f98a2 | 3056 | |
be20aa9d | 3057 | out: |
80ff3856 | 3058 | btrfs_free_path(path); |
f0486c68 YZ |
3059 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3060 | WARN_ON(ret > 0); | |
f321e491 | 3061 | return ret; |
be20aa9d | 3062 | } |
c5739bba | 3063 | |
5d4f98a2 | 3064 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3065 | struct btrfs_root *root, |
5d4f98a2 | 3066 | struct extent_buffer *buf, |
e339a6b0 | 3067 | int full_backref, int inc) |
31840ae1 | 3068 | { |
0b246afa | 3069 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 3070 | u64 bytenr; |
5d4f98a2 YZ |
3071 | u64 num_bytes; |
3072 | u64 parent; | |
31840ae1 | 3073 | u64 ref_root; |
31840ae1 | 3074 | u32 nritems; |
31840ae1 ZY |
3075 | struct btrfs_key key; |
3076 | struct btrfs_file_extent_item *fi; | |
82fa113f QW |
3077 | struct btrfs_ref generic_ref = { 0 }; |
3078 | bool for_reloc = btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC); | |
31840ae1 | 3079 | int i; |
82fa113f | 3080 | int action; |
31840ae1 ZY |
3081 | int level; |
3082 | int ret = 0; | |
fccb84c9 | 3083 | |
0b246afa | 3084 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3085 | return 0; |
fccb84c9 | 3086 | |
31840ae1 | 3087 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3088 | nritems = btrfs_header_nritems(buf); |
3089 | level = btrfs_header_level(buf); | |
3090 | ||
27cdeb70 | 3091 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3092 | return 0; |
31840ae1 | 3093 | |
5d4f98a2 YZ |
3094 | if (full_backref) |
3095 | parent = buf->start; | |
3096 | else | |
3097 | parent = 0; | |
82fa113f QW |
3098 | if (inc) |
3099 | action = BTRFS_ADD_DELAYED_REF; | |
3100 | else | |
3101 | action = BTRFS_DROP_DELAYED_REF; | |
5d4f98a2 YZ |
3102 | |
3103 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3104 | if (level == 0) { |
5d4f98a2 | 3105 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3106 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3107 | continue; |
5d4f98a2 | 3108 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3109 | struct btrfs_file_extent_item); |
3110 | if (btrfs_file_extent_type(buf, fi) == | |
3111 | BTRFS_FILE_EXTENT_INLINE) | |
3112 | continue; | |
3113 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3114 | if (bytenr == 0) | |
3115 | continue; | |
5d4f98a2 YZ |
3116 | |
3117 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3118 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
82fa113f QW |
3119 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
3120 | num_bytes, parent); | |
3121 | generic_ref.real_root = root->root_key.objectid; | |
3122 | btrfs_init_data_ref(&generic_ref, ref_root, key.objectid, | |
3123 | key.offset); | |
3124 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 3125 | if (inc) |
82fa113f | 3126 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 3127 | else |
ffd4bb2a | 3128 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
3129 | if (ret) |
3130 | goto fail; | |
3131 | } else { | |
5d4f98a2 | 3132 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 3133 | num_bytes = fs_info->nodesize; |
82fa113f QW |
3134 | btrfs_init_generic_ref(&generic_ref, action, bytenr, |
3135 | num_bytes, parent); | |
3136 | generic_ref.real_root = root->root_key.objectid; | |
3137 | btrfs_init_tree_ref(&generic_ref, level - 1, ref_root); | |
3138 | generic_ref.skip_qgroup = for_reloc; | |
dd28b6a5 | 3139 | if (inc) |
82fa113f | 3140 | ret = btrfs_inc_extent_ref(trans, &generic_ref); |
dd28b6a5 | 3141 | else |
ffd4bb2a | 3142 | ret = btrfs_free_extent(trans, &generic_ref); |
31840ae1 ZY |
3143 | if (ret) |
3144 | goto fail; | |
3145 | } | |
3146 | } | |
3147 | return 0; | |
3148 | fail: | |
5d4f98a2 YZ |
3149 | return ret; |
3150 | } | |
3151 | ||
3152 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3153 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3154 | { |
e339a6b0 | 3155 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3156 | } |
3157 | ||
3158 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3159 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3160 | { |
e339a6b0 | 3161 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3162 | } |
3163 | ||
9078a3e1 | 3164 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
9078a3e1 CM |
3165 | struct btrfs_path *path, |
3166 | struct btrfs_block_group_cache *cache) | |
3167 | { | |
39db232d | 3168 | struct btrfs_fs_info *fs_info = trans->fs_info; |
9078a3e1 | 3169 | int ret; |
0b246afa | 3170 | struct btrfs_root *extent_root = fs_info->extent_root; |
5f39d397 CM |
3171 | unsigned long bi; |
3172 | struct extent_buffer *leaf; | |
9078a3e1 | 3173 | |
9078a3e1 | 3174 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3175 | if (ret) { |
3176 | if (ret > 0) | |
3177 | ret = -ENOENT; | |
54aa1f4d | 3178 | goto fail; |
df95e7f0 | 3179 | } |
5f39d397 CM |
3180 | |
3181 | leaf = path->nodes[0]; | |
3182 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3183 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3184 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3185 | fail: |
24b89d08 | 3186 | btrfs_release_path(path); |
df95e7f0 | 3187 | return ret; |
9078a3e1 CM |
3188 | |
3189 | } | |
3190 | ||
f87b7eb8 DS |
3191 | static struct btrfs_block_group_cache *next_block_group( |
3192 | struct btrfs_block_group_cache *cache) | |
4a8c9a62 | 3193 | { |
f87b7eb8 | 3194 | struct btrfs_fs_info *fs_info = cache->fs_info; |
4a8c9a62 | 3195 | struct rb_node *node; |
292cbd51 | 3196 | |
0b246afa | 3197 | spin_lock(&fs_info->block_group_cache_lock); |
292cbd51 FM |
3198 | |
3199 | /* If our block group was removed, we need a full search. */ | |
3200 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3201 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3202 | ||
0b246afa | 3203 | spin_unlock(&fs_info->block_group_cache_lock); |
292cbd51 | 3204 | btrfs_put_block_group(cache); |
0b246afa | 3205 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; |
292cbd51 | 3206 | } |
4a8c9a62 YZ |
3207 | node = rb_next(&cache->cache_node); |
3208 | btrfs_put_block_group(cache); | |
3209 | if (node) { | |
3210 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3211 | cache_node); | |
11dfe35a | 3212 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3213 | } else |
3214 | cache = NULL; | |
0b246afa | 3215 | spin_unlock(&fs_info->block_group_cache_lock); |
4a8c9a62 YZ |
3216 | return cache; |
3217 | } | |
3218 | ||
0af3d00b JB |
3219 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3220 | struct btrfs_trans_handle *trans, | |
3221 | struct btrfs_path *path) | |
3222 | { | |
0b246afa JM |
3223 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
3224 | struct btrfs_root *root = fs_info->tree_root; | |
0af3d00b | 3225 | struct inode *inode = NULL; |
364ecf36 | 3226 | struct extent_changeset *data_reserved = NULL; |
0af3d00b | 3227 | u64 alloc_hint = 0; |
2b20982e | 3228 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3229 | u64 num_pages = 0; |
0af3d00b JB |
3230 | int retries = 0; |
3231 | int ret = 0; | |
3232 | ||
3233 | /* | |
3234 | * If this block group is smaller than 100 megs don't bother caching the | |
3235 | * block group. | |
3236 | */ | |
ee22184b | 3237 | if (block_group->key.offset < (100 * SZ_1M)) { |
0af3d00b JB |
3238 | spin_lock(&block_group->lock); |
3239 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3240 | spin_unlock(&block_group->lock); | |
3241 | return 0; | |
3242 | } | |
3243 | ||
0c0ef4bc JB |
3244 | if (trans->aborted) |
3245 | return 0; | |
0af3d00b | 3246 | again: |
7949f339 | 3247 | inode = lookup_free_space_inode(block_group, path); |
0af3d00b JB |
3248 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { |
3249 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3250 | btrfs_release_path(path); |
0af3d00b JB |
3251 | goto out; |
3252 | } | |
3253 | ||
3254 | if (IS_ERR(inode)) { | |
3255 | BUG_ON(retries); | |
3256 | retries++; | |
3257 | ||
3258 | if (block_group->ro) | |
3259 | goto out_free; | |
3260 | ||
4ca75f1b | 3261 | ret = create_free_space_inode(trans, block_group, path); |
0af3d00b JB |
3262 | if (ret) |
3263 | goto out_free; | |
3264 | goto again; | |
3265 | } | |
3266 | ||
3267 | /* | |
3268 | * We want to set the generation to 0, that way if anything goes wrong | |
3269 | * from here on out we know not to trust this cache when we load up next | |
3270 | * time. | |
3271 | */ | |
3272 | BTRFS_I(inode)->generation = 0; | |
3273 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3274 | if (ret) { |
3275 | /* | |
3276 | * So theoretically we could recover from this, simply set the | |
3277 | * super cache generation to 0 so we know to invalidate the | |
3278 | * cache, but then we'd have to keep track of the block groups | |
3279 | * that fail this way so we know we _have_ to reset this cache | |
3280 | * before the next commit or risk reading stale cache. So to | |
3281 | * limit our exposure to horrible edge cases lets just abort the | |
3282 | * transaction, this only happens in really bad situations | |
3283 | * anyway. | |
3284 | */ | |
66642832 | 3285 | btrfs_abort_transaction(trans, ret); |
0c0ef4bc JB |
3286 | goto out_put; |
3287 | } | |
0af3d00b JB |
3288 | WARN_ON(ret); |
3289 | ||
8e138e0d JB |
3290 | /* We've already setup this transaction, go ahead and exit */ |
3291 | if (block_group->cache_generation == trans->transid && | |
3292 | i_size_read(inode)) { | |
3293 | dcs = BTRFS_DC_SETUP; | |
3294 | goto out_put; | |
3295 | } | |
3296 | ||
0af3d00b | 3297 | if (i_size_read(inode) > 0) { |
2ff7e61e | 3298 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
0b246afa | 3299 | &fs_info->global_block_rsv); |
7b61cd92 MX |
3300 | if (ret) |
3301 | goto out_put; | |
3302 | ||
77ab86bf | 3303 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); |
0af3d00b JB |
3304 | if (ret) |
3305 | goto out_put; | |
3306 | } | |
3307 | ||
3308 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3309 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
0b246afa | 3310 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cf7c1ef6 LB |
3311 | /* |
3312 | * don't bother trying to write stuff out _if_ | |
3313 | * a) we're not cached, | |
1a79c1f2 LB |
3314 | * b) we're with nospace_cache mount option, |
3315 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
cf7c1ef6 | 3316 | */ |
2b20982e | 3317 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3318 | spin_unlock(&block_group->lock); |
3319 | goto out_put; | |
3320 | } | |
3321 | spin_unlock(&block_group->lock); | |
3322 | ||
2968b1f4 JB |
3323 | /* |
3324 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3325 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3326 | */ | |
3327 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3328 | ret = -ENOSPC; | |
3329 | goto out_put; | |
3330 | } | |
3331 | ||
6fc823b1 JB |
3332 | /* |
3333 | * Try to preallocate enough space based on how big the block group is. | |
3334 | * Keep in mind this has to include any pinned space which could end up | |
3335 | * taking up quite a bit since it's not folded into the other space | |
3336 | * cache. | |
3337 | */ | |
ee22184b | 3338 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
0af3d00b JB |
3339 | if (!num_pages) |
3340 | num_pages = 1; | |
3341 | ||
0af3d00b | 3342 | num_pages *= 16; |
09cbfeaf | 3343 | num_pages *= PAGE_SIZE; |
0af3d00b | 3344 | |
364ecf36 | 3345 | ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages); |
0af3d00b JB |
3346 | if (ret) |
3347 | goto out_put; | |
3348 | ||
3349 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3350 | num_pages, num_pages, | |
3351 | &alloc_hint); | |
2968b1f4 JB |
3352 | /* |
3353 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3354 | * of metadata or split extents when writing the cache out, which means | |
3355 | * we can enospc if we are heavily fragmented in addition to just normal | |
3356 | * out of space conditions. So if we hit this just skip setting up any | |
3357 | * other block groups for this transaction, maybe we'll unpin enough | |
3358 | * space the next time around. | |
3359 | */ | |
2b20982e JB |
3360 | if (!ret) |
3361 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3362 | else if (ret == -ENOSPC) |
3363 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
c09544e0 | 3364 | |
0af3d00b JB |
3365 | out_put: |
3366 | iput(inode); | |
3367 | out_free: | |
b3b4aa74 | 3368 | btrfs_release_path(path); |
0af3d00b JB |
3369 | out: |
3370 | spin_lock(&block_group->lock); | |
e65cbb94 | 3371 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3372 | block_group->cache_generation = trans->transid; |
2b20982e | 3373 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3374 | spin_unlock(&block_group->lock); |
3375 | ||
364ecf36 | 3376 | extent_changeset_free(data_reserved); |
0af3d00b JB |
3377 | return ret; |
3378 | } | |
3379 | ||
bbebb3e0 | 3380 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans) |
dcdf7f6d | 3381 | { |
bbebb3e0 | 3382 | struct btrfs_fs_info *fs_info = trans->fs_info; |
dcdf7f6d JB |
3383 | struct btrfs_block_group_cache *cache, *tmp; |
3384 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3385 | struct btrfs_path *path; | |
3386 | ||
3387 | if (list_empty(&cur_trans->dirty_bgs) || | |
0b246afa | 3388 | !btrfs_test_opt(fs_info, SPACE_CACHE)) |
dcdf7f6d JB |
3389 | return 0; |
3390 | ||
3391 | path = btrfs_alloc_path(); | |
3392 | if (!path) | |
3393 | return -ENOMEM; | |
3394 | ||
3395 | /* Could add new block groups, use _safe just in case */ | |
3396 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3397 | dirty_list) { | |
3398 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3399 | cache_save_setup(cache, trans, path); | |
3400 | } | |
3401 | ||
3402 | btrfs_free_path(path); | |
3403 | return 0; | |
3404 | } | |
3405 | ||
1bbc621e CM |
3406 | /* |
3407 | * transaction commit does final block group cache writeback during a | |
3408 | * critical section where nothing is allowed to change the FS. This is | |
3409 | * required in order for the cache to actually match the block group, | |
3410 | * but can introduce a lot of latency into the commit. | |
3411 | * | |
3412 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3413 | * cache IO. There's a chance we'll have to redo some of it if the | |
3414 | * block group changes again during the commit, but it greatly reduces | |
3415 | * the commit latency by getting rid of the easy block groups while | |
3416 | * we're still allowing others to join the commit. | |
3417 | */ | |
21217054 | 3418 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) |
9078a3e1 | 3419 | { |
21217054 | 3420 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4a8c9a62 | 3421 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3422 | struct btrfs_transaction *cur_trans = trans->transaction; |
3423 | int ret = 0; | |
c9dc4c65 | 3424 | int should_put; |
1bbc621e CM |
3425 | struct btrfs_path *path = NULL; |
3426 | LIST_HEAD(dirty); | |
3427 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3428 | int num_started = 0; |
1bbc621e CM |
3429 | int loops = 0; |
3430 | ||
3431 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3432 | if (list_empty(&cur_trans->dirty_bgs)) { |
3433 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3434 | return 0; | |
1bbc621e | 3435 | } |
b58d1a9e | 3436 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3437 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3438 | |
1bbc621e | 3439 | again: |
1bbc621e CM |
3440 | /* |
3441 | * make sure all the block groups on our dirty list actually | |
3442 | * exist | |
3443 | */ | |
6c686b35 | 3444 | btrfs_create_pending_block_groups(trans); |
1bbc621e CM |
3445 | |
3446 | if (!path) { | |
3447 | path = btrfs_alloc_path(); | |
3448 | if (!path) | |
3449 | return -ENOMEM; | |
3450 | } | |
3451 | ||
b58d1a9e FM |
3452 | /* |
3453 | * cache_write_mutex is here only to save us from balance or automatic | |
3454 | * removal of empty block groups deleting this block group while we are | |
3455 | * writing out the cache | |
3456 | */ | |
3457 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3458 | while (!list_empty(&dirty)) { |
ba2c4d4e JB |
3459 | bool drop_reserve = true; |
3460 | ||
1bbc621e CM |
3461 | cache = list_first_entry(&dirty, |
3462 | struct btrfs_block_group_cache, | |
3463 | dirty_list); | |
1bbc621e CM |
3464 | /* |
3465 | * this can happen if something re-dirties a block | |
3466 | * group that is already under IO. Just wait for it to | |
3467 | * finish and then do it all again | |
3468 | */ | |
3469 | if (!list_empty(&cache->io_list)) { | |
3470 | list_del_init(&cache->io_list); | |
afdb5718 | 3471 | btrfs_wait_cache_io(trans, cache, path); |
1bbc621e CM |
3472 | btrfs_put_block_group(cache); |
3473 | } | |
3474 | ||
3475 | ||
3476 | /* | |
3477 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3478 | * if it should update the cache_state. Don't delete | |
3479 | * until after we wait. | |
3480 | * | |
3481 | * Since we're not running in the commit critical section | |
3482 | * we need the dirty_bgs_lock to protect from update_block_group | |
3483 | */ | |
3484 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3485 | list_del_init(&cache->dirty_list); | |
3486 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3487 | ||
3488 | should_put = 1; | |
3489 | ||
3490 | cache_save_setup(cache, trans, path); | |
3491 | ||
3492 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3493 | cache->io_ctl.inode = NULL; | |
fe041534 | 3494 | ret = btrfs_write_out_cache(trans, cache, path); |
1bbc621e CM |
3495 | if (ret == 0 && cache->io_ctl.inode) { |
3496 | num_started++; | |
3497 | should_put = 0; | |
3498 | ||
3499 | /* | |
45ae2c18 NB |
3500 | * The cache_write_mutex is protecting the |
3501 | * io_list, also refer to the definition of | |
3502 | * btrfs_transaction::io_bgs for more details | |
1bbc621e CM |
3503 | */ |
3504 | list_add_tail(&cache->io_list, io); | |
3505 | } else { | |
3506 | /* | |
3507 | * if we failed to write the cache, the | |
3508 | * generation will be bad and life goes on | |
3509 | */ | |
3510 | ret = 0; | |
3511 | } | |
3512 | } | |
ff1f8250 | 3513 | if (!ret) { |
39db232d | 3514 | ret = write_one_cache_group(trans, path, cache); |
ff1f8250 FM |
3515 | /* |
3516 | * Our block group might still be attached to the list | |
3517 | * of new block groups in the transaction handle of some | |
3518 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3519 | * means its block group item isn't yet in the extent | |
3520 | * tree. If this happens ignore the error, as we will | |
3521 | * try again later in the critical section of the | |
3522 | * transaction commit. | |
3523 | */ | |
3524 | if (ret == -ENOENT) { | |
3525 | ret = 0; | |
3526 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3527 | if (list_empty(&cache->dirty_list)) { | |
3528 | list_add_tail(&cache->dirty_list, | |
3529 | &cur_trans->dirty_bgs); | |
3530 | btrfs_get_block_group(cache); | |
ba2c4d4e | 3531 | drop_reserve = false; |
ff1f8250 FM |
3532 | } |
3533 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3534 | } else if (ret) { | |
66642832 | 3535 | btrfs_abort_transaction(trans, ret); |
ff1f8250 FM |
3536 | } |
3537 | } | |
1bbc621e | 3538 | |
52042d8e | 3539 | /* if it's not on the io list, we need to put the block group */ |
1bbc621e CM |
3540 | if (should_put) |
3541 | btrfs_put_block_group(cache); | |
ba2c4d4e JB |
3542 | if (drop_reserve) |
3543 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
1bbc621e CM |
3544 | |
3545 | if (ret) | |
3546 | break; | |
b58d1a9e FM |
3547 | |
3548 | /* | |
3549 | * Avoid blocking other tasks for too long. It might even save | |
3550 | * us from writing caches for block groups that are going to be | |
3551 | * removed. | |
3552 | */ | |
3553 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3554 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3555 | } |
b58d1a9e | 3556 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3557 | |
3558 | /* | |
3559 | * go through delayed refs for all the stuff we've just kicked off | |
3560 | * and then loop back (just once) | |
3561 | */ | |
c79a70b1 | 3562 | ret = btrfs_run_delayed_refs(trans, 0); |
1bbc621e CM |
3563 | if (!ret && loops == 0) { |
3564 | loops++; | |
3565 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3566 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3567 | /* |
3568 | * dirty_bgs_lock protects us from concurrent block group | |
3569 | * deletes too (not just cache_write_mutex). | |
3570 | */ | |
3571 | if (!list_empty(&dirty)) { | |
3572 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3573 | goto again; | |
3574 | } | |
1bbc621e | 3575 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c79a1751 | 3576 | } else if (ret < 0) { |
2ff7e61e | 3577 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
1bbc621e CM |
3578 | } |
3579 | ||
3580 | btrfs_free_path(path); | |
3581 | return ret; | |
3582 | } | |
3583 | ||
5742d15f | 3584 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans) |
1bbc621e | 3585 | { |
5742d15f | 3586 | struct btrfs_fs_info *fs_info = trans->fs_info; |
1bbc621e CM |
3587 | struct btrfs_block_group_cache *cache; |
3588 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3589 | int ret = 0; | |
3590 | int should_put; | |
3591 | struct btrfs_path *path; | |
3592 | struct list_head *io = &cur_trans->io_bgs; | |
3593 | int num_started = 0; | |
9078a3e1 CM |
3594 | |
3595 | path = btrfs_alloc_path(); | |
3596 | if (!path) | |
3597 | return -ENOMEM; | |
3598 | ||
ce93ec54 | 3599 | /* |
e44081ef FM |
3600 | * Even though we are in the critical section of the transaction commit, |
3601 | * we can still have concurrent tasks adding elements to this | |
3602 | * transaction's list of dirty block groups. These tasks correspond to | |
3603 | * endio free space workers started when writeback finishes for a | |
3604 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3605 | * allocate new block groups as a result of COWing nodes of the root | |
3606 | * tree when updating the free space inode. The writeback for the space | |
3607 | * caches is triggered by an earlier call to | |
3608 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3609 | * loop. | |
3610 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3611 | * delayed refs to make sure we have the best chance at doing this all |
3612 | * in one shot. | |
3613 | */ | |
e44081ef | 3614 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3615 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3616 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3617 | struct btrfs_block_group_cache, | |
3618 | dirty_list); | |
c9dc4c65 CM |
3619 | |
3620 | /* | |
3621 | * this can happen if cache_save_setup re-dirties a block | |
3622 | * group that is already under IO. Just wait for it to | |
3623 | * finish and then do it all again | |
3624 | */ | |
3625 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3626 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3627 | list_del_init(&cache->io_list); |
afdb5718 | 3628 | btrfs_wait_cache_io(trans, cache, path); |
c9dc4c65 | 3629 | btrfs_put_block_group(cache); |
e44081ef | 3630 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3631 | } |
3632 | ||
1bbc621e CM |
3633 | /* |
3634 | * don't remove from the dirty list until after we've waited | |
3635 | * on any pending IO | |
3636 | */ | |
ce93ec54 | 3637 | list_del_init(&cache->dirty_list); |
e44081ef | 3638 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3639 | should_put = 1; |
3640 | ||
1bbc621e | 3641 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3642 | |
ce93ec54 | 3643 | if (!ret) |
c79a70b1 | 3644 | ret = btrfs_run_delayed_refs(trans, |
2ff7e61e | 3645 | (unsigned long) -1); |
c9dc4c65 CM |
3646 | |
3647 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3648 | cache->io_ctl.inode = NULL; | |
fe041534 | 3649 | ret = btrfs_write_out_cache(trans, cache, path); |
c9dc4c65 CM |
3650 | if (ret == 0 && cache->io_ctl.inode) { |
3651 | num_started++; | |
3652 | should_put = 0; | |
1bbc621e | 3653 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3654 | } else { |
3655 | /* | |
3656 | * if we failed to write the cache, the | |
3657 | * generation will be bad and life goes on | |
3658 | */ | |
3659 | ret = 0; | |
3660 | } | |
3661 | } | |
ff1f8250 | 3662 | if (!ret) { |
39db232d | 3663 | ret = write_one_cache_group(trans, path, cache); |
2bc0bb5f FM |
3664 | /* |
3665 | * One of the free space endio workers might have | |
3666 | * created a new block group while updating a free space | |
3667 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3668 | * and hasn't released its transaction handle yet, in | |
3669 | * which case the new block group is still attached to | |
3670 | * its transaction handle and its creation has not | |
3671 | * finished yet (no block group item in the extent tree | |
3672 | * yet, etc). If this is the case, wait for all free | |
3673 | * space endio workers to finish and retry. This is a | |
3674 | * a very rare case so no need for a more efficient and | |
3675 | * complex approach. | |
3676 | */ | |
3677 | if (ret == -ENOENT) { | |
3678 | wait_event(cur_trans->writer_wait, | |
3679 | atomic_read(&cur_trans->num_writers) == 1); | |
39db232d | 3680 | ret = write_one_cache_group(trans, path, cache); |
2bc0bb5f | 3681 | } |
ff1f8250 | 3682 | if (ret) |
66642832 | 3683 | btrfs_abort_transaction(trans, ret); |
ff1f8250 | 3684 | } |
c9dc4c65 CM |
3685 | |
3686 | /* if its not on the io list, we need to put the block group */ | |
3687 | if (should_put) | |
3688 | btrfs_put_block_group(cache); | |
ba2c4d4e | 3689 | btrfs_delayed_refs_rsv_release(fs_info, 1); |
e44081ef | 3690 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3691 | } |
e44081ef | 3692 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3693 | |
45ae2c18 NB |
3694 | /* |
3695 | * Refer to the definition of io_bgs member for details why it's safe | |
3696 | * to use it without any locking | |
3697 | */ | |
1bbc621e CM |
3698 | while (!list_empty(io)) { |
3699 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3700 | io_list); |
3701 | list_del_init(&cache->io_list); | |
afdb5718 | 3702 | btrfs_wait_cache_io(trans, cache, path); |
0cb59c99 JB |
3703 | btrfs_put_block_group(cache); |
3704 | } | |
3705 | ||
9078a3e1 | 3706 | btrfs_free_path(path); |
ce93ec54 | 3707 | return ret; |
9078a3e1 CM |
3708 | } |
3709 | ||
2ff7e61e | 3710 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 YZ |
3711 | { |
3712 | struct btrfs_block_group_cache *block_group; | |
3713 | int readonly = 0; | |
3714 | ||
0b246afa | 3715 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
3716 | if (!block_group || block_group->ro) |
3717 | readonly = 1; | |
3718 | if (block_group) | |
fa9c0d79 | 3719 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3720 | return readonly; |
3721 | } | |
3722 | ||
f78c436c FM |
3723 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
3724 | { | |
3725 | struct btrfs_block_group_cache *bg; | |
3726 | bool ret = true; | |
3727 | ||
3728 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3729 | if (!bg) | |
3730 | return false; | |
3731 | ||
3732 | spin_lock(&bg->lock); | |
3733 | if (bg->ro) | |
3734 | ret = false; | |
3735 | else | |
3736 | atomic_inc(&bg->nocow_writers); | |
3737 | spin_unlock(&bg->lock); | |
3738 | ||
3739 | /* no put on block group, done by btrfs_dec_nocow_writers */ | |
3740 | if (!ret) | |
3741 | btrfs_put_block_group(bg); | |
3742 | ||
3743 | return ret; | |
3744 | ||
3745 | } | |
3746 | ||
3747 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
3748 | { | |
3749 | struct btrfs_block_group_cache *bg; | |
3750 | ||
3751 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3752 | ASSERT(bg); | |
3753 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
4625956a | 3754 | wake_up_var(&bg->nocow_writers); |
f78c436c FM |
3755 | /* |
3756 | * Once for our lookup and once for the lookup done by a previous call | |
3757 | * to btrfs_inc_nocow_writers() | |
3758 | */ | |
3759 | btrfs_put_block_group(bg); | |
3760 | btrfs_put_block_group(bg); | |
3761 | } | |
3762 | ||
f78c436c FM |
3763 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) |
3764 | { | |
4625956a | 3765 | wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers)); |
f78c436c FM |
3766 | } |
3767 | ||
8790d502 CM |
3768 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3769 | { | |
899c81ea ID |
3770 | u64 extra_flags = chunk_to_extended(flags) & |
3771 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 3772 | |
de98ced9 | 3773 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
3774 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3775 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3776 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3777 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3778 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3779 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 3780 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 3781 | } |
593060d7 | 3782 | |
fc67c450 ID |
3783 | /* |
3784 | * returns target flags in extended format or 0 if restripe for this | |
3785 | * chunk_type is not in progress | |
c6664b42 | 3786 | * |
dccdb07b | 3787 | * should be called with balance_lock held |
fc67c450 ID |
3788 | */ |
3789 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
3790 | { | |
3791 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3792 | u64 target = 0; | |
3793 | ||
fc67c450 ID |
3794 | if (!bctl) |
3795 | return 0; | |
3796 | ||
3797 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
3798 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3799 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3800 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
3801 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3802 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3803 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
3804 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3805 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3806 | } | |
3807 | ||
3808 | return target; | |
3809 | } | |
3810 | ||
a46d11a8 ID |
3811 | /* |
3812 | * @flags: available profiles in extended format (see ctree.h) | |
3813 | * | |
e4d8ec0f ID |
3814 | * Returns reduced profile in chunk format. If profile changing is in |
3815 | * progress (either running or paused) picks the target profile (if it's | |
3816 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 3817 | */ |
2ff7e61e | 3818 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c | 3819 | { |
0b246afa | 3820 | u64 num_devices = fs_info->fs_devices->rw_devices; |
fc67c450 | 3821 | u64 target; |
9c170b26 ZL |
3822 | u64 raid_type; |
3823 | u64 allowed = 0; | |
a061fc8d | 3824 | |
fc67c450 ID |
3825 | /* |
3826 | * see if restripe for this chunk_type is in progress, if so | |
3827 | * try to reduce to the target profile | |
3828 | */ | |
0b246afa JM |
3829 | spin_lock(&fs_info->balance_lock); |
3830 | target = get_restripe_target(fs_info, flags); | |
fc67c450 ID |
3831 | if (target) { |
3832 | /* pick target profile only if it's already available */ | |
3833 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
0b246afa | 3834 | spin_unlock(&fs_info->balance_lock); |
fc67c450 | 3835 | return extended_to_chunk(target); |
e4d8ec0f ID |
3836 | } |
3837 | } | |
0b246afa | 3838 | spin_unlock(&fs_info->balance_lock); |
e4d8ec0f | 3839 | |
53b381b3 | 3840 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
3841 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
3842 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
41a6e891 | 3843 | allowed |= btrfs_raid_array[raid_type].bg_flag; |
9c170b26 ZL |
3844 | } |
3845 | allowed &= flags; | |
3846 | ||
3847 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
3848 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
3849 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
3850 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
3851 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
3852 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
3853 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
3854 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
3855 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
3856 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
3857 | ||
3858 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
3859 | ||
3860 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
3861 | } |
3862 | ||
2ff7e61e | 3863 | static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
6a63209f | 3864 | { |
de98ced9 | 3865 | unsigned seq; |
f8213bdc | 3866 | u64 flags; |
de98ced9 MX |
3867 | |
3868 | do { | |
f8213bdc | 3869 | flags = orig_flags; |
0b246afa | 3870 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 MX |
3871 | |
3872 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
0b246afa | 3873 | flags |= fs_info->avail_data_alloc_bits; |
de98ced9 | 3874 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
0b246afa | 3875 | flags |= fs_info->avail_system_alloc_bits; |
de98ced9 | 3876 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
0b246afa JM |
3877 | flags |= fs_info->avail_metadata_alloc_bits; |
3878 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
6fef8df1 | 3879 | |
2ff7e61e | 3880 | return btrfs_reduce_alloc_profile(fs_info, flags); |
6a63209f JB |
3881 | } |
3882 | ||
1b86826d | 3883 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 3884 | { |
0b246afa | 3885 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 3886 | u64 flags; |
53b381b3 | 3887 | u64 ret; |
9ed74f2d | 3888 | |
b742bb82 YZ |
3889 | if (data) |
3890 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 3891 | else if (root == fs_info->chunk_root) |
b742bb82 | 3892 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 3893 | else |
b742bb82 | 3894 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 3895 | |
2ff7e61e | 3896 | ret = get_alloc_profile(fs_info, flags); |
53b381b3 | 3897 | return ret; |
6a63209f | 3898 | } |
9ed74f2d | 3899 | |
1b86826d JM |
3900 | u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info) |
3901 | { | |
3902 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA); | |
3903 | } | |
3904 | ||
3905 | u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info) | |
3906 | { | |
3907 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
3908 | } | |
3909 | ||
3910 | u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info) | |
3911 | { | |
3912 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
3913 | } | |
3914 | ||
04f4f916 | 3915 | int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes) |
6a63209f | 3916 | { |
04f4f916 | 3917 | struct btrfs_root *root = inode->root; |
b4d7c3c9 | 3918 | struct btrfs_fs_info *fs_info = root->fs_info; |
1174cade | 3919 | struct btrfs_space_info *data_sinfo = fs_info->data_sinfo; |
ab6e2410 | 3920 | u64 used; |
94b947b2 | 3921 | int ret = 0; |
c99f1b0c ZL |
3922 | int need_commit = 2; |
3923 | int have_pinned_space; | |
6a63209f | 3924 | |
6a63209f | 3925 | /* make sure bytes are sectorsize aligned */ |
0b246afa | 3926 | bytes = ALIGN(bytes, fs_info->sectorsize); |
6a63209f | 3927 | |
9dced186 | 3928 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 3929 | need_commit = 0; |
9dced186 | 3930 | ASSERT(current->journal_info); |
0af3d00b JB |
3931 | } |
3932 | ||
6a63209f JB |
3933 | again: |
3934 | /* make sure we have enough space to handle the data first */ | |
3935 | spin_lock(&data_sinfo->lock); | |
4136135b | 3936 | used = btrfs_space_info_used(data_sinfo, true); |
ab6e2410 JB |
3937 | |
3938 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 3939 | struct btrfs_trans_handle *trans; |
9ed74f2d | 3940 | |
6a63209f JB |
3941 | /* |
3942 | * if we don't have enough free bytes in this space then we need | |
3943 | * to alloc a new chunk. | |
3944 | */ | |
b9fd47cd | 3945 | if (!data_sinfo->full) { |
6a63209f | 3946 | u64 alloc_target; |
9ed74f2d | 3947 | |
0e4f8f88 | 3948 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 3949 | spin_unlock(&data_sinfo->lock); |
1174cade | 3950 | |
1b86826d | 3951 | alloc_target = btrfs_data_alloc_profile(fs_info); |
9dced186 MX |
3952 | /* |
3953 | * It is ugly that we don't call nolock join | |
3954 | * transaction for the free space inode case here. | |
3955 | * But it is safe because we only do the data space | |
3956 | * reservation for the free space cache in the | |
3957 | * transaction context, the common join transaction | |
3958 | * just increase the counter of the current transaction | |
3959 | * handler, doesn't try to acquire the trans_lock of | |
3960 | * the fs. | |
3961 | */ | |
7a7eaa40 | 3962 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
3963 | if (IS_ERR(trans)) |
3964 | return PTR_ERR(trans); | |
9ed74f2d | 3965 | |
fc471cb0 JB |
3966 | ret = btrfs_chunk_alloc(trans, alloc_target, |
3967 | CHUNK_ALLOC_NO_FORCE); | |
3a45bb20 | 3968 | btrfs_end_transaction(trans); |
d52a5b5f MX |
3969 | if (ret < 0) { |
3970 | if (ret != -ENOSPC) | |
3971 | return ret; | |
c99f1b0c ZL |
3972 | else { |
3973 | have_pinned_space = 1; | |
d52a5b5f | 3974 | goto commit_trans; |
c99f1b0c | 3975 | } |
d52a5b5f | 3976 | } |
9ed74f2d | 3977 | |
6a63209f JB |
3978 | goto again; |
3979 | } | |
f2bb8f5c JB |
3980 | |
3981 | /* | |
b150a4f1 | 3982 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
3983 | * allocation, and no removed chunk in current transaction, |
3984 | * don't bother committing the transaction. | |
f2bb8f5c | 3985 | */ |
dec59fa3 | 3986 | have_pinned_space = __percpu_counter_compare( |
c99f1b0c | 3987 | &data_sinfo->total_bytes_pinned, |
dec59fa3 EL |
3988 | used + bytes - data_sinfo->total_bytes, |
3989 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
6a63209f | 3990 | spin_unlock(&data_sinfo->lock); |
6a63209f | 3991 | |
4e06bdd6 | 3992 | /* commit the current transaction and try again */ |
d52a5b5f | 3993 | commit_trans: |
92e2f7e3 | 3994 | if (need_commit) { |
c99f1b0c | 3995 | need_commit--; |
b150a4f1 | 3996 | |
e1746e83 | 3997 | if (need_commit > 0) { |
82b3e53b | 3998 | btrfs_start_delalloc_roots(fs_info, -1); |
6374e57a | 3999 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, |
0b246afa | 4000 | (u64)-1); |
e1746e83 | 4001 | } |
9a4e7276 | 4002 | |
7a7eaa40 | 4003 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4004 | if (IS_ERR(trans)) |
4005 | return PTR_ERR(trans); | |
c99f1b0c | 4006 | if (have_pinned_space >= 0 || |
3204d33c JB |
4007 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4008 | &trans->transaction->flags) || | |
c99f1b0c | 4009 | need_commit > 0) { |
3a45bb20 | 4010 | ret = btrfs_commit_transaction(trans); |
94b947b2 ZL |
4011 | if (ret) |
4012 | return ret; | |
d7c15171 | 4013 | /* |
c2d6cb16 FM |
4014 | * The cleaner kthread might still be doing iput |
4015 | * operations. Wait for it to finish so that | |
034f784d JB |
4016 | * more space is released. We don't need to |
4017 | * explicitly run the delayed iputs here because | |
4018 | * the commit_transaction would have woken up | |
4019 | * the cleaner. | |
d7c15171 | 4020 | */ |
034f784d JB |
4021 | ret = btrfs_wait_on_delayed_iputs(fs_info); |
4022 | if (ret) | |
4023 | return ret; | |
94b947b2 ZL |
4024 | goto again; |
4025 | } else { | |
3a45bb20 | 4026 | btrfs_end_transaction(trans); |
94b947b2 | 4027 | } |
4e06bdd6 | 4028 | } |
9ed74f2d | 4029 | |
0b246afa | 4030 | trace_btrfs_space_reservation(fs_info, |
cab45e22 JM |
4031 | "space_info:enospc", |
4032 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4033 | return -ENOSPC; |
4034 | } | |
bb96c4e5 | 4035 | btrfs_space_info_update_bytes_may_use(fs_info, data_sinfo, bytes); |
0b246afa | 4036 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4037 | data_sinfo->flags, bytes, 1); |
6a63209f | 4038 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4039 | |
4559b0a7 | 4040 | return 0; |
9ed74f2d | 4041 | } |
6a63209f | 4042 | |
364ecf36 QW |
4043 | int btrfs_check_data_free_space(struct inode *inode, |
4044 | struct extent_changeset **reserved, u64 start, u64 len) | |
4ceff079 | 4045 | { |
0b246afa | 4046 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4047 | int ret; |
4048 | ||
4049 | /* align the range */ | |
0b246afa JM |
4050 | len = round_up(start + len, fs_info->sectorsize) - |
4051 | round_down(start, fs_info->sectorsize); | |
4052 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4053 | |
04f4f916 | 4054 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len); |
4ceff079 QW |
4055 | if (ret < 0) |
4056 | return ret; | |
4057 | ||
1e5ec2e7 | 4058 | /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */ |
364ecf36 | 4059 | ret = btrfs_qgroup_reserve_data(inode, reserved, start, len); |
7bc329c1 | 4060 | if (ret < 0) |
1e5ec2e7 | 4061 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
364ecf36 QW |
4062 | else |
4063 | ret = 0; | |
4ceff079 QW |
4064 | return ret; |
4065 | } | |
4066 | ||
4ceff079 QW |
4067 | /* |
4068 | * Called if we need to clear a data reservation for this inode | |
4069 | * Normally in a error case. | |
4070 | * | |
51773bec QW |
4071 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4072 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4073 | * Like clear_bit_hook(). | |
4ceff079 | 4074 | */ |
51773bec QW |
4075 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4076 | u64 len) | |
4ceff079 | 4077 | { |
0b246afa | 4078 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4079 | struct btrfs_space_info *data_sinfo; |
4080 | ||
4081 | /* Make sure the range is aligned to sectorsize */ | |
0b246afa JM |
4082 | len = round_up(start + len, fs_info->sectorsize) - |
4083 | round_down(start, fs_info->sectorsize); | |
4084 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4085 | |
0b246afa | 4086 | data_sinfo = fs_info->data_sinfo; |
4ceff079 | 4087 | spin_lock(&data_sinfo->lock); |
bb96c4e5 | 4088 | btrfs_space_info_update_bytes_may_use(fs_info, data_sinfo, -len); |
0b246afa | 4089 | trace_btrfs_space_reservation(fs_info, "space_info", |
4ceff079 QW |
4090 | data_sinfo->flags, len, 0); |
4091 | spin_unlock(&data_sinfo->lock); | |
4092 | } | |
4093 | ||
51773bec QW |
4094 | /* |
4095 | * Called if we need to clear a data reservation for this inode | |
4096 | * Normally in a error case. | |
4097 | * | |
01327610 | 4098 | * This one will handle the per-inode data rsv map for accurate reserved |
51773bec QW |
4099 | * space framework. |
4100 | */ | |
bc42bda2 QW |
4101 | void btrfs_free_reserved_data_space(struct inode *inode, |
4102 | struct extent_changeset *reserved, u64 start, u64 len) | |
51773bec | 4103 | { |
0c476a5d JM |
4104 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4105 | ||
4106 | /* Make sure the range is aligned to sectorsize */ | |
da17066c JM |
4107 | len = round_up(start + len, root->fs_info->sectorsize) - |
4108 | round_down(start, root->fs_info->sectorsize); | |
4109 | start = round_down(start, root->fs_info->sectorsize); | |
0c476a5d | 4110 | |
51773bec | 4111 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
bc42bda2 | 4112 | btrfs_qgroup_free_data(inode, reserved, start, len); |
51773bec QW |
4113 | } |
4114 | ||
97e728d4 | 4115 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4116 | { |
97e728d4 JB |
4117 | struct list_head *head = &info->space_info; |
4118 | struct btrfs_space_info *found; | |
e3ccfa98 | 4119 | |
97e728d4 JB |
4120 | rcu_read_lock(); |
4121 | list_for_each_entry_rcu(found, head, list) { | |
4122 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4123 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4124 | } |
97e728d4 | 4125 | rcu_read_unlock(); |
e3ccfa98 JB |
4126 | } |
4127 | ||
2ff7e61e | 4128 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, |
698d0082 | 4129 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4130 | { |
8d8aafee | 4131 | u64 bytes_used = btrfs_space_info_used(sinfo, false); |
e5bc2458 | 4132 | u64 thresh; |
e3ccfa98 | 4133 | |
0e4f8f88 CM |
4134 | if (force == CHUNK_ALLOC_FORCE) |
4135 | return 1; | |
4136 | ||
4137 | /* | |
4138 | * in limited mode, we want to have some free space up to | |
4139 | * about 1% of the FS size. | |
4140 | */ | |
4141 | if (force == CHUNK_ALLOC_LIMITED) { | |
0b246afa | 4142 | thresh = btrfs_super_total_bytes(fs_info->super_copy); |
ee22184b | 4143 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
0e4f8f88 | 4144 | |
8d8aafee | 4145 | if (sinfo->total_bytes - bytes_used < thresh) |
0e4f8f88 CM |
4146 | return 1; |
4147 | } | |
0e4f8f88 | 4148 | |
8d8aafee | 4149 | if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) |
14ed0ca6 | 4150 | return 0; |
424499db | 4151 | return 1; |
32c00aff JB |
4152 | } |
4153 | ||
2ff7e61e | 4154 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4155 | { |
4156 | u64 num_dev; | |
4157 | ||
9fa02ac7 DS |
4158 | num_dev = btrfs_raid_array[btrfs_bg_flags_to_raid_index(type)].devs_max; |
4159 | if (!num_dev) | |
0b246afa | 4160 | num_dev = fs_info->fs_devices->rw_devices; |
15d1ff81 | 4161 | |
39c2d7fa | 4162 | return num_dev; |
15d1ff81 LB |
4163 | } |
4164 | ||
39c2d7fa FM |
4165 | /* |
4166 | * If @is_allocation is true, reserve space in the system space info necessary | |
4167 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4168 | * removing a chunk. | |
4169 | */ | |
451a2c13 | 4170 | void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) |
15d1ff81 | 4171 | { |
451a2c13 | 4172 | struct btrfs_fs_info *fs_info = trans->fs_info; |
15d1ff81 LB |
4173 | struct btrfs_space_info *info; |
4174 | u64 left; | |
4175 | u64 thresh; | |
4fbcdf66 | 4176 | int ret = 0; |
39c2d7fa | 4177 | u64 num_devs; |
4fbcdf66 FM |
4178 | |
4179 | /* | |
4180 | * Needed because we can end up allocating a system chunk and for an | |
4181 | * atomic and race free space reservation in the chunk block reserve. | |
4182 | */ | |
a32bf9a3 | 4183 | lockdep_assert_held(&fs_info->chunk_mutex); |
15d1ff81 | 4184 | |
280c2908 | 4185 | info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
15d1ff81 | 4186 | spin_lock(&info->lock); |
4136135b | 4187 | left = info->total_bytes - btrfs_space_info_used(info, true); |
15d1ff81 LB |
4188 | spin_unlock(&info->lock); |
4189 | ||
2ff7e61e | 4190 | num_devs = get_profile_num_devs(fs_info, type); |
39c2d7fa FM |
4191 | |
4192 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
0b246afa JM |
4193 | thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) + |
4194 | btrfs_calc_trans_metadata_size(fs_info, 1); | |
39c2d7fa | 4195 | |
0b246afa JM |
4196 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
4197 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
4198 | left, thresh, type); | |
4199 | dump_space_info(fs_info, info, 0, 0); | |
15d1ff81 LB |
4200 | } |
4201 | ||
4202 | if (left < thresh) { | |
1b86826d | 4203 | u64 flags = btrfs_system_alloc_profile(fs_info); |
15d1ff81 | 4204 | |
4fbcdf66 FM |
4205 | /* |
4206 | * Ignore failure to create system chunk. We might end up not | |
4207 | * needing it, as we might not need to COW all nodes/leafs from | |
4208 | * the paths we visit in the chunk tree (they were already COWed | |
4209 | * or created in the current transaction for example). | |
4210 | */ | |
c216b203 | 4211 | ret = btrfs_alloc_chunk(trans, flags); |
4fbcdf66 FM |
4212 | } |
4213 | ||
4214 | if (!ret) { | |
0b246afa JM |
4215 | ret = btrfs_block_rsv_add(fs_info->chunk_root, |
4216 | &fs_info->chunk_block_rsv, | |
4fbcdf66 FM |
4217 | thresh, BTRFS_RESERVE_NO_FLUSH); |
4218 | if (!ret) | |
4219 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4220 | } |
4221 | } | |
4222 | ||
28b737f6 LB |
4223 | /* |
4224 | * If force is CHUNK_ALLOC_FORCE: | |
4225 | * - return 1 if it successfully allocates a chunk, | |
4226 | * - return errors including -ENOSPC otherwise. | |
4227 | * If force is NOT CHUNK_ALLOC_FORCE: | |
4228 | * - return 0 if it doesn't need to allocate a new chunk, | |
4229 | * - return 1 if it successfully allocates a chunk, | |
4230 | * - return errors including -ENOSPC otherwise. | |
4231 | */ | |
fc471cb0 JB |
4232 | int btrfs_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, |
4233 | enum btrfs_chunk_alloc_enum force) | |
9ed74f2d | 4234 | { |
01458828 | 4235 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6324fbf3 | 4236 | struct btrfs_space_info *space_info; |
2556fbb0 NB |
4237 | bool wait_for_alloc = false; |
4238 | bool should_alloc = false; | |
9ed74f2d | 4239 | int ret = 0; |
9ed74f2d | 4240 | |
c6b305a8 JB |
4241 | /* Don't re-enter if we're already allocating a chunk */ |
4242 | if (trans->allocating_chunk) | |
4243 | return -ENOSPC; | |
4244 | ||
280c2908 | 4245 | space_info = btrfs_find_space_info(fs_info, flags); |
dc2d3005 | 4246 | ASSERT(space_info); |
9ed74f2d | 4247 | |
2556fbb0 NB |
4248 | do { |
4249 | spin_lock(&space_info->lock); | |
4250 | if (force < space_info->force_alloc) | |
4251 | force = space_info->force_alloc; | |
4252 | should_alloc = should_alloc_chunk(fs_info, space_info, force); | |
4253 | if (space_info->full) { | |
4254 | /* No more free physical space */ | |
4255 | if (should_alloc) | |
4256 | ret = -ENOSPC; | |
4257 | else | |
4258 | ret = 0; | |
4259 | spin_unlock(&space_info->lock); | |
4260 | return ret; | |
4261 | } else if (!should_alloc) { | |
4262 | spin_unlock(&space_info->lock); | |
4263 | return 0; | |
4264 | } else if (space_info->chunk_alloc) { | |
4265 | /* | |
4266 | * Someone is already allocating, so we need to block | |
4267 | * until this someone is finished and then loop to | |
4268 | * recheck if we should continue with our allocation | |
4269 | * attempt. | |
4270 | */ | |
4271 | wait_for_alloc = true; | |
4272 | spin_unlock(&space_info->lock); | |
4273 | mutex_lock(&fs_info->chunk_mutex); | |
4274 | mutex_unlock(&fs_info->chunk_mutex); | |
4275 | } else { | |
4276 | /* Proceed with allocation */ | |
4277 | space_info->chunk_alloc = 1; | |
4278 | wait_for_alloc = false; | |
4279 | spin_unlock(&space_info->lock); | |
4280 | } | |
6d74119f | 4281 | |
1e1c50a9 | 4282 | cond_resched(); |
2556fbb0 | 4283 | } while (wait_for_alloc); |
6d74119f | 4284 | |
2556fbb0 | 4285 | mutex_lock(&fs_info->chunk_mutex); |
c6b305a8 JB |
4286 | trans->allocating_chunk = true; |
4287 | ||
67377734 JB |
4288 | /* |
4289 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4290 | * allocating mixed chunks instead of individual chunks. | |
4291 | */ | |
4292 | if (btrfs_mixed_space_info(space_info)) | |
4293 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4294 | ||
97e728d4 JB |
4295 | /* |
4296 | * if we're doing a data chunk, go ahead and make sure that | |
4297 | * we keep a reasonable number of metadata chunks allocated in the | |
4298 | * FS as well. | |
4299 | */ | |
9ed74f2d | 4300 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4301 | fs_info->data_chunk_allocations++; |
4302 | if (!(fs_info->data_chunk_allocations % | |
4303 | fs_info->metadata_ratio)) | |
4304 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4305 | } |
4306 | ||
15d1ff81 LB |
4307 | /* |
4308 | * Check if we have enough space in SYSTEM chunk because we may need | |
4309 | * to update devices. | |
4310 | */ | |
451a2c13 | 4311 | check_system_chunk(trans, flags); |
15d1ff81 | 4312 | |
c216b203 | 4313 | ret = btrfs_alloc_chunk(trans, flags); |
c6b305a8 | 4314 | trans->allocating_chunk = false; |
92b8e897 | 4315 | |
9ed74f2d | 4316 | spin_lock(&space_info->lock); |
57f1642e NB |
4317 | if (ret < 0) { |
4318 | if (ret == -ENOSPC) | |
4319 | space_info->full = 1; | |
4320 | else | |
4321 | goto out; | |
4322 | } else { | |
424499db | 4323 | ret = 1; |
21a94f7a | 4324 | space_info->max_extent_size = 0; |
57f1642e | 4325 | } |
6d74119f | 4326 | |
0e4f8f88 | 4327 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4328 | out: |
6d74119f | 4329 | space_info->chunk_alloc = 0; |
9ed74f2d | 4330 | spin_unlock(&space_info->lock); |
a25c75d5 | 4331 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4332 | /* |
4333 | * When we allocate a new chunk we reserve space in the chunk block | |
4334 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4335 | * add new nodes/leafs to it if we end up needing to do it when | |
4336 | * inserting the chunk item and updating device items as part of the | |
4337 | * second phase of chunk allocation, performed by | |
4338 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4339 | * large number of new block groups to create in our transaction | |
4340 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4341 | * in extreme cases - like having a single transaction create many new | |
4342 | * block groups when starting to write out the free space caches of all | |
4343 | * the block groups that were made dirty during the lifetime of the | |
4344 | * transaction. | |
4345 | */ | |
5ce55557 | 4346 | if (trans->chunk_bytes_reserved >= (u64)SZ_2M) |
6c686b35 | 4347 | btrfs_create_pending_block_groups(trans); |
5ce55557 | 4348 | |
0f9dd46c | 4349 | return ret; |
6324fbf3 | 4350 | } |
9ed74f2d | 4351 | |
2ff7e61e | 4352 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info, |
6c255e67 | 4353 | unsigned long nr_pages, int nr_items) |
da633a42 | 4354 | { |
0b246afa | 4355 | struct super_block *sb = fs_info->sb; |
da633a42 | 4356 | |
925a6efb JB |
4357 | if (down_read_trylock(&sb->s_umount)) { |
4358 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4359 | up_read(&sb->s_umount); | |
4360 | } else { | |
da633a42 MX |
4361 | /* |
4362 | * We needn't worry the filesystem going from r/w to r/o though | |
4363 | * we don't acquire ->s_umount mutex, because the filesystem | |
4364 | * should guarantee the delalloc inodes list be empty after | |
4365 | * the filesystem is readonly(all dirty pages are written to | |
4366 | * the disk). | |
4367 | */ | |
82b3e53b | 4368 | btrfs_start_delalloc_roots(fs_info, nr_items); |
98ad69cf | 4369 | if (!current->journal_info) |
0b246afa | 4370 | btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1); |
da633a42 MX |
4371 | } |
4372 | } | |
4373 | ||
6374e57a | 4374 | static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info, |
2ff7e61e | 4375 | u64 to_reclaim) |
18cd8ea6 MX |
4376 | { |
4377 | u64 bytes; | |
6374e57a | 4378 | u64 nr; |
18cd8ea6 | 4379 | |
2ff7e61e | 4380 | bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
6374e57a | 4381 | nr = div64_u64(to_reclaim, bytes); |
18cd8ea6 MX |
4382 | if (!nr) |
4383 | nr = 1; | |
4384 | return nr; | |
4385 | } | |
4386 | ||
ee22184b | 4387 | #define EXTENT_SIZE_PER_ITEM SZ_256K |
c61a16a7 | 4388 | |
9ed74f2d | 4389 | /* |
5da9d01b | 4390 | * shrink metadata reservation for delalloc |
9ed74f2d | 4391 | */ |
c1c4919b JM |
4392 | static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim, |
4393 | u64 orig, bool wait_ordered) | |
5da9d01b | 4394 | { |
0019f10d | 4395 | struct btrfs_space_info *space_info; |
663350ac | 4396 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4397 | u64 delalloc_bytes; |
4297ff84 | 4398 | u64 dio_bytes; |
420829d8 | 4399 | u64 async_pages; |
6374e57a | 4400 | u64 items; |
b1953bce | 4401 | long time_left; |
d3ee29e3 MX |
4402 | unsigned long nr_pages; |
4403 | int loops; | |
5da9d01b | 4404 | |
c61a16a7 | 4405 | /* Calc the number of the pages we need flush for space reservation */ |
2ff7e61e | 4406 | items = calc_reclaim_items_nr(fs_info, to_reclaim); |
6374e57a | 4407 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; |
c61a16a7 | 4408 | |
663350ac | 4409 | trans = (struct btrfs_trans_handle *)current->journal_info; |
280c2908 | 4410 | space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
bf9022e0 | 4411 | |
963d678b | 4412 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4413 | &fs_info->delalloc_bytes); |
4297ff84 JB |
4414 | dio_bytes = percpu_counter_sum_positive(&fs_info->dio_bytes); |
4415 | if (delalloc_bytes == 0 && dio_bytes == 0) { | |
fdb5effd | 4416 | if (trans) |
f4c738c2 | 4417 | return; |
38c135af | 4418 | if (wait_ordered) |
0b246afa | 4419 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f4c738c2 | 4420 | return; |
fdb5effd JB |
4421 | } |
4422 | ||
4297ff84 JB |
4423 | /* |
4424 | * If we are doing more ordered than delalloc we need to just wait on | |
4425 | * ordered extents, otherwise we'll waste time trying to flush delalloc | |
4426 | * that likely won't give us the space back we need. | |
4427 | */ | |
4428 | if (dio_bytes > delalloc_bytes) | |
4429 | wait_ordered = true; | |
4430 | ||
d3ee29e3 | 4431 | loops = 0; |
4297ff84 | 4432 | while ((delalloc_bytes || dio_bytes) && loops < 3) { |
420829d8 NB |
4433 | nr_pages = min(delalloc_bytes, to_reclaim) >> PAGE_SHIFT; |
4434 | ||
4435 | /* | |
4436 | * Triggers inode writeback for up to nr_pages. This will invoke | |
4437 | * ->writepages callback and trigger delalloc filling | |
4438 | * (btrfs_run_delalloc_range()). | |
4439 | */ | |
2ff7e61e | 4440 | btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items); |
420829d8 | 4441 | |
dea31f52 | 4442 | /* |
420829d8 NB |
4443 | * We need to wait for the compressed pages to start before |
4444 | * we continue. | |
dea31f52 | 4445 | */ |
420829d8 NB |
4446 | async_pages = atomic_read(&fs_info->async_delalloc_pages); |
4447 | if (!async_pages) | |
9f3a074d MX |
4448 | goto skip_async; |
4449 | ||
420829d8 NB |
4450 | /* |
4451 | * Calculate how many compressed pages we want to be written | |
4452 | * before we continue. I.e if there are more async pages than we | |
4453 | * require wait_event will wait until nr_pages are written. | |
4454 | */ | |
4455 | if (async_pages <= nr_pages) | |
4456 | async_pages = 0; | |
9f3a074d | 4457 | else |
420829d8 | 4458 | async_pages -= nr_pages; |
dea31f52 | 4459 | |
0b246afa JM |
4460 | wait_event(fs_info->async_submit_wait, |
4461 | atomic_read(&fs_info->async_delalloc_pages) <= | |
420829d8 | 4462 | (int)async_pages); |
9f3a074d | 4463 | skip_async: |
0019f10d | 4464 | spin_lock(&space_info->lock); |
957780eb JB |
4465 | if (list_empty(&space_info->tickets) && |
4466 | list_empty(&space_info->priority_tickets)) { | |
4467 | spin_unlock(&space_info->lock); | |
4468 | break; | |
4469 | } | |
0019f10d | 4470 | spin_unlock(&space_info->lock); |
5da9d01b | 4471 | |
36e39c40 | 4472 | loops++; |
f104d044 | 4473 | if (wait_ordered && !trans) { |
0b246afa | 4474 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f104d044 | 4475 | } else { |
f4c738c2 | 4476 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4477 | if (time_left) |
4478 | break; | |
4479 | } | |
963d678b | 4480 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4481 | &fs_info->delalloc_bytes); |
4297ff84 | 4482 | dio_bytes = percpu_counter_sum_positive(&fs_info->dio_bytes); |
5da9d01b | 4483 | } |
5da9d01b YZ |
4484 | } |
4485 | ||
663350ac JB |
4486 | /** |
4487 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4488 | * @root - the root we're allocating for | |
4489 | * @bytes - the number of bytes we want to reserve | |
4490 | * @force - force the commit | |
8bb8ab2e | 4491 | * |
663350ac JB |
4492 | * This will check to make sure that committing the transaction will actually |
4493 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4494 | * will return -ENOSPC. | |
8bb8ab2e | 4495 | */ |
0c9ab349 | 4496 | static int may_commit_transaction(struct btrfs_fs_info *fs_info, |
996478ca | 4497 | struct btrfs_space_info *space_info) |
663350ac | 4498 | { |
996478ca | 4499 | struct reserve_ticket *ticket = NULL; |
0b246afa | 4500 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv; |
4c8edbc7 | 4501 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; |
663350ac | 4502 | struct btrfs_trans_handle *trans; |
4c8edbc7 JB |
4503 | u64 bytes_needed; |
4504 | u64 reclaim_bytes = 0; | |
663350ac JB |
4505 | |
4506 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4507 | if (trans) | |
4508 | return -EAGAIN; | |
4509 | ||
996478ca JB |
4510 | spin_lock(&space_info->lock); |
4511 | if (!list_empty(&space_info->priority_tickets)) | |
4512 | ticket = list_first_entry(&space_info->priority_tickets, | |
4513 | struct reserve_ticket, list); | |
4514 | else if (!list_empty(&space_info->tickets)) | |
4515 | ticket = list_first_entry(&space_info->tickets, | |
4516 | struct reserve_ticket, list); | |
4c8edbc7 | 4517 | bytes_needed = (ticket) ? ticket->bytes : 0; |
996478ca JB |
4518 | spin_unlock(&space_info->lock); |
4519 | ||
4c8edbc7 | 4520 | if (!bytes_needed) |
996478ca | 4521 | return 0; |
663350ac | 4522 | |
d89dbefb JB |
4523 | trans = btrfs_join_transaction(fs_info->extent_root); |
4524 | if (IS_ERR(trans)) | |
4525 | return PTR_ERR(trans); | |
4526 | ||
4527 | /* | |
4528 | * See if there is enough pinned space to make this reservation, or if | |
4529 | * we have block groups that are going to be freed, allowing us to | |
4530 | * possibly do a chunk allocation the next loop through. | |
4531 | */ | |
4532 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags) || | |
4533 | __percpu_counter_compare(&space_info->total_bytes_pinned, | |
4534 | bytes_needed, | |
4535 | BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0) | |
663350ac | 4536 | goto commit; |
663350ac JB |
4537 | |
4538 | /* | |
4539 | * See if there is some space in the delayed insertion reservation for | |
4540 | * this reservation. | |
4541 | */ | |
4542 | if (space_info != delayed_rsv->space_info) | |
d89dbefb | 4543 | goto enospc; |
663350ac JB |
4544 | |
4545 | spin_lock(&delayed_rsv->lock); | |
4c8edbc7 | 4546 | reclaim_bytes += delayed_rsv->reserved; |
057aac3e NB |
4547 | spin_unlock(&delayed_rsv->lock); |
4548 | ||
4c8edbc7 JB |
4549 | spin_lock(&delayed_refs_rsv->lock); |
4550 | reclaim_bytes += delayed_refs_rsv->reserved; | |
4551 | spin_unlock(&delayed_refs_rsv->lock); | |
4552 | if (reclaim_bytes >= bytes_needed) | |
4553 | goto commit; | |
4554 | bytes_needed -= reclaim_bytes; | |
4555 | ||
dec59fa3 | 4556 | if (__percpu_counter_compare(&space_info->total_bytes_pinned, |
4c8edbc7 | 4557 | bytes_needed, |
d89dbefb JB |
4558 | BTRFS_TOTAL_BYTES_PINNED_BATCH) < 0) |
4559 | goto enospc; | |
663350ac JB |
4560 | |
4561 | commit: | |
3a45bb20 | 4562 | return btrfs_commit_transaction(trans); |
d89dbefb JB |
4563 | enospc: |
4564 | btrfs_end_transaction(trans); | |
4565 | return -ENOSPC; | |
663350ac JB |
4566 | } |
4567 | ||
e38ae7a0 NB |
4568 | /* |
4569 | * Try to flush some data based on policy set by @state. This is only advisory | |
4570 | * and may fail for various reasons. The caller is supposed to examine the | |
4571 | * state of @space_info to detect the outcome. | |
4572 | */ | |
4573 | static void flush_space(struct btrfs_fs_info *fs_info, | |
96c3f433 | 4574 | struct btrfs_space_info *space_info, u64 num_bytes, |
7bdd6277 | 4575 | int state) |
96c3f433 | 4576 | { |
a9b3311e | 4577 | struct btrfs_root *root = fs_info->extent_root; |
96c3f433 JB |
4578 | struct btrfs_trans_handle *trans; |
4579 | int nr; | |
f4c738c2 | 4580 | int ret = 0; |
96c3f433 JB |
4581 | |
4582 | switch (state) { | |
96c3f433 JB |
4583 | case FLUSH_DELAYED_ITEMS_NR: |
4584 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 | 4585 | if (state == FLUSH_DELAYED_ITEMS_NR) |
2ff7e61e | 4586 | nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2; |
18cd8ea6 | 4587 | else |
96c3f433 | 4588 | nr = -1; |
18cd8ea6 | 4589 | |
96c3f433 JB |
4590 | trans = btrfs_join_transaction(root); |
4591 | if (IS_ERR(trans)) { | |
4592 | ret = PTR_ERR(trans); | |
4593 | break; | |
4594 | } | |
e5c304e6 | 4595 | ret = btrfs_run_delayed_items_nr(trans, nr); |
3a45bb20 | 4596 | btrfs_end_transaction(trans); |
96c3f433 | 4597 | break; |
67b0fd63 JB |
4598 | case FLUSH_DELALLOC: |
4599 | case FLUSH_DELALLOC_WAIT: | |
7bdd6277 | 4600 | shrink_delalloc(fs_info, num_bytes * 2, num_bytes, |
67b0fd63 JB |
4601 | state == FLUSH_DELALLOC_WAIT); |
4602 | break; | |
413df725 JB |
4603 | case FLUSH_DELAYED_REFS_NR: |
4604 | case FLUSH_DELAYED_REFS: | |
4605 | trans = btrfs_join_transaction(root); | |
4606 | if (IS_ERR(trans)) { | |
4607 | ret = PTR_ERR(trans); | |
4608 | break; | |
4609 | } | |
4610 | if (state == FLUSH_DELAYED_REFS_NR) | |
4611 | nr = calc_reclaim_items_nr(fs_info, num_bytes); | |
4612 | else | |
4613 | nr = 0; | |
4614 | btrfs_run_delayed_refs(trans, nr); | |
4615 | btrfs_end_transaction(trans); | |
4616 | break; | |
ea658bad | 4617 | case ALLOC_CHUNK: |
450114fc | 4618 | case ALLOC_CHUNK_FORCE: |
ea658bad JB |
4619 | trans = btrfs_join_transaction(root); |
4620 | if (IS_ERR(trans)) { | |
4621 | ret = PTR_ERR(trans); | |
4622 | break; | |
4623 | } | |
fc471cb0 JB |
4624 | ret = btrfs_chunk_alloc(trans, |
4625 | btrfs_metadata_alloc_profile(fs_info), | |
4626 | (state == ALLOC_CHUNK) ? CHUNK_ALLOC_NO_FORCE : | |
4627 | CHUNK_ALLOC_FORCE); | |
3a45bb20 | 4628 | btrfs_end_transaction(trans); |
eecba891 | 4629 | if (ret > 0 || ret == -ENOSPC) |
ea658bad JB |
4630 | ret = 0; |
4631 | break; | |
96c3f433 | 4632 | case COMMIT_TRANS: |
3ec9a4c8 JB |
4633 | /* |
4634 | * If we have pending delayed iputs then we could free up a | |
4635 | * bunch of pinned space, so make sure we run the iputs before | |
4636 | * we do our pinned bytes check below. | |
4637 | */ | |
3ec9a4c8 | 4638 | btrfs_run_delayed_iputs(fs_info); |
034f784d | 4639 | btrfs_wait_on_delayed_iputs(fs_info); |
3ec9a4c8 | 4640 | |
996478ca | 4641 | ret = may_commit_transaction(fs_info, space_info); |
96c3f433 JB |
4642 | break; |
4643 | default: | |
4644 | ret = -ENOSPC; | |
4645 | break; | |
4646 | } | |
4647 | ||
7bdd6277 NB |
4648 | trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state, |
4649 | ret); | |
e38ae7a0 | 4650 | return; |
96c3f433 | 4651 | } |
21c7e756 MX |
4652 | |
4653 | static inline u64 | |
c1c4919b JM |
4654 | btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info, |
4655 | struct btrfs_space_info *space_info, | |
4656 | bool system_chunk) | |
21c7e756 | 4657 | { |
957780eb | 4658 | struct reserve_ticket *ticket; |
21c7e756 MX |
4659 | u64 used; |
4660 | u64 expected; | |
957780eb | 4661 | u64 to_reclaim = 0; |
21c7e756 | 4662 | |
957780eb JB |
4663 | list_for_each_entry(ticket, &space_info->tickets, list) |
4664 | to_reclaim += ticket->bytes; | |
4665 | list_for_each_entry(ticket, &space_info->priority_tickets, list) | |
4666 | to_reclaim += ticket->bytes; | |
4667 | if (to_reclaim) | |
4668 | return to_reclaim; | |
21c7e756 | 4669 | |
e0af2484 | 4670 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); |
41783ef2 JB |
4671 | if (btrfs_can_overcommit(fs_info, space_info, to_reclaim, |
4672 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
e0af2484 WX |
4673 | return 0; |
4674 | ||
0eee8a49 NB |
4675 | used = btrfs_space_info_used(space_info, true); |
4676 | ||
41783ef2 JB |
4677 | if (btrfs_can_overcommit(fs_info, space_info, SZ_1M, |
4678 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
21c7e756 MX |
4679 | expected = div_factor_fine(space_info->total_bytes, 95); |
4680 | else | |
4681 | expected = div_factor_fine(space_info->total_bytes, 90); | |
4682 | ||
4683 | if (used > expected) | |
4684 | to_reclaim = used - expected; | |
4685 | else | |
4686 | to_reclaim = 0; | |
4687 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
4688 | space_info->bytes_reserved); | |
21c7e756 MX |
4689 | return to_reclaim; |
4690 | } | |
4691 | ||
c1c4919b JM |
4692 | static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info, |
4693 | struct btrfs_space_info *space_info, | |
4694 | u64 used, bool system_chunk) | |
21c7e756 | 4695 | { |
365c5313 JB |
4696 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
4697 | ||
4698 | /* If we're just plain full then async reclaim just slows us down. */ | |
baee8790 | 4699 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) |
365c5313 JB |
4700 | return 0; |
4701 | ||
c1c4919b JM |
4702 | if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
4703 | system_chunk)) | |
d38b349c JB |
4704 | return 0; |
4705 | ||
0b246afa JM |
4706 | return (used >= thresh && !btrfs_fs_closing(fs_info) && |
4707 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); | |
21c7e756 MX |
4708 | } |
4709 | ||
f91587e4 | 4710 | static bool wake_all_tickets(struct list_head *head) |
21c7e756 | 4711 | { |
957780eb | 4712 | struct reserve_ticket *ticket; |
25ce459c | 4713 | |
957780eb JB |
4714 | while (!list_empty(head)) { |
4715 | ticket = list_first_entry(head, struct reserve_ticket, list); | |
4716 | list_del_init(&ticket->list); | |
4717 | ticket->error = -ENOSPC; | |
4718 | wake_up(&ticket->wait); | |
f91587e4 JB |
4719 | if (ticket->bytes != ticket->orig_bytes) |
4720 | return true; | |
21c7e756 | 4721 | } |
f91587e4 | 4722 | return false; |
21c7e756 MX |
4723 | } |
4724 | ||
957780eb JB |
4725 | /* |
4726 | * This is for normal flushers, we can wait all goddamned day if we want to. We | |
4727 | * will loop and continuously try to flush as long as we are making progress. | |
4728 | * We count progress as clearing off tickets each time we have to loop. | |
4729 | */ | |
21c7e756 MX |
4730 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) |
4731 | { | |
4732 | struct btrfs_fs_info *fs_info; | |
4733 | struct btrfs_space_info *space_info; | |
4734 | u64 to_reclaim; | |
4735 | int flush_state; | |
957780eb | 4736 | int commit_cycles = 0; |
ce129655 | 4737 | u64 last_tickets_id; |
21c7e756 MX |
4738 | |
4739 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
280c2908 | 4740 | space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
21c7e756 | 4741 | |
957780eb | 4742 | spin_lock(&space_info->lock); |
c1c4919b JM |
4743 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
4744 | false); | |
957780eb JB |
4745 | if (!to_reclaim) { |
4746 | space_info->flush = 0; | |
4747 | spin_unlock(&space_info->lock); | |
21c7e756 | 4748 | return; |
957780eb | 4749 | } |
ce129655 | 4750 | last_tickets_id = space_info->tickets_id; |
957780eb | 4751 | spin_unlock(&space_info->lock); |
21c7e756 MX |
4752 | |
4753 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
957780eb | 4754 | do { |
e38ae7a0 | 4755 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
957780eb JB |
4756 | spin_lock(&space_info->lock); |
4757 | if (list_empty(&space_info->tickets)) { | |
4758 | space_info->flush = 0; | |
4759 | spin_unlock(&space_info->lock); | |
4760 | return; | |
4761 | } | |
c1c4919b JM |
4762 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, |
4763 | space_info, | |
4764 | false); | |
ce129655 | 4765 | if (last_tickets_id == space_info->tickets_id) { |
957780eb JB |
4766 | flush_state++; |
4767 | } else { | |
ce129655 | 4768 | last_tickets_id = space_info->tickets_id; |
957780eb JB |
4769 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
4770 | if (commit_cycles) | |
4771 | commit_cycles--; | |
4772 | } | |
4773 | ||
450114fc JB |
4774 | /* |
4775 | * We don't want to force a chunk allocation until we've tried | |
4776 | * pretty hard to reclaim space. Think of the case where we | |
4777 | * freed up a bunch of space and so have a lot of pinned space | |
4778 | * to reclaim. We would rather use that than possibly create a | |
4779 | * underutilized metadata chunk. So if this is our first run | |
4780 | * through the flushing state machine skip ALLOC_CHUNK_FORCE and | |
4781 | * commit the transaction. If nothing has changed the next go | |
4782 | * around then we can force a chunk allocation. | |
4783 | */ | |
4784 | if (flush_state == ALLOC_CHUNK_FORCE && !commit_cycles) | |
4785 | flush_state++; | |
4786 | ||
957780eb JB |
4787 | if (flush_state > COMMIT_TRANS) { |
4788 | commit_cycles++; | |
4789 | if (commit_cycles > 2) { | |
f91587e4 JB |
4790 | if (wake_all_tickets(&space_info->tickets)) { |
4791 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
4792 | commit_cycles--; | |
4793 | } else { | |
4794 | space_info->flush = 0; | |
4795 | } | |
957780eb JB |
4796 | } else { |
4797 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
4798 | } | |
4799 | } | |
4800 | spin_unlock(&space_info->lock); | |
4801 | } while (flush_state <= COMMIT_TRANS); | |
4802 | } | |
4803 | ||
4804 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
4805 | { | |
4806 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
4807 | } | |
4808 | ||
8a1bbe1d JB |
4809 | static const enum btrfs_flush_state priority_flush_states[] = { |
4810 | FLUSH_DELAYED_ITEMS_NR, | |
4811 | FLUSH_DELAYED_ITEMS, | |
4812 | ALLOC_CHUNK, | |
4813 | }; | |
4814 | ||
957780eb JB |
4815 | static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, |
4816 | struct btrfs_space_info *space_info, | |
4817 | struct reserve_ticket *ticket) | |
4818 | { | |
4819 | u64 to_reclaim; | |
8a1bbe1d | 4820 | int flush_state; |
957780eb JB |
4821 | |
4822 | spin_lock(&space_info->lock); | |
c1c4919b JM |
4823 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
4824 | false); | |
957780eb JB |
4825 | if (!to_reclaim) { |
4826 | spin_unlock(&space_info->lock); | |
4827 | return; | |
4828 | } | |
4829 | spin_unlock(&space_info->lock); | |
4830 | ||
8a1bbe1d | 4831 | flush_state = 0; |
21c7e756 | 4832 | do { |
8a1bbe1d JB |
4833 | flush_space(fs_info, space_info, to_reclaim, |
4834 | priority_flush_states[flush_state]); | |
21c7e756 | 4835 | flush_state++; |
957780eb JB |
4836 | spin_lock(&space_info->lock); |
4837 | if (ticket->bytes == 0) { | |
4838 | spin_unlock(&space_info->lock); | |
21c7e756 | 4839 | return; |
957780eb JB |
4840 | } |
4841 | spin_unlock(&space_info->lock); | |
8a1bbe1d | 4842 | } while (flush_state < ARRAY_SIZE(priority_flush_states)); |
21c7e756 MX |
4843 | } |
4844 | ||
957780eb JB |
4845 | static int wait_reserve_ticket(struct btrfs_fs_info *fs_info, |
4846 | struct btrfs_space_info *space_info, | |
f91587e4 | 4847 | struct reserve_ticket *ticket) |
957780eb | 4848 | |
21c7e756 | 4849 | { |
957780eb | 4850 | DEFINE_WAIT(wait); |
f91587e4 | 4851 | u64 reclaim_bytes = 0; |
957780eb JB |
4852 | int ret = 0; |
4853 | ||
4854 | spin_lock(&space_info->lock); | |
4855 | while (ticket->bytes > 0 && ticket->error == 0) { | |
4856 | ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE); | |
4857 | if (ret) { | |
4858 | ret = -EINTR; | |
4859 | break; | |
4860 | } | |
4861 | spin_unlock(&space_info->lock); | |
4862 | ||
4863 | schedule(); | |
4864 | ||
4865 | finish_wait(&ticket->wait, &wait); | |
4866 | spin_lock(&space_info->lock); | |
4867 | } | |
4868 | if (!ret) | |
4869 | ret = ticket->error; | |
4870 | if (!list_empty(&ticket->list)) | |
4871 | list_del_init(&ticket->list); | |
f91587e4 JB |
4872 | if (ticket->bytes && ticket->bytes < ticket->orig_bytes) |
4873 | reclaim_bytes = ticket->orig_bytes - ticket->bytes; | |
957780eb JB |
4874 | spin_unlock(&space_info->lock); |
4875 | ||
f91587e4 | 4876 | if (reclaim_bytes) |
d44b72aa JB |
4877 | btrfs_space_info_add_old_bytes(fs_info, space_info, |
4878 | reclaim_bytes); | |
957780eb | 4879 | return ret; |
21c7e756 MX |
4880 | } |
4881 | ||
4a92b1b8 JB |
4882 | /** |
4883 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
4884 | * @root - the root we're allocating for | |
957780eb | 4885 | * @space_info - the space info we want to allocate from |
4a92b1b8 | 4886 | * @orig_bytes - the number of bytes we want |
48fc7f7e | 4887 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 4888 | * |
01327610 | 4889 | * This will reserve orig_bytes number of bytes from the space info associated |
4a92b1b8 JB |
4890 | * with the block_rsv. If there is not enough space it will make an attempt to |
4891 | * flush out space to make room. It will do this by flushing delalloc if | |
4892 | * possible or committing the transaction. If flush is 0 then no attempts to | |
4893 | * regain reservations will be made and this will fail if there is not enough | |
4894 | * space already. | |
8bb8ab2e | 4895 | */ |
c1c4919b | 4896 | static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info, |
957780eb JB |
4897 | struct btrfs_space_info *space_info, |
4898 | u64 orig_bytes, | |
c1c4919b JM |
4899 | enum btrfs_reserve_flush_enum flush, |
4900 | bool system_chunk) | |
9ed74f2d | 4901 | { |
957780eb | 4902 | struct reserve_ticket ticket; |
2bf64758 | 4903 | u64 used; |
f91587e4 | 4904 | u64 reclaim_bytes = 0; |
8bb8ab2e | 4905 | int ret = 0; |
9ed74f2d | 4906 | |
957780eb | 4907 | ASSERT(orig_bytes); |
8ca17f0f | 4908 | ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL); |
fdb5effd | 4909 | |
8bb8ab2e | 4910 | spin_lock(&space_info->lock); |
fdb5effd | 4911 | ret = -ENOSPC; |
4136135b | 4912 | used = btrfs_space_info_used(space_info, true); |
9ed74f2d | 4913 | |
8bb8ab2e | 4914 | /* |
957780eb JB |
4915 | * If we have enough space then hooray, make our reservation and carry |
4916 | * on. If not see if we can overcommit, and if we can, hooray carry on. | |
4917 | * If not things get more complicated. | |
8bb8ab2e | 4918 | */ |
957780eb | 4919 | if (used + orig_bytes <= space_info->total_bytes) { |
bb96c4e5 JB |
4920 | btrfs_space_info_update_bytes_may_use(fs_info, space_info, |
4921 | orig_bytes); | |
0b246afa JM |
4922 | trace_btrfs_space_reservation(fs_info, "space_info", |
4923 | space_info->flags, orig_bytes, 1); | |
957780eb | 4924 | ret = 0; |
41783ef2 JB |
4925 | } else if (btrfs_can_overcommit(fs_info, space_info, orig_bytes, flush, |
4926 | system_chunk)) { | |
bb96c4e5 JB |
4927 | btrfs_space_info_update_bytes_may_use(fs_info, space_info, |
4928 | orig_bytes); | |
0b246afa JM |
4929 | trace_btrfs_space_reservation(fs_info, "space_info", |
4930 | space_info->flags, orig_bytes, 1); | |
44734ed1 | 4931 | ret = 0; |
2bf64758 JB |
4932 | } |
4933 | ||
8bb8ab2e | 4934 | /* |
957780eb JB |
4935 | * If we couldn't make a reservation then setup our reservation ticket |
4936 | * and kick the async worker if it's not already running. | |
08e007d2 | 4937 | * |
957780eb JB |
4938 | * If we are a priority flusher then we just need to add our ticket to |
4939 | * the list and we will do our own flushing further down. | |
8bb8ab2e | 4940 | */ |
72bcd99d | 4941 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
f91587e4 | 4942 | ticket.orig_bytes = orig_bytes; |
957780eb JB |
4943 | ticket.bytes = orig_bytes; |
4944 | ticket.error = 0; | |
4945 | init_waitqueue_head(&ticket.wait); | |
4946 | if (flush == BTRFS_RESERVE_FLUSH_ALL) { | |
4947 | list_add_tail(&ticket.list, &space_info->tickets); | |
4948 | if (!space_info->flush) { | |
4949 | space_info->flush = 1; | |
0b246afa | 4950 | trace_btrfs_trigger_flush(fs_info, |
f376df2b JB |
4951 | space_info->flags, |
4952 | orig_bytes, flush, | |
4953 | "enospc"); | |
957780eb | 4954 | queue_work(system_unbound_wq, |
c1c4919b | 4955 | &fs_info->async_reclaim_work); |
957780eb JB |
4956 | } |
4957 | } else { | |
4958 | list_add_tail(&ticket.list, | |
4959 | &space_info->priority_tickets); | |
4960 | } | |
21c7e756 MX |
4961 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
4962 | used += orig_bytes; | |
f6acfd50 JB |
4963 | /* |
4964 | * We will do the space reservation dance during log replay, | |
4965 | * which means we won't have fs_info->fs_root set, so don't do | |
4966 | * the async reclaim as we will panic. | |
4967 | */ | |
0b246afa | 4968 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) && |
c1c4919b JM |
4969 | need_do_async_reclaim(fs_info, space_info, |
4970 | used, system_chunk) && | |
0b246afa JM |
4971 | !work_busy(&fs_info->async_reclaim_work)) { |
4972 | trace_btrfs_trigger_flush(fs_info, space_info->flags, | |
4973 | orig_bytes, flush, "preempt"); | |
21c7e756 | 4974 | queue_work(system_unbound_wq, |
0b246afa | 4975 | &fs_info->async_reclaim_work); |
f376df2b | 4976 | } |
8bb8ab2e | 4977 | } |
f0486c68 | 4978 | spin_unlock(&space_info->lock); |
08e007d2 | 4979 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
957780eb | 4980 | return ret; |
f0486c68 | 4981 | |
957780eb | 4982 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
f91587e4 | 4983 | return wait_reserve_ticket(fs_info, space_info, &ticket); |
08e007d2 | 4984 | |
957780eb | 4985 | ret = 0; |
0b246afa | 4986 | priority_reclaim_metadata_space(fs_info, space_info, &ticket); |
957780eb JB |
4987 | spin_lock(&space_info->lock); |
4988 | if (ticket.bytes) { | |
f91587e4 JB |
4989 | if (ticket.bytes < orig_bytes) |
4990 | reclaim_bytes = orig_bytes - ticket.bytes; | |
957780eb JB |
4991 | list_del_init(&ticket.list); |
4992 | ret = -ENOSPC; | |
4993 | } | |
4994 | spin_unlock(&space_info->lock); | |
f91587e4 JB |
4995 | |
4996 | if (reclaim_bytes) | |
d44b72aa JB |
4997 | btrfs_space_info_add_old_bytes(fs_info, space_info, |
4998 | reclaim_bytes); | |
957780eb JB |
4999 | ASSERT(list_empty(&ticket.list)); |
5000 | return ret; | |
5001 | } | |
8bb8ab2e | 5002 | |
957780eb JB |
5003 | /** |
5004 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5005 | * @root - the root we're allocating for | |
5006 | * @block_rsv - the block_rsv we're allocating for | |
5007 | * @orig_bytes - the number of bytes we want | |
5008 | * @flush - whether or not we can flush to make our reservation | |
5009 | * | |
52042d8e | 5010 | * This will reserve orig_bytes number of bytes from the space info associated |
957780eb JB |
5011 | * with the block_rsv. If there is not enough space it will make an attempt to |
5012 | * flush out space to make room. It will do this by flushing delalloc if | |
5013 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5014 | * regain reservations will be made and this will fail if there is not enough | |
5015 | * space already. | |
5016 | */ | |
5017 | static int reserve_metadata_bytes(struct btrfs_root *root, | |
5018 | struct btrfs_block_rsv *block_rsv, | |
5019 | u64 orig_bytes, | |
5020 | enum btrfs_reserve_flush_enum flush) | |
5021 | { | |
0b246afa JM |
5022 | struct btrfs_fs_info *fs_info = root->fs_info; |
5023 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
957780eb | 5024 | int ret; |
c1c4919b | 5025 | bool system_chunk = (root == fs_info->chunk_root); |
957780eb | 5026 | |
c1c4919b JM |
5027 | ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info, |
5028 | orig_bytes, flush, system_chunk); | |
5d80366e JB |
5029 | if (ret == -ENOSPC && |
5030 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5d80366e | 5031 | if (block_rsv != global_rsv && |
c2a67a76 | 5032 | !btrfs_block_rsv_use_bytes(global_rsv, orig_bytes)) |
5d80366e JB |
5033 | ret = 0; |
5034 | } | |
9a3daff3 | 5035 | if (ret == -ENOSPC) { |
0b246afa | 5036 | trace_btrfs_space_reservation(fs_info, "space_info:enospc", |
957780eb JB |
5037 | block_rsv->space_info->flags, |
5038 | orig_bytes, 1); | |
9a3daff3 NB |
5039 | |
5040 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) | |
5041 | dump_space_info(fs_info, block_rsv->space_info, | |
5042 | orig_bytes, 0); | |
5043 | } | |
f0486c68 YZ |
5044 | return ret; |
5045 | } | |
5046 | ||
79787eaa JM |
5047 | static struct btrfs_block_rsv *get_block_rsv( |
5048 | const struct btrfs_trans_handle *trans, | |
5049 | const struct btrfs_root *root) | |
f0486c68 | 5050 | { |
0b246afa | 5051 | struct btrfs_fs_info *fs_info = root->fs_info; |
4c13d758 JB |
5052 | struct btrfs_block_rsv *block_rsv = NULL; |
5053 | ||
e9cf439f | 5054 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa JM |
5055 | (root == fs_info->csum_root && trans->adding_csums) || |
5056 | (root == fs_info->uuid_root)) | |
f7a81ea4 SB |
5057 | block_rsv = trans->block_rsv; |
5058 | ||
4c13d758 | 5059 | if (!block_rsv) |
f0486c68 YZ |
5060 | block_rsv = root->block_rsv; |
5061 | ||
5062 | if (!block_rsv) | |
0b246afa | 5063 | block_rsv = &fs_info->empty_block_rsv; |
f0486c68 YZ |
5064 | |
5065 | return block_rsv; | |
5066 | } | |
5067 | ||
c2a67a76 | 5068 | int btrfs_block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, u64 num_bytes) |
f0486c68 YZ |
5069 | { |
5070 | int ret = -ENOSPC; | |
5071 | spin_lock(&block_rsv->lock); | |
5072 | if (block_rsv->reserved >= num_bytes) { | |
5073 | block_rsv->reserved -= num_bytes; | |
5074 | if (block_rsv->reserved < block_rsv->size) | |
5075 | block_rsv->full = 0; | |
5076 | ret = 0; | |
5077 | } | |
5078 | spin_unlock(&block_rsv->lock); | |
5079 | return ret; | |
5080 | } | |
5081 | ||
5082 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
3a584174 | 5083 | u64 num_bytes, bool update_size) |
f0486c68 YZ |
5084 | { |
5085 | spin_lock(&block_rsv->lock); | |
5086 | block_rsv->reserved += num_bytes; | |
5087 | if (update_size) | |
5088 | block_rsv->size += num_bytes; | |
5089 | else if (block_rsv->reserved >= block_rsv->size) | |
5090 | block_rsv->full = 1; | |
5091 | spin_unlock(&block_rsv->lock); | |
5092 | } | |
5093 | ||
d52be818 JB |
5094 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
5095 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5096 | int min_factor) | |
5097 | { | |
5098 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5099 | u64 min_bytes; | |
5100 | ||
5101 | if (global_rsv->space_info != dest->space_info) | |
5102 | return -ENOSPC; | |
5103 | ||
5104 | spin_lock(&global_rsv->lock); | |
5105 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5106 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5107 | spin_unlock(&global_rsv->lock); | |
5108 | return -ENOSPC; | |
5109 | } | |
5110 | global_rsv->reserved -= num_bytes; | |
5111 | if (global_rsv->reserved < global_rsv->size) | |
5112 | global_rsv->full = 0; | |
5113 | spin_unlock(&global_rsv->lock); | |
5114 | ||
3a584174 | 5115 | block_rsv_add_bytes(dest, num_bytes, true); |
d52be818 JB |
5116 | return 0; |
5117 | } | |
5118 | ||
ba2c4d4e JB |
5119 | /** |
5120 | * btrfs_migrate_to_delayed_refs_rsv - transfer bytes to our delayed refs rsv. | |
5121 | * @fs_info - the fs info for our fs. | |
5122 | * @src - the source block rsv to transfer from. | |
5123 | * @num_bytes - the number of bytes to transfer. | |
5124 | * | |
5125 | * This transfers up to the num_bytes amount from the src rsv to the | |
5126 | * delayed_refs_rsv. Any extra bytes are returned to the space info. | |
5127 | */ | |
5128 | void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info, | |
5129 | struct btrfs_block_rsv *src, | |
5130 | u64 num_bytes) | |
5131 | { | |
5132 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; | |
5133 | u64 to_free = 0; | |
5134 | ||
5135 | spin_lock(&src->lock); | |
5136 | src->reserved -= num_bytes; | |
5137 | src->size -= num_bytes; | |
5138 | spin_unlock(&src->lock); | |
5139 | ||
5140 | spin_lock(&delayed_refs_rsv->lock); | |
5141 | if (delayed_refs_rsv->size > delayed_refs_rsv->reserved) { | |
5142 | u64 delta = delayed_refs_rsv->size - | |
5143 | delayed_refs_rsv->reserved; | |
5144 | if (num_bytes > delta) { | |
5145 | to_free = num_bytes - delta; | |
5146 | num_bytes = delta; | |
5147 | } | |
5148 | } else { | |
5149 | to_free = num_bytes; | |
5150 | num_bytes = 0; | |
5151 | } | |
5152 | ||
5153 | if (num_bytes) | |
5154 | delayed_refs_rsv->reserved += num_bytes; | |
5155 | if (delayed_refs_rsv->reserved >= delayed_refs_rsv->size) | |
5156 | delayed_refs_rsv->full = 1; | |
5157 | spin_unlock(&delayed_refs_rsv->lock); | |
5158 | ||
5159 | if (num_bytes) | |
5160 | trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv", | |
5161 | 0, num_bytes, 1); | |
5162 | if (to_free) | |
d44b72aa JB |
5163 | btrfs_space_info_add_old_bytes(fs_info, |
5164 | delayed_refs_rsv->space_info, to_free); | |
ba2c4d4e JB |
5165 | } |
5166 | ||
5167 | /** | |
5168 | * btrfs_delayed_refs_rsv_refill - refill based on our delayed refs usage. | |
5169 | * @fs_info - the fs_info for our fs. | |
5170 | * @flush - control how we can flush for this reservation. | |
5171 | * | |
5172 | * This will refill the delayed block_rsv up to 1 items size worth of space and | |
5173 | * will return -ENOSPC if we can't make the reservation. | |
5174 | */ | |
5175 | int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info, | |
5176 | enum btrfs_reserve_flush_enum flush) | |
5177 | { | |
5178 | struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv; | |
5179 | u64 limit = btrfs_calc_trans_metadata_size(fs_info, 1); | |
5180 | u64 num_bytes = 0; | |
5181 | int ret = -ENOSPC; | |
5182 | ||
5183 | spin_lock(&block_rsv->lock); | |
5184 | if (block_rsv->reserved < block_rsv->size) { | |
5185 | num_bytes = block_rsv->size - block_rsv->reserved; | |
5186 | num_bytes = min(num_bytes, limit); | |
5187 | } | |
5188 | spin_unlock(&block_rsv->lock); | |
5189 | ||
5190 | if (!num_bytes) | |
5191 | return 0; | |
5192 | ||
5193 | ret = reserve_metadata_bytes(fs_info->extent_root, block_rsv, | |
5194 | num_bytes, flush); | |
5195 | if (ret) | |
5196 | return ret; | |
5197 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
5198 | trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv", | |
5199 | 0, num_bytes, 1); | |
5200 | return 0; | |
5201 | } | |
5202 | ||
69fe2d75 | 5203 | static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
8c2a3ca2 | 5204 | struct btrfs_block_rsv *block_rsv, |
ff6bc37e QW |
5205 | struct btrfs_block_rsv *dest, u64 num_bytes, |
5206 | u64 *qgroup_to_release_ret) | |
f0486c68 YZ |
5207 | { |
5208 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
ff6bc37e | 5209 | u64 qgroup_to_release = 0; |
69fe2d75 | 5210 | u64 ret; |
f0486c68 YZ |
5211 | |
5212 | spin_lock(&block_rsv->lock); | |
ff6bc37e | 5213 | if (num_bytes == (u64)-1) { |
f0486c68 | 5214 | num_bytes = block_rsv->size; |
ff6bc37e QW |
5215 | qgroup_to_release = block_rsv->qgroup_rsv_size; |
5216 | } | |
f0486c68 YZ |
5217 | block_rsv->size -= num_bytes; |
5218 | if (block_rsv->reserved >= block_rsv->size) { | |
5219 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5220 | block_rsv->reserved = block_rsv->size; | |
5221 | block_rsv->full = 1; | |
5222 | } else { | |
5223 | num_bytes = 0; | |
5224 | } | |
ff6bc37e QW |
5225 | if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) { |
5226 | qgroup_to_release = block_rsv->qgroup_rsv_reserved - | |
5227 | block_rsv->qgroup_rsv_size; | |
5228 | block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size; | |
5229 | } else { | |
5230 | qgroup_to_release = 0; | |
5231 | } | |
f0486c68 YZ |
5232 | spin_unlock(&block_rsv->lock); |
5233 | ||
69fe2d75 | 5234 | ret = num_bytes; |
f0486c68 YZ |
5235 | if (num_bytes > 0) { |
5236 | if (dest) { | |
e9e22899 JB |
5237 | spin_lock(&dest->lock); |
5238 | if (!dest->full) { | |
5239 | u64 bytes_to_add; | |
5240 | ||
5241 | bytes_to_add = dest->size - dest->reserved; | |
5242 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5243 | dest->reserved += bytes_to_add; | |
5244 | if (dest->reserved >= dest->size) | |
5245 | dest->full = 1; | |
5246 | num_bytes -= bytes_to_add; | |
5247 | } | |
5248 | spin_unlock(&dest->lock); | |
5249 | } | |
957780eb | 5250 | if (num_bytes) |
d44b72aa JB |
5251 | btrfs_space_info_add_old_bytes(fs_info, space_info, |
5252 | num_bytes); | |
9ed74f2d | 5253 | } |
ff6bc37e QW |
5254 | if (qgroup_to_release_ret) |
5255 | *qgroup_to_release_ret = qgroup_to_release; | |
69fe2d75 | 5256 | return ret; |
f0486c68 | 5257 | } |
4e06bdd6 | 5258 | |
25d609f8 JB |
5259 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src, |
5260 | struct btrfs_block_rsv *dst, u64 num_bytes, | |
3a584174 | 5261 | bool update_size) |
f0486c68 YZ |
5262 | { |
5263 | int ret; | |
9ed74f2d | 5264 | |
c2a67a76 | 5265 | ret = btrfs_block_rsv_use_bytes(src, num_bytes); |
f0486c68 YZ |
5266 | if (ret) |
5267 | return ret; | |
9ed74f2d | 5268 | |
25d609f8 | 5269 | block_rsv_add_bytes(dst, num_bytes, update_size); |
9ed74f2d JB |
5270 | return 0; |
5271 | } | |
5272 | ||
66d8f3dd | 5273 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 5274 | { |
f0486c68 YZ |
5275 | memset(rsv, 0, sizeof(*rsv)); |
5276 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 5277 | rsv->type = type; |
f0486c68 YZ |
5278 | } |
5279 | ||
69fe2d75 JB |
5280 | void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info, |
5281 | struct btrfs_block_rsv *rsv, | |
5282 | unsigned short type) | |
5283 | { | |
5284 | btrfs_init_block_rsv(rsv, type); | |
280c2908 | 5285 | rsv->space_info = btrfs_find_space_info(fs_info, |
69fe2d75 JB |
5286 | BTRFS_BLOCK_GROUP_METADATA); |
5287 | } | |
5288 | ||
2ff7e61e | 5289 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info, |
66d8f3dd | 5290 | unsigned short type) |
f0486c68 YZ |
5291 | { |
5292 | struct btrfs_block_rsv *block_rsv; | |
9ed74f2d | 5293 | |
f0486c68 YZ |
5294 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
5295 | if (!block_rsv) | |
5296 | return NULL; | |
9ed74f2d | 5297 | |
69fe2d75 | 5298 | btrfs_init_metadata_block_rsv(fs_info, block_rsv, type); |
f0486c68 YZ |
5299 | return block_rsv; |
5300 | } | |
9ed74f2d | 5301 | |
2ff7e61e | 5302 | void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5303 | struct btrfs_block_rsv *rsv) |
5304 | { | |
2aaa6655 JB |
5305 | if (!rsv) |
5306 | return; | |
2ff7e61e | 5307 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
dabdb640 | 5308 | kfree(rsv); |
9ed74f2d JB |
5309 | } |
5310 | ||
08e007d2 MX |
5311 | int btrfs_block_rsv_add(struct btrfs_root *root, |
5312 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5313 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5314 | { |
f0486c68 | 5315 | int ret; |
9ed74f2d | 5316 | |
f0486c68 YZ |
5317 | if (num_bytes == 0) |
5318 | return 0; | |
8bb8ab2e | 5319 | |
61b520a9 | 5320 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
5a2cb25a | 5321 | if (!ret) |
3a584174 | 5322 | block_rsv_add_bytes(block_rsv, num_bytes, true); |
9ed74f2d | 5323 | |
f0486c68 | 5324 | return ret; |
f0486c68 | 5325 | } |
9ed74f2d | 5326 | |
2ff7e61e | 5327 | int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
5328 | { |
5329 | u64 num_bytes = 0; | |
f0486c68 | 5330 | int ret = -ENOSPC; |
9ed74f2d | 5331 | |
f0486c68 YZ |
5332 | if (!block_rsv) |
5333 | return 0; | |
9ed74f2d | 5334 | |
f0486c68 | 5335 | spin_lock(&block_rsv->lock); |
36ba022a JB |
5336 | num_bytes = div_factor(block_rsv->size, min_factor); |
5337 | if (block_rsv->reserved >= num_bytes) | |
5338 | ret = 0; | |
5339 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 5340 | |
36ba022a JB |
5341 | return ret; |
5342 | } | |
5343 | ||
08e007d2 MX |
5344 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
5345 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5346 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
5347 | { |
5348 | u64 num_bytes = 0; | |
5349 | int ret = -ENOSPC; | |
5350 | ||
5351 | if (!block_rsv) | |
5352 | return 0; | |
5353 | ||
5354 | spin_lock(&block_rsv->lock); | |
5355 | num_bytes = min_reserved; | |
13553e52 | 5356 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 5357 | ret = 0; |
13553e52 | 5358 | else |
f0486c68 | 5359 | num_bytes -= block_rsv->reserved; |
f0486c68 | 5360 | spin_unlock(&block_rsv->lock); |
13553e52 | 5361 | |
f0486c68 YZ |
5362 | if (!ret) |
5363 | return 0; | |
5364 | ||
aa38a711 | 5365 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 | 5366 | if (!ret) { |
3a584174 | 5367 | block_rsv_add_bytes(block_rsv, num_bytes, false); |
f0486c68 | 5368 | return 0; |
6a63209f | 5369 | } |
9ed74f2d | 5370 | |
13553e52 | 5371 | return ret; |
f0486c68 YZ |
5372 | } |
5373 | ||
ba2c4d4e JB |
5374 | static u64 __btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, |
5375 | struct btrfs_block_rsv *block_rsv, | |
5376 | u64 num_bytes, u64 *qgroup_to_release) | |
5377 | { | |
5378 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5379 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv; | |
5380 | struct btrfs_block_rsv *target = delayed_rsv; | |
5381 | ||
5382 | if (target->full || target == block_rsv) | |
5383 | target = global_rsv; | |
5384 | ||
5385 | if (block_rsv->space_info != target->space_info) | |
5386 | target = NULL; | |
5387 | ||
5388 | return block_rsv_release_bytes(fs_info, block_rsv, target, num_bytes, | |
5389 | qgroup_to_release); | |
5390 | } | |
5391 | ||
5392 | void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, | |
5393 | struct btrfs_block_rsv *block_rsv, | |
5394 | u64 num_bytes) | |
5395 | { | |
5396 | __btrfs_block_rsv_release(fs_info, block_rsv, num_bytes, NULL); | |
5397 | } | |
5398 | ||
69fe2d75 JB |
5399 | /** |
5400 | * btrfs_inode_rsv_release - release any excessive reservation. | |
5401 | * @inode - the inode we need to release from. | |
43b18595 QW |
5402 | * @qgroup_free - free or convert qgroup meta. |
5403 | * Unlike normal operation, qgroup meta reservation needs to know if we are | |
5404 | * freeing qgroup reservation or just converting it into per-trans. Normally | |
5405 | * @qgroup_free is true for error handling, and false for normal release. | |
69fe2d75 JB |
5406 | * |
5407 | * This is the same as btrfs_block_rsv_release, except that it handles the | |
5408 | * tracepoint for the reservation. | |
5409 | */ | |
43b18595 | 5410 | static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free) |
69fe2d75 JB |
5411 | { |
5412 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
69fe2d75 JB |
5413 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
5414 | u64 released = 0; | |
ff6bc37e | 5415 | u64 qgroup_to_release = 0; |
69fe2d75 JB |
5416 | |
5417 | /* | |
5418 | * Since we statically set the block_rsv->size we just want to say we | |
5419 | * are releasing 0 bytes, and then we'll just get the reservation over | |
5420 | * the size free'd. | |
5421 | */ | |
ba2c4d4e JB |
5422 | released = __btrfs_block_rsv_release(fs_info, block_rsv, 0, |
5423 | &qgroup_to_release); | |
69fe2d75 JB |
5424 | if (released > 0) |
5425 | trace_btrfs_space_reservation(fs_info, "delalloc", | |
5426 | btrfs_ino(inode), released, 0); | |
43b18595 | 5427 | if (qgroup_free) |
ff6bc37e | 5428 | btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release); |
43b18595 | 5429 | else |
ff6bc37e QW |
5430 | btrfs_qgroup_convert_reserved_meta(inode->root, |
5431 | qgroup_to_release); | |
69fe2d75 JB |
5432 | } |
5433 | ||
ba2c4d4e JB |
5434 | /** |
5435 | * btrfs_delayed_refs_rsv_release - release a ref head's reservation. | |
5436 | * @fs_info - the fs_info for our fs. | |
5437 | * @nr - the number of items to drop. | |
5438 | * | |
5439 | * This drops the delayed ref head's count from the delayed refs rsv and frees | |
5440 | * any excess reservation we had. | |
5441 | */ | |
5442 | void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr) | |
f0486c68 | 5443 | { |
ba2c4d4e | 5444 | struct btrfs_block_rsv *block_rsv = &fs_info->delayed_refs_rsv; |
0b246afa | 5445 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
ba2c4d4e JB |
5446 | u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, nr); |
5447 | u64 released = 0; | |
0b246afa | 5448 | |
ba2c4d4e JB |
5449 | released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, |
5450 | num_bytes, NULL); | |
5451 | if (released) | |
5452 | trace_btrfs_space_reservation(fs_info, "delayed_refs_rsv", | |
5453 | 0, released, 0); | |
6a63209f JB |
5454 | } |
5455 | ||
8929ecfa YZ |
5456 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5457 | { | |
5458 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5459 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5460 | u64 num_bytes; | |
6a63209f | 5461 | |
ae2e4728 JB |
5462 | /* |
5463 | * The global block rsv is based on the size of the extent tree, the | |
5464 | * checksum tree and the root tree. If the fs is empty we want to set | |
5465 | * it to a minimal amount for safety. | |
5466 | */ | |
5467 | num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) + | |
5468 | btrfs_root_used(&fs_info->csum_root->root_item) + | |
5469 | btrfs_root_used(&fs_info->tree_root->root_item); | |
5470 | num_bytes = max_t(u64, num_bytes, SZ_16M); | |
33b4d47f | 5471 | |
8929ecfa | 5472 | spin_lock(&sinfo->lock); |
1f699d38 | 5473 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5474 | |
ee22184b | 5475 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); |
4e06bdd6 | 5476 | |
fb4b10e5 | 5477 | if (block_rsv->reserved < block_rsv->size) { |
4136135b | 5478 | num_bytes = btrfs_space_info_used(sinfo, true); |
fb4b10e5 JB |
5479 | if (sinfo->total_bytes > num_bytes) { |
5480 | num_bytes = sinfo->total_bytes - num_bytes; | |
5481 | num_bytes = min(num_bytes, | |
5482 | block_rsv->size - block_rsv->reserved); | |
5483 | block_rsv->reserved += num_bytes; | |
bb96c4e5 JB |
5484 | btrfs_space_info_update_bytes_may_use(fs_info, sinfo, |
5485 | num_bytes); | |
fb4b10e5 JB |
5486 | trace_btrfs_space_reservation(fs_info, "space_info", |
5487 | sinfo->flags, num_bytes, | |
5488 | 1); | |
5489 | } | |
5490 | } else if (block_rsv->reserved > block_rsv->size) { | |
8929ecfa | 5491 | num_bytes = block_rsv->reserved - block_rsv->size; |
bb96c4e5 JB |
5492 | btrfs_space_info_update_bytes_may_use(fs_info, sinfo, |
5493 | -num_bytes); | |
8c2a3ca2 | 5494 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5495 | sinfo->flags, num_bytes, 0); |
8929ecfa | 5496 | block_rsv->reserved = block_rsv->size; |
8929ecfa | 5497 | } |
182608c8 | 5498 | |
fb4b10e5 JB |
5499 | if (block_rsv->reserved == block_rsv->size) |
5500 | block_rsv->full = 1; | |
5501 | else | |
5502 | block_rsv->full = 0; | |
5503 | ||
8929ecfa | 5504 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5505 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5506 | } |
5507 | ||
f0486c68 | 5508 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5509 | { |
f0486c68 | 5510 | struct btrfs_space_info *space_info; |
6a63209f | 5511 | |
280c2908 | 5512 | space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
f0486c68 | 5513 | fs_info->chunk_block_rsv.space_info = space_info; |
6a63209f | 5514 | |
280c2908 | 5515 | space_info = btrfs_find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5516 | fs_info->global_block_rsv.space_info = space_info; |
f0486c68 YZ |
5517 | fs_info->trans_block_rsv.space_info = space_info; |
5518 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5519 | fs_info->delayed_block_rsv.space_info = space_info; |
ba2c4d4e | 5520 | fs_info->delayed_refs_rsv.space_info = space_info; |
f0486c68 | 5521 | |
ba2c4d4e JB |
5522 | fs_info->extent_root->block_rsv = &fs_info->delayed_refs_rsv; |
5523 | fs_info->csum_root->block_rsv = &fs_info->delayed_refs_rsv; | |
8929ecfa YZ |
5524 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; |
5525 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5526 | if (fs_info->quota_root) |
5527 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5528 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5529 | |
8929ecfa | 5530 | update_global_block_rsv(fs_info); |
6a63209f JB |
5531 | } |
5532 | ||
8929ecfa | 5533 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5534 | { |
8c2a3ca2 | 5535 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
ff6bc37e | 5536 | (u64)-1, NULL); |
8929ecfa YZ |
5537 | WARN_ON(fs_info->trans_block_rsv.size > 0); |
5538 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5539 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5540 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5541 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5542 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
ba2c4d4e JB |
5543 | WARN_ON(fs_info->delayed_refs_rsv.reserved > 0); |
5544 | WARN_ON(fs_info->delayed_refs_rsv.size > 0); | |
fcb80c2a JB |
5545 | } |
5546 | ||
ba2c4d4e JB |
5547 | /* |
5548 | * btrfs_update_delayed_refs_rsv - adjust the size of the delayed refs rsv | |
5549 | * @trans - the trans that may have generated delayed refs | |
5550 | * | |
5551 | * This is to be called anytime we may have adjusted trans->delayed_ref_updates, | |
5552 | * it'll calculate the additional size and add it to the delayed_refs_rsv. | |
5553 | */ | |
5554 | void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans) | |
5555 | { | |
5556 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
5557 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_refs_rsv; | |
5558 | u64 num_bytes; | |
5559 | ||
5560 | if (!trans->delayed_ref_updates) | |
5561 | return; | |
5562 | ||
5563 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, | |
5564 | trans->delayed_ref_updates); | |
5565 | spin_lock(&delayed_rsv->lock); | |
5566 | delayed_rsv->size += num_bytes; | |
5567 | delayed_rsv->full = 0; | |
5568 | spin_unlock(&delayed_rsv->lock); | |
5569 | trans->delayed_ref_updates = 0; | |
5570 | } | |
6a63209f | 5571 | |
4fbcdf66 FM |
5572 | /* |
5573 | * To be called after all the new block groups attached to the transaction | |
5574 | * handle have been created (btrfs_create_pending_block_groups()). | |
5575 | */ | |
5576 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5577 | { | |
64b63580 | 5578 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4fbcdf66 FM |
5579 | |
5580 | if (!trans->chunk_bytes_reserved) | |
5581 | return; | |
5582 | ||
5583 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5584 | ||
5585 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
ff6bc37e | 5586 | trans->chunk_bytes_reserved, NULL); |
4fbcdf66 FM |
5587 | trans->chunk_bytes_reserved = 0; |
5588 | } | |
5589 | ||
d5c12070 MX |
5590 | /* |
5591 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5592 | * root: the root of the parent directory | |
5593 | * rsv: block reservation | |
5594 | * items: the number of items that we need do reservation | |
a5b7f429 | 5595 | * use_global_rsv: allow fallback to the global block reservation |
d5c12070 MX |
5596 | * |
5597 | * This function is used to reserve the space for snapshot/subvolume | |
5598 | * creation and deletion. Those operations are different with the | |
5599 | * common file/directory operations, they change two fs/file trees | |
5600 | * and root tree, the number of items that the qgroup reserves is | |
5601 | * different with the free space reservation. So we can not use | |
01327610 | 5602 | * the space reservation mechanism in start_transaction(). |
d5c12070 MX |
5603 | */ |
5604 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
a5b7f429 | 5605 | struct btrfs_block_rsv *rsv, int items, |
ee3441b4 | 5606 | bool use_global_rsv) |
a22285a6 | 5607 | { |
a5b7f429 | 5608 | u64 qgroup_num_bytes = 0; |
d5c12070 MX |
5609 | u64 num_bytes; |
5610 | int ret; | |
0b246afa JM |
5611 | struct btrfs_fs_info *fs_info = root->fs_info; |
5612 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d5c12070 | 5613 | |
0b246afa | 5614 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
d5c12070 | 5615 | /* One for parent inode, two for dir entries */ |
a5b7f429 LF |
5616 | qgroup_num_bytes = 3 * fs_info->nodesize; |
5617 | ret = btrfs_qgroup_reserve_meta_prealloc(root, | |
5618 | qgroup_num_bytes, true); | |
d5c12070 MX |
5619 | if (ret) |
5620 | return ret; | |
d5c12070 MX |
5621 | } |
5622 | ||
0b246afa | 5623 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, items); |
280c2908 | 5624 | rsv->space_info = btrfs_find_space_info(fs_info, |
d5c12070 MX |
5625 | BTRFS_BLOCK_GROUP_METADATA); |
5626 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
5627 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
5628 | |
5629 | if (ret == -ENOSPC && use_global_rsv) | |
3a584174 | 5630 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true); |
ee3441b4 | 5631 | |
a5b7f429 LF |
5632 | if (ret && qgroup_num_bytes) |
5633 | btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes); | |
d5c12070 MX |
5634 | |
5635 | return ret; | |
5636 | } | |
5637 | ||
2ff7e61e | 5638 | void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info, |
7775c818 | 5639 | struct btrfs_block_rsv *rsv) |
d5c12070 | 5640 | { |
2ff7e61e | 5641 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
97e728d4 JB |
5642 | } |
5643 | ||
69fe2d75 JB |
5644 | static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info, |
5645 | struct btrfs_inode *inode) | |
9e0baf60 | 5646 | { |
69fe2d75 JB |
5647 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
5648 | u64 reserve_size = 0; | |
ff6bc37e | 5649 | u64 qgroup_rsv_size = 0; |
69fe2d75 JB |
5650 | u64 csum_leaves; |
5651 | unsigned outstanding_extents; | |
9e0baf60 | 5652 | |
69fe2d75 JB |
5653 | lockdep_assert_held(&inode->lock); |
5654 | outstanding_extents = inode->outstanding_extents; | |
5655 | if (outstanding_extents) | |
5656 | reserve_size = btrfs_calc_trans_metadata_size(fs_info, | |
5657 | outstanding_extents + 1); | |
5658 | csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, | |
5659 | inode->csum_bytes); | |
5660 | reserve_size += btrfs_calc_trans_metadata_size(fs_info, | |
5661 | csum_leaves); | |
ff6bc37e QW |
5662 | /* |
5663 | * For qgroup rsv, the calculation is very simple: | |
5664 | * account one nodesize for each outstanding extent | |
5665 | * | |
5666 | * This is overestimating in most cases. | |
5667 | */ | |
139a5617 | 5668 | qgroup_rsv_size = (u64)outstanding_extents * fs_info->nodesize; |
9e0baf60 | 5669 | |
69fe2d75 JB |
5670 | spin_lock(&block_rsv->lock); |
5671 | block_rsv->size = reserve_size; | |
ff6bc37e | 5672 | block_rsv->qgroup_rsv_size = qgroup_rsv_size; |
69fe2d75 | 5673 | spin_unlock(&block_rsv->lock); |
0ca1f7ce | 5674 | } |
c146afad | 5675 | |
c8eaeac7 JB |
5676 | static void calc_inode_reservations(struct btrfs_fs_info *fs_info, |
5677 | u64 num_bytes, u64 *meta_reserve, | |
5678 | u64 *qgroup_reserve) | |
5679 | { | |
5680 | u64 nr_extents = count_max_extents(num_bytes); | |
5681 | u64 csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, num_bytes); | |
5682 | ||
5683 | /* We add one for the inode update at finish ordered time */ | |
5684 | *meta_reserve = btrfs_calc_trans_metadata_size(fs_info, | |
5685 | nr_extents + csum_leaves + 1); | |
5686 | *qgroup_reserve = nr_extents * fs_info->nodesize; | |
5687 | } | |
5688 | ||
9f3db423 | 5689 | int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 5690 | { |
c8eaeac7 JB |
5691 | struct btrfs_root *root = inode->root; |
5692 | struct btrfs_fs_info *fs_info = root->fs_info; | |
5693 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; | |
5694 | u64 meta_reserve, qgroup_reserve; | |
69fe2d75 | 5695 | unsigned nr_extents; |
08e007d2 | 5696 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 5697 | int ret = 0; |
c64c2bd8 | 5698 | bool delalloc_lock = true; |
6324fbf3 | 5699 | |
c64c2bd8 JB |
5700 | /* If we are a free space inode we need to not flush since we will be in |
5701 | * the middle of a transaction commit. We also don't need the delalloc | |
5702 | * mutex since we won't race with anybody. We need this mostly to make | |
5703 | * lockdep shut its filthy mouth. | |
bac357dc JB |
5704 | * |
5705 | * If we have a transaction open (can happen if we call truncate_block | |
5706 | * from truncate), then we need FLUSH_LIMIT so we don't deadlock. | |
c64c2bd8 JB |
5707 | */ |
5708 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 5709 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 | 5710 | delalloc_lock = false; |
da07d4ab NB |
5711 | } else { |
5712 | if (current->journal_info) | |
5713 | flush = BTRFS_RESERVE_FLUSH_LIMIT; | |
c09544e0 | 5714 | |
da07d4ab NB |
5715 | if (btrfs_transaction_in_commit(fs_info)) |
5716 | schedule_timeout(1); | |
5717 | } | |
ec44a35c | 5718 | |
c64c2bd8 | 5719 | if (delalloc_lock) |
9f3db423 | 5720 | mutex_lock(&inode->delalloc_mutex); |
c64c2bd8 | 5721 | |
0b246afa | 5722 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
69fe2d75 | 5723 | |
c8eaeac7 JB |
5724 | /* |
5725 | * We always want to do it this way, every other way is wrong and ends | |
5726 | * in tears. Pre-reserving the amount we are going to add will always | |
5727 | * be the right way, because otherwise if we have enough parallelism we | |
5728 | * could end up with thousands of inodes all holding little bits of | |
5729 | * reservations they were able to make previously and the only way to | |
5730 | * reclaim that space is to ENOSPC out the operations and clear | |
5731 | * everything out and try again, which is bad. This way we just | |
5732 | * over-reserve slightly, and clean up the mess when we are done. | |
5733 | */ | |
5734 | calc_inode_reservations(fs_info, num_bytes, &meta_reserve, | |
5735 | &qgroup_reserve); | |
5736 | ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_reserve, true); | |
5737 | if (ret) | |
5738 | goto out_fail; | |
5739 | ret = reserve_metadata_bytes(root, block_rsv, meta_reserve, flush); | |
5740 | if (ret) | |
5741 | goto out_qgroup; | |
5742 | ||
5743 | /* | |
5744 | * Now we need to update our outstanding extents and csum bytes _first_ | |
5745 | * and then add the reservation to the block_rsv. This keeps us from | |
5746 | * racing with an ordered completion or some such that would think it | |
5747 | * needs to free the reservation we just made. | |
5748 | */ | |
9f3db423 | 5749 | spin_lock(&inode->lock); |
69fe2d75 | 5750 | nr_extents = count_max_extents(num_bytes); |
8b62f87b | 5751 | btrfs_mod_outstanding_extents(inode, nr_extents); |
69fe2d75 JB |
5752 | inode->csum_bytes += num_bytes; |
5753 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 5754 | spin_unlock(&inode->lock); |
57a45ced | 5755 | |
c8eaeac7 JB |
5756 | /* Now we can safely add our space to our block rsv */ |
5757 | block_rsv_add_bytes(block_rsv, meta_reserve, false); | |
5758 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
5759 | btrfs_ino(inode), meta_reserve, 1); | |
5760 | ||
5761 | spin_lock(&block_rsv->lock); | |
5762 | block_rsv->qgroup_rsv_reserved += qgroup_reserve; | |
5763 | spin_unlock(&block_rsv->lock); | |
25179201 | 5764 | |
c64c2bd8 | 5765 | if (delalloc_lock) |
9f3db423 | 5766 | mutex_unlock(&inode->delalloc_mutex); |
0ca1f7ce | 5767 | return 0; |
c8eaeac7 JB |
5768 | out_qgroup: |
5769 | btrfs_qgroup_free_meta_prealloc(root, qgroup_reserve); | |
88e081bf | 5770 | out_fail: |
43b18595 | 5771 | btrfs_inode_rsv_release(inode, true); |
88e081bf | 5772 | if (delalloc_lock) |
9f3db423 | 5773 | mutex_unlock(&inode->delalloc_mutex); |
88e081bf | 5774 | return ret; |
0ca1f7ce YZ |
5775 | } |
5776 | ||
7709cde3 JB |
5777 | /** |
5778 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
8b62f87b JB |
5779 | * @inode: the inode to release the reservation for. |
5780 | * @num_bytes: the number of bytes we are releasing. | |
43b18595 | 5781 | * @qgroup_free: free qgroup reservation or convert it to per-trans reservation |
7709cde3 JB |
5782 | * |
5783 | * This will release the metadata reservation for an inode. This can be called | |
5784 | * once we complete IO for a given set of bytes to release their metadata | |
8b62f87b | 5785 | * reservations, or on error for the same reason. |
7709cde3 | 5786 | */ |
43b18595 QW |
5787 | void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes, |
5788 | bool qgroup_free) | |
0ca1f7ce | 5789 | { |
3ffbd68c | 5790 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
0ca1f7ce | 5791 | |
0b246afa | 5792 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
691fa059 | 5793 | spin_lock(&inode->lock); |
69fe2d75 JB |
5794 | inode->csum_bytes -= num_bytes; |
5795 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
691fa059 | 5796 | spin_unlock(&inode->lock); |
0ca1f7ce | 5797 | |
0b246afa | 5798 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
5799 | return; |
5800 | ||
43b18595 | 5801 | btrfs_inode_rsv_release(inode, qgroup_free); |
0ca1f7ce YZ |
5802 | } |
5803 | ||
8b62f87b JB |
5804 | /** |
5805 | * btrfs_delalloc_release_extents - release our outstanding_extents | |
5806 | * @inode: the inode to balance the reservation for. | |
5807 | * @num_bytes: the number of bytes we originally reserved with | |
43b18595 | 5808 | * @qgroup_free: do we need to free qgroup meta reservation or convert them. |
8b62f87b JB |
5809 | * |
5810 | * When we reserve space we increase outstanding_extents for the extents we may | |
5811 | * add. Once we've set the range as delalloc or created our ordered extents we | |
5812 | * have outstanding_extents to track the real usage, so we use this to free our | |
5813 | * temporarily tracked outstanding_extents. This _must_ be used in conjunction | |
5814 | * with btrfs_delalloc_reserve_metadata. | |
5815 | */ | |
43b18595 QW |
5816 | void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes, |
5817 | bool qgroup_free) | |
8b62f87b | 5818 | { |
3ffbd68c | 5819 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
8b62f87b | 5820 | unsigned num_extents; |
8b62f87b JB |
5821 | |
5822 | spin_lock(&inode->lock); | |
5823 | num_extents = count_max_extents(num_bytes); | |
5824 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
69fe2d75 | 5825 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); |
8b62f87b JB |
5826 | spin_unlock(&inode->lock); |
5827 | ||
8b62f87b JB |
5828 | if (btrfs_is_testing(fs_info)) |
5829 | return; | |
5830 | ||
43b18595 | 5831 | btrfs_inode_rsv_release(inode, qgroup_free); |
8b62f87b JB |
5832 | } |
5833 | ||
1ada3a62 | 5834 | /** |
7cf5b976 | 5835 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
5836 | * delalloc |
5837 | * @inode: inode we're writing to | |
5838 | * @start: start range we are writing to | |
5839 | * @len: how long the range we are writing to | |
364ecf36 QW |
5840 | * @reserved: mandatory parameter, record actually reserved qgroup ranges of |
5841 | * current reservation. | |
1ada3a62 | 5842 | * |
1ada3a62 QW |
5843 | * This will do the following things |
5844 | * | |
5845 | * o reserve space in data space info for num bytes | |
5846 | * and reserve precious corresponding qgroup space | |
5847 | * (Done in check_data_free_space) | |
5848 | * | |
5849 | * o reserve space for metadata space, based on the number of outstanding | |
5850 | * extents and how much csums will be needed | |
5851 | * also reserve metadata space in a per root over-reserve method. | |
5852 | * o add to the inodes->delalloc_bytes | |
5853 | * o add it to the fs_info's delalloc inodes list. | |
5854 | * (Above 3 all done in delalloc_reserve_metadata) | |
5855 | * | |
5856 | * Return 0 for success | |
5857 | * Return <0 for error(-ENOSPC or -EQUOT) | |
5858 | */ | |
364ecf36 QW |
5859 | int btrfs_delalloc_reserve_space(struct inode *inode, |
5860 | struct extent_changeset **reserved, u64 start, u64 len) | |
1ada3a62 QW |
5861 | { |
5862 | int ret; | |
5863 | ||
364ecf36 | 5864 | ret = btrfs_check_data_free_space(inode, reserved, start, len); |
1ada3a62 QW |
5865 | if (ret < 0) |
5866 | return ret; | |
9f3db423 | 5867 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len); |
1ada3a62 | 5868 | if (ret < 0) |
bc42bda2 | 5869 | btrfs_free_reserved_data_space(inode, *reserved, start, len); |
1ada3a62 QW |
5870 | return ret; |
5871 | } | |
5872 | ||
7709cde3 | 5873 | /** |
7cf5b976 | 5874 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
5875 | * @inode: inode we're releasing space for |
5876 | * @start: start position of the space already reserved | |
5877 | * @len: the len of the space already reserved | |
8b62f87b | 5878 | * @release_bytes: the len of the space we consumed or didn't use |
1ada3a62 QW |
5879 | * |
5880 | * This function will release the metadata space that was not used and will | |
5881 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
5882 | * list if there are no delalloc bytes left. | |
5883 | * Also it will handle the qgroup reserved space. | |
5884 | */ | |
bc42bda2 | 5885 | void btrfs_delalloc_release_space(struct inode *inode, |
8b62f87b | 5886 | struct extent_changeset *reserved, |
43b18595 | 5887 | u64 start, u64 len, bool qgroup_free) |
1ada3a62 | 5888 | { |
43b18595 | 5889 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free); |
bc42bda2 | 5890 | btrfs_free_reserved_data_space(inode, reserved, start, len); |
6324fbf3 CM |
5891 | } |
5892 | ||
ce93ec54 | 5893 | static int update_block_group(struct btrfs_trans_handle *trans, |
6b279408 | 5894 | u64 bytenr, u64 num_bytes, int alloc) |
9078a3e1 | 5895 | { |
6b279408 | 5896 | struct btrfs_fs_info *info = trans->fs_info; |
0af3d00b | 5897 | struct btrfs_block_group_cache *cache = NULL; |
db94535d | 5898 | u64 total = num_bytes; |
9078a3e1 | 5899 | u64 old_val; |
db94535d | 5900 | u64 byte_in_group; |
0af3d00b | 5901 | int factor; |
ba2c4d4e | 5902 | int ret = 0; |
3e1ad54f | 5903 | |
5d4f98a2 | 5904 | /* block accounting for super block */ |
eb73c1b7 | 5905 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 5906 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
5907 | if (alloc) |
5908 | old_val += num_bytes; | |
5909 | else | |
5910 | old_val -= num_bytes; | |
6c41761f | 5911 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 5912 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 5913 | |
d397712b | 5914 | while (total) { |
db94535d | 5915 | cache = btrfs_lookup_block_group(info, bytenr); |
ba2c4d4e JB |
5916 | if (!cache) { |
5917 | ret = -ENOENT; | |
5918 | break; | |
5919 | } | |
46df06b8 DS |
5920 | factor = btrfs_bg_type_to_factor(cache->flags); |
5921 | ||
9d66e233 JB |
5922 | /* |
5923 | * If this block group has free space cache written out, we | |
5924 | * need to make sure to load it if we are removing space. This | |
5925 | * is because we need the unpinning stage to actually add the | |
5926 | * space back to the block group, otherwise we will leak space. | |
5927 | */ | |
5928 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 5929 | cache_block_group(cache, 1); |
0af3d00b | 5930 | |
db94535d CM |
5931 | byte_in_group = bytenr - cache->key.objectid; |
5932 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 5933 | |
25179201 | 5934 | spin_lock(&cache->space_info->lock); |
c286ac48 | 5935 | spin_lock(&cache->lock); |
0af3d00b | 5936 | |
6202df69 | 5937 | if (btrfs_test_opt(info, SPACE_CACHE) && |
0af3d00b JB |
5938 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
5939 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
5940 | ||
9078a3e1 | 5941 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 5942 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 5943 | if (alloc) { |
db94535d | 5944 | old_val += num_bytes; |
11833d66 YZ |
5945 | btrfs_set_block_group_used(&cache->item, old_val); |
5946 | cache->reserved -= num_bytes; | |
11833d66 | 5947 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
5948 | cache->space_info->bytes_used += num_bytes; |
5949 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 5950 | spin_unlock(&cache->lock); |
25179201 | 5951 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 5952 | } else { |
db94535d | 5953 | old_val -= num_bytes; |
ae0ab003 FM |
5954 | btrfs_set_block_group_used(&cache->item, old_val); |
5955 | cache->pinned += num_bytes; | |
bb96c4e5 JB |
5956 | btrfs_space_info_update_bytes_pinned(info, |
5957 | cache->space_info, num_bytes); | |
ae0ab003 FM |
5958 | cache->space_info->bytes_used -= num_bytes; |
5959 | cache->space_info->disk_used -= num_bytes * factor; | |
5960 | spin_unlock(&cache->lock); | |
5961 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 5962 | |
0b246afa | 5963 | trace_btrfs_space_reservation(info, "pinned", |
c51e7bb1 JB |
5964 | cache->space_info->flags, |
5965 | num_bytes, 1); | |
dec59fa3 EL |
5966 | percpu_counter_add_batch(&cache->space_info->total_bytes_pinned, |
5967 | num_bytes, | |
5968 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
ae0ab003 FM |
5969 | set_extent_dirty(info->pinned_extents, |
5970 | bytenr, bytenr + num_bytes - 1, | |
5971 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 5972 | } |
1bbc621e CM |
5973 | |
5974 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
5975 | if (list_empty(&cache->dirty_list)) { | |
5976 | list_add_tail(&cache->dirty_list, | |
5977 | &trans->transaction->dirty_bgs); | |
ba2c4d4e | 5978 | trans->delayed_ref_updates++; |
1bbc621e CM |
5979 | btrfs_get_block_group(cache); |
5980 | } | |
5981 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
5982 | ||
036a9348 FM |
5983 | /* |
5984 | * No longer have used bytes in this block group, queue it for | |
5985 | * deletion. We do this after adding the block group to the | |
5986 | * dirty list to avoid races between cleaner kthread and space | |
5987 | * cache writeout. | |
5988 | */ | |
031f24da QW |
5989 | if (!alloc && old_val == 0) |
5990 | btrfs_mark_bg_unused(cache); | |
036a9348 | 5991 | |
fa9c0d79 | 5992 | btrfs_put_block_group(cache); |
db94535d CM |
5993 | total -= num_bytes; |
5994 | bytenr += num_bytes; | |
9078a3e1 | 5995 | } |
ba2c4d4e JB |
5996 | |
5997 | /* Modified block groups are accounted for in the delayed_refs_rsv. */ | |
5998 | btrfs_update_delayed_refs_rsv(trans); | |
5999 | return ret; | |
9078a3e1 | 6000 | } |
6324fbf3 | 6001 | |
2ff7e61e | 6002 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 6003 | { |
0f9dd46c | 6004 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 6005 | u64 bytenr; |
0f9dd46c | 6006 | |
0b246afa JM |
6007 | spin_lock(&fs_info->block_group_cache_lock); |
6008 | bytenr = fs_info->first_logical_byte; | |
6009 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
6010 | |
6011 | if (bytenr < (u64)-1) | |
6012 | return bytenr; | |
6013 | ||
0b246afa | 6014 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 6015 | if (!cache) |
a061fc8d | 6016 | return 0; |
0f9dd46c | 6017 | |
d2fb3437 | 6018 | bytenr = cache->key.objectid; |
fa9c0d79 | 6019 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
6020 | |
6021 | return bytenr; | |
a061fc8d CM |
6022 | } |
6023 | ||
fdf08605 | 6024 | static int pin_down_extent(struct btrfs_block_group_cache *cache, |
f0486c68 | 6025 | u64 bytenr, u64 num_bytes, int reserved) |
324ae4df | 6026 | { |
fdf08605 DS |
6027 | struct btrfs_fs_info *fs_info = cache->fs_info; |
6028 | ||
11833d66 YZ |
6029 | spin_lock(&cache->space_info->lock); |
6030 | spin_lock(&cache->lock); | |
6031 | cache->pinned += num_bytes; | |
bb96c4e5 JB |
6032 | btrfs_space_info_update_bytes_pinned(fs_info, cache->space_info, |
6033 | num_bytes); | |
11833d66 YZ |
6034 | if (reserved) { |
6035 | cache->reserved -= num_bytes; | |
6036 | cache->space_info->bytes_reserved -= num_bytes; | |
6037 | } | |
6038 | spin_unlock(&cache->lock); | |
6039 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 6040 | |
0b246afa | 6041 | trace_btrfs_space_reservation(fs_info, "pinned", |
c51e7bb1 | 6042 | cache->space_info->flags, num_bytes, 1); |
dec59fa3 EL |
6043 | percpu_counter_add_batch(&cache->space_info->total_bytes_pinned, |
6044 | num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
0b246afa | 6045 | set_extent_dirty(fs_info->pinned_extents, bytenr, |
f0486c68 YZ |
6046 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
6047 | return 0; | |
6048 | } | |
68b38550 | 6049 | |
f0486c68 YZ |
6050 | /* |
6051 | * this function must be called within transaction | |
6052 | */ | |
2ff7e61e | 6053 | int btrfs_pin_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6054 | u64 bytenr, u64 num_bytes, int reserved) |
6055 | { | |
6056 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6057 | |
0b246afa | 6058 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
79787eaa | 6059 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 6060 | |
fdf08605 | 6061 | pin_down_extent(cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
6062 | |
6063 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6064 | return 0; |
6065 | } | |
6066 | ||
f0486c68 | 6067 | /* |
e688b725 CM |
6068 | * this function must be called within transaction |
6069 | */ | |
2ff7e61e | 6070 | int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info, |
e688b725 CM |
6071 | u64 bytenr, u64 num_bytes) |
6072 | { | |
6073 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6074 | int ret; |
e688b725 | 6075 | |
0b246afa | 6076 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
b50c6e25 JB |
6077 | if (!cache) |
6078 | return -EINVAL; | |
e688b725 CM |
6079 | |
6080 | /* | |
6081 | * pull in the free space cache (if any) so that our pin | |
6082 | * removes the free space from the cache. We have load_only set | |
6083 | * to one because the slow code to read in the free extents does check | |
6084 | * the pinned extents. | |
6085 | */ | |
f6373bf3 | 6086 | cache_block_group(cache, 1); |
e688b725 | 6087 | |
fdf08605 | 6088 | pin_down_extent(cache, bytenr, num_bytes, 0); |
e688b725 CM |
6089 | |
6090 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6091 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6092 | btrfs_put_block_group(cache); |
b50c6e25 | 6093 | return ret; |
e688b725 CM |
6094 | } |
6095 | ||
2ff7e61e JM |
6096 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
6097 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
6098 | { |
6099 | int ret; | |
6100 | struct btrfs_block_group_cache *block_group; | |
6101 | struct btrfs_caching_control *caching_ctl; | |
6102 | ||
0b246afa | 6103 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
6104 | if (!block_group) |
6105 | return -EINVAL; | |
6106 | ||
6107 | cache_block_group(block_group, 0); | |
6108 | caching_ctl = get_caching_control(block_group); | |
6109 | ||
6110 | if (!caching_ctl) { | |
6111 | /* Logic error */ | |
6112 | BUG_ON(!block_group_cache_done(block_group)); | |
6113 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6114 | } else { | |
6115 | mutex_lock(&caching_ctl->mutex); | |
6116 | ||
6117 | if (start >= caching_ctl->progress) { | |
2ff7e61e | 6118 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6119 | } else if (start + num_bytes <= caching_ctl->progress) { |
6120 | ret = btrfs_remove_free_space(block_group, | |
6121 | start, num_bytes); | |
6122 | } else { | |
6123 | num_bytes = caching_ctl->progress - start; | |
6124 | ret = btrfs_remove_free_space(block_group, | |
6125 | start, num_bytes); | |
6126 | if (ret) | |
6127 | goto out_lock; | |
6128 | ||
6129 | num_bytes = (start + num_bytes) - | |
6130 | caching_ctl->progress; | |
6131 | start = caching_ctl->progress; | |
2ff7e61e | 6132 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6133 | } |
6134 | out_lock: | |
6135 | mutex_unlock(&caching_ctl->mutex); | |
6136 | put_caching_control(caching_ctl); | |
6137 | } | |
6138 | btrfs_put_block_group(block_group); | |
6139 | return ret; | |
6140 | } | |
6141 | ||
bcdc428c | 6142 | int btrfs_exclude_logged_extents(struct extent_buffer *eb) |
8c2a1a30 | 6143 | { |
bcdc428c | 6144 | struct btrfs_fs_info *fs_info = eb->fs_info; |
8c2a1a30 JB |
6145 | struct btrfs_file_extent_item *item; |
6146 | struct btrfs_key key; | |
6147 | int found_type; | |
6148 | int i; | |
b89311ef | 6149 | int ret = 0; |
8c2a1a30 | 6150 | |
2ff7e61e | 6151 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
6152 | return 0; |
6153 | ||
6154 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6155 | btrfs_item_key_to_cpu(eb, &key, i); | |
6156 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6157 | continue; | |
6158 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6159 | found_type = btrfs_file_extent_type(eb, item); | |
6160 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6161 | continue; | |
6162 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6163 | continue; | |
6164 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6165 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
b89311ef GJ |
6166 | ret = __exclude_logged_extent(fs_info, key.objectid, key.offset); |
6167 | if (ret) | |
6168 | break; | |
8c2a1a30 JB |
6169 | } |
6170 | ||
b89311ef | 6171 | return ret; |
8c2a1a30 JB |
6172 | } |
6173 | ||
9cfa3e34 FM |
6174 | static void |
6175 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6176 | { | |
6177 | atomic_inc(&bg->reservations); | |
6178 | } | |
6179 | ||
6180 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
6181 | const u64 start) | |
6182 | { | |
6183 | struct btrfs_block_group_cache *bg; | |
6184 | ||
6185 | bg = btrfs_lookup_block_group(fs_info, start); | |
6186 | ASSERT(bg); | |
6187 | if (atomic_dec_and_test(&bg->reservations)) | |
4625956a | 6188 | wake_up_var(&bg->reservations); |
9cfa3e34 FM |
6189 | btrfs_put_block_group(bg); |
6190 | } | |
6191 | ||
9cfa3e34 FM |
6192 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) |
6193 | { | |
6194 | struct btrfs_space_info *space_info = bg->space_info; | |
6195 | ||
6196 | ASSERT(bg->ro); | |
6197 | ||
6198 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
6199 | return; | |
6200 | ||
6201 | /* | |
6202 | * Our block group is read only but before we set it to read only, | |
6203 | * some task might have had allocated an extent from it already, but it | |
6204 | * has not yet created a respective ordered extent (and added it to a | |
6205 | * root's list of ordered extents). | |
6206 | * Therefore wait for any task currently allocating extents, since the | |
6207 | * block group's reservations counter is incremented while a read lock | |
6208 | * on the groups' semaphore is held and decremented after releasing | |
6209 | * the read access on that semaphore and creating the ordered extent. | |
6210 | */ | |
6211 | down_write(&space_info->groups_sem); | |
6212 | up_write(&space_info->groups_sem); | |
6213 | ||
4625956a | 6214 | wait_var_event(&bg->reservations, !atomic_read(&bg->reservations)); |
9cfa3e34 FM |
6215 | } |
6216 | ||
fb25e914 | 6217 | /** |
4824f1f4 | 6218 | * btrfs_add_reserved_bytes - update the block_group and space info counters |
fb25e914 | 6219 | * @cache: The cache we are manipulating |
18513091 WX |
6220 | * @ram_bytes: The number of bytes of file content, and will be same to |
6221 | * @num_bytes except for the compress path. | |
fb25e914 | 6222 | * @num_bytes: The number of bytes in question |
e570fd27 | 6223 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 | 6224 | * |
745699ef XW |
6225 | * This is called by the allocator when it reserves space. If this is a |
6226 | * reservation and the block group has become read only we cannot make the | |
6227 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
f0486c68 | 6228 | */ |
4824f1f4 | 6229 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 6230 | u64 ram_bytes, u64 num_bytes, int delalloc) |
11833d66 | 6231 | { |
fb25e914 | 6232 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6233 | int ret = 0; |
79787eaa | 6234 | |
fb25e914 JB |
6235 | spin_lock(&space_info->lock); |
6236 | spin_lock(&cache->lock); | |
4824f1f4 WX |
6237 | if (cache->ro) { |
6238 | ret = -EAGAIN; | |
fb25e914 | 6239 | } else { |
4824f1f4 WX |
6240 | cache->reserved += num_bytes; |
6241 | space_info->bytes_reserved += num_bytes; | |
bb96c4e5 JB |
6242 | btrfs_space_info_update_bytes_may_use(cache->fs_info, |
6243 | space_info, -ram_bytes); | |
e570fd27 | 6244 | if (delalloc) |
4824f1f4 | 6245 | cache->delalloc_bytes += num_bytes; |
324ae4df | 6246 | } |
fb25e914 JB |
6247 | spin_unlock(&cache->lock); |
6248 | spin_unlock(&space_info->lock); | |
f0486c68 | 6249 | return ret; |
324ae4df | 6250 | } |
9078a3e1 | 6251 | |
4824f1f4 WX |
6252 | /** |
6253 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
6254 | * @cache: The cache we are manipulating | |
6255 | * @num_bytes: The number of bytes in question | |
6256 | * @delalloc: The blocks are allocated for the delalloc write | |
6257 | * | |
6258 | * This is called by somebody who is freeing space that was never actually used | |
6259 | * on disk. For example if you reserve some space for a new leaf in transaction | |
6260 | * A and before transaction A commits you free that leaf, you call this with | |
6261 | * reserve set to 0 in order to clear the reservation. | |
6262 | */ | |
6263 | ||
556f3ca8 | 6264 | static void btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, |
6265 | u64 num_bytes, int delalloc) | |
4824f1f4 WX |
6266 | { |
6267 | struct btrfs_space_info *space_info = cache->space_info; | |
4824f1f4 WX |
6268 | |
6269 | spin_lock(&space_info->lock); | |
6270 | spin_lock(&cache->lock); | |
6271 | if (cache->ro) | |
6272 | space_info->bytes_readonly += num_bytes; | |
6273 | cache->reserved -= num_bytes; | |
6274 | space_info->bytes_reserved -= num_bytes; | |
21a94f7a | 6275 | space_info->max_extent_size = 0; |
4824f1f4 WX |
6276 | |
6277 | if (delalloc) | |
6278 | cache->delalloc_bytes -= num_bytes; | |
6279 | spin_unlock(&cache->lock); | |
6280 | spin_unlock(&space_info->lock); | |
4824f1f4 | 6281 | } |
8b74c03e | 6282 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 6283 | { |
11833d66 YZ |
6284 | struct btrfs_caching_control *next; |
6285 | struct btrfs_caching_control *caching_ctl; | |
6286 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6287 | |
9e351cc8 | 6288 | down_write(&fs_info->commit_root_sem); |
25179201 | 6289 | |
11833d66 YZ |
6290 | list_for_each_entry_safe(caching_ctl, next, |
6291 | &fs_info->caching_block_groups, list) { | |
6292 | cache = caching_ctl->block_group; | |
6293 | if (block_group_cache_done(cache)) { | |
6294 | cache->last_byte_to_unpin = (u64)-1; | |
6295 | list_del_init(&caching_ctl->list); | |
6296 | put_caching_control(caching_ctl); | |
e8569813 | 6297 | } else { |
11833d66 | 6298 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6299 | } |
e8569813 | 6300 | } |
11833d66 YZ |
6301 | |
6302 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6303 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6304 | else | |
6305 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6306 | ||
9e351cc8 | 6307 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6308 | |
6309 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6310 | } |
6311 | ||
c759c4e1 JB |
6312 | /* |
6313 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6314 | * what it should be based on the mount options. | |
6315 | */ | |
6316 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
6317 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
6318 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
6319 | { |
6320 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
6321 | |
6322 | *empty_cluster = 0; | |
6323 | if (btrfs_mixed_space_info(space_info)) | |
6324 | return ret; | |
6325 | ||
c759c4e1 | 6326 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 6327 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
6328 | if (btrfs_test_opt(fs_info, SSD)) |
6329 | *empty_cluster = SZ_2M; | |
6330 | else | |
ee22184b | 6331 | *empty_cluster = SZ_64K; |
583b7231 HK |
6332 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
6333 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
6334 | *empty_cluster = SZ_2M; | |
0b246afa | 6335 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
6336 | } |
6337 | ||
6338 | return ret; | |
6339 | } | |
6340 | ||
2ff7e61e JM |
6341 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
6342 | u64 start, u64 end, | |
678886bd | 6343 | const bool return_free_space) |
ccd467d6 | 6344 | { |
11833d66 | 6345 | struct btrfs_block_group_cache *cache = NULL; |
7b398f8e JB |
6346 | struct btrfs_space_info *space_info; |
6347 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6348 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6349 | u64 len; |
c759c4e1 JB |
6350 | u64 total_unpinned = 0; |
6351 | u64 empty_cluster = 0; | |
7b398f8e | 6352 | bool readonly; |
ccd467d6 | 6353 | |
11833d66 | 6354 | while (start <= end) { |
7b398f8e | 6355 | readonly = false; |
11833d66 YZ |
6356 | if (!cache || |
6357 | start >= cache->key.objectid + cache->key.offset) { | |
6358 | if (cache) | |
6359 | btrfs_put_block_group(cache); | |
c759c4e1 | 6360 | total_unpinned = 0; |
11833d66 | 6361 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6362 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 6363 | |
2ff7e61e | 6364 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
6365 | cache->space_info, |
6366 | &empty_cluster); | |
6367 | empty_cluster <<= 1; | |
11833d66 YZ |
6368 | } |
6369 | ||
6370 | len = cache->key.objectid + cache->key.offset - start; | |
6371 | len = min(len, end + 1 - start); | |
6372 | ||
6373 | if (start < cache->last_byte_to_unpin) { | |
6374 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6375 | if (return_free_space) |
6376 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6377 | } |
6378 | ||
f0486c68 | 6379 | start += len; |
c759c4e1 | 6380 | total_unpinned += len; |
7b398f8e | 6381 | space_info = cache->space_info; |
f0486c68 | 6382 | |
c759c4e1 JB |
6383 | /* |
6384 | * If this space cluster has been marked as fragmented and we've | |
6385 | * unpinned enough in this block group to potentially allow a | |
6386 | * cluster to be created inside of it go ahead and clear the | |
6387 | * fragmented check. | |
6388 | */ | |
6389 | if (cluster && cluster->fragmented && | |
6390 | total_unpinned > empty_cluster) { | |
6391 | spin_lock(&cluster->lock); | |
6392 | cluster->fragmented = 0; | |
6393 | spin_unlock(&cluster->lock); | |
6394 | } | |
6395 | ||
7b398f8e | 6396 | spin_lock(&space_info->lock); |
11833d66 YZ |
6397 | spin_lock(&cache->lock); |
6398 | cache->pinned -= len; | |
bb96c4e5 | 6399 | btrfs_space_info_update_bytes_pinned(fs_info, space_info, -len); |
c51e7bb1 JB |
6400 | |
6401 | trace_btrfs_space_reservation(fs_info, "pinned", | |
6402 | space_info->flags, len, 0); | |
4f4db217 | 6403 | space_info->max_extent_size = 0; |
dec59fa3 EL |
6404 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
6405 | -len, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
7b398f8e JB |
6406 | if (cache->ro) { |
6407 | space_info->bytes_readonly += len; | |
6408 | readonly = true; | |
6409 | } | |
11833d66 | 6410 | spin_unlock(&cache->lock); |
957780eb JB |
6411 | if (!readonly && return_free_space && |
6412 | global_rsv->space_info == space_info) { | |
6413 | u64 to_add = len; | |
92ac58ec | 6414 | |
7b398f8e JB |
6415 | spin_lock(&global_rsv->lock); |
6416 | if (!global_rsv->full) { | |
957780eb JB |
6417 | to_add = min(len, global_rsv->size - |
6418 | global_rsv->reserved); | |
6419 | global_rsv->reserved += to_add; | |
bb96c4e5 JB |
6420 | btrfs_space_info_update_bytes_may_use(fs_info, |
6421 | space_info, to_add); | |
7b398f8e JB |
6422 | if (global_rsv->reserved >= global_rsv->size) |
6423 | global_rsv->full = 1; | |
957780eb JB |
6424 | trace_btrfs_space_reservation(fs_info, |
6425 | "space_info", | |
6426 | space_info->flags, | |
6427 | to_add, 1); | |
6428 | len -= to_add; | |
7b398f8e JB |
6429 | } |
6430 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
6431 | /* Add to any tickets we may have */ |
6432 | if (len) | |
d44b72aa JB |
6433 | btrfs_space_info_add_new_bytes(fs_info, |
6434 | space_info, len); | |
7b398f8e JB |
6435 | } |
6436 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6437 | } |
11833d66 YZ |
6438 | |
6439 | if (cache) | |
6440 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6441 | return 0; |
6442 | } | |
6443 | ||
5ead2dd0 | 6444 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 6445 | { |
5ead2dd0 | 6446 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e33e17ee JM |
6447 | struct btrfs_block_group_cache *block_group, *tmp; |
6448 | struct list_head *deleted_bgs; | |
11833d66 | 6449 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6450 | u64 start; |
6451 | u64 end; | |
a28ec197 | 6452 | int ret; |
a28ec197 | 6453 | |
11833d66 YZ |
6454 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6455 | unpin = &fs_info->freed_extents[1]; | |
6456 | else | |
6457 | unpin = &fs_info->freed_extents[0]; | |
6458 | ||
e33e17ee | 6459 | while (!trans->aborted) { |
0e6ec385 FM |
6460 | struct extent_state *cached_state = NULL; |
6461 | ||
d4b450cd | 6462 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6463 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
0e6ec385 | 6464 | EXTENT_DIRTY, &cached_state); |
d4b450cd FM |
6465 | if (ret) { |
6466 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6467 | break; |
d4b450cd | 6468 | } |
1f3c79a2 | 6469 | |
0b246afa | 6470 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 6471 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 6472 | end + 1 - start, NULL); |
1f3c79a2 | 6473 | |
0e6ec385 | 6474 | clear_extent_dirty(unpin, start, end, &cached_state); |
2ff7e61e | 6475 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 6476 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
0e6ec385 | 6477 | free_extent_state(cached_state); |
b9473439 | 6478 | cond_resched(); |
a28ec197 | 6479 | } |
817d52f8 | 6480 | |
e33e17ee JM |
6481 | /* |
6482 | * Transaction is finished. We don't need the lock anymore. We | |
6483 | * do need to clean up the block groups in case of a transaction | |
6484 | * abort. | |
6485 | */ | |
6486 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6487 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6488 | u64 trimmed = 0; | |
6489 | ||
6490 | ret = -EROFS; | |
6491 | if (!trans->aborted) | |
2ff7e61e | 6492 | ret = btrfs_discard_extent(fs_info, |
e33e17ee JM |
6493 | block_group->key.objectid, |
6494 | block_group->key.offset, | |
6495 | &trimmed); | |
6496 | ||
6497 | list_del_init(&block_group->bg_list); | |
6498 | btrfs_put_block_group_trimming(block_group); | |
6499 | btrfs_put_block_group(block_group); | |
6500 | ||
6501 | if (ret) { | |
6502 | const char *errstr = btrfs_decode_error(ret); | |
6503 | btrfs_warn(fs_info, | |
913e1535 | 6504 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
6505 | ret, errstr); |
6506 | } | |
6507 | } | |
6508 | ||
e20d96d6 CM |
6509 | return 0; |
6510 | } | |
6511 | ||
5d4f98a2 | 6512 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
6513 | struct btrfs_delayed_ref_node *node, u64 parent, |
6514 | u64 root_objectid, u64 owner_objectid, | |
6515 | u64 owner_offset, int refs_to_drop, | |
6516 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 6517 | { |
e72cb923 | 6518 | struct btrfs_fs_info *info = trans->fs_info; |
e2fa7227 | 6519 | struct btrfs_key key; |
5d4f98a2 | 6520 | struct btrfs_path *path; |
1261ec42 | 6521 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 6522 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6523 | struct btrfs_extent_item *ei; |
6524 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6525 | int ret; |
5d4f98a2 | 6526 | int is_data; |
952fccac CM |
6527 | int extent_slot = 0; |
6528 | int found_extent = 0; | |
6529 | int num_to_del = 1; | |
5d4f98a2 YZ |
6530 | u32 item_size; |
6531 | u64 refs; | |
c682f9b3 QW |
6532 | u64 bytenr = node->bytenr; |
6533 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6534 | int last_ref = 0; |
0b246afa | 6535 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 6536 | |
5caf2a00 | 6537 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6538 | if (!path) |
6539 | return -ENOMEM; | |
5f26f772 | 6540 | |
e4058b54 | 6541 | path->reada = READA_FORWARD; |
b9473439 | 6542 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6543 | |
6544 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6545 | BUG_ON(!is_data && refs_to_drop != 1); | |
6546 | ||
3173a18f | 6547 | if (is_data) |
897ca819 | 6548 | skinny_metadata = false; |
3173a18f | 6549 | |
fbe4801b NB |
6550 | ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes, |
6551 | parent, root_objectid, owner_objectid, | |
5d4f98a2 | 6552 | owner_offset); |
7bb86316 | 6553 | if (ret == 0) { |
952fccac | 6554 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6555 | while (extent_slot >= 0) { |
6556 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6557 | extent_slot); |
5d4f98a2 | 6558 | if (key.objectid != bytenr) |
952fccac | 6559 | break; |
5d4f98a2 YZ |
6560 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6561 | key.offset == num_bytes) { | |
952fccac CM |
6562 | found_extent = 1; |
6563 | break; | |
6564 | } | |
3173a18f JB |
6565 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
6566 | key.offset == owner_objectid) { | |
6567 | found_extent = 1; | |
6568 | break; | |
6569 | } | |
952fccac CM |
6570 | if (path->slots[0] - extent_slot > 5) |
6571 | break; | |
5d4f98a2 | 6572 | extent_slot--; |
952fccac | 6573 | } |
a79865c6 | 6574 | |
31840ae1 | 6575 | if (!found_extent) { |
5d4f98a2 | 6576 | BUG_ON(iref); |
87cc7a8a | 6577 | ret = remove_extent_backref(trans, path, NULL, |
87bde3cd | 6578 | refs_to_drop, |
fcebe456 | 6579 | is_data, &last_ref); |
005d6427 | 6580 | if (ret) { |
66642832 | 6581 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6582 | goto out; |
6583 | } | |
b3b4aa74 | 6584 | btrfs_release_path(path); |
b9473439 | 6585 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6586 | |
6587 | key.objectid = bytenr; | |
6588 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6589 | key.offset = num_bytes; | |
6590 | ||
3173a18f JB |
6591 | if (!is_data && skinny_metadata) { |
6592 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6593 | key.offset = owner_objectid; | |
6594 | } | |
6595 | ||
31840ae1 ZY |
6596 | ret = btrfs_search_slot(trans, extent_root, |
6597 | &key, path, -1, 1); | |
3173a18f JB |
6598 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
6599 | /* | |
6600 | * Couldn't find our skinny metadata item, | |
6601 | * see if we have ye olde extent item. | |
6602 | */ | |
6603 | path->slots[0]--; | |
6604 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6605 | path->slots[0]); | |
6606 | if (key.objectid == bytenr && | |
6607 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6608 | key.offset == num_bytes) | |
6609 | ret = 0; | |
6610 | } | |
6611 | ||
6612 | if (ret > 0 && skinny_metadata) { | |
6613 | skinny_metadata = false; | |
9ce49a0b | 6614 | key.objectid = bytenr; |
3173a18f JB |
6615 | key.type = BTRFS_EXTENT_ITEM_KEY; |
6616 | key.offset = num_bytes; | |
6617 | btrfs_release_path(path); | |
6618 | ret = btrfs_search_slot(trans, extent_root, | |
6619 | &key, path, -1, 1); | |
6620 | } | |
6621 | ||
f3465ca4 | 6622 | if (ret) { |
5d163e0e JM |
6623 | btrfs_err(info, |
6624 | "umm, got %d back from search, was looking for %llu", | |
6625 | ret, bytenr); | |
b783e62d | 6626 | if (ret > 0) |
a4f78750 | 6627 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 6628 | } |
005d6427 | 6629 | if (ret < 0) { |
66642832 | 6630 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6631 | goto out; |
6632 | } | |
31840ae1 ZY |
6633 | extent_slot = path->slots[0]; |
6634 | } | |
fae7f21c | 6635 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 6636 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
6637 | btrfs_err(info, |
6638 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
6639 | bytenr, parent, root_objectid, owner_objectid, |
6640 | owner_offset); | |
66642832 | 6641 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 6642 | goto out; |
79787eaa | 6643 | } else { |
66642832 | 6644 | btrfs_abort_transaction(trans, ret); |
005d6427 | 6645 | goto out; |
7bb86316 | 6646 | } |
5f39d397 CM |
6647 | |
6648 | leaf = path->nodes[0]; | |
5d4f98a2 | 6649 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6d8ff4e4 | 6650 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
6651 | ret = -EINVAL; |
6652 | btrfs_print_v0_err(info); | |
6653 | btrfs_abort_transaction(trans, ret); | |
6654 | goto out; | |
6655 | } | |
952fccac | 6656 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 6657 | struct btrfs_extent_item); |
3173a18f JB |
6658 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
6659 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
6660 | struct btrfs_tree_block_info *bi; |
6661 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
6662 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
6663 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
6664 | } | |
56bec294 | 6665 | |
5d4f98a2 | 6666 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 6667 | if (refs < refs_to_drop) { |
5d163e0e JM |
6668 | btrfs_err(info, |
6669 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
6670 | refs_to_drop, refs, bytenr); | |
32b02538 | 6671 | ret = -EINVAL; |
66642832 | 6672 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
6673 | goto out; |
6674 | } | |
56bec294 | 6675 | refs -= refs_to_drop; |
5f39d397 | 6676 | |
5d4f98a2 YZ |
6677 | if (refs > 0) { |
6678 | if (extent_op) | |
6679 | __run_delayed_extent_op(extent_op, leaf, ei); | |
6680 | /* | |
6681 | * In the case of inline back ref, reference count will | |
6682 | * be updated by remove_extent_backref | |
952fccac | 6683 | */ |
5d4f98a2 YZ |
6684 | if (iref) { |
6685 | BUG_ON(!found_extent); | |
6686 | } else { | |
6687 | btrfs_set_extent_refs(leaf, ei, refs); | |
6688 | btrfs_mark_buffer_dirty(leaf); | |
6689 | } | |
6690 | if (found_extent) { | |
87cc7a8a NB |
6691 | ret = remove_extent_backref(trans, path, iref, |
6692 | refs_to_drop, is_data, | |
6693 | &last_ref); | |
005d6427 | 6694 | if (ret) { |
66642832 | 6695 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6696 | goto out; |
6697 | } | |
952fccac | 6698 | } |
5d4f98a2 | 6699 | } else { |
5d4f98a2 YZ |
6700 | if (found_extent) { |
6701 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 6702 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
6703 | if (iref) { |
6704 | BUG_ON(path->slots[0] != extent_slot); | |
6705 | } else { | |
6706 | BUG_ON(path->slots[0] != extent_slot + 1); | |
6707 | path->slots[0] = extent_slot; | |
6708 | num_to_del = 2; | |
6709 | } | |
78fae27e | 6710 | } |
b9473439 | 6711 | |
fcebe456 | 6712 | last_ref = 1; |
952fccac CM |
6713 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
6714 | num_to_del); | |
005d6427 | 6715 | if (ret) { |
66642832 | 6716 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6717 | goto out; |
6718 | } | |
b3b4aa74 | 6719 | btrfs_release_path(path); |
21af804c | 6720 | |
5d4f98a2 | 6721 | if (is_data) { |
5b4aacef | 6722 | ret = btrfs_del_csums(trans, info, bytenr, num_bytes); |
005d6427 | 6723 | if (ret) { |
66642832 | 6724 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6725 | goto out; |
6726 | } | |
459931ec CM |
6727 | } |
6728 | ||
e7355e50 | 6729 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
1e144fb8 | 6730 | if (ret) { |
66642832 | 6731 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
6732 | goto out; |
6733 | } | |
6734 | ||
6b279408 | 6735 | ret = update_block_group(trans, bytenr, num_bytes, 0); |
005d6427 | 6736 | if (ret) { |
66642832 | 6737 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6738 | goto out; |
6739 | } | |
a28ec197 | 6740 | } |
fcebe456 JB |
6741 | btrfs_release_path(path); |
6742 | ||
79787eaa | 6743 | out: |
5caf2a00 | 6744 | btrfs_free_path(path); |
a28ec197 CM |
6745 | return ret; |
6746 | } | |
6747 | ||
1887be66 | 6748 | /* |
f0486c68 | 6749 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
6750 | * delayed ref for that extent as well. This searches the delayed ref tree for |
6751 | * a given extent, and if there are no other delayed refs to be processed, it | |
6752 | * removes it from the tree. | |
6753 | */ | |
6754 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 6755 | u64 bytenr) |
1887be66 CM |
6756 | { |
6757 | struct btrfs_delayed_ref_head *head; | |
6758 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 6759 | int ret = 0; |
1887be66 CM |
6760 | |
6761 | delayed_refs = &trans->transaction->delayed_refs; | |
6762 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 6763 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 6764 | if (!head) |
cf93da7b | 6765 | goto out_delayed_unlock; |
1887be66 | 6766 | |
d7df2c79 | 6767 | spin_lock(&head->lock); |
e3d03965 | 6768 | if (!RB_EMPTY_ROOT(&head->ref_tree.rb_root)) |
1887be66 CM |
6769 | goto out; |
6770 | ||
bedc6617 JB |
6771 | if (cleanup_extent_op(head) != NULL) |
6772 | goto out; | |
5d4f98a2 | 6773 | |
1887be66 CM |
6774 | /* |
6775 | * waiting for the lock here would deadlock. If someone else has it | |
6776 | * locked they are already in the process of dropping it anyway | |
6777 | */ | |
6778 | if (!mutex_trylock(&head->mutex)) | |
6779 | goto out; | |
6780 | ||
d7baffda | 6781 | btrfs_delete_ref_head(delayed_refs, head); |
d7df2c79 | 6782 | head->processing = 0; |
d7baffda | 6783 | |
d7df2c79 | 6784 | spin_unlock(&head->lock); |
1887be66 CM |
6785 | spin_unlock(&delayed_refs->lock); |
6786 | ||
f0486c68 YZ |
6787 | BUG_ON(head->extent_op); |
6788 | if (head->must_insert_reserved) | |
6789 | ret = 1; | |
6790 | ||
31890da0 | 6791 | btrfs_cleanup_ref_head_accounting(trans->fs_info, delayed_refs, head); |
f0486c68 | 6792 | mutex_unlock(&head->mutex); |
d278850e | 6793 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 6794 | return ret; |
1887be66 | 6795 | out: |
d7df2c79 | 6796 | spin_unlock(&head->lock); |
cf93da7b CM |
6797 | |
6798 | out_delayed_unlock: | |
1887be66 CM |
6799 | spin_unlock(&delayed_refs->lock); |
6800 | return 0; | |
6801 | } | |
6802 | ||
f0486c68 YZ |
6803 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
6804 | struct btrfs_root *root, | |
6805 | struct extent_buffer *buf, | |
5581a51a | 6806 | u64 parent, int last_ref) |
f0486c68 | 6807 | { |
0b246afa | 6808 | struct btrfs_fs_info *fs_info = root->fs_info; |
ed4f255b | 6809 | struct btrfs_ref generic_ref = { 0 }; |
b150a4f1 | 6810 | int pin = 1; |
f0486c68 YZ |
6811 | int ret; |
6812 | ||
ed4f255b QW |
6813 | btrfs_init_generic_ref(&generic_ref, BTRFS_DROP_DELAYED_REF, |
6814 | buf->start, buf->len, parent); | |
6815 | btrfs_init_tree_ref(&generic_ref, btrfs_header_level(buf), | |
6816 | root->root_key.objectid); | |
6817 | ||
f0486c68 | 6818 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
d7eae340 OS |
6819 | int old_ref_mod, new_ref_mod; |
6820 | ||
8a5040f7 | 6821 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
ed4f255b | 6822 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, NULL, |
d7eae340 | 6823 | &old_ref_mod, &new_ref_mod); |
79787eaa | 6824 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 6825 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
6826 | } |
6827 | ||
0a16c7d7 | 6828 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
6829 | struct btrfs_block_group_cache *cache; |
6830 | ||
f0486c68 | 6831 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 6832 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 6833 | if (!ret) |
37be25bc | 6834 | goto out; |
f0486c68 YZ |
6835 | } |
6836 | ||
4da8b76d | 6837 | pin = 0; |
0b246afa | 6838 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 6839 | |
f0486c68 | 6840 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
fdf08605 | 6841 | pin_down_extent(cache, buf->start, buf->len, 1); |
6219872d | 6842 | btrfs_put_block_group(cache); |
37be25bc | 6843 | goto out; |
f0486c68 YZ |
6844 | } |
6845 | ||
6846 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
6847 | ||
6848 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 6849 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 6850 | btrfs_put_block_group(cache); |
71ff6437 | 6851 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
6852 | } |
6853 | out: | |
b150a4f1 | 6854 | if (pin) |
78192442 | 6855 | add_pinned_bytes(fs_info, &generic_ref); |
b150a4f1 | 6856 | |
0a16c7d7 OS |
6857 | if (last_ref) { |
6858 | /* | |
6859 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
6860 | * matter anymore. | |
6861 | */ | |
6862 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
6863 | } | |
f0486c68 YZ |
6864 | } |
6865 | ||
79787eaa | 6866 | /* Can return -ENOMEM */ |
ffd4bb2a | 6867 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_ref *ref) |
925baedd | 6868 | { |
ffd4bb2a | 6869 | struct btrfs_fs_info *fs_info = trans->fs_info; |
d7eae340 | 6870 | int old_ref_mod, new_ref_mod; |
925baedd CM |
6871 | int ret; |
6872 | ||
f5ee5c9a | 6873 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 6874 | return 0; |
fccb84c9 | 6875 | |
56bec294 CM |
6876 | /* |
6877 | * tree log blocks never actually go into the extent allocation | |
6878 | * tree, just update pinning info and exit early. | |
56bec294 | 6879 | */ |
ffd4bb2a QW |
6880 | if ((ref->type == BTRFS_REF_METADATA && |
6881 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
6882 | (ref->type == BTRFS_REF_DATA && | |
6883 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID)) { | |
b9473439 | 6884 | /* unlocks the pinned mutex */ |
ffd4bb2a | 6885 | btrfs_pin_extent(fs_info, ref->bytenr, ref->len, 1); |
d7eae340 | 6886 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 6887 | ret = 0; |
ffd4bb2a QW |
6888 | } else if (ref->type == BTRFS_REF_METADATA) { |
6889 | ret = btrfs_add_delayed_tree_ref(trans, ref, NULL, | |
d7eae340 | 6890 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 6891 | } else { |
ffd4bb2a | 6892 | ret = btrfs_add_delayed_data_ref(trans, ref, 0, |
d7eae340 | 6893 | &old_ref_mod, &new_ref_mod); |
56bec294 | 6894 | } |
d7eae340 | 6895 | |
ffd4bb2a QW |
6896 | if (!((ref->type == BTRFS_REF_METADATA && |
6897 | ref->tree_ref.root == BTRFS_TREE_LOG_OBJECTID) || | |
6898 | (ref->type == BTRFS_REF_DATA && | |
6899 | ref->data_ref.ref_root == BTRFS_TREE_LOG_OBJECTID))) | |
6900 | btrfs_ref_tree_mod(fs_info, ref); | |
8a5040f7 | 6901 | |
ddf30cf0 | 6902 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) |
78192442 | 6903 | add_pinned_bytes(fs_info, ref); |
d7eae340 | 6904 | |
925baedd CM |
6905 | return ret; |
6906 | } | |
6907 | ||
817d52f8 JB |
6908 | /* |
6909 | * when we wait for progress in the block group caching, its because | |
6910 | * our allocation attempt failed at least once. So, we must sleep | |
6911 | * and let some progress happen before we try again. | |
6912 | * | |
6913 | * This function will sleep at least once waiting for new free space to | |
6914 | * show up, and then it will check the block group free space numbers | |
6915 | * for our min num_bytes. Another option is to have it go ahead | |
6916 | * and look in the rbtree for a free extent of a given size, but this | |
6917 | * is a good start. | |
36cce922 JB |
6918 | * |
6919 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
6920 | * any of the information in this block group. | |
817d52f8 | 6921 | */ |
36cce922 | 6922 | static noinline void |
817d52f8 JB |
6923 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
6924 | u64 num_bytes) | |
6925 | { | |
11833d66 | 6926 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 6927 | |
11833d66 YZ |
6928 | caching_ctl = get_caching_control(cache); |
6929 | if (!caching_ctl) | |
36cce922 | 6930 | return; |
817d52f8 | 6931 | |
11833d66 | 6932 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 6933 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
6934 | |
6935 | put_caching_control(caching_ctl); | |
11833d66 YZ |
6936 | } |
6937 | ||
6938 | static noinline int | |
6939 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
6940 | { | |
6941 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 6942 | int ret = 0; |
11833d66 YZ |
6943 | |
6944 | caching_ctl = get_caching_control(cache); | |
6945 | if (!caching_ctl) | |
36cce922 | 6946 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
6947 | |
6948 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
6949 | if (cache->cached == BTRFS_CACHE_ERROR) |
6950 | ret = -EIO; | |
11833d66 | 6951 | put_caching_control(caching_ctl); |
36cce922 | 6952 | return ret; |
817d52f8 JB |
6953 | } |
6954 | ||
6955 | enum btrfs_loop_type { | |
f262fa8d DS |
6956 | LOOP_CACHING_NOWAIT, |
6957 | LOOP_CACHING_WAIT, | |
6958 | LOOP_ALLOC_CHUNK, | |
6959 | LOOP_NO_EMPTY_SIZE, | |
817d52f8 JB |
6960 | }; |
6961 | ||
e570fd27 MX |
6962 | static inline void |
6963 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
6964 | int delalloc) | |
6965 | { | |
6966 | if (delalloc) | |
6967 | down_read(&cache->data_rwsem); | |
6968 | } | |
6969 | ||
6970 | static inline void | |
6971 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
6972 | int delalloc) | |
6973 | { | |
6974 | btrfs_get_block_group(cache); | |
6975 | if (delalloc) | |
6976 | down_read(&cache->data_rwsem); | |
6977 | } | |
6978 | ||
6979 | static struct btrfs_block_group_cache * | |
6980 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
6981 | struct btrfs_free_cluster *cluster, | |
6982 | int delalloc) | |
6983 | { | |
89771cc9 | 6984 | struct btrfs_block_group_cache *used_bg = NULL; |
6719afdc | 6985 | |
e570fd27 | 6986 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
6987 | while (1) { |
6988 | used_bg = cluster->block_group; | |
6989 | if (!used_bg) | |
6990 | return NULL; | |
6991 | ||
6992 | if (used_bg == block_group) | |
e570fd27 MX |
6993 | return used_bg; |
6994 | ||
6719afdc | 6995 | btrfs_get_block_group(used_bg); |
e570fd27 | 6996 | |
6719afdc GU |
6997 | if (!delalloc) |
6998 | return used_bg; | |
e570fd27 | 6999 | |
6719afdc GU |
7000 | if (down_read_trylock(&used_bg->data_rwsem)) |
7001 | return used_bg; | |
e570fd27 | 7002 | |
6719afdc | 7003 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 7004 | |
e321f8a8 LB |
7005 | /* We should only have one-level nested. */ |
7006 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 7007 | |
6719afdc GU |
7008 | spin_lock(&cluster->refill_lock); |
7009 | if (used_bg == cluster->block_group) | |
7010 | return used_bg; | |
e570fd27 | 7011 | |
6719afdc GU |
7012 | up_read(&used_bg->data_rwsem); |
7013 | btrfs_put_block_group(used_bg); | |
7014 | } | |
e570fd27 MX |
7015 | } |
7016 | ||
7017 | static inline void | |
7018 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7019 | int delalloc) | |
7020 | { | |
7021 | if (delalloc) | |
7022 | up_read(&cache->data_rwsem); | |
7023 | btrfs_put_block_group(cache); | |
7024 | } | |
7025 | ||
b4bd745d QW |
7026 | /* |
7027 | * Structure used internally for find_free_extent() function. Wraps needed | |
7028 | * parameters. | |
7029 | */ | |
7030 | struct find_free_extent_ctl { | |
7031 | /* Basic allocation info */ | |
7032 | u64 ram_bytes; | |
7033 | u64 num_bytes; | |
7034 | u64 empty_size; | |
7035 | u64 flags; | |
7036 | int delalloc; | |
7037 | ||
7038 | /* Where to start the search inside the bg */ | |
7039 | u64 search_start; | |
7040 | ||
7041 | /* For clustered allocation */ | |
7042 | u64 empty_cluster; | |
7043 | ||
7044 | bool have_caching_bg; | |
7045 | bool orig_have_caching_bg; | |
7046 | ||
7047 | /* RAID index, converted from flags */ | |
7048 | int index; | |
7049 | ||
e72d79d6 QW |
7050 | /* |
7051 | * Current loop number, check find_free_extent_update_loop() for details | |
7052 | */ | |
b4bd745d QW |
7053 | int loop; |
7054 | ||
7055 | /* | |
7056 | * Whether we're refilling a cluster, if true we need to re-search | |
7057 | * current block group but don't try to refill the cluster again. | |
7058 | */ | |
7059 | bool retry_clustered; | |
7060 | ||
7061 | /* | |
7062 | * Whether we're updating free space cache, if true we need to re-search | |
7063 | * current block group but don't try updating free space cache again. | |
7064 | */ | |
7065 | bool retry_unclustered; | |
7066 | ||
7067 | /* If current block group is cached */ | |
7068 | int cached; | |
7069 | ||
7070 | /* Max contiguous hole found */ | |
7071 | u64 max_extent_size; | |
7072 | ||
7073 | /* Total free space from free space cache, not always contiguous */ | |
7074 | u64 total_free_space; | |
7075 | ||
7076 | /* Found result */ | |
7077 | u64 found_offset; | |
7078 | }; | |
7079 | ||
d06e3bb6 QW |
7080 | |
7081 | /* | |
7082 | * Helper function for find_free_extent(). | |
7083 | * | |
7084 | * Return -ENOENT to inform caller that we need fallback to unclustered mode. | |
7085 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
7086 | * Return >0 to inform caller that we find nothing | |
7087 | * Return 0 means we have found a location and set ffe_ctl->found_offset. | |
7088 | */ | |
7089 | static int find_free_extent_clustered(struct btrfs_block_group_cache *bg, | |
7090 | struct btrfs_free_cluster *last_ptr, | |
7091 | struct find_free_extent_ctl *ffe_ctl, | |
7092 | struct btrfs_block_group_cache **cluster_bg_ret) | |
7093 | { | |
d06e3bb6 QW |
7094 | struct btrfs_block_group_cache *cluster_bg; |
7095 | u64 aligned_cluster; | |
7096 | u64 offset; | |
7097 | int ret; | |
7098 | ||
7099 | cluster_bg = btrfs_lock_cluster(bg, last_ptr, ffe_ctl->delalloc); | |
7100 | if (!cluster_bg) | |
7101 | goto refill_cluster; | |
7102 | if (cluster_bg != bg && (cluster_bg->ro || | |
7103 | !block_group_bits(cluster_bg, ffe_ctl->flags))) | |
7104 | goto release_cluster; | |
7105 | ||
7106 | offset = btrfs_alloc_from_cluster(cluster_bg, last_ptr, | |
7107 | ffe_ctl->num_bytes, cluster_bg->key.objectid, | |
7108 | &ffe_ctl->max_extent_size); | |
7109 | if (offset) { | |
7110 | /* We have a block, we're done */ | |
7111 | spin_unlock(&last_ptr->refill_lock); | |
7112 | trace_btrfs_reserve_extent_cluster(cluster_bg, | |
7113 | ffe_ctl->search_start, ffe_ctl->num_bytes); | |
7114 | *cluster_bg_ret = cluster_bg; | |
7115 | ffe_ctl->found_offset = offset; | |
7116 | return 0; | |
7117 | } | |
7118 | WARN_ON(last_ptr->block_group != cluster_bg); | |
7119 | ||
7120 | release_cluster: | |
7121 | /* | |
7122 | * If we are on LOOP_NO_EMPTY_SIZE, we can't set up a new clusters, so | |
7123 | * lets just skip it and let the allocator find whatever block it can | |
7124 | * find. If we reach this point, we will have tried the cluster | |
7125 | * allocator plenty of times and not have found anything, so we are | |
7126 | * likely way too fragmented for the clustering stuff to find anything. | |
7127 | * | |
7128 | * However, if the cluster is taken from the current block group, | |
7129 | * release the cluster first, so that we stand a better chance of | |
7130 | * succeeding in the unclustered allocation. | |
7131 | */ | |
7132 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE && cluster_bg != bg) { | |
7133 | spin_unlock(&last_ptr->refill_lock); | |
7134 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
7135 | return -ENOENT; | |
7136 | } | |
7137 | ||
7138 | /* This cluster didn't work out, free it and start over */ | |
7139 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7140 | ||
7141 | if (cluster_bg != bg) | |
7142 | btrfs_release_block_group(cluster_bg, ffe_ctl->delalloc); | |
7143 | ||
7144 | refill_cluster: | |
7145 | if (ffe_ctl->loop >= LOOP_NO_EMPTY_SIZE) { | |
7146 | spin_unlock(&last_ptr->refill_lock); | |
7147 | return -ENOENT; | |
7148 | } | |
7149 | ||
7150 | aligned_cluster = max_t(u64, | |
7151 | ffe_ctl->empty_cluster + ffe_ctl->empty_size, | |
7152 | bg->full_stripe_len); | |
2ceeae2e DS |
7153 | ret = btrfs_find_space_cluster(bg, last_ptr, ffe_ctl->search_start, |
7154 | ffe_ctl->num_bytes, aligned_cluster); | |
d06e3bb6 QW |
7155 | if (ret == 0) { |
7156 | /* Now pull our allocation out of this cluster */ | |
7157 | offset = btrfs_alloc_from_cluster(bg, last_ptr, | |
7158 | ffe_ctl->num_bytes, ffe_ctl->search_start, | |
7159 | &ffe_ctl->max_extent_size); | |
7160 | if (offset) { | |
7161 | /* We found one, proceed */ | |
7162 | spin_unlock(&last_ptr->refill_lock); | |
7163 | trace_btrfs_reserve_extent_cluster(bg, | |
7164 | ffe_ctl->search_start, | |
7165 | ffe_ctl->num_bytes); | |
7166 | ffe_ctl->found_offset = offset; | |
7167 | return 0; | |
7168 | } | |
7169 | } else if (!ffe_ctl->cached && ffe_ctl->loop > LOOP_CACHING_NOWAIT && | |
7170 | !ffe_ctl->retry_clustered) { | |
7171 | spin_unlock(&last_ptr->refill_lock); | |
7172 | ||
7173 | ffe_ctl->retry_clustered = true; | |
7174 | wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + | |
7175 | ffe_ctl->empty_cluster + ffe_ctl->empty_size); | |
7176 | return -EAGAIN; | |
7177 | } | |
7178 | /* | |
7179 | * At this point we either didn't find a cluster or we weren't able to | |
7180 | * allocate a block from our cluster. Free the cluster we've been | |
7181 | * trying to use, and go to the next block group. | |
7182 | */ | |
7183 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7184 | spin_unlock(&last_ptr->refill_lock); | |
7185 | return 1; | |
7186 | } | |
7187 | ||
e1a41848 QW |
7188 | /* |
7189 | * Return >0 to inform caller that we find nothing | |
7190 | * Return 0 when we found an free extent and set ffe_ctrl->found_offset | |
7191 | * Return -EAGAIN to inform caller that we need to re-search this block group | |
7192 | */ | |
7193 | static int find_free_extent_unclustered(struct btrfs_block_group_cache *bg, | |
7194 | struct btrfs_free_cluster *last_ptr, | |
7195 | struct find_free_extent_ctl *ffe_ctl) | |
7196 | { | |
7197 | u64 offset; | |
7198 | ||
7199 | /* | |
7200 | * We are doing an unclustered allocation, set the fragmented flag so | |
7201 | * we don't bother trying to setup a cluster again until we get more | |
7202 | * space. | |
7203 | */ | |
7204 | if (unlikely(last_ptr)) { | |
7205 | spin_lock(&last_ptr->lock); | |
7206 | last_ptr->fragmented = 1; | |
7207 | spin_unlock(&last_ptr->lock); | |
7208 | } | |
7209 | if (ffe_ctl->cached) { | |
7210 | struct btrfs_free_space_ctl *free_space_ctl; | |
7211 | ||
7212 | free_space_ctl = bg->free_space_ctl; | |
7213 | spin_lock(&free_space_ctl->tree_lock); | |
7214 | if (free_space_ctl->free_space < | |
7215 | ffe_ctl->num_bytes + ffe_ctl->empty_cluster + | |
7216 | ffe_ctl->empty_size) { | |
7217 | ffe_ctl->total_free_space = max_t(u64, | |
7218 | ffe_ctl->total_free_space, | |
7219 | free_space_ctl->free_space); | |
7220 | spin_unlock(&free_space_ctl->tree_lock); | |
7221 | return 1; | |
7222 | } | |
7223 | spin_unlock(&free_space_ctl->tree_lock); | |
7224 | } | |
7225 | ||
7226 | offset = btrfs_find_space_for_alloc(bg, ffe_ctl->search_start, | |
7227 | ffe_ctl->num_bytes, ffe_ctl->empty_size, | |
7228 | &ffe_ctl->max_extent_size); | |
7229 | ||
7230 | /* | |
7231 | * If we didn't find a chunk, and we haven't failed on this block group | |
7232 | * before, and this block group is in the middle of caching and we are | |
7233 | * ok with waiting, then go ahead and wait for progress to be made, and | |
7234 | * set @retry_unclustered to true. | |
7235 | * | |
7236 | * If @retry_unclustered is true then we've already waited on this | |
7237 | * block group once and should move on to the next block group. | |
7238 | */ | |
7239 | if (!offset && !ffe_ctl->retry_unclustered && !ffe_ctl->cached && | |
7240 | ffe_ctl->loop > LOOP_CACHING_NOWAIT) { | |
7241 | wait_block_group_cache_progress(bg, ffe_ctl->num_bytes + | |
7242 | ffe_ctl->empty_size); | |
7243 | ffe_ctl->retry_unclustered = true; | |
7244 | return -EAGAIN; | |
7245 | } else if (!offset) { | |
7246 | return 1; | |
7247 | } | |
7248 | ffe_ctl->found_offset = offset; | |
7249 | return 0; | |
7250 | } | |
7251 | ||
e72d79d6 QW |
7252 | /* |
7253 | * Return >0 means caller needs to re-search for free extent | |
7254 | * Return 0 means we have the needed free extent. | |
7255 | * Return <0 means we failed to locate any free extent. | |
7256 | */ | |
7257 | static int find_free_extent_update_loop(struct btrfs_fs_info *fs_info, | |
7258 | struct btrfs_free_cluster *last_ptr, | |
7259 | struct btrfs_key *ins, | |
7260 | struct find_free_extent_ctl *ffe_ctl, | |
7261 | int full_search, bool use_cluster) | |
7262 | { | |
7263 | struct btrfs_root *root = fs_info->extent_root; | |
7264 | int ret; | |
7265 | ||
7266 | if ((ffe_ctl->loop == LOOP_CACHING_NOWAIT) && | |
7267 | ffe_ctl->have_caching_bg && !ffe_ctl->orig_have_caching_bg) | |
7268 | ffe_ctl->orig_have_caching_bg = true; | |
7269 | ||
7270 | if (!ins->objectid && ffe_ctl->loop >= LOOP_CACHING_WAIT && | |
7271 | ffe_ctl->have_caching_bg) | |
7272 | return 1; | |
7273 | ||
7274 | if (!ins->objectid && ++(ffe_ctl->index) < BTRFS_NR_RAID_TYPES) | |
7275 | return 1; | |
7276 | ||
7277 | if (ins->objectid) { | |
7278 | if (!use_cluster && last_ptr) { | |
7279 | spin_lock(&last_ptr->lock); | |
7280 | last_ptr->window_start = ins->objectid; | |
7281 | spin_unlock(&last_ptr->lock); | |
7282 | } | |
7283 | return 0; | |
7284 | } | |
7285 | ||
7286 | /* | |
7287 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking | |
7288 | * caching kthreads as we move along | |
7289 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching | |
7290 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7291 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7292 | * again | |
7293 | */ | |
7294 | if (ffe_ctl->loop < LOOP_NO_EMPTY_SIZE) { | |
7295 | ffe_ctl->index = 0; | |
7296 | if (ffe_ctl->loop == LOOP_CACHING_NOWAIT) { | |
7297 | /* | |
7298 | * We want to skip the LOOP_CACHING_WAIT step if we | |
7299 | * don't have any uncached bgs and we've already done a | |
7300 | * full search through. | |
7301 | */ | |
7302 | if (ffe_ctl->orig_have_caching_bg || !full_search) | |
7303 | ffe_ctl->loop = LOOP_CACHING_WAIT; | |
7304 | else | |
7305 | ffe_ctl->loop = LOOP_ALLOC_CHUNK; | |
7306 | } else { | |
7307 | ffe_ctl->loop++; | |
7308 | } | |
7309 | ||
7310 | if (ffe_ctl->loop == LOOP_ALLOC_CHUNK) { | |
7311 | struct btrfs_trans_handle *trans; | |
7312 | int exist = 0; | |
7313 | ||
7314 | trans = current->journal_info; | |
7315 | if (trans) | |
7316 | exist = 1; | |
7317 | else | |
7318 | trans = btrfs_join_transaction(root); | |
7319 | ||
7320 | if (IS_ERR(trans)) { | |
7321 | ret = PTR_ERR(trans); | |
7322 | return ret; | |
7323 | } | |
7324 | ||
fc471cb0 JB |
7325 | ret = btrfs_chunk_alloc(trans, ffe_ctl->flags, |
7326 | CHUNK_ALLOC_FORCE); | |
e72d79d6 QW |
7327 | |
7328 | /* | |
7329 | * If we can't allocate a new chunk we've already looped | |
7330 | * through at least once, move on to the NO_EMPTY_SIZE | |
7331 | * case. | |
7332 | */ | |
7333 | if (ret == -ENOSPC) | |
7334 | ffe_ctl->loop = LOOP_NO_EMPTY_SIZE; | |
7335 | ||
7336 | /* Do not bail out on ENOSPC since we can do more. */ | |
7337 | if (ret < 0 && ret != -ENOSPC) | |
7338 | btrfs_abort_transaction(trans, ret); | |
7339 | else | |
7340 | ret = 0; | |
7341 | if (!exist) | |
7342 | btrfs_end_transaction(trans); | |
7343 | if (ret) | |
7344 | return ret; | |
7345 | } | |
7346 | ||
7347 | if (ffe_ctl->loop == LOOP_NO_EMPTY_SIZE) { | |
7348 | /* | |
7349 | * Don't loop again if we already have no empty_size and | |
7350 | * no empty_cluster. | |
7351 | */ | |
7352 | if (ffe_ctl->empty_size == 0 && | |
7353 | ffe_ctl->empty_cluster == 0) | |
7354 | return -ENOSPC; | |
7355 | ffe_ctl->empty_size = 0; | |
7356 | ffe_ctl->empty_cluster = 0; | |
7357 | } | |
7358 | return 1; | |
7359 | } | |
7360 | return -ENOSPC; | |
7361 | } | |
7362 | ||
fec577fb CM |
7363 | /* |
7364 | * walks the btree of allocated extents and find a hole of a given size. | |
7365 | * The key ins is changed to record the hole: | |
a4820398 | 7366 | * ins->objectid == start position |
62e2749e | 7367 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7368 | * ins->offset == the size of the hole. |
fec577fb | 7369 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7370 | * |
7371 | * If there is no suitable free space, we will record the max size of | |
7372 | * the free space extent currently. | |
e72d79d6 QW |
7373 | * |
7374 | * The overall logic and call chain: | |
7375 | * | |
7376 | * find_free_extent() | |
7377 | * |- Iterate through all block groups | |
7378 | * | |- Get a valid block group | |
7379 | * | |- Try to do clustered allocation in that block group | |
7380 | * | |- Try to do unclustered allocation in that block group | |
7381 | * | |- Check if the result is valid | |
7382 | * | | |- If valid, then exit | |
7383 | * | |- Jump to next block group | |
7384 | * | | |
7385 | * |- Push harder to find free extents | |
7386 | * |- If not found, re-iterate all block groups | |
fec577fb | 7387 | */ |
87bde3cd | 7388 | static noinline int find_free_extent(struct btrfs_fs_info *fs_info, |
18513091 WX |
7389 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
7390 | u64 hint_byte, struct btrfs_key *ins, | |
7391 | u64 flags, int delalloc) | |
fec577fb | 7392 | { |
80eb234a | 7393 | int ret = 0; |
fa9c0d79 | 7394 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7395 | struct btrfs_block_group_cache *block_group = NULL; |
b4bd745d | 7396 | struct find_free_extent_ctl ffe_ctl = {0}; |
80eb234a | 7397 | struct btrfs_space_info *space_info; |
67377734 | 7398 | bool use_cluster = true; |
a5e681d9 | 7399 | bool full_search = false; |
fec577fb | 7400 | |
0b246afa | 7401 | WARN_ON(num_bytes < fs_info->sectorsize); |
b4bd745d QW |
7402 | |
7403 | ffe_ctl.ram_bytes = ram_bytes; | |
7404 | ffe_ctl.num_bytes = num_bytes; | |
7405 | ffe_ctl.empty_size = empty_size; | |
7406 | ffe_ctl.flags = flags; | |
7407 | ffe_ctl.search_start = 0; | |
7408 | ffe_ctl.retry_clustered = false; | |
7409 | ffe_ctl.retry_unclustered = false; | |
7410 | ffe_ctl.delalloc = delalloc; | |
7411 | ffe_ctl.index = btrfs_bg_flags_to_raid_index(flags); | |
7412 | ffe_ctl.have_caching_bg = false; | |
7413 | ffe_ctl.orig_have_caching_bg = false; | |
7414 | ffe_ctl.found_offset = 0; | |
7415 | ||
962a298f | 7416 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7417 | ins->objectid = 0; |
7418 | ins->offset = 0; | |
b1a4d965 | 7419 | |
71ff6437 | 7420 | trace_find_free_extent(fs_info, num_bytes, empty_size, flags); |
3f7de037 | 7421 | |
280c2908 | 7422 | space_info = btrfs_find_space_info(fs_info, flags); |
1b1d1f66 | 7423 | if (!space_info) { |
0b246afa | 7424 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7425 | return -ENOSPC; |
7426 | } | |
2552d17e | 7427 | |
67377734 | 7428 | /* |
4f4db217 JB |
7429 | * If our free space is heavily fragmented we may not be able to make |
7430 | * big contiguous allocations, so instead of doing the expensive search | |
7431 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7432 | * can go ahead and search for a more manageable chunk. | |
7433 | * | |
7434 | * If our max_extent_size is large enough for our allocation simply | |
7435 | * disable clustering since we will likely not be able to find enough | |
7436 | * space to create a cluster and induce latency trying. | |
67377734 | 7437 | */ |
4f4db217 JB |
7438 | if (unlikely(space_info->max_extent_size)) { |
7439 | spin_lock(&space_info->lock); | |
7440 | if (space_info->max_extent_size && | |
7441 | num_bytes > space_info->max_extent_size) { | |
7442 | ins->offset = space_info->max_extent_size; | |
7443 | spin_unlock(&space_info->lock); | |
7444 | return -ENOSPC; | |
7445 | } else if (space_info->max_extent_size) { | |
7446 | use_cluster = false; | |
7447 | } | |
7448 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7449 | } |
0f9dd46c | 7450 | |
b4bd745d QW |
7451 | last_ptr = fetch_cluster_info(fs_info, space_info, |
7452 | &ffe_ctl.empty_cluster); | |
239b14b3 | 7453 | if (last_ptr) { |
fa9c0d79 CM |
7454 | spin_lock(&last_ptr->lock); |
7455 | if (last_ptr->block_group) | |
7456 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7457 | if (last_ptr->fragmented) { |
7458 | /* | |
7459 | * We still set window_start so we can keep track of the | |
7460 | * last place we found an allocation to try and save | |
7461 | * some time. | |
7462 | */ | |
7463 | hint_byte = last_ptr->window_start; | |
7464 | use_cluster = false; | |
7465 | } | |
fa9c0d79 | 7466 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7467 | } |
fa9c0d79 | 7468 | |
b4bd745d QW |
7469 | ffe_ctl.search_start = max(ffe_ctl.search_start, |
7470 | first_logical_byte(fs_info, 0)); | |
7471 | ffe_ctl.search_start = max(ffe_ctl.search_start, hint_byte); | |
7472 | if (ffe_ctl.search_start == hint_byte) { | |
7473 | block_group = btrfs_lookup_block_group(fs_info, | |
7474 | ffe_ctl.search_start); | |
817d52f8 JB |
7475 | /* |
7476 | * we don't want to use the block group if it doesn't match our | |
7477 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7478 | * |
7479 | * However if we are re-searching with an ideal block group | |
7480 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7481 | */ |
b6919a58 | 7482 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7483 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7484 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7485 | if (list_empty(&block_group->list) || |
7486 | block_group->ro) { | |
7487 | /* | |
7488 | * someone is removing this block group, | |
7489 | * we can't jump into the have_block_group | |
7490 | * target because our list pointers are not | |
7491 | * valid | |
7492 | */ | |
7493 | btrfs_put_block_group(block_group); | |
7494 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7495 | } else { |
b4bd745d | 7496 | ffe_ctl.index = btrfs_bg_flags_to_raid_index( |
3e72ee88 | 7497 | block_group->flags); |
e570fd27 | 7498 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7499 | goto have_block_group; |
ccf0e725 | 7500 | } |
2552d17e | 7501 | } else if (block_group) { |
fa9c0d79 | 7502 | btrfs_put_block_group(block_group); |
2552d17e | 7503 | } |
42e70e7a | 7504 | } |
2552d17e | 7505 | search: |
b4bd745d QW |
7506 | ffe_ctl.have_caching_bg = false; |
7507 | if (ffe_ctl.index == btrfs_bg_flags_to_raid_index(flags) || | |
7508 | ffe_ctl.index == 0) | |
a5e681d9 | 7509 | full_search = true; |
80eb234a | 7510 | down_read(&space_info->groups_sem); |
b4bd745d QW |
7511 | list_for_each_entry(block_group, |
7512 | &space_info->block_groups[ffe_ctl.index], list) { | |
14443937 JM |
7513 | /* If the block group is read-only, we can skip it entirely. */ |
7514 | if (unlikely(block_group->ro)) | |
7515 | continue; | |
7516 | ||
e570fd27 | 7517 | btrfs_grab_block_group(block_group, delalloc); |
b4bd745d | 7518 | ffe_ctl.search_start = block_group->key.objectid; |
42e70e7a | 7519 | |
83a50de9 CM |
7520 | /* |
7521 | * this can happen if we end up cycling through all the | |
7522 | * raid types, but we want to make sure we only allocate | |
7523 | * for the proper type. | |
7524 | */ | |
b6919a58 | 7525 | if (!block_group_bits(block_group, flags)) { |
bece2e82 | 7526 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
c7369b3f | 7527 | BTRFS_BLOCK_GROUP_RAID1_MASK | |
a07e8a46 | 7528 | BTRFS_BLOCK_GROUP_RAID56_MASK | |
83a50de9 CM |
7529 | BTRFS_BLOCK_GROUP_RAID10; |
7530 | ||
7531 | /* | |
7532 | * if they asked for extra copies and this block group | |
7533 | * doesn't provide them, bail. This does allow us to | |
7534 | * fill raid0 from raid1. | |
7535 | */ | |
b6919a58 | 7536 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
7537 | goto loop; |
7538 | } | |
7539 | ||
2552d17e | 7540 | have_block_group: |
b4bd745d QW |
7541 | ffe_ctl.cached = block_group_cache_done(block_group); |
7542 | if (unlikely(!ffe_ctl.cached)) { | |
7543 | ffe_ctl.have_caching_bg = true; | |
f6373bf3 | 7544 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7545 | BUG_ON(ret < 0); |
7546 | ret = 0; | |
817d52f8 JB |
7547 | } |
7548 | ||
36cce922 JB |
7549 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7550 | goto loop; | |
0f9dd46c | 7551 | |
0a24325e | 7552 | /* |
062c05c4 AO |
7553 | * Ok we want to try and use the cluster allocator, so |
7554 | * lets look there | |
0a24325e | 7555 | */ |
c759c4e1 | 7556 | if (last_ptr && use_cluster) { |
d06e3bb6 | 7557 | struct btrfs_block_group_cache *cluster_bg = NULL; |
fa9c0d79 | 7558 | |
d06e3bb6 QW |
7559 | ret = find_free_extent_clustered(block_group, last_ptr, |
7560 | &ffe_ctl, &cluster_bg); | |
062c05c4 | 7561 | |
fa9c0d79 | 7562 | if (ret == 0) { |
d06e3bb6 QW |
7563 | if (cluster_bg && cluster_bg != block_group) { |
7564 | btrfs_release_block_group(block_group, | |
7565 | delalloc); | |
7566 | block_group = cluster_bg; | |
fa9c0d79 | 7567 | } |
d06e3bb6 QW |
7568 | goto checks; |
7569 | } else if (ret == -EAGAIN) { | |
817d52f8 | 7570 | goto have_block_group; |
d06e3bb6 QW |
7571 | } else if (ret > 0) { |
7572 | goto loop; | |
fa9c0d79 | 7573 | } |
d06e3bb6 | 7574 | /* ret == -ENOENT case falls through */ |
fa9c0d79 CM |
7575 | } |
7576 | ||
e1a41848 QW |
7577 | ret = find_free_extent_unclustered(block_group, last_ptr, |
7578 | &ffe_ctl); | |
7579 | if (ret == -EAGAIN) | |
817d52f8 | 7580 | goto have_block_group; |
e1a41848 | 7581 | else if (ret > 0) |
1cdda9b8 | 7582 | goto loop; |
e1a41848 | 7583 | /* ret == 0 case falls through */ |
fa9c0d79 | 7584 | checks: |
b4bd745d QW |
7585 | ffe_ctl.search_start = round_up(ffe_ctl.found_offset, |
7586 | fs_info->stripesize); | |
25179201 | 7587 | |
2552d17e | 7588 | /* move on to the next group */ |
b4bd745d | 7589 | if (ffe_ctl.search_start + num_bytes > |
215a63d1 | 7590 | block_group->key.objectid + block_group->key.offset) { |
b4bd745d QW |
7591 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
7592 | num_bytes); | |
2552d17e | 7593 | goto loop; |
6226cb0a | 7594 | } |
f5a31e16 | 7595 | |
b4bd745d QW |
7596 | if (ffe_ctl.found_offset < ffe_ctl.search_start) |
7597 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, | |
7598 | ffe_ctl.search_start - ffe_ctl.found_offset); | |
2552d17e | 7599 | |
18513091 WX |
7600 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
7601 | num_bytes, delalloc); | |
f0486c68 | 7602 | if (ret == -EAGAIN) { |
b4bd745d QW |
7603 | btrfs_add_free_space(block_group, ffe_ctl.found_offset, |
7604 | num_bytes); | |
2552d17e | 7605 | goto loop; |
0f9dd46c | 7606 | } |
9cfa3e34 | 7607 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 7608 | |
f0486c68 | 7609 | /* we are all good, lets return */ |
b4bd745d | 7610 | ins->objectid = ffe_ctl.search_start; |
2552d17e | 7611 | ins->offset = num_bytes; |
d2fb3437 | 7612 | |
b4bd745d QW |
7613 | trace_btrfs_reserve_extent(block_group, ffe_ctl.search_start, |
7614 | num_bytes); | |
e570fd27 | 7615 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7616 | break; |
7617 | loop: | |
b4bd745d QW |
7618 | ffe_ctl.retry_clustered = false; |
7619 | ffe_ctl.retry_unclustered = false; | |
3e72ee88 | 7620 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != |
b4bd745d | 7621 | ffe_ctl.index); |
e570fd27 | 7622 | btrfs_release_block_group(block_group, delalloc); |
14443937 | 7623 | cond_resched(); |
2552d17e JB |
7624 | } |
7625 | up_read(&space_info->groups_sem); | |
7626 | ||
e72d79d6 QW |
7627 | ret = find_free_extent_update_loop(fs_info, last_ptr, ins, &ffe_ctl, |
7628 | full_search, use_cluster); | |
7629 | if (ret > 0) | |
b742bb82 YZ |
7630 | goto search; |
7631 | ||
4f4db217 | 7632 | if (ret == -ENOSPC) { |
b4bd745d QW |
7633 | /* |
7634 | * Use ffe_ctl->total_free_space as fallback if we can't find | |
7635 | * any contiguous hole. | |
7636 | */ | |
7637 | if (!ffe_ctl.max_extent_size) | |
7638 | ffe_ctl.max_extent_size = ffe_ctl.total_free_space; | |
4f4db217 | 7639 | spin_lock(&space_info->lock); |
b4bd745d | 7640 | space_info->max_extent_size = ffe_ctl.max_extent_size; |
4f4db217 | 7641 | spin_unlock(&space_info->lock); |
b4bd745d | 7642 | ins->offset = ffe_ctl.max_extent_size; |
4f4db217 | 7643 | } |
0f70abe2 | 7644 | return ret; |
fec577fb | 7645 | } |
ec44a35c | 7646 | |
b78e5616 JB |
7647 | #define DUMP_BLOCK_RSV(fs_info, rsv_name) \ |
7648 | do { \ | |
7649 | struct btrfs_block_rsv *__rsv = &(fs_info)->rsv_name; \ | |
7650 | spin_lock(&__rsv->lock); \ | |
7651 | btrfs_info(fs_info, #rsv_name ": size %llu reserved %llu", \ | |
7652 | __rsv->size, __rsv->reserved); \ | |
7653 | spin_unlock(&__rsv->lock); \ | |
7654 | } while (0) | |
7655 | ||
ab8d0fc4 JM |
7656 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
7657 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 7658 | int dump_block_groups) |
0f9dd46c JB |
7659 | { |
7660 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 7661 | int index = 0; |
0f9dd46c | 7662 | |
9ed74f2d | 7663 | spin_lock(&info->lock); |
ab8d0fc4 JM |
7664 | btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull", |
7665 | info->flags, | |
4136135b LB |
7666 | info->total_bytes - btrfs_space_info_used(info, true), |
7667 | info->full ? "" : "not "); | |
ab8d0fc4 JM |
7668 | btrfs_info(fs_info, |
7669 | "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu", | |
7670 | info->total_bytes, info->bytes_used, info->bytes_pinned, | |
7671 | info->bytes_reserved, info->bytes_may_use, | |
7672 | info->bytes_readonly); | |
9ed74f2d JB |
7673 | spin_unlock(&info->lock); |
7674 | ||
b78e5616 JB |
7675 | DUMP_BLOCK_RSV(fs_info, global_block_rsv); |
7676 | DUMP_BLOCK_RSV(fs_info, trans_block_rsv); | |
7677 | DUMP_BLOCK_RSV(fs_info, chunk_block_rsv); | |
7678 | DUMP_BLOCK_RSV(fs_info, delayed_block_rsv); | |
7679 | DUMP_BLOCK_RSV(fs_info, delayed_refs_rsv); | |
7680 | ||
9ed74f2d JB |
7681 | if (!dump_block_groups) |
7682 | return; | |
0f9dd46c | 7683 | |
80eb234a | 7684 | down_read(&info->groups_sem); |
b742bb82 YZ |
7685 | again: |
7686 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 7687 | spin_lock(&cache->lock); |
ab8d0fc4 JM |
7688 | btrfs_info(fs_info, |
7689 | "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s", | |
7690 | cache->key.objectid, cache->key.offset, | |
7691 | btrfs_block_group_used(&cache->item), cache->pinned, | |
7692 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
7693 | btrfs_dump_free_space(cache, bytes); |
7694 | spin_unlock(&cache->lock); | |
7695 | } | |
b742bb82 YZ |
7696 | if (++index < BTRFS_NR_RAID_TYPES) |
7697 | goto again; | |
80eb234a | 7698 | up_read(&info->groups_sem); |
0f9dd46c | 7699 | } |
e8569813 | 7700 | |
6f47c706 NB |
7701 | /* |
7702 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a | |
7703 | * hole that is at least as big as @num_bytes. | |
7704 | * | |
7705 | * @root - The root that will contain this extent | |
7706 | * | |
7707 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
7708 | * is used for accounting purposes. This value differs | |
7709 | * from @num_bytes only in the case of compressed extents. | |
7710 | * | |
7711 | * @num_bytes - Number of bytes to allocate on-disk. | |
7712 | * | |
7713 | * @min_alloc_size - Indicates the minimum amount of space that the | |
7714 | * allocator should try to satisfy. In some cases | |
7715 | * @num_bytes may be larger than what is required and if | |
7716 | * the filesystem is fragmented then allocation fails. | |
7717 | * However, the presence of @min_alloc_size gives a | |
7718 | * chance to try and satisfy the smaller allocation. | |
7719 | * | |
7720 | * @empty_size - A hint that you plan on doing more COW. This is the | |
7721 | * size in bytes the allocator should try to find free | |
7722 | * next to the block it returns. This is just a hint and | |
7723 | * may be ignored by the allocator. | |
7724 | * | |
7725 | * @hint_byte - Hint to the allocator to start searching above the byte | |
7726 | * address passed. It might be ignored. | |
7727 | * | |
7728 | * @ins - This key is modified to record the found hole. It will | |
7729 | * have the following values: | |
7730 | * ins->objectid == start position | |
7731 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
7732 | * ins->offset == the size of the hole. | |
7733 | * | |
7734 | * @is_data - Boolean flag indicating whether an extent is | |
7735 | * allocated for data (true) or metadata (false) | |
7736 | * | |
7737 | * @delalloc - Boolean flag indicating whether this allocation is for | |
7738 | * delalloc or not. If 'true' data_rwsem of block groups | |
7739 | * is going to be acquired. | |
7740 | * | |
7741 | * | |
7742 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
7743 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
7744 | * largest available hole the allocator managed to find. | |
7745 | */ | |
18513091 | 7746 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
7747 | u64 num_bytes, u64 min_alloc_size, |
7748 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 7749 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 7750 | { |
ab8d0fc4 | 7751 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 7752 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 7753 | u64 flags; |
fec577fb | 7754 | int ret; |
925baedd | 7755 | |
1b86826d | 7756 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 7757 | again: |
0b246afa | 7758 | WARN_ON(num_bytes < fs_info->sectorsize); |
87bde3cd | 7759 | ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size, |
18513091 | 7760 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 7761 | if (!ret && !is_data) { |
ab8d0fc4 | 7762 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 7763 | } else if (ret == -ENOSPC) { |
a4820398 MX |
7764 | if (!final_tried && ins->offset) { |
7765 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 7766 | num_bytes = round_down(num_bytes, |
0b246afa | 7767 | fs_info->sectorsize); |
9e622d6b | 7768 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 7769 | ram_bytes = num_bytes; |
9e622d6b MX |
7770 | if (num_bytes == min_alloc_size) |
7771 | final_tried = true; | |
7772 | goto again; | |
ab8d0fc4 | 7773 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
7774 | struct btrfs_space_info *sinfo; |
7775 | ||
280c2908 | 7776 | sinfo = btrfs_find_space_info(fs_info, flags); |
0b246afa | 7777 | btrfs_err(fs_info, |
5d163e0e JM |
7778 | "allocation failed flags %llu, wanted %llu", |
7779 | flags, num_bytes); | |
53804280 | 7780 | if (sinfo) |
ab8d0fc4 | 7781 | dump_space_info(fs_info, sinfo, num_bytes, 1); |
9e622d6b | 7782 | } |
925baedd | 7783 | } |
0f9dd46c JB |
7784 | |
7785 | return ret; | |
e6dcd2dc CM |
7786 | } |
7787 | ||
2ff7e61e | 7788 | static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
7789 | u64 start, u64 len, |
7790 | int pin, int delalloc) | |
65b51a00 | 7791 | { |
0f9dd46c | 7792 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 7793 | int ret = 0; |
0f9dd46c | 7794 | |
0b246afa | 7795 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 7796 | if (!cache) { |
0b246afa JM |
7797 | btrfs_err(fs_info, "Unable to find block group for %llu", |
7798 | start); | |
0f9dd46c JB |
7799 | return -ENOSPC; |
7800 | } | |
1f3c79a2 | 7801 | |
e688b725 | 7802 | if (pin) |
fdf08605 | 7803 | pin_down_extent(cache, start, len, 1); |
e688b725 | 7804 | else { |
0b246afa | 7805 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 7806 | ret = btrfs_discard_extent(fs_info, start, len, NULL); |
e688b725 | 7807 | btrfs_add_free_space(cache, start, len); |
4824f1f4 | 7808 | btrfs_free_reserved_bytes(cache, len, delalloc); |
71ff6437 | 7809 | trace_btrfs_reserved_extent_free(fs_info, start, len); |
e688b725 | 7810 | } |
31193213 | 7811 | |
fa9c0d79 | 7812 | btrfs_put_block_group(cache); |
e6dcd2dc CM |
7813 | return ret; |
7814 | } | |
7815 | ||
2ff7e61e | 7816 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 | 7817 | u64 start, u64 len, int delalloc) |
e688b725 | 7818 | { |
2ff7e61e | 7819 | return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc); |
e688b725 CM |
7820 | } |
7821 | ||
2ff7e61e | 7822 | int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info, |
e688b725 CM |
7823 | u64 start, u64 len) |
7824 | { | |
2ff7e61e | 7825 | return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0); |
e688b725 CM |
7826 | } |
7827 | ||
5d4f98a2 | 7828 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
7829 | u64 parent, u64 root_objectid, |
7830 | u64 flags, u64 owner, u64 offset, | |
7831 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc | 7832 | { |
ef89b824 | 7833 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 7834 | int ret; |
e6dcd2dc | 7835 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 7836 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 7837 | struct btrfs_path *path; |
5d4f98a2 YZ |
7838 | struct extent_buffer *leaf; |
7839 | int type; | |
7840 | u32 size; | |
26b8003f | 7841 | |
5d4f98a2 YZ |
7842 | if (parent > 0) |
7843 | type = BTRFS_SHARED_DATA_REF_KEY; | |
7844 | else | |
7845 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 7846 | |
5d4f98a2 | 7847 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
7848 | |
7849 | path = btrfs_alloc_path(); | |
db5b493a TI |
7850 | if (!path) |
7851 | return -ENOMEM; | |
47e4bb98 | 7852 | |
b9473439 | 7853 | path->leave_spinning = 1; |
5d4f98a2 YZ |
7854 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
7855 | ins, size); | |
79787eaa JM |
7856 | if (ret) { |
7857 | btrfs_free_path(path); | |
7858 | return ret; | |
7859 | } | |
0f9dd46c | 7860 | |
5d4f98a2 YZ |
7861 | leaf = path->nodes[0]; |
7862 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 7863 | struct btrfs_extent_item); |
5d4f98a2 YZ |
7864 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
7865 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
7866 | btrfs_set_extent_flags(leaf, extent_item, | |
7867 | flags | BTRFS_EXTENT_FLAG_DATA); | |
7868 | ||
7869 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
7870 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
7871 | if (parent > 0) { | |
7872 | struct btrfs_shared_data_ref *ref; | |
7873 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
7874 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
7875 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
7876 | } else { | |
7877 | struct btrfs_extent_data_ref *ref; | |
7878 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
7879 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
7880 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
7881 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
7882 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
7883 | } | |
47e4bb98 CM |
7884 | |
7885 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 7886 | btrfs_free_path(path); |
f510cfec | 7887 | |
25a356d3 | 7888 | ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset); |
1e144fb8 OS |
7889 | if (ret) |
7890 | return ret; | |
7891 | ||
6b279408 | 7892 | ret = update_block_group(trans, ins->objectid, ins->offset, 1); |
79787eaa | 7893 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 7894 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 7895 | ins->objectid, ins->offset); |
f5947066 CM |
7896 | BUG(); |
7897 | } | |
71ff6437 | 7898 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
7899 | return ret; |
7900 | } | |
7901 | ||
5d4f98a2 | 7902 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 7903 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 7904 | struct btrfs_delayed_extent_op *extent_op) |
e6dcd2dc | 7905 | { |
9dcdbe01 | 7906 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 7907 | int ret; |
5d4f98a2 | 7908 | struct btrfs_extent_item *extent_item; |
4e6bd4e0 | 7909 | struct btrfs_key extent_key; |
5d4f98a2 YZ |
7910 | struct btrfs_tree_block_info *block_info; |
7911 | struct btrfs_extent_inline_ref *iref; | |
7912 | struct btrfs_path *path; | |
7913 | struct extent_buffer *leaf; | |
4e6bd4e0 | 7914 | struct btrfs_delayed_tree_ref *ref; |
3173a18f | 7915 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
4e6bd4e0 | 7916 | u64 num_bytes; |
21ebfbe7 | 7917 | u64 flags = extent_op->flags_to_set; |
0b246afa | 7918 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f | 7919 | |
4e6bd4e0 NB |
7920 | ref = btrfs_delayed_node_to_tree_ref(node); |
7921 | ||
4e6bd4e0 NB |
7922 | extent_key.objectid = node->bytenr; |
7923 | if (skinny_metadata) { | |
7924 | extent_key.offset = ref->level; | |
7925 | extent_key.type = BTRFS_METADATA_ITEM_KEY; | |
7926 | num_bytes = fs_info->nodesize; | |
7927 | } else { | |
7928 | extent_key.offset = node->num_bytes; | |
7929 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
3173a18f | 7930 | size += sizeof(*block_info); |
4e6bd4e0 NB |
7931 | num_bytes = node->num_bytes; |
7932 | } | |
1c2308f8 | 7933 | |
5d4f98a2 | 7934 | path = btrfs_alloc_path(); |
80ee54bf | 7935 | if (!path) |
d8926bb3 | 7936 | return -ENOMEM; |
56bec294 | 7937 | |
5d4f98a2 YZ |
7938 | path->leave_spinning = 1; |
7939 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
4e6bd4e0 | 7940 | &extent_key, size); |
79787eaa | 7941 | if (ret) { |
dd825259 | 7942 | btrfs_free_path(path); |
79787eaa JM |
7943 | return ret; |
7944 | } | |
5d4f98a2 YZ |
7945 | |
7946 | leaf = path->nodes[0]; | |
7947 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
7948 | struct btrfs_extent_item); | |
7949 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
7950 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
7951 | btrfs_set_extent_flags(leaf, extent_item, | |
7952 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 7953 | |
3173a18f JB |
7954 | if (skinny_metadata) { |
7955 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
7956 | } else { | |
7957 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
21ebfbe7 | 7958 | btrfs_set_tree_block_key(leaf, block_info, &extent_op->key); |
4e6bd4e0 | 7959 | btrfs_set_tree_block_level(leaf, block_info, ref->level); |
3173a18f JB |
7960 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
7961 | } | |
5d4f98a2 | 7962 | |
d4b20733 | 7963 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) { |
5d4f98a2 YZ |
7964 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); |
7965 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
7966 | BTRFS_SHARED_BLOCK_REF_KEY); | |
d4b20733 | 7967 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent); |
5d4f98a2 YZ |
7968 | } else { |
7969 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
7970 | BTRFS_TREE_BLOCK_REF_KEY); | |
4e6bd4e0 | 7971 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root); |
5d4f98a2 YZ |
7972 | } |
7973 | ||
7974 | btrfs_mark_buffer_dirty(leaf); | |
7975 | btrfs_free_path(path); | |
7976 | ||
4e6bd4e0 NB |
7977 | ret = remove_from_free_space_tree(trans, extent_key.objectid, |
7978 | num_bytes); | |
1e144fb8 OS |
7979 | if (ret) |
7980 | return ret; | |
7981 | ||
6b279408 | 7982 | ret = update_block_group(trans, extent_key.objectid, |
6202df69 | 7983 | fs_info->nodesize, 1); |
79787eaa | 7984 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 7985 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
4e6bd4e0 | 7986 | extent_key.objectid, extent_key.offset); |
5d4f98a2 YZ |
7987 | BUG(); |
7988 | } | |
0be5dc67 | 7989 | |
4e6bd4e0 | 7990 | trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid, |
0b246afa | 7991 | fs_info->nodesize); |
5d4f98a2 YZ |
7992 | return ret; |
7993 | } | |
7994 | ||
7995 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 7996 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
7997 | u64 offset, u64 ram_bytes, |
7998 | struct btrfs_key *ins) | |
5d4f98a2 | 7999 | { |
76675593 | 8000 | struct btrfs_ref generic_ref = { 0 }; |
5d4f98a2 YZ |
8001 | int ret; |
8002 | ||
84f7d8e6 | 8003 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 8004 | |
76675593 QW |
8005 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
8006 | ins->objectid, ins->offset, 0); | |
8007 | btrfs_init_data_ref(&generic_ref, root->root_key.objectid, owner, offset); | |
8a5040f7 | 8008 | btrfs_ref_tree_mod(root->fs_info, &generic_ref); |
76675593 QW |
8009 | ret = btrfs_add_delayed_data_ref(trans, &generic_ref, |
8010 | ram_bytes, NULL, NULL); | |
e6dcd2dc CM |
8011 | return ret; |
8012 | } | |
e02119d5 CM |
8013 | |
8014 | /* | |
8015 | * this is used by the tree logging recovery code. It records that | |
8016 | * an extent has been allocated and makes sure to clear the free | |
8017 | * space cache bits as well | |
8018 | */ | |
5d4f98a2 | 8019 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
8020 | u64 root_objectid, u64 owner, u64 offset, |
8021 | struct btrfs_key *ins) | |
e02119d5 | 8022 | { |
61da2abf | 8023 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e02119d5 CM |
8024 | int ret; |
8025 | struct btrfs_block_group_cache *block_group; | |
ed7a6948 | 8026 | struct btrfs_space_info *space_info; |
11833d66 | 8027 | |
8c2a1a30 JB |
8028 | /* |
8029 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 8030 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 8031 | */ |
0b246afa | 8032 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
8033 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
8034 | ins->offset); | |
b50c6e25 | 8035 | if (ret) |
8c2a1a30 | 8036 | return ret; |
11833d66 YZ |
8037 | } |
8038 | ||
0b246afa | 8039 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
8040 | if (!block_group) |
8041 | return -EINVAL; | |
8042 | ||
ed7a6948 WX |
8043 | space_info = block_group->space_info; |
8044 | spin_lock(&space_info->lock); | |
8045 | spin_lock(&block_group->lock); | |
8046 | space_info->bytes_reserved += ins->offset; | |
8047 | block_group->reserved += ins->offset; | |
8048 | spin_unlock(&block_group->lock); | |
8049 | spin_unlock(&space_info->lock); | |
8050 | ||
ef89b824 NB |
8051 | ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner, |
8052 | offset, ins, 1); | |
b50c6e25 | 8053 | btrfs_put_block_group(block_group); |
e02119d5 CM |
8054 | return ret; |
8055 | } | |
8056 | ||
48a3b636 ES |
8057 | static struct extent_buffer * |
8058 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
bc877d28 | 8059 | u64 bytenr, int level, u64 owner) |
65b51a00 | 8060 | { |
0b246afa | 8061 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
8062 | struct extent_buffer *buf; |
8063 | ||
2ff7e61e | 8064 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8065 | if (IS_ERR(buf)) |
8066 | return buf; | |
8067 | ||
b72c3aba QW |
8068 | /* |
8069 | * Extra safety check in case the extent tree is corrupted and extent | |
8070 | * allocator chooses to use a tree block which is already used and | |
8071 | * locked. | |
8072 | */ | |
8073 | if (buf->lock_owner == current->pid) { | |
8074 | btrfs_err_rl(fs_info, | |
8075 | "tree block %llu owner %llu already locked by pid=%d, extent tree corruption detected", | |
8076 | buf->start, btrfs_header_owner(buf), current->pid); | |
8077 | free_extent_buffer(buf); | |
8078 | return ERR_PTR(-EUCLEAN); | |
8079 | } | |
8080 | ||
85d4e461 | 8081 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 8082 | btrfs_tree_lock(buf); |
6a884d7d | 8083 | btrfs_clean_tree_block(buf); |
3083ee2e | 8084 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de | 8085 | |
8bead258 | 8086 | btrfs_set_lock_blocking_write(buf); |
4db8c528 | 8087 | set_extent_buffer_uptodate(buf); |
b4ce94de | 8088 | |
bc877d28 NB |
8089 | memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header)); |
8090 | btrfs_set_header_level(buf, level); | |
8091 | btrfs_set_header_bytenr(buf, buf->start); | |
8092 | btrfs_set_header_generation(buf, trans->transid); | |
8093 | btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV); | |
8094 | btrfs_set_header_owner(buf, owner); | |
de37aa51 | 8095 | write_extent_buffer_fsid(buf, fs_info->fs_devices->metadata_uuid); |
bc877d28 | 8096 | write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid); |
d0c803c4 | 8097 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 8098 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
8099 | /* |
8100 | * we allow two log transactions at a time, use different | |
52042d8e | 8101 | * EXTENT bit to differentiate dirty pages. |
8cef4e16 | 8102 | */ |
656f30db | 8103 | if (buf->log_index == 0) |
8cef4e16 YZ |
8104 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
8105 | buf->start + buf->len - 1, GFP_NOFS); | |
8106 | else | |
8107 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 8108 | buf->start + buf->len - 1); |
d0c803c4 | 8109 | } else { |
656f30db | 8110 | buf->log_index = -1; |
d0c803c4 | 8111 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 8112 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 8113 | } |
64c12921 | 8114 | trans->dirty = true; |
b4ce94de | 8115 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
8116 | return buf; |
8117 | } | |
8118 | ||
f0486c68 YZ |
8119 | static struct btrfs_block_rsv * |
8120 | use_block_rsv(struct btrfs_trans_handle *trans, | |
8121 | struct btrfs_root *root, u32 blocksize) | |
8122 | { | |
0b246afa | 8123 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8124 | struct btrfs_block_rsv *block_rsv; |
0b246afa | 8125 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
f0486c68 | 8126 | int ret; |
d88033db | 8127 | bool global_updated = false; |
f0486c68 YZ |
8128 | |
8129 | block_rsv = get_block_rsv(trans, root); | |
8130 | ||
b586b323 MX |
8131 | if (unlikely(block_rsv->size == 0)) |
8132 | goto try_reserve; | |
d88033db | 8133 | again: |
c2a67a76 | 8134 | ret = btrfs_block_rsv_use_bytes(block_rsv, blocksize); |
f0486c68 YZ |
8135 | if (!ret) |
8136 | return block_rsv; | |
8137 | ||
b586b323 MX |
8138 | if (block_rsv->failfast) |
8139 | return ERR_PTR(ret); | |
8140 | ||
d88033db MX |
8141 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
8142 | global_updated = true; | |
0b246afa | 8143 | update_global_block_rsv(fs_info); |
d88033db MX |
8144 | goto again; |
8145 | } | |
8146 | ||
ba2c4d4e JB |
8147 | /* |
8148 | * The global reserve still exists to save us from ourselves, so don't | |
8149 | * warn_on if we are short on our delayed refs reserve. | |
8150 | */ | |
8151 | if (block_rsv->type != BTRFS_BLOCK_RSV_DELREFS && | |
8152 | btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { | |
b586b323 MX |
8153 | static DEFINE_RATELIMIT_STATE(_rs, |
8154 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
8155 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
8156 | if (__ratelimit(&_rs)) | |
8157 | WARN(1, KERN_DEBUG | |
efe120a0 | 8158 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
8159 | } |
8160 | try_reserve: | |
8161 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
8162 | BTRFS_RESERVE_NO_FLUSH); | |
8163 | if (!ret) | |
8164 | return block_rsv; | |
8165 | /* | |
8166 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
8167 | * the global reserve if its space type is the same as the global |
8168 | * reservation. | |
b586b323 | 8169 | */ |
5881cfc9 MX |
8170 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
8171 | block_rsv->space_info == global_rsv->space_info) { | |
c2a67a76 | 8172 | ret = btrfs_block_rsv_use_bytes(global_rsv, blocksize); |
b586b323 MX |
8173 | if (!ret) |
8174 | return global_rsv; | |
8175 | } | |
8176 | return ERR_PTR(ret); | |
f0486c68 YZ |
8177 | } |
8178 | ||
8c2a3ca2 JB |
8179 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
8180 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 | 8181 | { |
3a584174 | 8182 | block_rsv_add_bytes(block_rsv, blocksize, false); |
ff6bc37e | 8183 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL); |
f0486c68 YZ |
8184 | } |
8185 | ||
fec577fb | 8186 | /* |
f0486c68 | 8187 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 8188 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 8189 | */ |
4d75f8a9 | 8190 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
8191 | struct btrfs_root *root, |
8192 | u64 parent, u64 root_objectid, | |
8193 | const struct btrfs_disk_key *key, | |
8194 | int level, u64 hint, | |
8195 | u64 empty_size) | |
fec577fb | 8196 | { |
0b246afa | 8197 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 8198 | struct btrfs_key ins; |
f0486c68 | 8199 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 8200 | struct extent_buffer *buf; |
67b7859e | 8201 | struct btrfs_delayed_extent_op *extent_op; |
ed4f255b | 8202 | struct btrfs_ref generic_ref = { 0 }; |
f0486c68 YZ |
8203 | u64 flags = 0; |
8204 | int ret; | |
0b246afa JM |
8205 | u32 blocksize = fs_info->nodesize; |
8206 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 8207 | |
05653ef3 | 8208 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 8209 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 8210 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
bc877d28 | 8211 | level, root_objectid); |
faa2dbf0 JB |
8212 | if (!IS_ERR(buf)) |
8213 | root->alloc_bytenr += blocksize; | |
8214 | return buf; | |
8215 | } | |
05653ef3 | 8216 | #endif |
fccb84c9 | 8217 | |
f0486c68 YZ |
8218 | block_rsv = use_block_rsv(trans, root, blocksize); |
8219 | if (IS_ERR(block_rsv)) | |
8220 | return ERR_CAST(block_rsv); | |
8221 | ||
18513091 | 8222 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 8223 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
8224 | if (ret) |
8225 | goto out_unuse; | |
55c69072 | 8226 | |
bc877d28 NB |
8227 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level, |
8228 | root_objectid); | |
67b7859e OS |
8229 | if (IS_ERR(buf)) { |
8230 | ret = PTR_ERR(buf); | |
8231 | goto out_free_reserved; | |
8232 | } | |
f0486c68 YZ |
8233 | |
8234 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8235 | if (parent == 0) | |
8236 | parent = ins.objectid; | |
8237 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8238 | } else | |
8239 | BUG_ON(parent > 0); | |
8240 | ||
8241 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 8242 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
8243 | if (!extent_op) { |
8244 | ret = -ENOMEM; | |
8245 | goto out_free_buf; | |
8246 | } | |
f0486c68 YZ |
8247 | if (key) |
8248 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8249 | else | |
8250 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8251 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
8252 | extent_op->update_key = skinny_metadata ? false : true; |
8253 | extent_op->update_flags = true; | |
8254 | extent_op->is_data = false; | |
b1c79e09 | 8255 | extent_op->level = level; |
f0486c68 | 8256 | |
ed4f255b QW |
8257 | btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT, |
8258 | ins.objectid, ins.offset, parent); | |
8259 | generic_ref.real_root = root->root_key.objectid; | |
8260 | btrfs_init_tree_ref(&generic_ref, level, root_objectid); | |
8a5040f7 | 8261 | btrfs_ref_tree_mod(fs_info, &generic_ref); |
ed4f255b | 8262 | ret = btrfs_add_delayed_tree_ref(trans, &generic_ref, |
7be07912 | 8263 | extent_op, NULL, NULL); |
67b7859e OS |
8264 | if (ret) |
8265 | goto out_free_delayed; | |
f0486c68 | 8266 | } |
fec577fb | 8267 | return buf; |
67b7859e OS |
8268 | |
8269 | out_free_delayed: | |
8270 | btrfs_free_delayed_extent_op(extent_op); | |
8271 | out_free_buf: | |
8272 | free_extent_buffer(buf); | |
8273 | out_free_reserved: | |
2ff7e61e | 8274 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 8275 | out_unuse: |
0b246afa | 8276 | unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 8277 | return ERR_PTR(ret); |
fec577fb | 8278 | } |
a28ec197 | 8279 | |
2c47e605 YZ |
8280 | struct walk_control { |
8281 | u64 refs[BTRFS_MAX_LEVEL]; | |
8282 | u64 flags[BTRFS_MAX_LEVEL]; | |
8283 | struct btrfs_key update_progress; | |
aea6f028 JB |
8284 | struct btrfs_key drop_progress; |
8285 | int drop_level; | |
2c47e605 YZ |
8286 | int stage; |
8287 | int level; | |
8288 | int shared_level; | |
8289 | int update_ref; | |
8290 | int keep_locks; | |
1c4850e2 YZ |
8291 | int reada_slot; |
8292 | int reada_count; | |
78c52d9e | 8293 | int restarted; |
2c47e605 YZ |
8294 | }; |
8295 | ||
8296 | #define DROP_REFERENCE 1 | |
8297 | #define UPDATE_BACKREF 2 | |
8298 | ||
1c4850e2 YZ |
8299 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8300 | struct btrfs_root *root, | |
8301 | struct walk_control *wc, | |
8302 | struct btrfs_path *path) | |
6407bf6d | 8303 | { |
0b246afa | 8304 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8305 | u64 bytenr; |
8306 | u64 generation; | |
8307 | u64 refs; | |
94fcca9f | 8308 | u64 flags; |
5d4f98a2 | 8309 | u32 nritems; |
1c4850e2 YZ |
8310 | struct btrfs_key key; |
8311 | struct extent_buffer *eb; | |
6407bf6d | 8312 | int ret; |
1c4850e2 YZ |
8313 | int slot; |
8314 | int nread = 0; | |
6407bf6d | 8315 | |
1c4850e2 YZ |
8316 | if (path->slots[wc->level] < wc->reada_slot) { |
8317 | wc->reada_count = wc->reada_count * 2 / 3; | |
8318 | wc->reada_count = max(wc->reada_count, 2); | |
8319 | } else { | |
8320 | wc->reada_count = wc->reada_count * 3 / 2; | |
8321 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 8322 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 8323 | } |
7bb86316 | 8324 | |
1c4850e2 YZ |
8325 | eb = path->nodes[wc->level]; |
8326 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 8327 | |
1c4850e2 YZ |
8328 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8329 | if (nread >= wc->reada_count) | |
8330 | break; | |
bd56b302 | 8331 | |
2dd3e67b | 8332 | cond_resched(); |
1c4850e2 YZ |
8333 | bytenr = btrfs_node_blockptr(eb, slot); |
8334 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8335 | |
1c4850e2 YZ |
8336 | if (slot == path->slots[wc->level]) |
8337 | goto reada; | |
5d4f98a2 | 8338 | |
1c4850e2 YZ |
8339 | if (wc->stage == UPDATE_BACKREF && |
8340 | generation <= root->root_key.offset) | |
bd56b302 CM |
8341 | continue; |
8342 | ||
94fcca9f | 8343 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 8344 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
8345 | wc->level - 1, 1, &refs, |
8346 | &flags); | |
79787eaa JM |
8347 | /* We don't care about errors in readahead. */ |
8348 | if (ret < 0) | |
8349 | continue; | |
94fcca9f YZ |
8350 | BUG_ON(refs == 0); |
8351 | ||
1c4850e2 | 8352 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8353 | if (refs == 1) |
8354 | goto reada; | |
bd56b302 | 8355 | |
94fcca9f YZ |
8356 | if (wc->level == 1 && |
8357 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8358 | continue; | |
1c4850e2 YZ |
8359 | if (!wc->update_ref || |
8360 | generation <= root->root_key.offset) | |
8361 | continue; | |
8362 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8363 | ret = btrfs_comp_cpu_keys(&key, | |
8364 | &wc->update_progress); | |
8365 | if (ret < 0) | |
8366 | continue; | |
94fcca9f YZ |
8367 | } else { |
8368 | if (wc->level == 1 && | |
8369 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8370 | continue; | |
6407bf6d | 8371 | } |
1c4850e2 | 8372 | reada: |
2ff7e61e | 8373 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 8374 | nread++; |
20524f02 | 8375 | } |
1c4850e2 | 8376 | wc->reada_slot = slot; |
20524f02 | 8377 | } |
2c47e605 | 8378 | |
f82d02d9 | 8379 | /* |
2c016dc2 | 8380 | * helper to process tree block while walking down the tree. |
2c47e605 | 8381 | * |
2c47e605 YZ |
8382 | * when wc->stage == UPDATE_BACKREF, this function updates |
8383 | * back refs for pointers in the block. | |
8384 | * | |
8385 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8386 | */ |
2c47e605 | 8387 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8388 | struct btrfs_root *root, |
2c47e605 | 8389 | struct btrfs_path *path, |
94fcca9f | 8390 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8391 | { |
2ff7e61e | 8392 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
8393 | int level = wc->level; |
8394 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 8395 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
8396 | int ret; |
8397 | ||
2c47e605 YZ |
8398 | if (wc->stage == UPDATE_BACKREF && |
8399 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8400 | return 1; | |
f82d02d9 | 8401 | |
2c47e605 YZ |
8402 | /* |
8403 | * when reference count of tree block is 1, it won't increase | |
8404 | * again. once full backref flag is set, we never clear it. | |
8405 | */ | |
94fcca9f YZ |
8406 | if (lookup_info && |
8407 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8408 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 8409 | BUG_ON(!path->locks[level]); |
2ff7e61e | 8410 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8411 | eb->start, level, 1, |
2c47e605 YZ |
8412 | &wc->refs[level], |
8413 | &wc->flags[level]); | |
79787eaa JM |
8414 | BUG_ON(ret == -ENOMEM); |
8415 | if (ret) | |
8416 | return ret; | |
2c47e605 YZ |
8417 | BUG_ON(wc->refs[level] == 0); |
8418 | } | |
5d4f98a2 | 8419 | |
2c47e605 YZ |
8420 | if (wc->stage == DROP_REFERENCE) { |
8421 | if (wc->refs[level] > 1) | |
8422 | return 1; | |
f82d02d9 | 8423 | |
2c47e605 | 8424 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 8425 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8426 | path->locks[level] = 0; |
8427 | } | |
8428 | return 0; | |
8429 | } | |
f82d02d9 | 8430 | |
2c47e605 YZ |
8431 | /* wc->stage == UPDATE_BACKREF */ |
8432 | if (!(wc->flags[level] & flag)) { | |
8433 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 8434 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 8435 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 8436 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8437 | BUG_ON(ret); /* -ENOMEM */ |
f5c8daa5 | 8438 | ret = btrfs_set_disk_extent_flags(trans, eb->start, |
b1c79e09 JB |
8439 | eb->len, flag, |
8440 | btrfs_header_level(eb), 0); | |
79787eaa | 8441 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
8442 | wc->flags[level] |= flag; |
8443 | } | |
8444 | ||
8445 | /* | |
8446 | * the block is shared by multiple trees, so it's not good to | |
8447 | * keep the tree lock | |
8448 | */ | |
8449 | if (path->locks[level] && level > 0) { | |
bd681513 | 8450 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8451 | path->locks[level] = 0; |
8452 | } | |
8453 | return 0; | |
8454 | } | |
8455 | ||
78c52d9e JB |
8456 | /* |
8457 | * This is used to verify a ref exists for this root to deal with a bug where we | |
8458 | * would have a drop_progress key that hadn't been updated properly. | |
8459 | */ | |
8460 | static int check_ref_exists(struct btrfs_trans_handle *trans, | |
8461 | struct btrfs_root *root, u64 bytenr, u64 parent, | |
8462 | int level) | |
8463 | { | |
8464 | struct btrfs_path *path; | |
8465 | struct btrfs_extent_inline_ref *iref; | |
8466 | int ret; | |
8467 | ||
8468 | path = btrfs_alloc_path(); | |
8469 | if (!path) | |
8470 | return -ENOMEM; | |
8471 | ||
8472 | ret = lookup_extent_backref(trans, path, &iref, bytenr, | |
8473 | root->fs_info->nodesize, parent, | |
8474 | root->root_key.objectid, level, 0); | |
8475 | btrfs_free_path(path); | |
8476 | if (ret == -ENOENT) | |
8477 | return 0; | |
8478 | if (ret < 0) | |
8479 | return ret; | |
8480 | return 1; | |
8481 | } | |
8482 | ||
1c4850e2 | 8483 | /* |
2c016dc2 | 8484 | * helper to process tree block pointer. |
1c4850e2 YZ |
8485 | * |
8486 | * when wc->stage == DROP_REFERENCE, this function checks | |
8487 | * reference count of the block pointed to. if the block | |
8488 | * is shared and we need update back refs for the subtree | |
8489 | * rooted at the block, this function changes wc->stage to | |
8490 | * UPDATE_BACKREF. if the block is shared and there is no | |
8491 | * need to update back, this function drops the reference | |
8492 | * to the block. | |
8493 | * | |
8494 | * NOTE: return value 1 means we should stop walking down. | |
8495 | */ | |
8496 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8497 | struct btrfs_root *root, | |
8498 | struct btrfs_path *path, | |
94fcca9f | 8499 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 8500 | { |
0b246afa | 8501 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8502 | u64 bytenr; |
8503 | u64 generation; | |
8504 | u64 parent; | |
1c4850e2 | 8505 | struct btrfs_key key; |
581c1760 | 8506 | struct btrfs_key first_key; |
ffd4bb2a | 8507 | struct btrfs_ref ref = { 0 }; |
1c4850e2 YZ |
8508 | struct extent_buffer *next; |
8509 | int level = wc->level; | |
8510 | int reada = 0; | |
8511 | int ret = 0; | |
1152651a | 8512 | bool need_account = false; |
1c4850e2 YZ |
8513 | |
8514 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8515 | path->slots[level]); | |
8516 | /* | |
8517 | * if the lower level block was created before the snapshot | |
8518 | * was created, we know there is no need to update back refs | |
8519 | * for the subtree | |
8520 | */ | |
8521 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8522 | generation <= root->root_key.offset) { |
8523 | *lookup_info = 1; | |
1c4850e2 | 8524 | return 1; |
94fcca9f | 8525 | } |
1c4850e2 YZ |
8526 | |
8527 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
581c1760 QW |
8528 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
8529 | path->slots[level]); | |
1c4850e2 | 8530 | |
0b246afa | 8531 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 8532 | if (!next) { |
2ff7e61e | 8533 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8534 | if (IS_ERR(next)) |
8535 | return PTR_ERR(next); | |
8536 | ||
b2aaaa3b JB |
8537 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8538 | level - 1); | |
1c4850e2 YZ |
8539 | reada = 1; |
8540 | } | |
8541 | btrfs_tree_lock(next); | |
8bead258 | 8542 | btrfs_set_lock_blocking_write(next); |
1c4850e2 | 8543 | |
2ff7e61e | 8544 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
8545 | &wc->refs[level - 1], |
8546 | &wc->flags[level - 1]); | |
4867268c JB |
8547 | if (ret < 0) |
8548 | goto out_unlock; | |
79787eaa | 8549 | |
c2cf52eb | 8550 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 8551 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
8552 | ret = -EIO; |
8553 | goto out_unlock; | |
c2cf52eb | 8554 | } |
94fcca9f | 8555 | *lookup_info = 0; |
1c4850e2 | 8556 | |
94fcca9f | 8557 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8558 | if (wc->refs[level - 1] > 1) { |
1152651a | 8559 | need_account = true; |
94fcca9f YZ |
8560 | if (level == 1 && |
8561 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8562 | goto skip; | |
8563 | ||
1c4850e2 YZ |
8564 | if (!wc->update_ref || |
8565 | generation <= root->root_key.offset) | |
8566 | goto skip; | |
8567 | ||
8568 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8569 | path->slots[level]); | |
8570 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8571 | if (ret < 0) | |
8572 | goto skip; | |
8573 | ||
8574 | wc->stage = UPDATE_BACKREF; | |
8575 | wc->shared_level = level - 1; | |
8576 | } | |
94fcca9f YZ |
8577 | } else { |
8578 | if (level == 1 && | |
8579 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8580 | goto skip; | |
1c4850e2 YZ |
8581 | } |
8582 | ||
b9fab919 | 8583 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8584 | btrfs_tree_unlock(next); |
8585 | free_extent_buffer(next); | |
8586 | next = NULL; | |
94fcca9f | 8587 | *lookup_info = 1; |
1c4850e2 YZ |
8588 | } |
8589 | ||
8590 | if (!next) { | |
8591 | if (reada && level == 1) | |
8592 | reada_walk_down(trans, root, wc, path); | |
581c1760 QW |
8593 | next = read_tree_block(fs_info, bytenr, generation, level - 1, |
8594 | &first_key); | |
64c043de LB |
8595 | if (IS_ERR(next)) { |
8596 | return PTR_ERR(next); | |
8597 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8598 | free_extent_buffer(next); |
97d9a8a4 | 8599 | return -EIO; |
416bc658 | 8600 | } |
1c4850e2 | 8601 | btrfs_tree_lock(next); |
8bead258 | 8602 | btrfs_set_lock_blocking_write(next); |
1c4850e2 YZ |
8603 | } |
8604 | ||
8605 | level--; | |
4867268c JB |
8606 | ASSERT(level == btrfs_header_level(next)); |
8607 | if (level != btrfs_header_level(next)) { | |
8608 | btrfs_err(root->fs_info, "mismatched level"); | |
8609 | ret = -EIO; | |
8610 | goto out_unlock; | |
8611 | } | |
1c4850e2 YZ |
8612 | path->nodes[level] = next; |
8613 | path->slots[level] = 0; | |
bd681513 | 8614 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8615 | wc->level = level; |
8616 | if (wc->level == 1) | |
8617 | wc->reada_slot = 0; | |
8618 | return 0; | |
8619 | skip: | |
8620 | wc->refs[level - 1] = 0; | |
8621 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
8622 | if (wc->stage == DROP_REFERENCE) { |
8623 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8624 | parent = path->nodes[level]->start; | |
8625 | } else { | |
4867268c | 8626 | ASSERT(root->root_key.objectid == |
94fcca9f | 8627 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
8628 | if (root->root_key.objectid != |
8629 | btrfs_header_owner(path->nodes[level])) { | |
8630 | btrfs_err(root->fs_info, | |
8631 | "mismatched block owner"); | |
8632 | ret = -EIO; | |
8633 | goto out_unlock; | |
8634 | } | |
94fcca9f YZ |
8635 | parent = 0; |
8636 | } | |
1c4850e2 | 8637 | |
78c52d9e JB |
8638 | /* |
8639 | * If we had a drop_progress we need to verify the refs are set | |
8640 | * as expected. If we find our ref then we know that from here | |
8641 | * on out everything should be correct, and we can clear the | |
8642 | * ->restarted flag. | |
8643 | */ | |
8644 | if (wc->restarted) { | |
8645 | ret = check_ref_exists(trans, root, bytenr, parent, | |
8646 | level - 1); | |
8647 | if (ret < 0) | |
8648 | goto out_unlock; | |
8649 | if (ret == 0) | |
8650 | goto no_delete; | |
8651 | ret = 0; | |
8652 | wc->restarted = 0; | |
8653 | } | |
8654 | ||
2cd86d30 QW |
8655 | /* |
8656 | * Reloc tree doesn't contribute to qgroup numbers, and we have | |
8657 | * already accounted them at merge time (replace_path), | |
8658 | * thus we could skip expensive subtree trace here. | |
8659 | */ | |
8660 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID && | |
8661 | need_account) { | |
deb40627 | 8662 | ret = btrfs_qgroup_trace_subtree(trans, next, |
33d1f05c | 8663 | generation, level - 1); |
1152651a | 8664 | if (ret) { |
0b246afa | 8665 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8666 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
8667 | ret); | |
1152651a MF |
8668 | } |
8669 | } | |
aea6f028 JB |
8670 | |
8671 | /* | |
8672 | * We need to update the next key in our walk control so we can | |
8673 | * update the drop_progress key accordingly. We don't care if | |
8674 | * find_next_key doesn't find a key because that means we're at | |
8675 | * the end and are going to clean up now. | |
8676 | */ | |
8677 | wc->drop_level = level; | |
8678 | find_next_key(path, level, &wc->drop_progress); | |
8679 | ||
ffd4bb2a QW |
8680 | btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, |
8681 | fs_info->nodesize, parent); | |
8682 | btrfs_init_tree_ref(&ref, level - 1, root->root_key.objectid); | |
8683 | ret = btrfs_free_extent(trans, &ref); | |
4867268c JB |
8684 | if (ret) |
8685 | goto out_unlock; | |
1c4850e2 | 8686 | } |
78c52d9e | 8687 | no_delete: |
4867268c JB |
8688 | *lookup_info = 1; |
8689 | ret = 1; | |
8690 | ||
8691 | out_unlock: | |
1c4850e2 YZ |
8692 | btrfs_tree_unlock(next); |
8693 | free_extent_buffer(next); | |
4867268c JB |
8694 | |
8695 | return ret; | |
1c4850e2 YZ |
8696 | } |
8697 | ||
2c47e605 | 8698 | /* |
2c016dc2 | 8699 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
8700 | * |
8701 | * when wc->stage == DROP_REFERENCE, this function drops | |
8702 | * reference count on the block. | |
8703 | * | |
8704 | * when wc->stage == UPDATE_BACKREF, this function changes | |
8705 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
8706 | * to UPDATE_BACKREF previously while processing the block. | |
8707 | * | |
8708 | * NOTE: return value 1 means we should stop walking up. | |
8709 | */ | |
8710 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
8711 | struct btrfs_root *root, | |
8712 | struct btrfs_path *path, | |
8713 | struct walk_control *wc) | |
8714 | { | |
0b246afa | 8715 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8716 | int ret; |
2c47e605 YZ |
8717 | int level = wc->level; |
8718 | struct extent_buffer *eb = path->nodes[level]; | |
8719 | u64 parent = 0; | |
8720 | ||
8721 | if (wc->stage == UPDATE_BACKREF) { | |
8722 | BUG_ON(wc->shared_level < level); | |
8723 | if (level < wc->shared_level) | |
8724 | goto out; | |
8725 | ||
2c47e605 YZ |
8726 | ret = find_next_key(path, level + 1, &wc->update_progress); |
8727 | if (ret > 0) | |
8728 | wc->update_ref = 0; | |
8729 | ||
8730 | wc->stage = DROP_REFERENCE; | |
8731 | wc->shared_level = -1; | |
8732 | path->slots[level] = 0; | |
8733 | ||
8734 | /* | |
8735 | * check reference count again if the block isn't locked. | |
8736 | * we should start walking down the tree again if reference | |
8737 | * count is one. | |
8738 | */ | |
8739 | if (!path->locks[level]) { | |
8740 | BUG_ON(level == 0); | |
8741 | btrfs_tree_lock(eb); | |
8bead258 | 8742 | btrfs_set_lock_blocking_write(eb); |
bd681513 | 8743 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8744 | |
2ff7e61e | 8745 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8746 | eb->start, level, 1, |
2c47e605 YZ |
8747 | &wc->refs[level], |
8748 | &wc->flags[level]); | |
79787eaa JM |
8749 | if (ret < 0) { |
8750 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 8751 | path->locks[level] = 0; |
79787eaa JM |
8752 | return ret; |
8753 | } | |
2c47e605 YZ |
8754 | BUG_ON(wc->refs[level] == 0); |
8755 | if (wc->refs[level] == 1) { | |
bd681513 | 8756 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 8757 | path->locks[level] = 0; |
2c47e605 YZ |
8758 | return 1; |
8759 | } | |
f82d02d9 | 8760 | } |
2c47e605 | 8761 | } |
f82d02d9 | 8762 | |
2c47e605 YZ |
8763 | /* wc->stage == DROP_REFERENCE */ |
8764 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 8765 | |
2c47e605 YZ |
8766 | if (wc->refs[level] == 1) { |
8767 | if (level == 0) { | |
8768 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 8769 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 8770 | else |
e339a6b0 | 8771 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8772 | BUG_ON(ret); /* -ENOMEM */ |
c4140cbf QW |
8773 | if (is_fstree(root->root_key.objectid)) { |
8774 | ret = btrfs_qgroup_trace_leaf_items(trans, eb); | |
8775 | if (ret) { | |
8776 | btrfs_err_rl(fs_info, | |
8777 | "error %d accounting leaf items, quota is out of sync, rescan required", | |
5d163e0e | 8778 | ret); |
c4140cbf | 8779 | } |
1152651a | 8780 | } |
2c47e605 | 8781 | } |
6a884d7d | 8782 | /* make block locked assertion in btrfs_clean_tree_block happy */ |
2c47e605 YZ |
8783 | if (!path->locks[level] && |
8784 | btrfs_header_generation(eb) == trans->transid) { | |
8785 | btrfs_tree_lock(eb); | |
8bead258 | 8786 | btrfs_set_lock_blocking_write(eb); |
bd681513 | 8787 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8788 | } |
6a884d7d | 8789 | btrfs_clean_tree_block(eb); |
2c47e605 YZ |
8790 | } |
8791 | ||
8792 | if (eb == root->node) { | |
8793 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8794 | parent = eb->start; | |
65c6e82b QW |
8795 | else if (root->root_key.objectid != btrfs_header_owner(eb)) |
8796 | goto owner_mismatch; | |
2c47e605 YZ |
8797 | } else { |
8798 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8799 | parent = path->nodes[level + 1]->start; | |
65c6e82b QW |
8800 | else if (root->root_key.objectid != |
8801 | btrfs_header_owner(path->nodes[level + 1])) | |
8802 | goto owner_mismatch; | |
f82d02d9 | 8803 | } |
f82d02d9 | 8804 | |
5581a51a | 8805 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
8806 | out: |
8807 | wc->refs[level] = 0; | |
8808 | wc->flags[level] = 0; | |
f0486c68 | 8809 | return 0; |
65c6e82b QW |
8810 | |
8811 | owner_mismatch: | |
8812 | btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu", | |
8813 | btrfs_header_owner(eb), root->root_key.objectid); | |
8814 | return -EUCLEAN; | |
2c47e605 YZ |
8815 | } |
8816 | ||
8817 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
8818 | struct btrfs_root *root, | |
8819 | struct btrfs_path *path, | |
8820 | struct walk_control *wc) | |
8821 | { | |
2c47e605 | 8822 | int level = wc->level; |
94fcca9f | 8823 | int lookup_info = 1; |
2c47e605 YZ |
8824 | int ret; |
8825 | ||
8826 | while (level >= 0) { | |
94fcca9f | 8827 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
8828 | if (ret > 0) |
8829 | break; | |
8830 | ||
8831 | if (level == 0) | |
8832 | break; | |
8833 | ||
7a7965f8 YZ |
8834 | if (path->slots[level] >= |
8835 | btrfs_header_nritems(path->nodes[level])) | |
8836 | break; | |
8837 | ||
94fcca9f | 8838 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
8839 | if (ret > 0) { |
8840 | path->slots[level]++; | |
8841 | continue; | |
90d2c51d MX |
8842 | } else if (ret < 0) |
8843 | return ret; | |
1c4850e2 | 8844 | level = wc->level; |
f82d02d9 | 8845 | } |
f82d02d9 YZ |
8846 | return 0; |
8847 | } | |
8848 | ||
d397712b | 8849 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 8850 | struct btrfs_root *root, |
f82d02d9 | 8851 | struct btrfs_path *path, |
2c47e605 | 8852 | struct walk_control *wc, int max_level) |
20524f02 | 8853 | { |
2c47e605 | 8854 | int level = wc->level; |
20524f02 | 8855 | int ret; |
9f3a7427 | 8856 | |
2c47e605 YZ |
8857 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
8858 | while (level < max_level && path->nodes[level]) { | |
8859 | wc->level = level; | |
8860 | if (path->slots[level] + 1 < | |
8861 | btrfs_header_nritems(path->nodes[level])) { | |
8862 | path->slots[level]++; | |
20524f02 CM |
8863 | return 0; |
8864 | } else { | |
2c47e605 YZ |
8865 | ret = walk_up_proc(trans, root, path, wc); |
8866 | if (ret > 0) | |
8867 | return 0; | |
65c6e82b QW |
8868 | if (ret < 0) |
8869 | return ret; | |
bd56b302 | 8870 | |
2c47e605 | 8871 | if (path->locks[level]) { |
bd681513 CM |
8872 | btrfs_tree_unlock_rw(path->nodes[level], |
8873 | path->locks[level]); | |
2c47e605 | 8874 | path->locks[level] = 0; |
f82d02d9 | 8875 | } |
2c47e605 YZ |
8876 | free_extent_buffer(path->nodes[level]); |
8877 | path->nodes[level] = NULL; | |
8878 | level++; | |
20524f02 CM |
8879 | } |
8880 | } | |
8881 | return 1; | |
8882 | } | |
8883 | ||
9aca1d51 | 8884 | /* |
2c47e605 YZ |
8885 | * drop a subvolume tree. |
8886 | * | |
8887 | * this function traverses the tree freeing any blocks that only | |
8888 | * referenced by the tree. | |
8889 | * | |
8890 | * when a shared tree block is found. this function decreases its | |
8891 | * reference count by one. if update_ref is true, this function | |
8892 | * also make sure backrefs for the shared block and all lower level | |
8893 | * blocks are properly updated. | |
9d1a2a3a DS |
8894 | * |
8895 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 8896 | */ |
2c536799 | 8897 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
8898 | struct btrfs_block_rsv *block_rsv, int update_ref, |
8899 | int for_reloc) | |
20524f02 | 8900 | { |
ab8d0fc4 | 8901 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 8902 | struct btrfs_path *path; |
2c47e605 | 8903 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 8904 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 8905 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
8906 | struct walk_control *wc; |
8907 | struct btrfs_key key; | |
8908 | int err = 0; | |
8909 | int ret; | |
8910 | int level; | |
d29a9f62 | 8911 | bool root_dropped = false; |
20524f02 | 8912 | |
4fd786e6 | 8913 | btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid); |
1152651a | 8914 | |
5caf2a00 | 8915 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
8916 | if (!path) { |
8917 | err = -ENOMEM; | |
8918 | goto out; | |
8919 | } | |
20524f02 | 8920 | |
2c47e605 | 8921 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
8922 | if (!wc) { |
8923 | btrfs_free_path(path); | |
cb1b69f4 TI |
8924 | err = -ENOMEM; |
8925 | goto out; | |
38a1a919 | 8926 | } |
2c47e605 | 8927 | |
a22285a6 | 8928 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
8929 | if (IS_ERR(trans)) { |
8930 | err = PTR_ERR(trans); | |
8931 | goto out_free; | |
8932 | } | |
98d5dc13 | 8933 | |
0568e82d JB |
8934 | err = btrfs_run_delayed_items(trans); |
8935 | if (err) | |
8936 | goto out_end_trans; | |
8937 | ||
3fd0a558 YZ |
8938 | if (block_rsv) |
8939 | trans->block_rsv = block_rsv; | |
2c47e605 | 8940 | |
83354f07 JB |
8941 | /* |
8942 | * This will help us catch people modifying the fs tree while we're | |
8943 | * dropping it. It is unsafe to mess with the fs tree while it's being | |
8944 | * dropped as we unlock the root node and parent nodes as we walk down | |
8945 | * the tree, assuming nothing will change. If something does change | |
8946 | * then we'll have stale information and drop references to blocks we've | |
8947 | * already dropped. | |
8948 | */ | |
8949 | set_bit(BTRFS_ROOT_DELETING, &root->state); | |
9f3a7427 | 8950 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 8951 | level = btrfs_header_level(root->node); |
5d4f98a2 | 8952 | path->nodes[level] = btrfs_lock_root_node(root); |
8bead258 | 8953 | btrfs_set_lock_blocking_write(path->nodes[level]); |
9f3a7427 | 8954 | path->slots[level] = 0; |
bd681513 | 8955 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
8956 | memset(&wc->update_progress, 0, |
8957 | sizeof(wc->update_progress)); | |
9f3a7427 | 8958 | } else { |
9f3a7427 | 8959 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
8960 | memcpy(&wc->update_progress, &key, |
8961 | sizeof(wc->update_progress)); | |
8962 | ||
6702ed49 | 8963 | level = root_item->drop_level; |
2c47e605 | 8964 | BUG_ON(level == 0); |
6702ed49 | 8965 | path->lowest_level = level; |
2c47e605 YZ |
8966 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
8967 | path->lowest_level = 0; | |
8968 | if (ret < 0) { | |
8969 | err = ret; | |
79787eaa | 8970 | goto out_end_trans; |
9f3a7427 | 8971 | } |
1c4850e2 | 8972 | WARN_ON(ret > 0); |
2c47e605 | 8973 | |
7d9eb12c CM |
8974 | /* |
8975 | * unlock our path, this is safe because only this | |
8976 | * function is allowed to delete this snapshot | |
8977 | */ | |
5d4f98a2 | 8978 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
8979 | |
8980 | level = btrfs_header_level(root->node); | |
8981 | while (1) { | |
8982 | btrfs_tree_lock(path->nodes[level]); | |
8bead258 | 8983 | btrfs_set_lock_blocking_write(path->nodes[level]); |
fec386ac | 8984 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8985 | |
2ff7e61e | 8986 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 8987 | path->nodes[level]->start, |
3173a18f | 8988 | level, 1, &wc->refs[level], |
2c47e605 | 8989 | &wc->flags[level]); |
79787eaa JM |
8990 | if (ret < 0) { |
8991 | err = ret; | |
8992 | goto out_end_trans; | |
8993 | } | |
2c47e605 YZ |
8994 | BUG_ON(wc->refs[level] == 0); |
8995 | ||
8996 | if (level == root_item->drop_level) | |
8997 | break; | |
8998 | ||
8999 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 9000 | path->locks[level] = 0; |
2c47e605 YZ |
9001 | WARN_ON(wc->refs[level] != 1); |
9002 | level--; | |
9003 | } | |
9f3a7427 | 9004 | } |
2c47e605 | 9005 | |
78c52d9e | 9006 | wc->restarted = test_bit(BTRFS_ROOT_DEAD_TREE, &root->state); |
2c47e605 YZ |
9007 | wc->level = level; |
9008 | wc->shared_level = -1; | |
9009 | wc->stage = DROP_REFERENCE; | |
9010 | wc->update_ref = update_ref; | |
9011 | wc->keep_locks = 0; | |
0b246afa | 9012 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 9013 | |
d397712b | 9014 | while (1) { |
9d1a2a3a | 9015 | |
2c47e605 YZ |
9016 | ret = walk_down_tree(trans, root, path, wc); |
9017 | if (ret < 0) { | |
9018 | err = ret; | |
20524f02 | 9019 | break; |
2c47e605 | 9020 | } |
9aca1d51 | 9021 | |
2c47e605 YZ |
9022 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
9023 | if (ret < 0) { | |
9024 | err = ret; | |
20524f02 | 9025 | break; |
2c47e605 YZ |
9026 | } |
9027 | ||
9028 | if (ret > 0) { | |
9029 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
9030 | break; |
9031 | } | |
2c47e605 YZ |
9032 | |
9033 | if (wc->stage == DROP_REFERENCE) { | |
aea6f028 JB |
9034 | wc->drop_level = wc->level; |
9035 | btrfs_node_key_to_cpu(path->nodes[wc->drop_level], | |
9036 | &wc->drop_progress, | |
9037 | path->slots[wc->drop_level]); | |
9038 | } | |
9039 | btrfs_cpu_key_to_disk(&root_item->drop_progress, | |
9040 | &wc->drop_progress); | |
9041 | root_item->drop_level = wc->drop_level; | |
2c47e605 YZ |
9042 | |
9043 | BUG_ON(wc->level == 0); | |
3a45bb20 | 9044 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 9045 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
9046 | ret = btrfs_update_root(trans, tree_root, |
9047 | &root->root_key, | |
9048 | root_item); | |
79787eaa | 9049 | if (ret) { |
66642832 | 9050 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9051 | err = ret; |
9052 | goto out_end_trans; | |
9053 | } | |
2c47e605 | 9054 | |
3a45bb20 | 9055 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 9056 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
9057 | btrfs_debug(fs_info, |
9058 | "drop snapshot early exit"); | |
3c8f2422 JB |
9059 | err = -EAGAIN; |
9060 | goto out_free; | |
9061 | } | |
9062 | ||
a22285a6 | 9063 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9064 | if (IS_ERR(trans)) { |
9065 | err = PTR_ERR(trans); | |
9066 | goto out_free; | |
9067 | } | |
3fd0a558 YZ |
9068 | if (block_rsv) |
9069 | trans->block_rsv = block_rsv; | |
c3e69d58 | 9070 | } |
20524f02 | 9071 | } |
b3b4aa74 | 9072 | btrfs_release_path(path); |
79787eaa JM |
9073 | if (err) |
9074 | goto out_end_trans; | |
2c47e605 | 9075 | |
ab9ce7d4 | 9076 | ret = btrfs_del_root(trans, &root->root_key); |
79787eaa | 9077 | if (ret) { |
66642832 | 9078 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 9079 | err = ret; |
79787eaa JM |
9080 | goto out_end_trans; |
9081 | } | |
2c47e605 | 9082 | |
76dda93c | 9083 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
9084 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
9085 | NULL, NULL); | |
79787eaa | 9086 | if (ret < 0) { |
66642832 | 9087 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9088 | err = ret; |
9089 | goto out_end_trans; | |
9090 | } else if (ret > 0) { | |
84cd948c JB |
9091 | /* if we fail to delete the orphan item this time |
9092 | * around, it'll get picked up the next time. | |
9093 | * | |
9094 | * The most common failure here is just -ENOENT. | |
9095 | */ | |
9096 | btrfs_del_orphan_item(trans, tree_root, | |
9097 | root->root_key.objectid); | |
76dda93c YZ |
9098 | } |
9099 | } | |
9100 | ||
27cdeb70 | 9101 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 9102 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
9103 | } else { |
9104 | free_extent_buffer(root->node); | |
9105 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 9106 | btrfs_put_fs_root(root); |
76dda93c | 9107 | } |
d29a9f62 | 9108 | root_dropped = true; |
79787eaa | 9109 | out_end_trans: |
3a45bb20 | 9110 | btrfs_end_transaction_throttle(trans); |
79787eaa | 9111 | out_free: |
2c47e605 | 9112 | kfree(wc); |
5caf2a00 | 9113 | btrfs_free_path(path); |
cb1b69f4 | 9114 | out: |
d29a9f62 JB |
9115 | /* |
9116 | * So if we need to stop dropping the snapshot for whatever reason we | |
9117 | * need to make sure to add it back to the dead root list so that we | |
9118 | * keep trying to do the work later. This also cleans up roots if we | |
9119 | * don't have it in the radix (like when we recover after a power fail | |
9120 | * or unmount) so we don't leak memory. | |
9121 | */ | |
897ca819 | 9122 | if (!for_reloc && !root_dropped) |
d29a9f62 | 9123 | btrfs_add_dead_root(root); |
90515e7f | 9124 | if (err && err != -EAGAIN) |
ab8d0fc4 | 9125 | btrfs_handle_fs_error(fs_info, err, NULL); |
2c536799 | 9126 | return err; |
20524f02 | 9127 | } |
9078a3e1 | 9128 | |
2c47e605 YZ |
9129 | /* |
9130 | * drop subtree rooted at tree block 'node'. | |
9131 | * | |
9132 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 9133 | * only used by relocation code |
2c47e605 | 9134 | */ |
f82d02d9 YZ |
9135 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
9136 | struct btrfs_root *root, | |
9137 | struct extent_buffer *node, | |
9138 | struct extent_buffer *parent) | |
9139 | { | |
0b246afa | 9140 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 9141 | struct btrfs_path *path; |
2c47e605 | 9142 | struct walk_control *wc; |
f82d02d9 YZ |
9143 | int level; |
9144 | int parent_level; | |
9145 | int ret = 0; | |
9146 | int wret; | |
9147 | ||
2c47e605 YZ |
9148 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
9149 | ||
f82d02d9 | 9150 | path = btrfs_alloc_path(); |
db5b493a TI |
9151 | if (!path) |
9152 | return -ENOMEM; | |
f82d02d9 | 9153 | |
2c47e605 | 9154 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
9155 | if (!wc) { |
9156 | btrfs_free_path(path); | |
9157 | return -ENOMEM; | |
9158 | } | |
2c47e605 | 9159 | |
b9447ef8 | 9160 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
9161 | parent_level = btrfs_header_level(parent); |
9162 | extent_buffer_get(parent); | |
9163 | path->nodes[parent_level] = parent; | |
9164 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9165 | ||
b9447ef8 | 9166 | btrfs_assert_tree_locked(node); |
f82d02d9 | 9167 | level = btrfs_header_level(node); |
f82d02d9 YZ |
9168 | path->nodes[level] = node; |
9169 | path->slots[level] = 0; | |
bd681513 | 9170 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9171 | |
9172 | wc->refs[parent_level] = 1; | |
9173 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9174 | wc->level = level; | |
9175 | wc->shared_level = -1; | |
9176 | wc->stage = DROP_REFERENCE; | |
9177 | wc->update_ref = 0; | |
9178 | wc->keep_locks = 1; | |
0b246afa | 9179 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
9180 | |
9181 | while (1) { | |
2c47e605 YZ |
9182 | wret = walk_down_tree(trans, root, path, wc); |
9183 | if (wret < 0) { | |
f82d02d9 | 9184 | ret = wret; |
f82d02d9 | 9185 | break; |
2c47e605 | 9186 | } |
f82d02d9 | 9187 | |
2c47e605 | 9188 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9189 | if (wret < 0) |
9190 | ret = wret; | |
9191 | if (wret != 0) | |
9192 | break; | |
9193 | } | |
9194 | ||
2c47e605 | 9195 | kfree(wc); |
f82d02d9 YZ |
9196 | btrfs_free_path(path); |
9197 | return ret; | |
9198 | } | |
9199 | ||
6202df69 | 9200 | static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c CM |
9201 | { |
9202 | u64 num_devices; | |
fc67c450 | 9203 | u64 stripped; |
e4d8ec0f | 9204 | |
fc67c450 ID |
9205 | /* |
9206 | * if restripe for this chunk_type is on pick target profile and | |
9207 | * return, otherwise do the usual balance | |
9208 | */ | |
6202df69 | 9209 | stripped = get_restripe_target(fs_info, flags); |
fc67c450 ID |
9210 | if (stripped) |
9211 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9212 | |
6202df69 | 9213 | num_devices = fs_info->fs_devices->rw_devices; |
cd02dca5 | 9214 | |
a07e8a46 | 9215 | stripped = BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID56_MASK | |
c7369b3f | 9216 | BTRFS_BLOCK_GROUP_RAID1_MASK | BTRFS_BLOCK_GROUP_RAID10; |
fc67c450 | 9217 | |
ec44a35c CM |
9218 | if (num_devices == 1) { |
9219 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9220 | stripped = flags & ~stripped; | |
9221 | ||
9222 | /* turn raid0 into single device chunks */ | |
9223 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9224 | return stripped; | |
9225 | ||
9226 | /* turn mirroring into duplication */ | |
c7369b3f | 9227 | if (flags & (BTRFS_BLOCK_GROUP_RAID1_MASK | |
ec44a35c CM |
9228 | BTRFS_BLOCK_GROUP_RAID10)) |
9229 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9230 | } else { |
9231 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9232 | if (flags & stripped) |
9233 | return flags; | |
9234 | ||
9235 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9236 | stripped = flags & ~stripped; | |
9237 | ||
9238 | /* switch duplicated blocks with raid1 */ | |
9239 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9240 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9241 | ||
e3176ca2 | 9242 | /* this is drive concat, leave it alone */ |
ec44a35c | 9243 | } |
e3176ca2 | 9244 | |
ec44a35c CM |
9245 | return flags; |
9246 | } | |
9247 | ||
868f401a | 9248 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9249 | { |
f0486c68 YZ |
9250 | struct btrfs_space_info *sinfo = cache->space_info; |
9251 | u64 num_bytes; | |
3ece54e5 | 9252 | u64 sinfo_used; |
199c36ea | 9253 | u64 min_allocable_bytes; |
f0486c68 | 9254 | int ret = -ENOSPC; |
0ef3e66b | 9255 | |
199c36ea MX |
9256 | /* |
9257 | * We need some metadata space and system metadata space for | |
9258 | * allocating chunks in some corner cases until we force to set | |
9259 | * it to be readonly. | |
9260 | */ | |
9261 | if ((sinfo->flags & | |
9262 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9263 | !force) | |
ee22184b | 9264 | min_allocable_bytes = SZ_1M; |
199c36ea MX |
9265 | else |
9266 | min_allocable_bytes = 0; | |
9267 | ||
f0486c68 YZ |
9268 | spin_lock(&sinfo->lock); |
9269 | spin_lock(&cache->lock); | |
61cfea9b W |
9270 | |
9271 | if (cache->ro) { | |
868f401a | 9272 | cache->ro++; |
61cfea9b W |
9273 | ret = 0; |
9274 | goto out; | |
9275 | } | |
9276 | ||
f0486c68 YZ |
9277 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9278 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
3ece54e5 | 9279 | sinfo_used = btrfs_space_info_used(sinfo, true); |
f0486c68 | 9280 | |
3ece54e5 QW |
9281 | if (sinfo_used + num_bytes + min_allocable_bytes <= |
9282 | sinfo->total_bytes) { | |
f0486c68 | 9283 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9284 | cache->ro++; |
633c0aad | 9285 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9286 | ret = 0; |
9287 | } | |
61cfea9b | 9288 | out: |
f0486c68 YZ |
9289 | spin_unlock(&cache->lock); |
9290 | spin_unlock(&sinfo->lock); | |
3ece54e5 QW |
9291 | if (ret == -ENOSPC && btrfs_test_opt(cache->fs_info, ENOSPC_DEBUG)) { |
9292 | btrfs_info(cache->fs_info, | |
9293 | "unable to make block group %llu ro", | |
9294 | cache->key.objectid); | |
9295 | btrfs_info(cache->fs_info, | |
9296 | "sinfo_used=%llu bg_num_bytes=%llu min_allocable=%llu", | |
9297 | sinfo_used, num_bytes, min_allocable_bytes); | |
9298 | dump_space_info(cache->fs_info, cache->space_info, 0, 0); | |
9299 | } | |
f0486c68 YZ |
9300 | return ret; |
9301 | } | |
7d9eb12c | 9302 | |
c83488af | 9303 | int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache) |
c286ac48 | 9304 | |
f0486c68 | 9305 | { |
c83488af | 9306 | struct btrfs_fs_info *fs_info = cache->fs_info; |
f0486c68 YZ |
9307 | struct btrfs_trans_handle *trans; |
9308 | u64 alloc_flags; | |
9309 | int ret; | |
7d9eb12c | 9310 | |
1bbc621e | 9311 | again: |
5e00f193 | 9312 | trans = btrfs_join_transaction(fs_info->extent_root); |
79787eaa JM |
9313 | if (IS_ERR(trans)) |
9314 | return PTR_ERR(trans); | |
5d4f98a2 | 9315 | |
1bbc621e CM |
9316 | /* |
9317 | * we're not allowed to set block groups readonly after the dirty | |
9318 | * block groups cache has started writing. If it already started, | |
9319 | * back off and let this transaction commit | |
9320 | */ | |
0b246afa | 9321 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c | 9322 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9323 | u64 transid = trans->transid; |
9324 | ||
0b246afa | 9325 | mutex_unlock(&fs_info->ro_block_group_mutex); |
3a45bb20 | 9326 | btrfs_end_transaction(trans); |
1bbc621e | 9327 | |
2ff7e61e | 9328 | ret = btrfs_wait_for_commit(fs_info, transid); |
1bbc621e CM |
9329 | if (ret) |
9330 | return ret; | |
9331 | goto again; | |
9332 | } | |
9333 | ||
153c35b6 CM |
9334 | /* |
9335 | * if we are changing raid levels, try to allocate a corresponding | |
9336 | * block group with the new raid level. | |
9337 | */ | |
0b246afa | 9338 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
153c35b6 | 9339 | if (alloc_flags != cache->flags) { |
fc471cb0 | 9340 | ret = btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); |
153c35b6 CM |
9341 | /* |
9342 | * ENOSPC is allowed here, we may have enough space | |
9343 | * already allocated at the new raid level to | |
9344 | * carry on | |
9345 | */ | |
9346 | if (ret == -ENOSPC) | |
9347 | ret = 0; | |
9348 | if (ret < 0) | |
9349 | goto out; | |
9350 | } | |
1bbc621e | 9351 | |
868f401a | 9352 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9353 | if (!ret) |
9354 | goto out; | |
2ff7e61e | 9355 | alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags); |
fc471cb0 | 9356 | ret = btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9357 | if (ret < 0) |
9358 | goto out; | |
868f401a | 9359 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9360 | out: |
2f081088 | 9361 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 9362 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
34441361 | 9363 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 9364 | check_system_chunk(trans, alloc_flags); |
34441361 | 9365 | mutex_unlock(&fs_info->chunk_mutex); |
2f081088 | 9366 | } |
0b246afa | 9367 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2f081088 | 9368 | |
3a45bb20 | 9369 | btrfs_end_transaction(trans); |
f0486c68 YZ |
9370 | return ret; |
9371 | } | |
5d4f98a2 | 9372 | |
43a7e99d | 9373 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type) |
c87f08ca | 9374 | { |
43a7e99d | 9375 | u64 alloc_flags = get_alloc_profile(trans->fs_info, type); |
2ff7e61e | 9376 | |
fc471cb0 | 9377 | return btrfs_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); |
c87f08ca CM |
9378 | } |
9379 | ||
6d07bcec MX |
9380 | /* |
9381 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9382 | * space_info. takes mirrors into account. |
6d07bcec | 9383 | */ |
633c0aad | 9384 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9385 | { |
9386 | struct btrfs_block_group_cache *block_group; | |
9387 | u64 free_bytes = 0; | |
9388 | int factor; | |
9389 | ||
01327610 | 9390 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
9391 | if (list_empty(&sinfo->ro_bgs)) |
9392 | return 0; | |
9393 | ||
9394 | spin_lock(&sinfo->lock); | |
9395 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9396 | spin_lock(&block_group->lock); |
9397 | ||
9398 | if (!block_group->ro) { | |
9399 | spin_unlock(&block_group->lock); | |
9400 | continue; | |
9401 | } | |
9402 | ||
46df06b8 | 9403 | factor = btrfs_bg_type_to_factor(block_group->flags); |
6d07bcec MX |
9404 | free_bytes += (block_group->key.offset - |
9405 | btrfs_block_group_used(&block_group->item)) * | |
9406 | factor; | |
9407 | ||
9408 | spin_unlock(&block_group->lock); | |
9409 | } | |
6d07bcec MX |
9410 | spin_unlock(&sinfo->lock); |
9411 | ||
9412 | return free_bytes; | |
9413 | } | |
9414 | ||
2ff7e61e | 9415 | void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9416 | { |
f0486c68 YZ |
9417 | struct btrfs_space_info *sinfo = cache->space_info; |
9418 | u64 num_bytes; | |
9419 | ||
9420 | BUG_ON(!cache->ro); | |
9421 | ||
9422 | spin_lock(&sinfo->lock); | |
9423 | spin_lock(&cache->lock); | |
868f401a Z |
9424 | if (!--cache->ro) { |
9425 | num_bytes = cache->key.offset - cache->reserved - | |
9426 | cache->pinned - cache->bytes_super - | |
9427 | btrfs_block_group_used(&cache->item); | |
9428 | sinfo->bytes_readonly -= num_bytes; | |
9429 | list_del_init(&cache->ro_list); | |
9430 | } | |
f0486c68 YZ |
9431 | spin_unlock(&cache->lock); |
9432 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9433 | } |
9434 | ||
ba1bf481 | 9435 | /* |
52042d8e | 9436 | * Checks to see if it's even possible to relocate this block group. |
ba1bf481 JB |
9437 | * |
9438 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9439 | * ok to go ahead and try. | |
9440 | */ | |
6bccf3ab | 9441 | int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr) |
1a40e23b | 9442 | { |
ba1bf481 JB |
9443 | struct btrfs_block_group_cache *block_group; |
9444 | struct btrfs_space_info *space_info; | |
0b246afa | 9445 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
ba1bf481 | 9446 | struct btrfs_device *device; |
cdcb725c | 9447 | u64 min_free; |
6719db6a JB |
9448 | u64 dev_min = 1; |
9449 | u64 dev_nr = 0; | |
4a5e98f5 | 9450 | u64 target; |
0305bc27 | 9451 | int debug; |
cdcb725c | 9452 | int index; |
ba1bf481 JB |
9453 | int full = 0; |
9454 | int ret = 0; | |
1a40e23b | 9455 | |
0b246afa | 9456 | debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG); |
0305bc27 | 9457 | |
0b246afa | 9458 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
1a40e23b | 9459 | |
ba1bf481 | 9460 | /* odd, couldn't find the block group, leave it alone */ |
0305bc27 QW |
9461 | if (!block_group) { |
9462 | if (debug) | |
0b246afa | 9463 | btrfs_warn(fs_info, |
0305bc27 QW |
9464 | "can't find block group for bytenr %llu", |
9465 | bytenr); | |
ba1bf481 | 9466 | return -1; |
0305bc27 | 9467 | } |
1a40e23b | 9468 | |
cdcb725c | 9469 | min_free = btrfs_block_group_used(&block_group->item); |
9470 | ||
ba1bf481 | 9471 | /* no bytes used, we're good */ |
cdcb725c | 9472 | if (!min_free) |
1a40e23b ZY |
9473 | goto out; |
9474 | ||
ba1bf481 JB |
9475 | space_info = block_group->space_info; |
9476 | spin_lock(&space_info->lock); | |
17d217fe | 9477 | |
ba1bf481 | 9478 | full = space_info->full; |
17d217fe | 9479 | |
ba1bf481 JB |
9480 | /* |
9481 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9482 | * relocate it unless we're able to allocate a new chunk below. |
9483 | * | |
9484 | * Otherwise, we need to make sure we have room in the space to handle | |
9485 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9486 | */ |
7ce618db | 9487 | if ((space_info->total_bytes != block_group->key.offset) && |
4136135b LB |
9488 | (btrfs_space_info_used(space_info, false) + min_free < |
9489 | space_info->total_bytes)) { | |
ba1bf481 JB |
9490 | spin_unlock(&space_info->lock); |
9491 | goto out; | |
17d217fe | 9492 | } |
ba1bf481 | 9493 | spin_unlock(&space_info->lock); |
ea8c2819 | 9494 | |
ba1bf481 JB |
9495 | /* |
9496 | * ok we don't have enough space, but maybe we have free space on our | |
9497 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9498 | * alloc devices and guess if we have enough space. if this block |
9499 | * group is going to be restriped, run checks against the target | |
9500 | * profile instead of the current one. | |
ba1bf481 JB |
9501 | */ |
9502 | ret = -1; | |
ea8c2819 | 9503 | |
cdcb725c | 9504 | /* |
9505 | * index: | |
9506 | * 0: raid10 | |
9507 | * 1: raid1 | |
9508 | * 2: dup | |
9509 | * 3: raid0 | |
9510 | * 4: single | |
9511 | */ | |
0b246afa | 9512 | target = get_restripe_target(fs_info, block_group->flags); |
4a5e98f5 | 9513 | if (target) { |
3e72ee88 | 9514 | index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9515 | } else { |
9516 | /* | |
9517 | * this is just a balance, so if we were marked as full | |
9518 | * we know there is no space for a new chunk | |
9519 | */ | |
0305bc27 QW |
9520 | if (full) { |
9521 | if (debug) | |
0b246afa JM |
9522 | btrfs_warn(fs_info, |
9523 | "no space to alloc new chunk for block group %llu", | |
9524 | block_group->key.objectid); | |
4a5e98f5 | 9525 | goto out; |
0305bc27 | 9526 | } |
4a5e98f5 | 9527 | |
3e72ee88 | 9528 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
4a5e98f5 ID |
9529 | } |
9530 | ||
e6ec716f | 9531 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9532 | dev_min = 4; |
6719db6a JB |
9533 | /* Divide by 2 */ |
9534 | min_free >>= 1; | |
e6ec716f | 9535 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9536 | dev_min = 2; |
e6ec716f | 9537 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9538 | /* Multiply by 2 */ |
9539 | min_free <<= 1; | |
e6ec716f | 9540 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9541 | dev_min = fs_devices->rw_devices; |
47c5713f | 9542 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9543 | } |
9544 | ||
0b246afa | 9545 | mutex_lock(&fs_info->chunk_mutex); |
ba1bf481 | 9546 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
7bfc837d | 9547 | u64 dev_offset; |
56bec294 | 9548 | |
ba1bf481 JB |
9549 | /* |
9550 | * check to make sure we can actually find a chunk with enough | |
9551 | * space to fit our block group in. | |
9552 | */ | |
63a212ab | 9553 | if (device->total_bytes > device->bytes_used + min_free && |
401e29c1 | 9554 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
60dfdf25 | 9555 | ret = find_free_dev_extent(device, min_free, |
7bfc837d | 9556 | &dev_offset, NULL); |
ba1bf481 | 9557 | if (!ret) |
cdcb725c | 9558 | dev_nr++; |
9559 | ||
9560 | if (dev_nr >= dev_min) | |
73e48b27 | 9561 | break; |
cdcb725c | 9562 | |
ba1bf481 | 9563 | ret = -1; |
725c8463 | 9564 | } |
edbd8d4e | 9565 | } |
0305bc27 | 9566 | if (debug && ret == -1) |
0b246afa JM |
9567 | btrfs_warn(fs_info, |
9568 | "no space to allocate a new chunk for block group %llu", | |
9569 | block_group->key.objectid); | |
9570 | mutex_unlock(&fs_info->chunk_mutex); | |
edbd8d4e | 9571 | out: |
ba1bf481 | 9572 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9573 | return ret; |
9574 | } | |
9575 | ||
6bccf3ab JM |
9576 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
9577 | struct btrfs_path *path, | |
9578 | struct btrfs_key *key) | |
0b86a832 | 9579 | { |
6bccf3ab | 9580 | struct btrfs_root *root = fs_info->extent_root; |
925baedd | 9581 | int ret = 0; |
0b86a832 CM |
9582 | struct btrfs_key found_key; |
9583 | struct extent_buffer *leaf; | |
514c7dca QW |
9584 | struct btrfs_block_group_item bg; |
9585 | u64 flags; | |
0b86a832 | 9586 | int slot; |
edbd8d4e | 9587 | |
0b86a832 CM |
9588 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9589 | if (ret < 0) | |
925baedd CM |
9590 | goto out; |
9591 | ||
d397712b | 9592 | while (1) { |
0b86a832 | 9593 | slot = path->slots[0]; |
edbd8d4e | 9594 | leaf = path->nodes[0]; |
0b86a832 CM |
9595 | if (slot >= btrfs_header_nritems(leaf)) { |
9596 | ret = btrfs_next_leaf(root, path); | |
9597 | if (ret == 0) | |
9598 | continue; | |
9599 | if (ret < 0) | |
925baedd | 9600 | goto out; |
0b86a832 | 9601 | break; |
edbd8d4e | 9602 | } |
0b86a832 | 9603 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9604 | |
0b86a832 | 9605 | if (found_key.objectid >= key->objectid && |
925baedd | 9606 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6fb37b75 LB |
9607 | struct extent_map_tree *em_tree; |
9608 | struct extent_map *em; | |
9609 | ||
c8bf1b67 | 9610 | em_tree = &root->fs_info->mapping_tree; |
6fb37b75 LB |
9611 | read_lock(&em_tree->lock); |
9612 | em = lookup_extent_mapping(em_tree, found_key.objectid, | |
9613 | found_key.offset); | |
9614 | read_unlock(&em_tree->lock); | |
9615 | if (!em) { | |
0b246afa | 9616 | btrfs_err(fs_info, |
6fb37b75 LB |
9617 | "logical %llu len %llu found bg but no related chunk", |
9618 | found_key.objectid, found_key.offset); | |
9619 | ret = -ENOENT; | |
514c7dca QW |
9620 | } else if (em->start != found_key.objectid || |
9621 | em->len != found_key.offset) { | |
9622 | btrfs_err(fs_info, | |
9623 | "block group %llu len %llu mismatch with chunk %llu len %llu", | |
9624 | found_key.objectid, found_key.offset, | |
9625 | em->start, em->len); | |
9626 | ret = -EUCLEAN; | |
6fb37b75 | 9627 | } else { |
514c7dca QW |
9628 | read_extent_buffer(leaf, &bg, |
9629 | btrfs_item_ptr_offset(leaf, slot), | |
9630 | sizeof(bg)); | |
9631 | flags = btrfs_block_group_flags(&bg) & | |
9632 | BTRFS_BLOCK_GROUP_TYPE_MASK; | |
9633 | ||
9634 | if (flags != (em->map_lookup->type & | |
9635 | BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
9636 | btrfs_err(fs_info, | |
9637 | "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx", | |
9638 | found_key.objectid, | |
9639 | found_key.offset, flags, | |
9640 | (BTRFS_BLOCK_GROUP_TYPE_MASK & | |
9641 | em->map_lookup->type)); | |
9642 | ret = -EUCLEAN; | |
9643 | } else { | |
9644 | ret = 0; | |
9645 | } | |
6fb37b75 | 9646 | } |
187ee58c | 9647 | free_extent_map(em); |
925baedd CM |
9648 | goto out; |
9649 | } | |
0b86a832 | 9650 | path->slots[0]++; |
edbd8d4e | 9651 | } |
925baedd | 9652 | out: |
0b86a832 | 9653 | return ret; |
edbd8d4e CM |
9654 | } |
9655 | ||
0af3d00b JB |
9656 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
9657 | { | |
9658 | struct btrfs_block_group_cache *block_group; | |
9659 | u64 last = 0; | |
9660 | ||
9661 | while (1) { | |
9662 | struct inode *inode; | |
9663 | ||
9664 | block_group = btrfs_lookup_first_block_group(info, last); | |
9665 | while (block_group) { | |
3aa7c7a3 | 9666 | wait_block_group_cache_done(block_group); |
0af3d00b JB |
9667 | spin_lock(&block_group->lock); |
9668 | if (block_group->iref) | |
9669 | break; | |
9670 | spin_unlock(&block_group->lock); | |
f87b7eb8 | 9671 | block_group = next_block_group(block_group); |
0af3d00b JB |
9672 | } |
9673 | if (!block_group) { | |
9674 | if (last == 0) | |
9675 | break; | |
9676 | last = 0; | |
9677 | continue; | |
9678 | } | |
9679 | ||
9680 | inode = block_group->inode; | |
9681 | block_group->iref = 0; | |
9682 | block_group->inode = NULL; | |
9683 | spin_unlock(&block_group->lock); | |
f3bca802 | 9684 | ASSERT(block_group->io_ctl.inode == NULL); |
0af3d00b JB |
9685 | iput(inode); |
9686 | last = block_group->key.objectid + block_group->key.offset; | |
9687 | btrfs_put_block_group(block_group); | |
9688 | } | |
9689 | } | |
9690 | ||
5cdd7db6 FM |
9691 | /* |
9692 | * Must be called only after stopping all workers, since we could have block | |
9693 | * group caching kthreads running, and therefore they could race with us if we | |
9694 | * freed the block groups before stopping them. | |
9695 | */ | |
1a40e23b ZY |
9696 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9697 | { | |
9698 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 9699 | struct btrfs_space_info *space_info; |
11833d66 | 9700 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
9701 | struct rb_node *n; |
9702 | ||
9e351cc8 | 9703 | down_write(&info->commit_root_sem); |
11833d66 YZ |
9704 | while (!list_empty(&info->caching_block_groups)) { |
9705 | caching_ctl = list_entry(info->caching_block_groups.next, | |
9706 | struct btrfs_caching_control, list); | |
9707 | list_del(&caching_ctl->list); | |
9708 | put_caching_control(caching_ctl); | |
9709 | } | |
9e351cc8 | 9710 | up_write(&info->commit_root_sem); |
11833d66 | 9711 | |
47ab2a6c JB |
9712 | spin_lock(&info->unused_bgs_lock); |
9713 | while (!list_empty(&info->unused_bgs)) { | |
9714 | block_group = list_first_entry(&info->unused_bgs, | |
9715 | struct btrfs_block_group_cache, | |
9716 | bg_list); | |
9717 | list_del_init(&block_group->bg_list); | |
9718 | btrfs_put_block_group(block_group); | |
9719 | } | |
9720 | spin_unlock(&info->unused_bgs_lock); | |
9721 | ||
1a40e23b ZY |
9722 | spin_lock(&info->block_group_cache_lock); |
9723 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
9724 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
9725 | cache_node); | |
1a40e23b ZY |
9726 | rb_erase(&block_group->cache_node, |
9727 | &info->block_group_cache_tree); | |
01eacb27 | 9728 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
9729 | spin_unlock(&info->block_group_cache_lock); |
9730 | ||
80eb234a | 9731 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 9732 | list_del(&block_group->list); |
80eb234a | 9733 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 9734 | |
3c14874a JB |
9735 | /* |
9736 | * We haven't cached this block group, which means we could | |
9737 | * possibly have excluded extents on this block group. | |
9738 | */ | |
36cce922 JB |
9739 | if (block_group->cached == BTRFS_CACHE_NO || |
9740 | block_group->cached == BTRFS_CACHE_ERROR) | |
9e715da8 | 9741 | free_excluded_extents(block_group); |
3c14874a | 9742 | |
817d52f8 | 9743 | btrfs_remove_free_space_cache(block_group); |
5cdd7db6 | 9744 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); |
f3bca802 LB |
9745 | ASSERT(list_empty(&block_group->dirty_list)); |
9746 | ASSERT(list_empty(&block_group->io_list)); | |
9747 | ASSERT(list_empty(&block_group->bg_list)); | |
9748 | ASSERT(atomic_read(&block_group->count) == 1); | |
11dfe35a | 9749 | btrfs_put_block_group(block_group); |
d899e052 YZ |
9750 | |
9751 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
9752 | } |
9753 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
9754 | |
9755 | /* now that all the block groups are freed, go through and | |
9756 | * free all the space_info structs. This is only called during | |
9757 | * the final stages of unmount, and so we know nobody is | |
9758 | * using them. We call synchronize_rcu() once before we start, | |
9759 | * just to be on the safe side. | |
9760 | */ | |
9761 | synchronize_rcu(); | |
9762 | ||
8929ecfa YZ |
9763 | release_global_block_rsv(info); |
9764 | ||
67871254 | 9765 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
9766 | int i; |
9767 | ||
4184ea7f CM |
9768 | space_info = list_entry(info->space_info.next, |
9769 | struct btrfs_space_info, | |
9770 | list); | |
d555b6c3 JB |
9771 | |
9772 | /* | |
9773 | * Do not hide this behind enospc_debug, this is actually | |
9774 | * important and indicates a real bug if this happens. | |
9775 | */ | |
9776 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
b069e0c3 | 9777 | space_info->bytes_reserved > 0 || |
d555b6c3 | 9778 | space_info->bytes_may_use > 0)) |
ab8d0fc4 | 9779 | dump_space_info(info, space_info, 0, 0); |
4184ea7f | 9780 | list_del(&space_info->list); |
6ab0a202 JM |
9781 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
9782 | struct kobject *kobj; | |
c1895442 JM |
9783 | kobj = space_info->block_group_kobjs[i]; |
9784 | space_info->block_group_kobjs[i] = NULL; | |
9785 | if (kobj) { | |
6ab0a202 JM |
9786 | kobject_del(kobj); |
9787 | kobject_put(kobj); | |
9788 | } | |
9789 | } | |
9790 | kobject_del(&space_info->kobj); | |
9791 | kobject_put(&space_info->kobj); | |
4184ea7f | 9792 | } |
1a40e23b ZY |
9793 | return 0; |
9794 | } | |
9795 | ||
75cb379d JM |
9796 | /* link_block_group will queue up kobjects to add when we're reclaim-safe */ |
9797 | void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info) | |
9798 | { | |
9799 | struct btrfs_space_info *space_info; | |
9800 | struct raid_kobject *rkobj; | |
9801 | LIST_HEAD(list); | |
75cb379d JM |
9802 | int ret = 0; |
9803 | ||
9804 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
9805 | list_splice_init(&fs_info->pending_raid_kobjs, &list); | |
9806 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
9807 | ||
9808 | list_for_each_entry(rkobj, &list, list) { | |
280c2908 | 9809 | space_info = btrfs_find_space_info(fs_info, rkobj->flags); |
75cb379d JM |
9810 | |
9811 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
158da513 | 9812 | "%s", btrfs_bg_type_to_raid_name(rkobj->flags)); |
75cb379d JM |
9813 | if (ret) { |
9814 | kobject_put(&rkobj->kobj); | |
9815 | break; | |
9816 | } | |
9817 | } | |
9818 | if (ret) | |
9819 | btrfs_warn(fs_info, | |
9820 | "failed to add kobject for block cache, ignoring"); | |
9821 | } | |
9822 | ||
c434d21c | 9823 | static void link_block_group(struct btrfs_block_group_cache *cache) |
b742bb82 | 9824 | { |
c434d21c | 9825 | struct btrfs_space_info *space_info = cache->space_info; |
75cb379d | 9826 | struct btrfs_fs_info *fs_info = cache->fs_info; |
3e72ee88 | 9827 | int index = btrfs_bg_flags_to_raid_index(cache->flags); |
ed55b6ac | 9828 | bool first = false; |
b742bb82 YZ |
9829 | |
9830 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
9831 | if (list_empty(&space_info->block_groups[index])) |
9832 | first = true; | |
9833 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
9834 | up_write(&space_info->groups_sem); | |
9835 | ||
9836 | if (first) { | |
75cb379d JM |
9837 | struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
9838 | if (!rkobj) { | |
9839 | btrfs_warn(cache->fs_info, | |
9840 | "couldn't alloc memory for raid level kobject"); | |
9841 | return; | |
6ab0a202 | 9842 | } |
75cb379d JM |
9843 | rkobj->flags = cache->flags; |
9844 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
9845 | ||
9846 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
9847 | list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs); | |
9848 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
c1895442 | 9849 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 9850 | } |
b742bb82 YZ |
9851 | } |
9852 | ||
920e4a58 | 9853 | static struct btrfs_block_group_cache * |
2ff7e61e JM |
9854 | btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info, |
9855 | u64 start, u64 size) | |
920e4a58 MX |
9856 | { |
9857 | struct btrfs_block_group_cache *cache; | |
9858 | ||
9859 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
9860 | if (!cache) | |
9861 | return NULL; | |
9862 | ||
9863 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
9864 | GFP_NOFS); | |
9865 | if (!cache->free_space_ctl) { | |
9866 | kfree(cache); | |
9867 | return NULL; | |
9868 | } | |
9869 | ||
9870 | cache->key.objectid = start; | |
9871 | cache->key.offset = size; | |
9872 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
9873 | ||
0b246afa | 9874 | cache->fs_info = fs_info; |
e4ff5fb5 | 9875 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); |
1e144fb8 OS |
9876 | set_free_space_tree_thresholds(cache); |
9877 | ||
920e4a58 MX |
9878 | atomic_set(&cache->count, 1); |
9879 | spin_lock_init(&cache->lock); | |
e570fd27 | 9880 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
9881 | INIT_LIST_HEAD(&cache->list); |
9882 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 9883 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 9884 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 9885 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 9886 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 9887 | btrfs_init_free_space_ctl(cache); |
04216820 | 9888 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 9889 | mutex_init(&cache->free_space_lock); |
0966a7b1 | 9890 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); |
920e4a58 MX |
9891 | |
9892 | return cache; | |
9893 | } | |
9894 | ||
7ef49515 QW |
9895 | |
9896 | /* | |
9897 | * Iterate all chunks and verify that each of them has the corresponding block | |
9898 | * group | |
9899 | */ | |
9900 | static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info) | |
9901 | { | |
c8bf1b67 | 9902 | struct extent_map_tree *map_tree = &fs_info->mapping_tree; |
7ef49515 QW |
9903 | struct extent_map *em; |
9904 | struct btrfs_block_group_cache *bg; | |
9905 | u64 start = 0; | |
9906 | int ret = 0; | |
9907 | ||
9908 | while (1) { | |
c8bf1b67 | 9909 | read_lock(&map_tree->lock); |
7ef49515 QW |
9910 | /* |
9911 | * lookup_extent_mapping will return the first extent map | |
9912 | * intersecting the range, so setting @len to 1 is enough to | |
9913 | * get the first chunk. | |
9914 | */ | |
c8bf1b67 DS |
9915 | em = lookup_extent_mapping(map_tree, start, 1); |
9916 | read_unlock(&map_tree->lock); | |
7ef49515 QW |
9917 | if (!em) |
9918 | break; | |
9919 | ||
9920 | bg = btrfs_lookup_block_group(fs_info, em->start); | |
9921 | if (!bg) { | |
9922 | btrfs_err(fs_info, | |
9923 | "chunk start=%llu len=%llu doesn't have corresponding block group", | |
9924 | em->start, em->len); | |
9925 | ret = -EUCLEAN; | |
9926 | free_extent_map(em); | |
9927 | break; | |
9928 | } | |
9929 | if (bg->key.objectid != em->start || | |
9930 | bg->key.offset != em->len || | |
9931 | (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) != | |
9932 | (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
9933 | btrfs_err(fs_info, | |
9934 | "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx", | |
9935 | em->start, em->len, | |
9936 | em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK, | |
9937 | bg->key.objectid, bg->key.offset, | |
9938 | bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK); | |
9939 | ret = -EUCLEAN; | |
9940 | free_extent_map(em); | |
9941 | btrfs_put_block_group(bg); | |
9942 | break; | |
9943 | } | |
9944 | start = em->start + em->len; | |
9945 | free_extent_map(em); | |
9946 | btrfs_put_block_group(bg); | |
9947 | } | |
9948 | return ret; | |
9949 | } | |
9950 | ||
5b4aacef | 9951 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
9078a3e1 CM |
9952 | { |
9953 | struct btrfs_path *path; | |
9954 | int ret; | |
9078a3e1 | 9955 | struct btrfs_block_group_cache *cache; |
6324fbf3 | 9956 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
9957 | struct btrfs_key key; |
9958 | struct btrfs_key found_key; | |
5f39d397 | 9959 | struct extent_buffer *leaf; |
0af3d00b JB |
9960 | int need_clear = 0; |
9961 | u64 cache_gen; | |
49303381 LB |
9962 | u64 feature; |
9963 | int mixed; | |
9964 | ||
9965 | feature = btrfs_super_incompat_flags(info->super_copy); | |
9966 | mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS); | |
96b5179d | 9967 | |
9078a3e1 | 9968 | key.objectid = 0; |
0b86a832 | 9969 | key.offset = 0; |
962a298f | 9970 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
9971 | path = btrfs_alloc_path(); |
9972 | if (!path) | |
9973 | return -ENOMEM; | |
e4058b54 | 9974 | path->reada = READA_FORWARD; |
9078a3e1 | 9975 | |
0b246afa JM |
9976 | cache_gen = btrfs_super_cache_generation(info->super_copy); |
9977 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
9978 | btrfs_super_generation(info->super_copy) != cache_gen) | |
0af3d00b | 9979 | need_clear = 1; |
0b246afa | 9980 | if (btrfs_test_opt(info, CLEAR_CACHE)) |
88c2ba3b | 9981 | need_clear = 1; |
0af3d00b | 9982 | |
d397712b | 9983 | while (1) { |
6bccf3ab | 9984 | ret = find_first_block_group(info, path, &key); |
b742bb82 YZ |
9985 | if (ret > 0) |
9986 | break; | |
0b86a832 CM |
9987 | if (ret != 0) |
9988 | goto error; | |
920e4a58 | 9989 | |
5f39d397 CM |
9990 | leaf = path->nodes[0]; |
9991 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 | 9992 | |
2ff7e61e | 9993 | cache = btrfs_create_block_group_cache(info, found_key.objectid, |
920e4a58 | 9994 | found_key.offset); |
9078a3e1 | 9995 | if (!cache) { |
0b86a832 | 9996 | ret = -ENOMEM; |
f0486c68 | 9997 | goto error; |
9078a3e1 | 9998 | } |
96303081 | 9999 | |
cf7c1ef6 LB |
10000 | if (need_clear) { |
10001 | /* | |
10002 | * When we mount with old space cache, we need to | |
10003 | * set BTRFS_DC_CLEAR and set dirty flag. | |
10004 | * | |
10005 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
10006 | * truncate the old free space cache inode and | |
10007 | * setup a new one. | |
10008 | * b) Setting 'dirty flag' makes sure that we flush | |
10009 | * the new space cache info onto disk. | |
10010 | */ | |
0b246afa | 10011 | if (btrfs_test_opt(info, SPACE_CACHE)) |
ce93ec54 | 10012 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 10013 | } |
0af3d00b | 10014 | |
5f39d397 CM |
10015 | read_extent_buffer(leaf, &cache->item, |
10016 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
10017 | sizeof(cache->item)); | |
920e4a58 | 10018 | cache->flags = btrfs_block_group_flags(&cache->item); |
49303381 LB |
10019 | if (!mixed && |
10020 | ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && | |
10021 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
10022 | btrfs_err(info, | |
10023 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
10024 | cache->key.objectid); | |
10025 | ret = -EINVAL; | |
10026 | goto error; | |
10027 | } | |
0b86a832 | 10028 | |
9078a3e1 | 10029 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 10030 | btrfs_release_path(path); |
34d52cb6 | 10031 | |
3c14874a JB |
10032 | /* |
10033 | * We need to exclude the super stripes now so that the space | |
10034 | * info has super bytes accounted for, otherwise we'll think | |
10035 | * we have more space than we actually do. | |
10036 | */ | |
3c4da657 | 10037 | ret = exclude_super_stripes(cache); |
835d974f JB |
10038 | if (ret) { |
10039 | /* | |
10040 | * We may have excluded something, so call this just in | |
10041 | * case. | |
10042 | */ | |
9e715da8 | 10043 | free_excluded_extents(cache); |
920e4a58 | 10044 | btrfs_put_block_group(cache); |
835d974f JB |
10045 | goto error; |
10046 | } | |
3c14874a | 10047 | |
817d52f8 JB |
10048 | /* |
10049 | * check for two cases, either we are full, and therefore | |
10050 | * don't need to bother with the caching work since we won't | |
10051 | * find any space, or we are empty, and we can just add all | |
52042d8e | 10052 | * the space in and be done with it. This saves us _a_lot_ of |
817d52f8 JB |
10053 | * time, particularly in the full case. |
10054 | */ | |
10055 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 10056 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10057 | cache->cached = BTRFS_CACHE_FINISHED; |
9e715da8 | 10058 | free_excluded_extents(cache); |
817d52f8 | 10059 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 10060 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10061 | cache->cached = BTRFS_CACHE_FINISHED; |
4457c1c7 | 10062 | add_new_free_space(cache, found_key.objectid, |
817d52f8 JB |
10063 | found_key.objectid + |
10064 | found_key.offset); | |
9e715da8 | 10065 | free_excluded_extents(cache); |
817d52f8 | 10066 | } |
96b5179d | 10067 | |
0b246afa | 10068 | ret = btrfs_add_block_group_cache(info, cache); |
8c579fe7 JB |
10069 | if (ret) { |
10070 | btrfs_remove_free_space_cache(cache); | |
10071 | btrfs_put_block_group(cache); | |
10072 | goto error; | |
10073 | } | |
10074 | ||
0b246afa | 10075 | trace_btrfs_add_block_group(info, cache, 0); |
280c2908 JB |
10076 | btrfs_update_space_info(info, cache->flags, found_key.offset, |
10077 | btrfs_block_group_used(&cache->item), | |
10078 | cache->bytes_super, &space_info); | |
8c579fe7 | 10079 | |
6324fbf3 | 10080 | cache->space_info = space_info; |
1b2da372 | 10081 | |
c434d21c | 10082 | link_block_group(cache); |
0f9dd46c | 10083 | |
0b246afa | 10084 | set_avail_alloc_bits(info, cache->flags); |
2ff7e61e | 10085 | if (btrfs_chunk_readonly(info, cache->key.objectid)) { |
868f401a | 10086 | inc_block_group_ro(cache, 1); |
47ab2a6c | 10087 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
031f24da QW |
10088 | ASSERT(list_empty(&cache->bg_list)); |
10089 | btrfs_mark_bg_unused(cache); | |
47ab2a6c | 10090 | } |
9078a3e1 | 10091 | } |
b742bb82 | 10092 | |
0b246afa | 10093 | list_for_each_entry_rcu(space_info, &info->space_info, list) { |
2ff7e61e | 10094 | if (!(get_alloc_profile(info, space_info->flags) & |
b742bb82 | 10095 | (BTRFS_BLOCK_GROUP_RAID10 | |
c7369b3f | 10096 | BTRFS_BLOCK_GROUP_RAID1_MASK | |
a07e8a46 | 10097 | BTRFS_BLOCK_GROUP_RAID56_MASK | |
b742bb82 YZ |
10098 | BTRFS_BLOCK_GROUP_DUP))) |
10099 | continue; | |
10100 | /* | |
10101 | * avoid allocating from un-mirrored block group if there are | |
10102 | * mirrored block groups. | |
10103 | */ | |
1095cc0d | 10104 | list_for_each_entry(cache, |
10105 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
10106 | list) | |
868f401a | 10107 | inc_block_group_ro(cache, 1); |
1095cc0d | 10108 | list_for_each_entry(cache, |
10109 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
10110 | list) | |
868f401a | 10111 | inc_block_group_ro(cache, 1); |
9078a3e1 | 10112 | } |
f0486c68 | 10113 | |
75cb379d | 10114 | btrfs_add_raid_kobjects(info); |
f0486c68 | 10115 | init_global_block_rsv(info); |
7ef49515 | 10116 | ret = check_chunk_block_group_mappings(info); |
0b86a832 | 10117 | error: |
9078a3e1 | 10118 | btrfs_free_path(path); |
0b86a832 | 10119 | return ret; |
9078a3e1 | 10120 | } |
6324fbf3 | 10121 | |
6c686b35 | 10122 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) |
ea658bad | 10123 | { |
6c686b35 | 10124 | struct btrfs_fs_info *fs_info = trans->fs_info; |
545e3366 | 10125 | struct btrfs_block_group_cache *block_group; |
0b246afa | 10126 | struct btrfs_root *extent_root = fs_info->extent_root; |
ea658bad JB |
10127 | struct btrfs_block_group_item item; |
10128 | struct btrfs_key key; | |
10129 | int ret = 0; | |
10130 | ||
5ce55557 FM |
10131 | if (!trans->can_flush_pending_bgs) |
10132 | return; | |
10133 | ||
545e3366 JB |
10134 | while (!list_empty(&trans->new_bgs)) { |
10135 | block_group = list_first_entry(&trans->new_bgs, | |
10136 | struct btrfs_block_group_cache, | |
10137 | bg_list); | |
ea658bad | 10138 | if (ret) |
c92f6be3 | 10139 | goto next; |
ea658bad JB |
10140 | |
10141 | spin_lock(&block_group->lock); | |
10142 | memcpy(&item, &block_group->item, sizeof(item)); | |
10143 | memcpy(&key, &block_group->key, sizeof(key)); | |
10144 | spin_unlock(&block_group->lock); | |
10145 | ||
10146 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
10147 | sizeof(item)); | |
10148 | if (ret) | |
66642832 | 10149 | btrfs_abort_transaction(trans, ret); |
97aff912 | 10150 | ret = btrfs_finish_chunk_alloc(trans, key.objectid, key.offset); |
6df9a95e | 10151 | if (ret) |
66642832 | 10152 | btrfs_abort_transaction(trans, ret); |
e4e0711c | 10153 | add_block_group_free_space(trans, block_group); |
1e144fb8 | 10154 | /* already aborted the transaction if it failed. */ |
c92f6be3 | 10155 | next: |
ba2c4d4e | 10156 | btrfs_delayed_refs_rsv_release(fs_info, 1); |
c92f6be3 | 10157 | list_del_init(&block_group->bg_list); |
ea658bad | 10158 | } |
5ce55557 | 10159 | btrfs_trans_release_chunk_metadata(trans); |
ea658bad JB |
10160 | } |
10161 | ||
e7e02096 | 10162 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used, |
0174484d | 10163 | u64 type, u64 chunk_offset, u64 size) |
6324fbf3 | 10164 | { |
e7e02096 | 10165 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6324fbf3 | 10166 | struct btrfs_block_group_cache *cache; |
0b246afa | 10167 | int ret; |
6324fbf3 | 10168 | |
90787766 | 10169 | btrfs_set_log_full_commit(trans); |
e02119d5 | 10170 | |
2ff7e61e | 10171 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size); |
0f9dd46c JB |
10172 | if (!cache) |
10173 | return -ENOMEM; | |
34d52cb6 | 10174 | |
6324fbf3 | 10175 | btrfs_set_block_group_used(&cache->item, bytes_used); |
0174484d NB |
10176 | btrfs_set_block_group_chunk_objectid(&cache->item, |
10177 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
6324fbf3 CM |
10178 | btrfs_set_block_group_flags(&cache->item, type); |
10179 | ||
920e4a58 | 10180 | cache->flags = type; |
11833d66 | 10181 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10182 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 10183 | cache->needs_free_space = 1; |
3c4da657 | 10184 | ret = exclude_super_stripes(cache); |
835d974f JB |
10185 | if (ret) { |
10186 | /* | |
10187 | * We may have excluded something, so call this just in | |
10188 | * case. | |
10189 | */ | |
9e715da8 | 10190 | free_excluded_extents(cache); |
920e4a58 | 10191 | btrfs_put_block_group(cache); |
835d974f JB |
10192 | return ret; |
10193 | } | |
96303081 | 10194 | |
4457c1c7 | 10195 | add_new_free_space(cache, chunk_offset, chunk_offset + size); |
817d52f8 | 10196 | |
9e715da8 | 10197 | free_excluded_extents(cache); |
11833d66 | 10198 | |
d0bd4560 | 10199 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 10200 | if (btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
10201 | u64 new_bytes_used = size - bytes_used; |
10202 | ||
10203 | bytes_used += new_bytes_used >> 1; | |
2ff7e61e | 10204 | fragment_free_space(cache); |
d0bd4560 JB |
10205 | } |
10206 | #endif | |
2e6e5183 | 10207 | /* |
2be12ef7 NB |
10208 | * Ensure the corresponding space_info object is created and |
10209 | * assigned to our block group. We want our bg to be added to the rbtree | |
10210 | * with its ->space_info set. | |
2e6e5183 | 10211 | */ |
280c2908 | 10212 | cache->space_info = btrfs_find_space_info(fs_info, cache->flags); |
dc2d3005 | 10213 | ASSERT(cache->space_info); |
2e6e5183 | 10214 | |
0b246afa | 10215 | ret = btrfs_add_block_group_cache(fs_info, cache); |
8c579fe7 JB |
10216 | if (ret) { |
10217 | btrfs_remove_free_space_cache(cache); | |
10218 | btrfs_put_block_group(cache); | |
10219 | return ret; | |
10220 | } | |
10221 | ||
2e6e5183 FM |
10222 | /* |
10223 | * Now that our block group has its ->space_info set and is inserted in | |
10224 | * the rbtree, update the space info's counters. | |
10225 | */ | |
0b246afa | 10226 | trace_btrfs_add_block_group(fs_info, cache, 1); |
280c2908 | 10227 | btrfs_update_space_info(fs_info, cache->flags, size, bytes_used, |
e40edf2d | 10228 | cache->bytes_super, &cache->space_info); |
0b246afa | 10229 | update_global_block_rsv(fs_info); |
1b2da372 | 10230 | |
c434d21c | 10231 | link_block_group(cache); |
6324fbf3 | 10232 | |
47ab2a6c | 10233 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
ba2c4d4e JB |
10234 | trans->delayed_ref_updates++; |
10235 | btrfs_update_delayed_refs_rsv(trans); | |
6324fbf3 | 10236 | |
0b246afa | 10237 | set_avail_alloc_bits(fs_info, type); |
6324fbf3 CM |
10238 | return 0; |
10239 | } | |
1a40e23b | 10240 | |
10ea00f5 ID |
10241 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
10242 | { | |
899c81ea ID |
10243 | u64 extra_flags = chunk_to_extended(flags) & |
10244 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 10245 | |
de98ced9 | 10246 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10247 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
10248 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10249 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10250 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10251 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10252 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 10253 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10254 | } |
10255 | ||
6d58a55a DS |
10256 | /* |
10257 | * Clear incompat bits for the following feature(s): | |
10258 | * | |
10259 | * - RAID56 - in case there's neither RAID5 nor RAID6 profile block group | |
10260 | * in the whole filesystem | |
10261 | */ | |
10262 | static void clear_incompat_bg_bits(struct btrfs_fs_info *fs_info, u64 flags) | |
10263 | { | |
10264 | if (flags & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
10265 | struct list_head *head = &fs_info->space_info; | |
10266 | struct btrfs_space_info *sinfo; | |
10267 | ||
10268 | list_for_each_entry_rcu(sinfo, head, list) { | |
10269 | bool found = false; | |
10270 | ||
10271 | down_read(&sinfo->groups_sem); | |
10272 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID5])) | |
10273 | found = true; | |
10274 | if (!list_empty(&sinfo->block_groups[BTRFS_RAID_RAID6])) | |
10275 | found = true; | |
10276 | up_read(&sinfo->groups_sem); | |
10277 | ||
10278 | if (found) | |
10279 | return; | |
10280 | } | |
10281 | btrfs_clear_fs_incompat(fs_info, RAID56); | |
10282 | } | |
10283 | } | |
10284 | ||
1a40e23b | 10285 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
5a98ec01 | 10286 | u64 group_start, struct extent_map *em) |
1a40e23b | 10287 | { |
5a98ec01 | 10288 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 10289 | struct btrfs_root *root = fs_info->extent_root; |
1a40e23b ZY |
10290 | struct btrfs_path *path; |
10291 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 10292 | struct btrfs_free_cluster *cluster; |
0b246afa | 10293 | struct btrfs_root *tree_root = fs_info->tree_root; |
1a40e23b | 10294 | struct btrfs_key key; |
0af3d00b | 10295 | struct inode *inode; |
c1895442 | 10296 | struct kobject *kobj = NULL; |
1a40e23b | 10297 | int ret; |
10ea00f5 | 10298 | int index; |
89a55897 | 10299 | int factor; |
4f69cb98 | 10300 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 10301 | bool remove_em; |
ba2c4d4e | 10302 | bool remove_rsv = false; |
1a40e23b | 10303 | |
6bccf3ab | 10304 | block_group = btrfs_lookup_block_group(fs_info, group_start); |
1a40e23b | 10305 | BUG_ON(!block_group); |
c146afad | 10306 | BUG_ON(!block_group->ro); |
1a40e23b | 10307 | |
4ed0a7a3 | 10308 | trace_btrfs_remove_block_group(block_group); |
9f7c43c9 | 10309 | /* |
10310 | * Free the reserved super bytes from this block group before | |
10311 | * remove it. | |
10312 | */ | |
9e715da8 | 10313 | free_excluded_extents(block_group); |
fd708b81 JB |
10314 | btrfs_free_ref_tree_range(fs_info, block_group->key.objectid, |
10315 | block_group->key.offset); | |
9f7c43c9 | 10316 | |
1a40e23b | 10317 | memcpy(&key, &block_group->key, sizeof(key)); |
3e72ee88 | 10318 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
46df06b8 | 10319 | factor = btrfs_bg_type_to_factor(block_group->flags); |
1a40e23b | 10320 | |
44fb5511 | 10321 | /* make sure this block group isn't part of an allocation cluster */ |
0b246afa | 10322 | cluster = &fs_info->data_alloc_cluster; |
44fb5511 CM |
10323 | spin_lock(&cluster->refill_lock); |
10324 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10325 | spin_unlock(&cluster->refill_lock); | |
10326 | ||
10327 | /* | |
10328 | * make sure this block group isn't part of a metadata | |
10329 | * allocation cluster | |
10330 | */ | |
0b246afa | 10331 | cluster = &fs_info->meta_alloc_cluster; |
44fb5511 CM |
10332 | spin_lock(&cluster->refill_lock); |
10333 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10334 | spin_unlock(&cluster->refill_lock); | |
10335 | ||
1a40e23b | 10336 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10337 | if (!path) { |
10338 | ret = -ENOMEM; | |
10339 | goto out; | |
10340 | } | |
1a40e23b | 10341 | |
1bbc621e CM |
10342 | /* |
10343 | * get the inode first so any iput calls done for the io_list | |
10344 | * aren't the final iput (no unlinks allowed now) | |
10345 | */ | |
7949f339 | 10346 | inode = lookup_free_space_inode(block_group, path); |
1bbc621e CM |
10347 | |
10348 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10349 | /* | |
52042d8e | 10350 | * Make sure our free space cache IO is done before removing the |
1bbc621e CM |
10351 | * free space inode |
10352 | */ | |
10353 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10354 | if (!list_empty(&block_group->io_list)) { | |
10355 | list_del_init(&block_group->io_list); | |
10356 | ||
10357 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10358 | ||
10359 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
afdb5718 | 10360 | btrfs_wait_cache_io(trans, block_group, path); |
1bbc621e CM |
10361 | btrfs_put_block_group(block_group); |
10362 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10363 | } | |
10364 | ||
10365 | if (!list_empty(&block_group->dirty_list)) { | |
10366 | list_del_init(&block_group->dirty_list); | |
ba2c4d4e | 10367 | remove_rsv = true; |
1bbc621e CM |
10368 | btrfs_put_block_group(block_group); |
10369 | } | |
10370 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10371 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10372 | ||
0af3d00b | 10373 | if (!IS_ERR(inode)) { |
73f2e545 | 10374 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
79787eaa JM |
10375 | if (ret) { |
10376 | btrfs_add_delayed_iput(inode); | |
10377 | goto out; | |
10378 | } | |
0af3d00b JB |
10379 | clear_nlink(inode); |
10380 | /* One for the block groups ref */ | |
10381 | spin_lock(&block_group->lock); | |
10382 | if (block_group->iref) { | |
10383 | block_group->iref = 0; | |
10384 | block_group->inode = NULL; | |
10385 | spin_unlock(&block_group->lock); | |
10386 | iput(inode); | |
10387 | } else { | |
10388 | spin_unlock(&block_group->lock); | |
10389 | } | |
10390 | /* One for our lookup ref */ | |
455757c3 | 10391 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10392 | } |
10393 | ||
10394 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10395 | key.offset = block_group->key.objectid; | |
10396 | key.type = 0; | |
10397 | ||
10398 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10399 | if (ret < 0) | |
10400 | goto out; | |
10401 | if (ret > 0) | |
b3b4aa74 | 10402 | btrfs_release_path(path); |
0af3d00b JB |
10403 | if (ret == 0) { |
10404 | ret = btrfs_del_item(trans, tree_root, path); | |
10405 | if (ret) | |
10406 | goto out; | |
b3b4aa74 | 10407 | btrfs_release_path(path); |
0af3d00b JB |
10408 | } |
10409 | ||
0b246afa | 10410 | spin_lock(&fs_info->block_group_cache_lock); |
1a40e23b | 10411 | rb_erase(&block_group->cache_node, |
0b246afa | 10412 | &fs_info->block_group_cache_tree); |
292cbd51 | 10413 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd | 10414 | |
0b246afa JM |
10415 | if (fs_info->first_logical_byte == block_group->key.objectid) |
10416 | fs_info->first_logical_byte = (u64)-1; | |
10417 | spin_unlock(&fs_info->block_group_cache_lock); | |
817d52f8 | 10418 | |
80eb234a | 10419 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10420 | /* |
10421 | * we must use list_del_init so people can check to see if they | |
10422 | * are still on the list after taking the semaphore | |
10423 | */ | |
10424 | list_del_init(&block_group->list); | |
6ab0a202 | 10425 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10426 | kobj = block_group->space_info->block_group_kobjs[index]; |
10427 | block_group->space_info->block_group_kobjs[index] = NULL; | |
0b246afa | 10428 | clear_avail_alloc_bits(fs_info, block_group->flags); |
6ab0a202 | 10429 | } |
80eb234a | 10430 | up_write(&block_group->space_info->groups_sem); |
6d58a55a | 10431 | clear_incompat_bg_bits(fs_info, block_group->flags); |
c1895442 JM |
10432 | if (kobj) { |
10433 | kobject_del(kobj); | |
10434 | kobject_put(kobj); | |
10435 | } | |
1a40e23b | 10436 | |
4f69cb98 FM |
10437 | if (block_group->has_caching_ctl) |
10438 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10439 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10440 | wait_block_group_cache_done(block_group); |
4f69cb98 | 10441 | if (block_group->has_caching_ctl) { |
0b246afa | 10442 | down_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10443 | if (!caching_ctl) { |
10444 | struct btrfs_caching_control *ctl; | |
10445 | ||
10446 | list_for_each_entry(ctl, | |
0b246afa | 10447 | &fs_info->caching_block_groups, list) |
4f69cb98 FM |
10448 | if (ctl->block_group == block_group) { |
10449 | caching_ctl = ctl; | |
1e4f4714 | 10450 | refcount_inc(&caching_ctl->count); |
4f69cb98 FM |
10451 | break; |
10452 | } | |
10453 | } | |
10454 | if (caching_ctl) | |
10455 | list_del_init(&caching_ctl->list); | |
0b246afa | 10456 | up_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10457 | if (caching_ctl) { |
10458 | /* Once for the caching bgs list and once for us. */ | |
10459 | put_caching_control(caching_ctl); | |
10460 | put_caching_control(caching_ctl); | |
10461 | } | |
10462 | } | |
817d52f8 | 10463 | |
ce93ec54 | 10464 | spin_lock(&trans->transaction->dirty_bgs_lock); |
9a0ec83d NB |
10465 | WARN_ON(!list_empty(&block_group->dirty_list)); |
10466 | WARN_ON(!list_empty(&block_group->io_list)); | |
ce93ec54 | 10467 | spin_unlock(&trans->transaction->dirty_bgs_lock); |
9a0ec83d | 10468 | |
817d52f8 JB |
10469 | btrfs_remove_free_space_cache(block_group); |
10470 | ||
c146afad | 10471 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10472 | list_del_init(&block_group->ro_list); |
18d018ad | 10473 | |
0b246afa | 10474 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
18d018ad ZL |
10475 | WARN_ON(block_group->space_info->total_bytes |
10476 | < block_group->key.offset); | |
10477 | WARN_ON(block_group->space_info->bytes_readonly | |
10478 | < block_group->key.offset); | |
10479 | WARN_ON(block_group->space_info->disk_total | |
10480 | < block_group->key.offset * factor); | |
10481 | } | |
c146afad YZ |
10482 | block_group->space_info->total_bytes -= block_group->key.offset; |
10483 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10484 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10485 | |
c146afad | 10486 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10487 | |
0af3d00b JB |
10488 | memcpy(&key, &block_group->key, sizeof(key)); |
10489 | ||
34441361 | 10490 | mutex_lock(&fs_info->chunk_mutex); |
04216820 FM |
10491 | spin_lock(&block_group->lock); |
10492 | block_group->removed = 1; | |
10493 | /* | |
10494 | * At this point trimming can't start on this block group, because we | |
10495 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10496 | * so no one can't find it anymore and even if someone already got this | |
10497 | * block group before we removed it from the rbtree, they have already | |
10498 | * incremented block_group->trimming - if they didn't, they won't find | |
10499 | * any free space entries because we already removed them all when we | |
10500 | * called btrfs_remove_free_space_cache(). | |
10501 | * | |
10502 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10503 | * to prevent the same logical address range and physical device space | |
10504 | * ranges from being reused for a new block group. This is because our | |
10505 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10506 | * completely transactionless, so while it is trimming a range the | |
10507 | * currently running transaction might finish and a new one start, | |
10508 | * allowing for new block groups to be created that can reuse the same | |
10509 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10510 | * |
10511 | * There may also be an implicit trim operation if the file system | |
10512 | * is mounted with -odiscard. The same protections must remain | |
10513 | * in place until the extents have been discarded completely when | |
10514 | * the transaction commit has completed. | |
04216820 FM |
10515 | */ |
10516 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
04216820 | 10517 | spin_unlock(&block_group->lock); |
04216820 | 10518 | |
34441361 | 10519 | mutex_unlock(&fs_info->chunk_mutex); |
8dbcd10f | 10520 | |
f3f72779 | 10521 | ret = remove_block_group_free_space(trans, block_group); |
1e144fb8 OS |
10522 | if (ret) |
10523 | goto out; | |
10524 | ||
fa9c0d79 CM |
10525 | btrfs_put_block_group(block_group); |
10526 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10527 | |
10528 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10529 | if (ret > 0) | |
10530 | ret = -EIO; | |
10531 | if (ret < 0) | |
10532 | goto out; | |
10533 | ||
10534 | ret = btrfs_del_item(trans, root, path); | |
8eaf40c0 FM |
10535 | if (ret) |
10536 | goto out; | |
10537 | ||
10538 | if (remove_em) { | |
10539 | struct extent_map_tree *em_tree; | |
10540 | ||
c8bf1b67 | 10541 | em_tree = &fs_info->mapping_tree; |
8eaf40c0 FM |
10542 | write_lock(&em_tree->lock); |
10543 | remove_extent_mapping(em_tree, em); | |
10544 | write_unlock(&em_tree->lock); | |
10545 | /* once for the tree */ | |
10546 | free_extent_map(em); | |
10547 | } | |
1a40e23b | 10548 | out: |
ba2c4d4e JB |
10549 | if (remove_rsv) |
10550 | btrfs_delayed_refs_rsv_release(fs_info, 1); | |
1a40e23b ZY |
10551 | btrfs_free_path(path); |
10552 | return ret; | |
10553 | } | |
acce952b | 10554 | |
8eab77ff | 10555 | struct btrfs_trans_handle * |
7fd01182 FM |
10556 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10557 | const u64 chunk_offset) | |
8eab77ff | 10558 | { |
c8bf1b67 | 10559 | struct extent_map_tree *em_tree = &fs_info->mapping_tree; |
7fd01182 FM |
10560 | struct extent_map *em; |
10561 | struct map_lookup *map; | |
10562 | unsigned int num_items; | |
10563 | ||
10564 | read_lock(&em_tree->lock); | |
10565 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10566 | read_unlock(&em_tree->lock); | |
10567 | ASSERT(em && em->start == chunk_offset); | |
10568 | ||
8eab77ff | 10569 | /* |
7fd01182 FM |
10570 | * We need to reserve 3 + N units from the metadata space info in order |
10571 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10572 | * btrfs_remove_block_group()), which are used for: | |
10573 | * | |
8eab77ff FM |
10574 | * 1 unit for adding the free space inode's orphan (located in the tree |
10575 | * of tree roots). | |
7fd01182 FM |
10576 | * 1 unit for deleting the block group item (located in the extent |
10577 | * tree). | |
10578 | * 1 unit for deleting the free space item (located in tree of tree | |
10579 | * roots). | |
10580 | * N units for deleting N device extent items corresponding to each | |
10581 | * stripe (located in the device tree). | |
10582 | * | |
10583 | * In order to remove a block group we also need to reserve units in the | |
10584 | * system space info in order to update the chunk tree (update one or | |
10585 | * more device items and remove one chunk item), but this is done at | |
10586 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10587 | */ |
95617d69 | 10588 | map = em->map_lookup; |
7fd01182 FM |
10589 | num_items = 3 + map->num_stripes; |
10590 | free_extent_map(em); | |
10591 | ||
8eab77ff | 10592 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10593 | num_items, 1); |
8eab77ff FM |
10594 | } |
10595 | ||
47ab2a6c JB |
10596 | /* |
10597 | * Process the unused_bgs list and remove any that don't have any allocated | |
10598 | * space inside of them. | |
10599 | */ | |
10600 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10601 | { | |
10602 | struct btrfs_block_group_cache *block_group; | |
10603 | struct btrfs_space_info *space_info; | |
47ab2a6c JB |
10604 | struct btrfs_trans_handle *trans; |
10605 | int ret = 0; | |
10606 | ||
afcdd129 | 10607 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) |
47ab2a6c JB |
10608 | return; |
10609 | ||
10610 | spin_lock(&fs_info->unused_bgs_lock); | |
10611 | while (!list_empty(&fs_info->unused_bgs)) { | |
10612 | u64 start, end; | |
e33e17ee | 10613 | int trimming; |
47ab2a6c JB |
10614 | |
10615 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10616 | struct btrfs_block_group_cache, | |
10617 | bg_list); | |
47ab2a6c | 10618 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10619 | |
10620 | space_info = block_group->space_info; | |
10621 | ||
47ab2a6c JB |
10622 | if (ret || btrfs_mixed_space_info(space_info)) { |
10623 | btrfs_put_block_group(block_group); | |
10624 | continue; | |
10625 | } | |
10626 | spin_unlock(&fs_info->unused_bgs_lock); | |
10627 | ||
d5f2e33b | 10628 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10629 | |
47ab2a6c JB |
10630 | /* Don't want to race with allocators so take the groups_sem */ |
10631 | down_write(&space_info->groups_sem); | |
10632 | spin_lock(&block_group->lock); | |
43794446 | 10633 | if (block_group->reserved || block_group->pinned || |
47ab2a6c | 10634 | btrfs_block_group_used(&block_group->item) || |
19c4d2f9 | 10635 | block_group->ro || |
aefbe9a6 | 10636 | list_is_singular(&block_group->list)) { |
47ab2a6c JB |
10637 | /* |
10638 | * We want to bail if we made new allocations or have | |
10639 | * outstanding allocations in this block group. We do | |
10640 | * the ro check in case balance is currently acting on | |
10641 | * this block group. | |
10642 | */ | |
4ed0a7a3 | 10643 | trace_btrfs_skip_unused_block_group(block_group); |
47ab2a6c JB |
10644 | spin_unlock(&block_group->lock); |
10645 | up_write(&space_info->groups_sem); | |
10646 | goto next; | |
10647 | } | |
10648 | spin_unlock(&block_group->lock); | |
10649 | ||
10650 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 10651 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
10652 | up_write(&space_info->groups_sem); |
10653 | if (ret < 0) { | |
10654 | ret = 0; | |
10655 | goto next; | |
10656 | } | |
10657 | ||
10658 | /* | |
10659 | * Want to do this before we do anything else so we can recover | |
10660 | * properly if we fail to join the transaction. | |
10661 | */ | |
7fd01182 FM |
10662 | trans = btrfs_start_trans_remove_block_group(fs_info, |
10663 | block_group->key.objectid); | |
47ab2a6c | 10664 | if (IS_ERR(trans)) { |
2ff7e61e | 10665 | btrfs_dec_block_group_ro(block_group); |
47ab2a6c JB |
10666 | ret = PTR_ERR(trans); |
10667 | goto next; | |
10668 | } | |
10669 | ||
10670 | /* | |
10671 | * We could have pending pinned extents for this block group, | |
10672 | * just delete them, we don't care about them anymore. | |
10673 | */ | |
10674 | start = block_group->key.objectid; | |
10675 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
10676 | /* |
10677 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
10678 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
10679 | * another task might be running finish_extent_commit() for the | |
10680 | * previous transaction N - 1, and have seen a range belonging | |
10681 | * to the block group in freed_extents[] before we were able to | |
10682 | * clear the whole block group range from freed_extents[]. This | |
10683 | * means that task can lookup for the block group after we | |
10684 | * unpinned it from freed_extents[] and removed it, leading to | |
10685 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
10686 | */ | |
10687 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 10688 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
91166212 | 10689 | EXTENT_DIRTY); |
758eb51e | 10690 | if (ret) { |
d4b450cd | 10691 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10692 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10693 | goto end_trans; |
10694 | } | |
10695 | ret = clear_extent_bits(&fs_info->freed_extents[1], 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 | } | |
d4b450cd | 10702 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
10703 | |
10704 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
10705 | spin_lock(&space_info->lock); |
10706 | spin_lock(&block_group->lock); | |
10707 | ||
bb96c4e5 JB |
10708 | btrfs_space_info_update_bytes_pinned(fs_info, space_info, |
10709 | -block_group->pinned); | |
c30666d4 | 10710 | space_info->bytes_readonly += block_group->pinned; |
dec59fa3 EL |
10711 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
10712 | -block_group->pinned, | |
10713 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
47ab2a6c JB |
10714 | block_group->pinned = 0; |
10715 | ||
c30666d4 ZL |
10716 | spin_unlock(&block_group->lock); |
10717 | spin_unlock(&space_info->lock); | |
10718 | ||
e33e17ee | 10719 | /* DISCARD can flip during remount */ |
0b246afa | 10720 | trimming = btrfs_test_opt(fs_info, DISCARD); |
e33e17ee JM |
10721 | |
10722 | /* Implicit trim during transaction commit. */ | |
10723 | if (trimming) | |
10724 | btrfs_get_block_group_trimming(block_group); | |
10725 | ||
47ab2a6c JB |
10726 | /* |
10727 | * Btrfs_remove_chunk will abort the transaction if things go | |
10728 | * horribly wrong. | |
10729 | */ | |
97aff912 | 10730 | ret = btrfs_remove_chunk(trans, block_group->key.objectid); |
e33e17ee JM |
10731 | |
10732 | if (ret) { | |
10733 | if (trimming) | |
10734 | btrfs_put_block_group_trimming(block_group); | |
10735 | goto end_trans; | |
10736 | } | |
10737 | ||
10738 | /* | |
10739 | * If we're not mounted with -odiscard, we can just forget | |
10740 | * about this block group. Otherwise we'll need to wait | |
10741 | * until transaction commit to do the actual discard. | |
10742 | */ | |
10743 | if (trimming) { | |
348a0013 FM |
10744 | spin_lock(&fs_info->unused_bgs_lock); |
10745 | /* | |
10746 | * A concurrent scrub might have added us to the list | |
10747 | * fs_info->unused_bgs, so use a list_move operation | |
10748 | * to add the block group to the deleted_bgs list. | |
10749 | */ | |
e33e17ee JM |
10750 | list_move(&block_group->bg_list, |
10751 | &trans->transaction->deleted_bgs); | |
348a0013 | 10752 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
10753 | btrfs_get_block_group(block_group); |
10754 | } | |
758eb51e | 10755 | end_trans: |
3a45bb20 | 10756 | btrfs_end_transaction(trans); |
47ab2a6c | 10757 | next: |
d5f2e33b | 10758 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
10759 | btrfs_put_block_group(block_group); |
10760 | spin_lock(&fs_info->unused_bgs_lock); | |
10761 | } | |
10762 | spin_unlock(&fs_info->unused_bgs_lock); | |
10763 | } | |
10764 | ||
2ff7e61e JM |
10765 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
10766 | u64 start, u64 end) | |
acce952b | 10767 | { |
2ff7e61e | 10768 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 10769 | } |
10770 | ||
499f377f JM |
10771 | /* |
10772 | * It used to be that old block groups would be left around forever. | |
10773 | * Iterating over them would be enough to trim unused space. Since we | |
10774 | * now automatically remove them, we also need to iterate over unallocated | |
10775 | * space. | |
10776 | * | |
10777 | * We don't want a transaction for this since the discard may take a | |
10778 | * substantial amount of time. We don't require that a transaction be | |
10779 | * running, but we do need to take a running transaction into account | |
fee7acc3 JM |
10780 | * to ensure that we're not discarding chunks that were released or |
10781 | * allocated in the current transaction. | |
499f377f JM |
10782 | * |
10783 | * Holding the chunks lock will prevent other threads from allocating | |
10784 | * or releasing chunks, but it won't prevent a running transaction | |
10785 | * from committing and releasing the memory that the pending chunks | |
10786 | * list head uses. For that, we need to take a reference to the | |
fee7acc3 JM |
10787 | * transaction and hold the commit root sem. We only need to hold |
10788 | * it while performing the free space search since we have already | |
10789 | * held back allocations. | |
499f377f | 10790 | */ |
8103d10b | 10791 | static int btrfs_trim_free_extents(struct btrfs_device *device, u64 *trimmed) |
499f377f | 10792 | { |
8103d10b | 10793 | u64 start = SZ_1M, len = 0, end = 0; |
499f377f JM |
10794 | int ret; |
10795 | ||
10796 | *trimmed = 0; | |
10797 | ||
0be88e36 JM |
10798 | /* Discard not supported = nothing to do. */ |
10799 | if (!blk_queue_discard(bdev_get_queue(device->bdev))) | |
10800 | return 0; | |
10801 | ||
52042d8e | 10802 | /* Not writable = nothing to do. */ |
ebbede42 | 10803 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
10804 | return 0; |
10805 | ||
10806 | /* No free space = nothing to do. */ | |
10807 | if (device->total_bytes <= device->bytes_used) | |
10808 | return 0; | |
10809 | ||
10810 | ret = 0; | |
10811 | ||
10812 | while (1) { | |
fb456252 | 10813 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
10814 | u64 bytes; |
10815 | ||
10816 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
10817 | if (ret) | |
fee7acc3 | 10818 | break; |
499f377f | 10819 | |
929be17a NB |
10820 | find_first_clear_extent_bit(&device->alloc_state, start, |
10821 | &start, &end, | |
10822 | CHUNK_TRIMMED | CHUNK_ALLOCATED); | |
53460a45 NB |
10823 | |
10824 | /* Ensure we skip the reserved area in the first 1M */ | |
10825 | start = max_t(u64, start, SZ_1M); | |
10826 | ||
929be17a NB |
10827 | /* |
10828 | * If find_first_clear_extent_bit find a range that spans the | |
10829 | * end of the device it will set end to -1, in this case it's up | |
10830 | * to the caller to trim the value to the size of the device. | |
10831 | */ | |
10832 | end = min(end, device->total_bytes - 1); | |
53460a45 | 10833 | |
929be17a | 10834 | len = end - start + 1; |
499f377f | 10835 | |
929be17a NB |
10836 | /* We didn't find any extents */ |
10837 | if (!len) { | |
499f377f | 10838 | mutex_unlock(&fs_info->chunk_mutex); |
929be17a | 10839 | ret = 0; |
499f377f JM |
10840 | break; |
10841 | } | |
10842 | ||
929be17a NB |
10843 | ret = btrfs_issue_discard(device->bdev, start, len, |
10844 | &bytes); | |
10845 | if (!ret) | |
10846 | set_extent_bits(&device->alloc_state, start, | |
10847 | start + bytes - 1, | |
10848 | CHUNK_TRIMMED); | |
499f377f JM |
10849 | mutex_unlock(&fs_info->chunk_mutex); |
10850 | ||
10851 | if (ret) | |
10852 | break; | |
10853 | ||
10854 | start += len; | |
10855 | *trimmed += bytes; | |
10856 | ||
10857 | if (fatal_signal_pending(current)) { | |
10858 | ret = -ERESTARTSYS; | |
10859 | break; | |
10860 | } | |
10861 | ||
10862 | cond_resched(); | |
10863 | } | |
10864 | ||
10865 | return ret; | |
10866 | } | |
10867 | ||
93bba24d QW |
10868 | /* |
10869 | * Trim the whole filesystem by: | |
10870 | * 1) trimming the free space in each block group | |
10871 | * 2) trimming the unallocated space on each device | |
10872 | * | |
10873 | * This will also continue trimming even if a block group or device encounters | |
10874 | * an error. The return value will be the last error, or 0 if nothing bad | |
10875 | * happens. | |
10876 | */ | |
2ff7e61e | 10877 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 10878 | { |
f7039b1d | 10879 | struct btrfs_block_group_cache *cache = NULL; |
499f377f JM |
10880 | struct btrfs_device *device; |
10881 | struct list_head *devices; | |
f7039b1d LD |
10882 | u64 group_trimmed; |
10883 | u64 start; | |
10884 | u64 end; | |
10885 | u64 trimmed = 0; | |
93bba24d QW |
10886 | u64 bg_failed = 0; |
10887 | u64 dev_failed = 0; | |
10888 | int bg_ret = 0; | |
10889 | int dev_ret = 0; | |
f7039b1d LD |
10890 | int ret = 0; |
10891 | ||
6ba9fc8e | 10892 | cache = btrfs_lookup_first_block_group(fs_info, range->start); |
f87b7eb8 | 10893 | for (; cache; cache = next_block_group(cache)) { |
f7039b1d LD |
10894 | if (cache->key.objectid >= (range->start + range->len)) { |
10895 | btrfs_put_block_group(cache); | |
10896 | break; | |
10897 | } | |
10898 | ||
10899 | start = max(range->start, cache->key.objectid); | |
10900 | end = min(range->start + range->len, | |
10901 | cache->key.objectid + cache->key.offset); | |
10902 | ||
10903 | if (end - start >= range->minlen) { | |
10904 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 10905 | ret = cache_block_group(cache, 0); |
1be41b78 | 10906 | if (ret) { |
93bba24d QW |
10907 | bg_failed++; |
10908 | bg_ret = ret; | |
10909 | continue; | |
1be41b78 JB |
10910 | } |
10911 | ret = wait_block_group_cache_done(cache); | |
10912 | if (ret) { | |
93bba24d QW |
10913 | bg_failed++; |
10914 | bg_ret = ret; | |
10915 | continue; | |
1be41b78 | 10916 | } |
f7039b1d LD |
10917 | } |
10918 | ret = btrfs_trim_block_group(cache, | |
10919 | &group_trimmed, | |
10920 | start, | |
10921 | end, | |
10922 | range->minlen); | |
10923 | ||
10924 | trimmed += group_trimmed; | |
10925 | if (ret) { | |
93bba24d QW |
10926 | bg_failed++; |
10927 | bg_ret = ret; | |
10928 | continue; | |
f7039b1d LD |
10929 | } |
10930 | } | |
f7039b1d LD |
10931 | } |
10932 | ||
93bba24d QW |
10933 | if (bg_failed) |
10934 | btrfs_warn(fs_info, | |
10935 | "failed to trim %llu block group(s), last error %d", | |
10936 | bg_failed, bg_ret); | |
0b246afa | 10937 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
d4e329de JM |
10938 | devices = &fs_info->fs_devices->devices; |
10939 | list_for_each_entry(device, devices, dev_list) { | |
8103d10b | 10940 | ret = btrfs_trim_free_extents(device, &group_trimmed); |
93bba24d QW |
10941 | if (ret) { |
10942 | dev_failed++; | |
10943 | dev_ret = ret; | |
499f377f | 10944 | break; |
93bba24d | 10945 | } |
499f377f JM |
10946 | |
10947 | trimmed += group_trimmed; | |
10948 | } | |
0b246afa | 10949 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 10950 | |
93bba24d QW |
10951 | if (dev_failed) |
10952 | btrfs_warn(fs_info, | |
10953 | "failed to trim %llu device(s), last error %d", | |
10954 | dev_failed, dev_ret); | |
f7039b1d | 10955 | range->len = trimmed; |
93bba24d QW |
10956 | if (bg_ret) |
10957 | return bg_ret; | |
10958 | return dev_ret; | |
f7039b1d | 10959 | } |
8257b2dc MX |
10960 | |
10961 | /* | |
ea14b57f | 10962 | * btrfs_{start,end}_write_no_snapshotting() are similar to |
9ea24bbe FM |
10963 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing |
10964 | * data into the page cache through nocow before the subvolume is snapshoted, | |
10965 | * but flush the data into disk after the snapshot creation, or to prevent | |
ea14b57f | 10966 | * operations while snapshotting is ongoing and that cause the snapshot to be |
9ea24bbe | 10967 | * inconsistent (writes followed by expanding truncates for example). |
8257b2dc | 10968 | */ |
ea14b57f | 10969 | void btrfs_end_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc MX |
10970 | { |
10971 | percpu_counter_dec(&root->subv_writers->counter); | |
093258e6 | 10972 | cond_wake_up(&root->subv_writers->wait); |
8257b2dc MX |
10973 | } |
10974 | ||
ea14b57f | 10975 | int btrfs_start_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc | 10976 | { |
ea14b57f | 10977 | if (atomic_read(&root->will_be_snapshotted)) |
8257b2dc MX |
10978 | return 0; |
10979 | ||
10980 | percpu_counter_inc(&root->subv_writers->counter); | |
10981 | /* | |
10982 | * Make sure counter is updated before we check for snapshot creation. | |
10983 | */ | |
10984 | smp_mb(); | |
ea14b57f DS |
10985 | if (atomic_read(&root->will_be_snapshotted)) { |
10986 | btrfs_end_write_no_snapshotting(root); | |
8257b2dc MX |
10987 | return 0; |
10988 | } | |
10989 | return 1; | |
10990 | } | |
0bc19f90 | 10991 | |
0bc19f90 ZL |
10992 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) |
10993 | { | |
10994 | while (true) { | |
10995 | int ret; | |
10996 | ||
ea14b57f | 10997 | ret = btrfs_start_write_no_snapshotting(root); |
0bc19f90 ZL |
10998 | if (ret) |
10999 | break; | |
4625956a PZ |
11000 | wait_var_event(&root->will_be_snapshotted, |
11001 | !atomic_read(&root->will_be_snapshotted)); | |
0bc19f90 ZL |
11002 | } |
11003 | } | |
031f24da QW |
11004 | |
11005 | void btrfs_mark_bg_unused(struct btrfs_block_group_cache *bg) | |
11006 | { | |
11007 | struct btrfs_fs_info *fs_info = bg->fs_info; | |
11008 | ||
11009 | spin_lock(&fs_info->unused_bgs_lock); | |
11010 | if (list_empty(&bg->bg_list)) { | |
11011 | btrfs_get_block_group(bg); | |
11012 | trace_btrfs_add_unused_block_group(bg); | |
11013 | list_add_tail(&bg->bg_list, &fs_info->unused_bgs); | |
11014 | } | |
11015 | spin_unlock(&fs_info->unused_bgs_lock); | |
11016 | } |