]>
Commit | Line | Data |
---|---|---|
c1d7c514 | 1 | // SPDX-License-Identifier: GPL-2.0 |
6cbd5570 CM |
2 | /* |
3 | * Copyright (C) 2007 Oracle. All rights reserved. | |
6cbd5570 | 4 | */ |
c1d7c514 | 5 | |
ec6b910f | 6 | #include <linux/sched.h> |
f361bf4a | 7 | #include <linux/sched/signal.h> |
edbd8d4e | 8 | #include <linux/pagemap.h> |
ec44a35c | 9 | #include <linux/writeback.h> |
21af804c | 10 | #include <linux/blkdev.h> |
b7a9f29f | 11 | #include <linux/sort.h> |
4184ea7f | 12 | #include <linux/rcupdate.h> |
817d52f8 | 13 | #include <linux/kthread.h> |
5a0e3ad6 | 14 | #include <linux/slab.h> |
dff51cd1 | 15 | #include <linux/ratelimit.h> |
b150a4f1 | 16 | #include <linux/percpu_counter.h> |
69fe2d75 | 17 | #include <linux/lockdep.h> |
9678c543 | 18 | #include <linux/crc32c.h> |
995946dd | 19 | #include "tree-log.h" |
fec577fb CM |
20 | #include "disk-io.h" |
21 | #include "print-tree.h" | |
0b86a832 | 22 | #include "volumes.h" |
53b381b3 | 23 | #include "raid56.h" |
925baedd | 24 | #include "locking.h" |
fa9c0d79 | 25 | #include "free-space-cache.h" |
1e144fb8 | 26 | #include "free-space-tree.h" |
3fed40cc | 27 | #include "math.h" |
6ab0a202 | 28 | #include "sysfs.h" |
fcebe456 | 29 | #include "qgroup.h" |
fd708b81 | 30 | #include "ref-verify.h" |
fec577fb | 31 | |
709c0486 AJ |
32 | #undef SCRAMBLE_DELAYED_REFS |
33 | ||
9e622d6b MX |
34 | /* |
35 | * control flags for do_chunk_alloc's force field | |
0e4f8f88 CM |
36 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
37 | * if we really need one. | |
38 | * | |
0e4f8f88 CM |
39 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
40 | * if we have very few chunks already allocated. This is | |
41 | * used as part of the clustering code to help make sure | |
42 | * we have a good pool of storage to cluster in, without | |
43 | * filling the FS with empty chunks | |
44 | * | |
9e622d6b MX |
45 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
46 | * | |
0e4f8f88 CM |
47 | */ |
48 | enum { | |
49 | CHUNK_ALLOC_NO_FORCE = 0, | |
9e622d6b MX |
50 | CHUNK_ALLOC_LIMITED = 1, |
51 | CHUNK_ALLOC_FORCE = 2, | |
0e4f8f88 CM |
52 | }; |
53 | ||
5d4f98a2 | 54 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 55 | struct btrfs_fs_info *fs_info, |
c682f9b3 | 56 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
57 | u64 root_objectid, u64 owner_objectid, |
58 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 59 | struct btrfs_delayed_extent_op *extra_op); |
5d4f98a2 YZ |
60 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
61 | struct extent_buffer *leaf, | |
62 | struct btrfs_extent_item *ei); | |
63 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
2ff7e61e | 64 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
65 | u64 parent, u64 root_objectid, |
66 | u64 flags, u64 owner, u64 offset, | |
67 | struct btrfs_key *ins, int ref_mod); | |
68 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
2ff7e61e | 69 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
70 | u64 parent, u64 root_objectid, |
71 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 72 | int level, struct btrfs_key *ins); |
6a63209f | 73 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 74 | struct btrfs_fs_info *fs_info, u64 flags, |
698d0082 | 75 | int force); |
11833d66 YZ |
76 | static int find_next_key(struct btrfs_path *path, int level, |
77 | struct btrfs_key *key); | |
ab8d0fc4 JM |
78 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
79 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 80 | int dump_block_groups); |
5d80366e JB |
81 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
82 | u64 num_bytes); | |
957780eb JB |
83 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, |
84 | struct btrfs_space_info *space_info, | |
85 | u64 num_bytes); | |
86 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
87 | struct btrfs_space_info *space_info, | |
88 | u64 num_bytes); | |
6a63209f | 89 | |
817d52f8 JB |
90 | static noinline int |
91 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
92 | { | |
93 | smp_mb(); | |
36cce922 JB |
94 | return cache->cached == BTRFS_CACHE_FINISHED || |
95 | cache->cached == BTRFS_CACHE_ERROR; | |
817d52f8 JB |
96 | } |
97 | ||
0f9dd46c JB |
98 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
99 | { | |
100 | return (cache->flags & bits) == bits; | |
101 | } | |
102 | ||
758f2dfc | 103 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
104 | { |
105 | atomic_inc(&cache->count); | |
106 | } | |
107 | ||
108 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
109 | { | |
f0486c68 YZ |
110 | if (atomic_dec_and_test(&cache->count)) { |
111 | WARN_ON(cache->pinned > 0); | |
112 | WARN_ON(cache->reserved > 0); | |
0966a7b1 QW |
113 | |
114 | /* | |
115 | * If not empty, someone is still holding mutex of | |
116 | * full_stripe_lock, which can only be released by caller. | |
117 | * And it will definitely cause use-after-free when caller | |
118 | * tries to release full stripe lock. | |
119 | * | |
120 | * No better way to resolve, but only to warn. | |
121 | */ | |
122 | WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); | |
34d52cb6 | 123 | kfree(cache->free_space_ctl); |
11dfe35a | 124 | kfree(cache); |
f0486c68 | 125 | } |
11dfe35a JB |
126 | } |
127 | ||
0f9dd46c JB |
128 | /* |
129 | * this adds the block group to the fs_info rb tree for the block group | |
130 | * cache | |
131 | */ | |
b2950863 | 132 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
133 | struct btrfs_block_group_cache *block_group) |
134 | { | |
135 | struct rb_node **p; | |
136 | struct rb_node *parent = NULL; | |
137 | struct btrfs_block_group_cache *cache; | |
138 | ||
139 | spin_lock(&info->block_group_cache_lock); | |
140 | p = &info->block_group_cache_tree.rb_node; | |
141 | ||
142 | while (*p) { | |
143 | parent = *p; | |
144 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
145 | cache_node); | |
146 | if (block_group->key.objectid < cache->key.objectid) { | |
147 | p = &(*p)->rb_left; | |
148 | } else if (block_group->key.objectid > cache->key.objectid) { | |
149 | p = &(*p)->rb_right; | |
150 | } else { | |
151 | spin_unlock(&info->block_group_cache_lock); | |
152 | return -EEXIST; | |
153 | } | |
154 | } | |
155 | ||
156 | rb_link_node(&block_group->cache_node, parent, p); | |
157 | rb_insert_color(&block_group->cache_node, | |
158 | &info->block_group_cache_tree); | |
a1897fdd LB |
159 | |
160 | if (info->first_logical_byte > block_group->key.objectid) | |
161 | info->first_logical_byte = block_group->key.objectid; | |
162 | ||
0f9dd46c JB |
163 | spin_unlock(&info->block_group_cache_lock); |
164 | ||
165 | return 0; | |
166 | } | |
167 | ||
168 | /* | |
169 | * This will return the block group at or after bytenr if contains is 0, else | |
170 | * it will return the block group that contains the bytenr | |
171 | */ | |
172 | static struct btrfs_block_group_cache * | |
173 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
174 | int contains) | |
175 | { | |
176 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
177 | struct rb_node *n; | |
178 | u64 end, start; | |
179 | ||
180 | spin_lock(&info->block_group_cache_lock); | |
181 | n = info->block_group_cache_tree.rb_node; | |
182 | ||
183 | while (n) { | |
184 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
185 | cache_node); | |
186 | end = cache->key.objectid + cache->key.offset - 1; | |
187 | start = cache->key.objectid; | |
188 | ||
189 | if (bytenr < start) { | |
190 | if (!contains && (!ret || start < ret->key.objectid)) | |
191 | ret = cache; | |
192 | n = n->rb_left; | |
193 | } else if (bytenr > start) { | |
194 | if (contains && bytenr <= end) { | |
195 | ret = cache; | |
196 | break; | |
197 | } | |
198 | n = n->rb_right; | |
199 | } else { | |
200 | ret = cache; | |
201 | break; | |
202 | } | |
203 | } | |
a1897fdd | 204 | if (ret) { |
11dfe35a | 205 | btrfs_get_block_group(ret); |
a1897fdd LB |
206 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
207 | info->first_logical_byte = ret->key.objectid; | |
208 | } | |
0f9dd46c JB |
209 | spin_unlock(&info->block_group_cache_lock); |
210 | ||
211 | return ret; | |
212 | } | |
213 | ||
2ff7e61e | 214 | static int add_excluded_extent(struct btrfs_fs_info *fs_info, |
11833d66 | 215 | u64 start, u64 num_bytes) |
817d52f8 | 216 | { |
11833d66 | 217 | u64 end = start + num_bytes - 1; |
0b246afa | 218 | set_extent_bits(&fs_info->freed_extents[0], |
ceeb0ae7 | 219 | start, end, EXTENT_UPTODATE); |
0b246afa | 220 | set_extent_bits(&fs_info->freed_extents[1], |
ceeb0ae7 | 221 | start, end, EXTENT_UPTODATE); |
11833d66 YZ |
222 | return 0; |
223 | } | |
817d52f8 | 224 | |
2ff7e61e | 225 | static void free_excluded_extents(struct btrfs_fs_info *fs_info, |
11833d66 YZ |
226 | struct btrfs_block_group_cache *cache) |
227 | { | |
228 | u64 start, end; | |
817d52f8 | 229 | |
11833d66 YZ |
230 | start = cache->key.objectid; |
231 | end = start + cache->key.offset - 1; | |
232 | ||
0b246afa | 233 | clear_extent_bits(&fs_info->freed_extents[0], |
91166212 | 234 | start, end, EXTENT_UPTODATE); |
0b246afa | 235 | clear_extent_bits(&fs_info->freed_extents[1], |
91166212 | 236 | start, end, EXTENT_UPTODATE); |
817d52f8 JB |
237 | } |
238 | ||
2ff7e61e | 239 | static int exclude_super_stripes(struct btrfs_fs_info *fs_info, |
11833d66 | 240 | struct btrfs_block_group_cache *cache) |
817d52f8 | 241 | { |
817d52f8 JB |
242 | u64 bytenr; |
243 | u64 *logical; | |
244 | int stripe_len; | |
245 | int i, nr, ret; | |
246 | ||
06b2331f YZ |
247 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
248 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
249 | cache->bytes_super += stripe_len; | |
2ff7e61e | 250 | ret = add_excluded_extent(fs_info, cache->key.objectid, |
06b2331f | 251 | stripe_len); |
835d974f JB |
252 | if (ret) |
253 | return ret; | |
06b2331f YZ |
254 | } |
255 | ||
817d52f8 JB |
256 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
257 | bytenr = btrfs_sb_offset(i); | |
0b246afa | 258 | ret = btrfs_rmap_block(fs_info, cache->key.objectid, |
ab8d0fc4 | 259 | bytenr, 0, &logical, &nr, &stripe_len); |
835d974f JB |
260 | if (ret) |
261 | return ret; | |
11833d66 | 262 | |
817d52f8 | 263 | while (nr--) { |
51bf5f0b JB |
264 | u64 start, len; |
265 | ||
266 | if (logical[nr] > cache->key.objectid + | |
267 | cache->key.offset) | |
268 | continue; | |
269 | ||
270 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
271 | continue; | |
272 | ||
273 | start = logical[nr]; | |
274 | if (start < cache->key.objectid) { | |
275 | start = cache->key.objectid; | |
276 | len = (logical[nr] + stripe_len) - start; | |
277 | } else { | |
278 | len = min_t(u64, stripe_len, | |
279 | cache->key.objectid + | |
280 | cache->key.offset - start); | |
281 | } | |
282 | ||
283 | cache->bytes_super += len; | |
2ff7e61e | 284 | ret = add_excluded_extent(fs_info, start, len); |
835d974f JB |
285 | if (ret) { |
286 | kfree(logical); | |
287 | return ret; | |
288 | } | |
817d52f8 | 289 | } |
11833d66 | 290 | |
817d52f8 JB |
291 | kfree(logical); |
292 | } | |
817d52f8 JB |
293 | return 0; |
294 | } | |
295 | ||
11833d66 YZ |
296 | static struct btrfs_caching_control * |
297 | get_caching_control(struct btrfs_block_group_cache *cache) | |
298 | { | |
299 | struct btrfs_caching_control *ctl; | |
300 | ||
301 | spin_lock(&cache->lock); | |
dde5abee JB |
302 | if (!cache->caching_ctl) { |
303 | spin_unlock(&cache->lock); | |
11833d66 YZ |
304 | return NULL; |
305 | } | |
306 | ||
307 | ctl = cache->caching_ctl; | |
1e4f4714 | 308 | refcount_inc(&ctl->count); |
11833d66 YZ |
309 | spin_unlock(&cache->lock); |
310 | return ctl; | |
311 | } | |
312 | ||
313 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
314 | { | |
1e4f4714 | 315 | if (refcount_dec_and_test(&ctl->count)) |
11833d66 YZ |
316 | kfree(ctl); |
317 | } | |
318 | ||
d0bd4560 | 319 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 320 | static void fragment_free_space(struct btrfs_block_group_cache *block_group) |
d0bd4560 | 321 | { |
2ff7e61e | 322 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
d0bd4560 JB |
323 | u64 start = block_group->key.objectid; |
324 | u64 len = block_group->key.offset; | |
325 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? | |
0b246afa | 326 | fs_info->nodesize : fs_info->sectorsize; |
d0bd4560 JB |
327 | u64 step = chunk << 1; |
328 | ||
329 | while (len > chunk) { | |
330 | btrfs_remove_free_space(block_group, start, chunk); | |
331 | start += step; | |
332 | if (len < step) | |
333 | len = 0; | |
334 | else | |
335 | len -= step; | |
336 | } | |
337 | } | |
338 | #endif | |
339 | ||
0f9dd46c JB |
340 | /* |
341 | * this is only called by cache_block_group, since we could have freed extents | |
342 | * we need to check the pinned_extents for any extents that can't be used yet | |
343 | * since their free space will be released as soon as the transaction commits. | |
344 | */ | |
a5ed9182 OS |
345 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
346 | struct btrfs_fs_info *info, u64 start, u64 end) | |
0f9dd46c | 347 | { |
817d52f8 | 348 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
349 | int ret; |
350 | ||
351 | while (start < end) { | |
11833d66 | 352 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 353 | &extent_start, &extent_end, |
e6138876 JB |
354 | EXTENT_DIRTY | EXTENT_UPTODATE, |
355 | NULL); | |
0f9dd46c JB |
356 | if (ret) |
357 | break; | |
358 | ||
06b2331f | 359 | if (extent_start <= start) { |
0f9dd46c JB |
360 | start = extent_end + 1; |
361 | } else if (extent_start > start && extent_start < end) { | |
362 | size = extent_start - start; | |
817d52f8 | 363 | total_added += size; |
ea6a478e JB |
364 | ret = btrfs_add_free_space(block_group, start, |
365 | size); | |
79787eaa | 366 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
367 | start = extent_end + 1; |
368 | } else { | |
369 | break; | |
370 | } | |
371 | } | |
372 | ||
373 | if (start < end) { | |
374 | size = end - start; | |
817d52f8 | 375 | total_added += size; |
ea6a478e | 376 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 377 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
378 | } |
379 | ||
817d52f8 | 380 | return total_added; |
0f9dd46c JB |
381 | } |
382 | ||
73fa48b6 | 383 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) |
e37c9e69 | 384 | { |
0b246afa JM |
385 | struct btrfs_block_group_cache *block_group = caching_ctl->block_group; |
386 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
387 | struct btrfs_root *extent_root = fs_info->extent_root; | |
e37c9e69 | 388 | struct btrfs_path *path; |
5f39d397 | 389 | struct extent_buffer *leaf; |
11833d66 | 390 | struct btrfs_key key; |
817d52f8 | 391 | u64 total_found = 0; |
11833d66 YZ |
392 | u64 last = 0; |
393 | u32 nritems; | |
73fa48b6 | 394 | int ret; |
d0bd4560 | 395 | bool wakeup = true; |
f510cfec | 396 | |
e37c9e69 CM |
397 | path = btrfs_alloc_path(); |
398 | if (!path) | |
73fa48b6 | 399 | return -ENOMEM; |
7d7d6068 | 400 | |
817d52f8 | 401 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 402 | |
d0bd4560 JB |
403 | #ifdef CONFIG_BTRFS_DEBUG |
404 | /* | |
405 | * If we're fragmenting we don't want to make anybody think we can | |
406 | * allocate from this block group until we've had a chance to fragment | |
407 | * the free space. | |
408 | */ | |
2ff7e61e | 409 | if (btrfs_should_fragment_free_space(block_group)) |
d0bd4560 JB |
410 | wakeup = false; |
411 | #endif | |
5cd57b2c | 412 | /* |
817d52f8 JB |
413 | * We don't want to deadlock with somebody trying to allocate a new |
414 | * extent for the extent root while also trying to search the extent | |
415 | * root to add free space. So we skip locking and search the commit | |
416 | * root, since its read-only | |
5cd57b2c CM |
417 | */ |
418 | path->skip_locking = 1; | |
817d52f8 | 419 | path->search_commit_root = 1; |
e4058b54 | 420 | path->reada = READA_FORWARD; |
817d52f8 | 421 | |
e4404d6e | 422 | key.objectid = last; |
e37c9e69 | 423 | key.offset = 0; |
11833d66 | 424 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 425 | |
52ee28d2 | 426 | next: |
11833d66 | 427 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 428 | if (ret < 0) |
73fa48b6 | 429 | goto out; |
a512bbf8 | 430 | |
11833d66 YZ |
431 | leaf = path->nodes[0]; |
432 | nritems = btrfs_header_nritems(leaf); | |
433 | ||
d397712b | 434 | while (1) { |
7841cb28 | 435 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 436 | last = (u64)-1; |
817d52f8 | 437 | break; |
f25784b3 | 438 | } |
817d52f8 | 439 | |
11833d66 YZ |
440 | if (path->slots[0] < nritems) { |
441 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
442 | } else { | |
443 | ret = find_next_key(path, 0, &key); | |
444 | if (ret) | |
e37c9e69 | 445 | break; |
817d52f8 | 446 | |
c9ea7b24 | 447 | if (need_resched() || |
9e351cc8 | 448 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
d0bd4560 JB |
449 | if (wakeup) |
450 | caching_ctl->progress = last; | |
ff5714cc | 451 | btrfs_release_path(path); |
9e351cc8 | 452 | up_read(&fs_info->commit_root_sem); |
589d8ade | 453 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 454 | cond_resched(); |
73fa48b6 OS |
455 | mutex_lock(&caching_ctl->mutex); |
456 | down_read(&fs_info->commit_root_sem); | |
457 | goto next; | |
589d8ade | 458 | } |
0a3896d0 JB |
459 | |
460 | ret = btrfs_next_leaf(extent_root, path); | |
461 | if (ret < 0) | |
73fa48b6 | 462 | goto out; |
0a3896d0 JB |
463 | if (ret) |
464 | break; | |
589d8ade JB |
465 | leaf = path->nodes[0]; |
466 | nritems = btrfs_header_nritems(leaf); | |
467 | continue; | |
11833d66 | 468 | } |
817d52f8 | 469 | |
52ee28d2 LB |
470 | if (key.objectid < last) { |
471 | key.objectid = last; | |
472 | key.offset = 0; | |
473 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
474 | ||
d0bd4560 JB |
475 | if (wakeup) |
476 | caching_ctl->progress = last; | |
52ee28d2 LB |
477 | btrfs_release_path(path); |
478 | goto next; | |
479 | } | |
480 | ||
11833d66 YZ |
481 | if (key.objectid < block_group->key.objectid) { |
482 | path->slots[0]++; | |
817d52f8 | 483 | continue; |
e37c9e69 | 484 | } |
0f9dd46c | 485 | |
e37c9e69 | 486 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 487 | block_group->key.offset) |
e37c9e69 | 488 | break; |
7d7d6068 | 489 | |
3173a18f JB |
490 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
491 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
817d52f8 JB |
492 | total_found += add_new_free_space(block_group, |
493 | fs_info, last, | |
494 | key.objectid); | |
3173a18f JB |
495 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
496 | last = key.objectid + | |
da17066c | 497 | fs_info->nodesize; |
3173a18f JB |
498 | else |
499 | last = key.objectid + key.offset; | |
817d52f8 | 500 | |
73fa48b6 | 501 | if (total_found > CACHING_CTL_WAKE_UP) { |
11833d66 | 502 | total_found = 0; |
d0bd4560 JB |
503 | if (wakeup) |
504 | wake_up(&caching_ctl->wait); | |
11833d66 | 505 | } |
817d52f8 | 506 | } |
e37c9e69 CM |
507 | path->slots[0]++; |
508 | } | |
817d52f8 | 509 | ret = 0; |
e37c9e69 | 510 | |
817d52f8 JB |
511 | total_found += add_new_free_space(block_group, fs_info, last, |
512 | block_group->key.objectid + | |
513 | block_group->key.offset); | |
11833d66 | 514 | caching_ctl->progress = (u64)-1; |
817d52f8 | 515 | |
73fa48b6 OS |
516 | out: |
517 | btrfs_free_path(path); | |
518 | return ret; | |
519 | } | |
520 | ||
521 | static noinline void caching_thread(struct btrfs_work *work) | |
522 | { | |
523 | struct btrfs_block_group_cache *block_group; | |
524 | struct btrfs_fs_info *fs_info; | |
525 | struct btrfs_caching_control *caching_ctl; | |
526 | int ret; | |
527 | ||
528 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
529 | block_group = caching_ctl->block_group; | |
530 | fs_info = block_group->fs_info; | |
531 | ||
532 | mutex_lock(&caching_ctl->mutex); | |
533 | down_read(&fs_info->commit_root_sem); | |
534 | ||
1e144fb8 OS |
535 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
536 | ret = load_free_space_tree(caching_ctl); | |
537 | else | |
538 | ret = load_extent_tree_free(caching_ctl); | |
73fa48b6 | 539 | |
817d52f8 | 540 | spin_lock(&block_group->lock); |
11833d66 | 541 | block_group->caching_ctl = NULL; |
73fa48b6 | 542 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; |
817d52f8 | 543 | spin_unlock(&block_group->lock); |
0f9dd46c | 544 | |
d0bd4560 | 545 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 546 | if (btrfs_should_fragment_free_space(block_group)) { |
d0bd4560 JB |
547 | u64 bytes_used; |
548 | ||
549 | spin_lock(&block_group->space_info->lock); | |
550 | spin_lock(&block_group->lock); | |
551 | bytes_used = block_group->key.offset - | |
552 | btrfs_block_group_used(&block_group->item); | |
553 | block_group->space_info->bytes_used += bytes_used >> 1; | |
554 | spin_unlock(&block_group->lock); | |
555 | spin_unlock(&block_group->space_info->lock); | |
2ff7e61e | 556 | fragment_free_space(block_group); |
d0bd4560 JB |
557 | } |
558 | #endif | |
559 | ||
560 | caching_ctl->progress = (u64)-1; | |
11833d66 | 561 | |
9e351cc8 | 562 | up_read(&fs_info->commit_root_sem); |
2ff7e61e | 563 | free_excluded_extents(fs_info, block_group); |
11833d66 | 564 | mutex_unlock(&caching_ctl->mutex); |
73fa48b6 | 565 | |
11833d66 YZ |
566 | wake_up(&caching_ctl->wait); |
567 | ||
568 | put_caching_control(caching_ctl); | |
11dfe35a | 569 | btrfs_put_block_group(block_group); |
817d52f8 JB |
570 | } |
571 | ||
9d66e233 | 572 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 573 | int load_cache_only) |
817d52f8 | 574 | { |
291c7d2f | 575 | DEFINE_WAIT(wait); |
11833d66 YZ |
576 | struct btrfs_fs_info *fs_info = cache->fs_info; |
577 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
578 | int ret = 0; |
579 | ||
291c7d2f | 580 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
581 | if (!caching_ctl) |
582 | return -ENOMEM; | |
291c7d2f JB |
583 | |
584 | INIT_LIST_HEAD(&caching_ctl->list); | |
585 | mutex_init(&caching_ctl->mutex); | |
586 | init_waitqueue_head(&caching_ctl->wait); | |
587 | caching_ctl->block_group = cache; | |
588 | caching_ctl->progress = cache->key.objectid; | |
1e4f4714 | 589 | refcount_set(&caching_ctl->count, 1); |
9e0af237 LB |
590 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, |
591 | caching_thread, NULL, NULL); | |
291c7d2f JB |
592 | |
593 | spin_lock(&cache->lock); | |
594 | /* | |
595 | * This should be a rare occasion, but this could happen I think in the | |
596 | * case where one thread starts to load the space cache info, and then | |
597 | * some other thread starts a transaction commit which tries to do an | |
598 | * allocation while the other thread is still loading the space cache | |
599 | * info. The previous loop should have kept us from choosing this block | |
600 | * group, but if we've moved to the state where we will wait on caching | |
601 | * block groups we need to first check if we're doing a fast load here, | |
602 | * so we can wait for it to finish, otherwise we could end up allocating | |
603 | * from a block group who's cache gets evicted for one reason or | |
604 | * another. | |
605 | */ | |
606 | while (cache->cached == BTRFS_CACHE_FAST) { | |
607 | struct btrfs_caching_control *ctl; | |
608 | ||
609 | ctl = cache->caching_ctl; | |
1e4f4714 | 610 | refcount_inc(&ctl->count); |
291c7d2f JB |
611 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); |
612 | spin_unlock(&cache->lock); | |
613 | ||
614 | schedule(); | |
615 | ||
616 | finish_wait(&ctl->wait, &wait); | |
617 | put_caching_control(ctl); | |
618 | spin_lock(&cache->lock); | |
619 | } | |
620 | ||
621 | if (cache->cached != BTRFS_CACHE_NO) { | |
622 | spin_unlock(&cache->lock); | |
623 | kfree(caching_ctl); | |
11833d66 | 624 | return 0; |
291c7d2f JB |
625 | } |
626 | WARN_ON(cache->caching_ctl); | |
627 | cache->caching_ctl = caching_ctl; | |
628 | cache->cached = BTRFS_CACHE_FAST; | |
629 | spin_unlock(&cache->lock); | |
11833d66 | 630 | |
d8953d69 | 631 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cb83b7b8 | 632 | mutex_lock(&caching_ctl->mutex); |
9d66e233 JB |
633 | ret = load_free_space_cache(fs_info, cache); |
634 | ||
635 | spin_lock(&cache->lock); | |
636 | if (ret == 1) { | |
291c7d2f | 637 | cache->caching_ctl = NULL; |
9d66e233 JB |
638 | cache->cached = BTRFS_CACHE_FINISHED; |
639 | cache->last_byte_to_unpin = (u64)-1; | |
cb83b7b8 | 640 | caching_ctl->progress = (u64)-1; |
9d66e233 | 641 | } else { |
291c7d2f JB |
642 | if (load_cache_only) { |
643 | cache->caching_ctl = NULL; | |
644 | cache->cached = BTRFS_CACHE_NO; | |
645 | } else { | |
646 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 647 | cache->has_caching_ctl = 1; |
291c7d2f | 648 | } |
9d66e233 JB |
649 | } |
650 | spin_unlock(&cache->lock); | |
d0bd4560 JB |
651 | #ifdef CONFIG_BTRFS_DEBUG |
652 | if (ret == 1 && | |
2ff7e61e | 653 | btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
654 | u64 bytes_used; |
655 | ||
656 | spin_lock(&cache->space_info->lock); | |
657 | spin_lock(&cache->lock); | |
658 | bytes_used = cache->key.offset - | |
659 | btrfs_block_group_used(&cache->item); | |
660 | cache->space_info->bytes_used += bytes_used >> 1; | |
661 | spin_unlock(&cache->lock); | |
662 | spin_unlock(&cache->space_info->lock); | |
2ff7e61e | 663 | fragment_free_space(cache); |
d0bd4560 JB |
664 | } |
665 | #endif | |
cb83b7b8 JB |
666 | mutex_unlock(&caching_ctl->mutex); |
667 | ||
291c7d2f | 668 | wake_up(&caching_ctl->wait); |
3c14874a | 669 | if (ret == 1) { |
291c7d2f | 670 | put_caching_control(caching_ctl); |
2ff7e61e | 671 | free_excluded_extents(fs_info, cache); |
9d66e233 | 672 | return 0; |
3c14874a | 673 | } |
291c7d2f JB |
674 | } else { |
675 | /* | |
1e144fb8 OS |
676 | * We're either using the free space tree or no caching at all. |
677 | * Set cached to the appropriate value and wakeup any waiters. | |
291c7d2f JB |
678 | */ |
679 | spin_lock(&cache->lock); | |
680 | if (load_cache_only) { | |
681 | cache->caching_ctl = NULL; | |
682 | cache->cached = BTRFS_CACHE_NO; | |
683 | } else { | |
684 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 685 | cache->has_caching_ctl = 1; |
291c7d2f JB |
686 | } |
687 | spin_unlock(&cache->lock); | |
688 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
689 | } |
690 | ||
291c7d2f JB |
691 | if (load_cache_only) { |
692 | put_caching_control(caching_ctl); | |
11833d66 | 693 | return 0; |
817d52f8 | 694 | } |
817d52f8 | 695 | |
9e351cc8 | 696 | down_write(&fs_info->commit_root_sem); |
1e4f4714 | 697 | refcount_inc(&caching_ctl->count); |
11833d66 | 698 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
9e351cc8 | 699 | up_write(&fs_info->commit_root_sem); |
11833d66 | 700 | |
11dfe35a | 701 | btrfs_get_block_group(cache); |
11833d66 | 702 | |
e66f0bb1 | 703 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 704 | |
ef8bbdfe | 705 | return ret; |
e37c9e69 CM |
706 | } |
707 | ||
0f9dd46c JB |
708 | /* |
709 | * return the block group that starts at or after bytenr | |
710 | */ | |
d397712b CM |
711 | static struct btrfs_block_group_cache * |
712 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 713 | { |
e2c89907 | 714 | return block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b CM |
715 | } |
716 | ||
0f9dd46c | 717 | /* |
9f55684c | 718 | * return the block group that contains the given bytenr |
0f9dd46c | 719 | */ |
d397712b CM |
720 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
721 | struct btrfs_fs_info *info, | |
722 | u64 bytenr) | |
be744175 | 723 | { |
e2c89907 | 724 | return block_group_cache_tree_search(info, bytenr, 1); |
be744175 | 725 | } |
0b86a832 | 726 | |
0f9dd46c JB |
727 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
728 | u64 flags) | |
6324fbf3 | 729 | { |
0f9dd46c | 730 | struct list_head *head = &info->space_info; |
0f9dd46c | 731 | struct btrfs_space_info *found; |
4184ea7f | 732 | |
52ba6929 | 733 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 734 | |
4184ea7f CM |
735 | rcu_read_lock(); |
736 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 737 | if (found->flags & flags) { |
4184ea7f | 738 | rcu_read_unlock(); |
0f9dd46c | 739 | return found; |
4184ea7f | 740 | } |
0f9dd46c | 741 | } |
4184ea7f | 742 | rcu_read_unlock(); |
0f9dd46c | 743 | return NULL; |
6324fbf3 CM |
744 | } |
745 | ||
0d9f824d | 746 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes, |
29d2b84c | 747 | bool metadata, u64 root_objectid) |
0d9f824d OS |
748 | { |
749 | struct btrfs_space_info *space_info; | |
750 | u64 flags; | |
751 | ||
29d2b84c | 752 | if (metadata) { |
0d9f824d OS |
753 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) |
754 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
755 | else | |
756 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
757 | } else { | |
758 | flags = BTRFS_BLOCK_GROUP_DATA; | |
759 | } | |
760 | ||
761 | space_info = __find_space_info(fs_info, flags); | |
55e8196a | 762 | ASSERT(space_info); |
0d9f824d OS |
763 | percpu_counter_add(&space_info->total_bytes_pinned, num_bytes); |
764 | } | |
765 | ||
4184ea7f CM |
766 | /* |
767 | * after adding space to the filesystem, we need to clear the full flags | |
768 | * on all the space infos. | |
769 | */ | |
770 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
771 | { | |
772 | struct list_head *head = &info->space_info; | |
773 | struct btrfs_space_info *found; | |
774 | ||
775 | rcu_read_lock(); | |
776 | list_for_each_entry_rcu(found, head, list) | |
777 | found->full = 0; | |
778 | rcu_read_unlock(); | |
779 | } | |
780 | ||
1a4ed8fd | 781 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 782 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
783 | { |
784 | int ret; | |
785 | struct btrfs_key key; | |
31840ae1 | 786 | struct btrfs_path *path; |
e02119d5 | 787 | |
31840ae1 | 788 | path = btrfs_alloc_path(); |
d8926bb3 MF |
789 | if (!path) |
790 | return -ENOMEM; | |
791 | ||
e02119d5 CM |
792 | key.objectid = start; |
793 | key.offset = len; | |
3173a18f | 794 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 795 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 796 | btrfs_free_path(path); |
7bb86316 CM |
797 | return ret; |
798 | } | |
799 | ||
a22285a6 | 800 | /* |
3173a18f | 801 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
802 | * |
803 | * the head node for delayed ref is used to store the sum of all the | |
804 | * reference count modifications queued up in the rbtree. the head | |
805 | * node may also store the extent flags to set. This way you can check | |
806 | * to see what the reference count and extent flags would be if all of | |
807 | * the delayed refs are not processed. | |
808 | */ | |
809 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 810 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 811 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
812 | { |
813 | struct btrfs_delayed_ref_head *head; | |
814 | struct btrfs_delayed_ref_root *delayed_refs; | |
815 | struct btrfs_path *path; | |
816 | struct btrfs_extent_item *ei; | |
817 | struct extent_buffer *leaf; | |
818 | struct btrfs_key key; | |
819 | u32 item_size; | |
820 | u64 num_refs; | |
821 | u64 extent_flags; | |
822 | int ret; | |
823 | ||
3173a18f JB |
824 | /* |
825 | * If we don't have skinny metadata, don't bother doing anything | |
826 | * different | |
827 | */ | |
0b246afa JM |
828 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
829 | offset = fs_info->nodesize; | |
3173a18f JB |
830 | metadata = 0; |
831 | } | |
832 | ||
a22285a6 YZ |
833 | path = btrfs_alloc_path(); |
834 | if (!path) | |
835 | return -ENOMEM; | |
836 | ||
a22285a6 YZ |
837 | if (!trans) { |
838 | path->skip_locking = 1; | |
839 | path->search_commit_root = 1; | |
840 | } | |
639eefc8 FDBM |
841 | |
842 | search_again: | |
843 | key.objectid = bytenr; | |
844 | key.offset = offset; | |
845 | if (metadata) | |
846 | key.type = BTRFS_METADATA_ITEM_KEY; | |
847 | else | |
848 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
849 | ||
0b246afa | 850 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
851 | if (ret < 0) |
852 | goto out_free; | |
853 | ||
3173a18f | 854 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
855 | if (path->slots[0]) { |
856 | path->slots[0]--; | |
857 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
858 | path->slots[0]); | |
859 | if (key.objectid == bytenr && | |
860 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 861 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
862 | ret = 0; |
863 | } | |
3173a18f JB |
864 | } |
865 | ||
a22285a6 YZ |
866 | if (ret == 0) { |
867 | leaf = path->nodes[0]; | |
868 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
869 | if (item_size >= sizeof(*ei)) { | |
870 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
871 | struct btrfs_extent_item); | |
872 | num_refs = btrfs_extent_refs(leaf, ei); | |
873 | extent_flags = btrfs_extent_flags(leaf, ei); | |
874 | } else { | |
875 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
876 | struct btrfs_extent_item_v0 *ei0; | |
877 | BUG_ON(item_size != sizeof(*ei0)); | |
878 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
879 | struct btrfs_extent_item_v0); | |
880 | num_refs = btrfs_extent_refs_v0(leaf, ei0); | |
881 | /* FIXME: this isn't correct for data */ | |
882 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
883 | #else | |
884 | BUG(); | |
885 | #endif | |
886 | } | |
887 | BUG_ON(num_refs == 0); | |
888 | } else { | |
889 | num_refs = 0; | |
890 | extent_flags = 0; | |
891 | ret = 0; | |
892 | } | |
893 | ||
894 | if (!trans) | |
895 | goto out; | |
896 | ||
897 | delayed_refs = &trans->transaction->delayed_refs; | |
898 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 899 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
900 | if (head) { |
901 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 902 | refcount_inc(&head->refs); |
a22285a6 YZ |
903 | spin_unlock(&delayed_refs->lock); |
904 | ||
b3b4aa74 | 905 | btrfs_release_path(path); |
a22285a6 | 906 | |
8cc33e5c DS |
907 | /* |
908 | * Mutex was contended, block until it's released and try | |
909 | * again | |
910 | */ | |
a22285a6 YZ |
911 | mutex_lock(&head->mutex); |
912 | mutex_unlock(&head->mutex); | |
d278850e | 913 | btrfs_put_delayed_ref_head(head); |
639eefc8 | 914 | goto search_again; |
a22285a6 | 915 | } |
d7df2c79 | 916 | spin_lock(&head->lock); |
a22285a6 YZ |
917 | if (head->extent_op && head->extent_op->update_flags) |
918 | extent_flags |= head->extent_op->flags_to_set; | |
919 | else | |
920 | BUG_ON(num_refs == 0); | |
921 | ||
d278850e | 922 | num_refs += head->ref_mod; |
d7df2c79 | 923 | spin_unlock(&head->lock); |
a22285a6 YZ |
924 | mutex_unlock(&head->mutex); |
925 | } | |
926 | spin_unlock(&delayed_refs->lock); | |
927 | out: | |
928 | WARN_ON(num_refs == 0); | |
929 | if (refs) | |
930 | *refs = num_refs; | |
931 | if (flags) | |
932 | *flags = extent_flags; | |
933 | out_free: | |
934 | btrfs_free_path(path); | |
935 | return ret; | |
936 | } | |
937 | ||
d8d5f3e1 CM |
938 | /* |
939 | * Back reference rules. Back refs have three main goals: | |
940 | * | |
941 | * 1) differentiate between all holders of references to an extent so that | |
942 | * when a reference is dropped we can make sure it was a valid reference | |
943 | * before freeing the extent. | |
944 | * | |
945 | * 2) Provide enough information to quickly find the holders of an extent | |
946 | * if we notice a given block is corrupted or bad. | |
947 | * | |
948 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
949 | * maintenance. This is actually the same as #2, but with a slightly | |
950 | * different use case. | |
951 | * | |
5d4f98a2 YZ |
952 | * There are two kinds of back refs. The implicit back refs is optimized |
953 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
954 | * back refs of this kind provide information about the block's owner tree | |
955 | * and the pointer's key. These information allow us to find the block by | |
956 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
957 | * referenced by their owner trees. The location of tree block is recorded | |
958 | * in the back refs. Actually the full back refs is generic, and can be | |
959 | * used in all cases the implicit back refs is used. The major shortcoming | |
960 | * of the full back refs is its overhead. Every time a tree block gets | |
961 | * COWed, we have to update back refs entry for all pointers in it. | |
962 | * | |
963 | * For a newly allocated tree block, we use implicit back refs for | |
964 | * pointers in it. This means most tree related operations only involve | |
965 | * implicit back refs. For a tree block created in old transaction, the | |
966 | * only way to drop a reference to it is COW it. So we can detect the | |
967 | * event that tree block loses its owner tree's reference and do the | |
968 | * back refs conversion. | |
969 | * | |
01327610 | 970 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
971 | * |
972 | * The reference count of the block is one and the tree is the block's | |
973 | * owner tree. Nothing to do in this case. | |
974 | * | |
975 | * The reference count of the block is one and the tree is not the | |
976 | * block's owner tree. In this case, full back refs is used for pointers | |
977 | * in the block. Remove these full back refs, add implicit back refs for | |
978 | * every pointers in the new block. | |
979 | * | |
980 | * The reference count of the block is greater than one and the tree is | |
981 | * the block's owner tree. In this case, implicit back refs is used for | |
982 | * pointers in the block. Add full back refs for every pointers in the | |
983 | * block, increase lower level extents' reference counts. The original | |
984 | * implicit back refs are entailed to the new block. | |
985 | * | |
986 | * The reference count of the block is greater than one and the tree is | |
987 | * not the block's owner tree. Add implicit back refs for every pointer in | |
988 | * the new block, increase lower level extents' reference count. | |
989 | * | |
990 | * Back Reference Key composing: | |
991 | * | |
992 | * The key objectid corresponds to the first byte in the extent, | |
993 | * The key type is used to differentiate between types of back refs. | |
994 | * There are different meanings of the key offset for different types | |
995 | * of back refs. | |
996 | * | |
d8d5f3e1 CM |
997 | * File extents can be referenced by: |
998 | * | |
999 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 1000 | * - different files inside a single subvolume |
d8d5f3e1 CM |
1001 | * - different offsets inside a file (bookend extents in file.c) |
1002 | * | |
5d4f98a2 | 1003 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
1004 | * |
1005 | * - Objectid of the subvolume root | |
d8d5f3e1 | 1006 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
1007 | * - original offset in the file |
1008 | * - how many bookend extents | |
d8d5f3e1 | 1009 | * |
5d4f98a2 YZ |
1010 | * The key offset for the implicit back refs is hash of the first |
1011 | * three fields. | |
d8d5f3e1 | 1012 | * |
5d4f98a2 | 1013 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 1014 | * |
5d4f98a2 | 1015 | * - number of pointers in the tree leaf |
d8d5f3e1 | 1016 | * |
5d4f98a2 YZ |
1017 | * The key offset for the implicit back refs is the first byte of |
1018 | * the tree leaf | |
d8d5f3e1 | 1019 | * |
5d4f98a2 YZ |
1020 | * When a file extent is allocated, The implicit back refs is used. |
1021 | * the fields are filled in: | |
d8d5f3e1 | 1022 | * |
5d4f98a2 | 1023 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 1024 | * |
5d4f98a2 YZ |
1025 | * When a file extent is removed file truncation, we find the |
1026 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 1027 | * |
5d4f98a2 | 1028 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 1029 | * |
5d4f98a2 | 1030 | * Btree extents can be referenced by: |
d8d5f3e1 | 1031 | * |
5d4f98a2 | 1032 | * - Different subvolumes |
d8d5f3e1 | 1033 | * |
5d4f98a2 YZ |
1034 | * Both the implicit back refs and the full back refs for tree blocks |
1035 | * only consist of key. The key offset for the implicit back refs is | |
1036 | * objectid of block's owner tree. The key offset for the full back refs | |
1037 | * is the first byte of parent block. | |
d8d5f3e1 | 1038 | * |
5d4f98a2 YZ |
1039 | * When implicit back refs is used, information about the lowest key and |
1040 | * level of the tree block are required. These information are stored in | |
1041 | * tree block info structure. | |
d8d5f3e1 | 1042 | */ |
31840ae1 | 1043 | |
5d4f98a2 YZ |
1044 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
1045 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, | |
87bde3cd | 1046 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1047 | struct btrfs_path *path, |
1048 | u64 owner, u32 extra_size) | |
7bb86316 | 1049 | { |
87bde3cd | 1050 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1051 | struct btrfs_extent_item *item; |
1052 | struct btrfs_extent_item_v0 *ei0; | |
1053 | struct btrfs_extent_ref_v0 *ref0; | |
1054 | struct btrfs_tree_block_info *bi; | |
1055 | struct extent_buffer *leaf; | |
7bb86316 | 1056 | struct btrfs_key key; |
5d4f98a2 YZ |
1057 | struct btrfs_key found_key; |
1058 | u32 new_size = sizeof(*item); | |
1059 | u64 refs; | |
1060 | int ret; | |
1061 | ||
1062 | leaf = path->nodes[0]; | |
1063 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); | |
1064 | ||
1065 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1066 | ei0 = btrfs_item_ptr(leaf, path->slots[0], | |
1067 | struct btrfs_extent_item_v0); | |
1068 | refs = btrfs_extent_refs_v0(leaf, ei0); | |
1069 | ||
1070 | if (owner == (u64)-1) { | |
1071 | while (1) { | |
1072 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1073 | ret = btrfs_next_leaf(root, path); | |
1074 | if (ret < 0) | |
1075 | return ret; | |
79787eaa | 1076 | BUG_ON(ret > 0); /* Corruption */ |
5d4f98a2 YZ |
1077 | leaf = path->nodes[0]; |
1078 | } | |
1079 | btrfs_item_key_to_cpu(leaf, &found_key, | |
1080 | path->slots[0]); | |
1081 | BUG_ON(key.objectid != found_key.objectid); | |
1082 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { | |
1083 | path->slots[0]++; | |
1084 | continue; | |
1085 | } | |
1086 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1087 | struct btrfs_extent_ref_v0); | |
1088 | owner = btrfs_ref_objectid_v0(leaf, ref0); | |
1089 | break; | |
1090 | } | |
1091 | } | |
b3b4aa74 | 1092 | btrfs_release_path(path); |
5d4f98a2 YZ |
1093 | |
1094 | if (owner < BTRFS_FIRST_FREE_OBJECTID) | |
1095 | new_size += sizeof(*bi); | |
1096 | ||
1097 | new_size -= sizeof(*ei0); | |
1098 | ret = btrfs_search_slot(trans, root, &key, path, | |
1099 | new_size + extra_size, 1); | |
1100 | if (ret < 0) | |
1101 | return ret; | |
79787eaa | 1102 | BUG_ON(ret); /* Corruption */ |
5d4f98a2 | 1103 | |
87bde3cd | 1104 | btrfs_extend_item(fs_info, path, new_size); |
5d4f98a2 YZ |
1105 | |
1106 | leaf = path->nodes[0]; | |
1107 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1108 | btrfs_set_extent_refs(leaf, item, refs); | |
1109 | /* FIXME: get real generation */ | |
1110 | btrfs_set_extent_generation(leaf, item, 0); | |
1111 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1112 | btrfs_set_extent_flags(leaf, item, | |
1113 | BTRFS_EXTENT_FLAG_TREE_BLOCK | | |
1114 | BTRFS_BLOCK_FLAG_FULL_BACKREF); | |
1115 | bi = (struct btrfs_tree_block_info *)(item + 1); | |
1116 | /* FIXME: get first key of the block */ | |
b159fa28 | 1117 | memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi)); |
5d4f98a2 YZ |
1118 | btrfs_set_tree_block_level(leaf, bi, (int)owner); |
1119 | } else { | |
1120 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); | |
1121 | } | |
1122 | btrfs_mark_buffer_dirty(leaf); | |
1123 | return 0; | |
1124 | } | |
1125 | #endif | |
1126 | ||
167ce953 LB |
1127 | /* |
1128 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
1129 | * is_data == BTRFS_REF_TYPE_DATA, data type is requried, | |
1130 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. | |
1131 | */ | |
1132 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
1133 | struct btrfs_extent_inline_ref *iref, | |
1134 | enum btrfs_inline_ref_type is_data) | |
1135 | { | |
1136 | int type = btrfs_extent_inline_ref_type(eb, iref); | |
64ecdb64 | 1137 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 LB |
1138 | |
1139 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
1140 | type == BTRFS_SHARED_BLOCK_REF_KEY || | |
1141 | type == BTRFS_SHARED_DATA_REF_KEY || | |
1142 | type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1143 | if (is_data == BTRFS_REF_TYPE_BLOCK) { | |
64ecdb64 | 1144 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
167ce953 | 1145 | return type; |
64ecdb64 LB |
1146 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
1147 | ASSERT(eb->fs_info); | |
1148 | /* | |
1149 | * Every shared one has parent tree | |
1150 | * block, which must be aligned to | |
1151 | * nodesize. | |
1152 | */ | |
1153 | if (offset && | |
1154 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1155 | return type; | |
1156 | } | |
167ce953 | 1157 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 1158 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 1159 | return type; |
64ecdb64 LB |
1160 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
1161 | ASSERT(eb->fs_info); | |
1162 | /* | |
1163 | * Every shared one has parent tree | |
1164 | * block, which must be aligned to | |
1165 | * nodesize. | |
1166 | */ | |
1167 | if (offset && | |
1168 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1169 | return type; | |
1170 | } | |
167ce953 LB |
1171 | } else { |
1172 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
1173 | return type; | |
1174 | } | |
1175 | } | |
1176 | ||
1177 | btrfs_print_leaf((struct extent_buffer *)eb); | |
1178 | btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d", | |
1179 | eb->start, type); | |
1180 | WARN_ON(1); | |
1181 | ||
1182 | return BTRFS_REF_TYPE_INVALID; | |
1183 | } | |
1184 | ||
5d4f98a2 YZ |
1185 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
1186 | { | |
1187 | u32 high_crc = ~(u32)0; | |
1188 | u32 low_crc = ~(u32)0; | |
1189 | __le64 lenum; | |
1190 | ||
1191 | lenum = cpu_to_le64(root_objectid); | |
9678c543 | 1192 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1193 | lenum = cpu_to_le64(owner); |
9678c543 | 1194 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1195 | lenum = cpu_to_le64(offset); |
9678c543 | 1196 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1197 | |
1198 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1199 | } | |
1200 | ||
1201 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1202 | struct btrfs_extent_data_ref *ref) | |
1203 | { | |
1204 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1205 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1206 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1207 | } | |
1208 | ||
1209 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1210 | struct btrfs_extent_data_ref *ref, | |
1211 | u64 root_objectid, u64 owner, u64 offset) | |
1212 | { | |
1213 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1214 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1215 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1216 | return 0; | |
1217 | return 1; | |
1218 | } | |
1219 | ||
1220 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1221 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1222 | struct btrfs_path *path, |
1223 | u64 bytenr, u64 parent, | |
1224 | u64 root_objectid, | |
1225 | u64 owner, u64 offset) | |
1226 | { | |
87bde3cd | 1227 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1228 | struct btrfs_key key; |
1229 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1230 | struct extent_buffer *leaf; |
5d4f98a2 | 1231 | u32 nritems; |
74493f7a | 1232 | int ret; |
5d4f98a2 YZ |
1233 | int recow; |
1234 | int err = -ENOENT; | |
74493f7a | 1235 | |
31840ae1 | 1236 | key.objectid = bytenr; |
5d4f98a2 YZ |
1237 | if (parent) { |
1238 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1239 | key.offset = parent; | |
1240 | } else { | |
1241 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1242 | key.offset = hash_extent_data_ref(root_objectid, | |
1243 | owner, offset); | |
1244 | } | |
1245 | again: | |
1246 | recow = 0; | |
1247 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1248 | if (ret < 0) { | |
1249 | err = ret; | |
1250 | goto fail; | |
1251 | } | |
31840ae1 | 1252 | |
5d4f98a2 YZ |
1253 | if (parent) { |
1254 | if (!ret) | |
1255 | return 0; | |
1256 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1257 | key.type = BTRFS_EXTENT_REF_V0_KEY; | |
b3b4aa74 | 1258 | btrfs_release_path(path); |
5d4f98a2 YZ |
1259 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1260 | if (ret < 0) { | |
1261 | err = ret; | |
1262 | goto fail; | |
1263 | } | |
1264 | if (!ret) | |
1265 | return 0; | |
1266 | #endif | |
1267 | goto fail; | |
31840ae1 ZY |
1268 | } |
1269 | ||
1270 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1271 | nritems = btrfs_header_nritems(leaf); |
1272 | while (1) { | |
1273 | if (path->slots[0] >= nritems) { | |
1274 | ret = btrfs_next_leaf(root, path); | |
1275 | if (ret < 0) | |
1276 | err = ret; | |
1277 | if (ret) | |
1278 | goto fail; | |
1279 | ||
1280 | leaf = path->nodes[0]; | |
1281 | nritems = btrfs_header_nritems(leaf); | |
1282 | recow = 1; | |
1283 | } | |
1284 | ||
1285 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1286 | if (key.objectid != bytenr || | |
1287 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1288 | goto fail; | |
1289 | ||
1290 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1291 | struct btrfs_extent_data_ref); | |
1292 | ||
1293 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1294 | owner, offset)) { | |
1295 | if (recow) { | |
b3b4aa74 | 1296 | btrfs_release_path(path); |
5d4f98a2 YZ |
1297 | goto again; |
1298 | } | |
1299 | err = 0; | |
1300 | break; | |
1301 | } | |
1302 | path->slots[0]++; | |
31840ae1 | 1303 | } |
5d4f98a2 YZ |
1304 | fail: |
1305 | return err; | |
31840ae1 ZY |
1306 | } |
1307 | ||
5d4f98a2 | 1308 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1309 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1310 | struct btrfs_path *path, |
1311 | u64 bytenr, u64 parent, | |
1312 | u64 root_objectid, u64 owner, | |
1313 | u64 offset, int refs_to_add) | |
31840ae1 | 1314 | { |
87bde3cd | 1315 | struct btrfs_root *root = fs_info->extent_root; |
31840ae1 ZY |
1316 | struct btrfs_key key; |
1317 | struct extent_buffer *leaf; | |
5d4f98a2 | 1318 | u32 size; |
31840ae1 ZY |
1319 | u32 num_refs; |
1320 | int ret; | |
74493f7a | 1321 | |
74493f7a | 1322 | key.objectid = bytenr; |
5d4f98a2 YZ |
1323 | if (parent) { |
1324 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1325 | key.offset = parent; | |
1326 | size = sizeof(struct btrfs_shared_data_ref); | |
1327 | } else { | |
1328 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1329 | key.offset = hash_extent_data_ref(root_objectid, | |
1330 | owner, offset); | |
1331 | size = sizeof(struct btrfs_extent_data_ref); | |
1332 | } | |
74493f7a | 1333 | |
5d4f98a2 YZ |
1334 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1335 | if (ret && ret != -EEXIST) | |
1336 | goto fail; | |
1337 | ||
1338 | leaf = path->nodes[0]; | |
1339 | if (parent) { | |
1340 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1341 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1342 | struct btrfs_shared_data_ref); |
1343 | if (ret == 0) { | |
1344 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1345 | } else { | |
1346 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1347 | num_refs += refs_to_add; | |
1348 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1349 | } |
5d4f98a2 YZ |
1350 | } else { |
1351 | struct btrfs_extent_data_ref *ref; | |
1352 | while (ret == -EEXIST) { | |
1353 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1354 | struct btrfs_extent_data_ref); | |
1355 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1356 | owner, offset)) | |
1357 | break; | |
b3b4aa74 | 1358 | btrfs_release_path(path); |
5d4f98a2 YZ |
1359 | key.offset++; |
1360 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1361 | size); | |
1362 | if (ret && ret != -EEXIST) | |
1363 | goto fail; | |
31840ae1 | 1364 | |
5d4f98a2 YZ |
1365 | leaf = path->nodes[0]; |
1366 | } | |
1367 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1368 | struct btrfs_extent_data_ref); | |
1369 | if (ret == 0) { | |
1370 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1371 | root_objectid); | |
1372 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1373 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1374 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1375 | } else { | |
1376 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1377 | num_refs += refs_to_add; | |
1378 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1379 | } |
31840ae1 | 1380 | } |
5d4f98a2 YZ |
1381 | btrfs_mark_buffer_dirty(leaf); |
1382 | ret = 0; | |
1383 | fail: | |
b3b4aa74 | 1384 | btrfs_release_path(path); |
7bb86316 | 1385 | return ret; |
74493f7a CM |
1386 | } |
1387 | ||
5d4f98a2 | 1388 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1389 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 1390 | struct btrfs_path *path, |
fcebe456 | 1391 | int refs_to_drop, int *last_ref) |
31840ae1 | 1392 | { |
5d4f98a2 YZ |
1393 | struct btrfs_key key; |
1394 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1395 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1396 | struct extent_buffer *leaf; |
5d4f98a2 | 1397 | u32 num_refs = 0; |
31840ae1 ZY |
1398 | int ret = 0; |
1399 | ||
1400 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1401 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1402 | ||
1403 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1404 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1405 | struct btrfs_extent_data_ref); | |
1406 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1407 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1408 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1409 | struct btrfs_shared_data_ref); | |
1410 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1411 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1412 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1413 | struct btrfs_extent_ref_v0 *ref0; | |
1414 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1415 | struct btrfs_extent_ref_v0); | |
1416 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
1417 | #endif | |
1418 | } else { | |
1419 | BUG(); | |
1420 | } | |
1421 | ||
56bec294 CM |
1422 | BUG_ON(num_refs < refs_to_drop); |
1423 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1424 | |
31840ae1 | 1425 | if (num_refs == 0) { |
87bde3cd | 1426 | ret = btrfs_del_item(trans, fs_info->extent_root, path); |
fcebe456 | 1427 | *last_ref = 1; |
31840ae1 | 1428 | } else { |
5d4f98a2 YZ |
1429 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1430 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1431 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1432 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
1433 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1434 | else { | |
1435 | struct btrfs_extent_ref_v0 *ref0; | |
1436 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1437 | struct btrfs_extent_ref_v0); | |
1438 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); | |
1439 | } | |
1440 | #endif | |
31840ae1 ZY |
1441 | btrfs_mark_buffer_dirty(leaf); |
1442 | } | |
31840ae1 ZY |
1443 | return ret; |
1444 | } | |
1445 | ||
9ed0dea0 | 1446 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1447 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1448 | { |
5d4f98a2 YZ |
1449 | struct btrfs_key key; |
1450 | struct extent_buffer *leaf; | |
1451 | struct btrfs_extent_data_ref *ref1; | |
1452 | struct btrfs_shared_data_ref *ref2; | |
1453 | u32 num_refs = 0; | |
3de28d57 | 1454 | int type; |
5d4f98a2 YZ |
1455 | |
1456 | leaf = path->nodes[0]; | |
1457 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1458 | if (iref) { | |
3de28d57 LB |
1459 | /* |
1460 | * If type is invalid, we should have bailed out earlier than | |
1461 | * this call. | |
1462 | */ | |
1463 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
1464 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
1465 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
1466 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
1467 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1468 | } else { | |
1469 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1470 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1471 | } | |
1472 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1473 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1474 | struct btrfs_extent_data_ref); | |
1475 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1476 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1477 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1478 | struct btrfs_shared_data_ref); | |
1479 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1480 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1481 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { | |
1482 | struct btrfs_extent_ref_v0 *ref0; | |
1483 | ref0 = btrfs_item_ptr(leaf, path->slots[0], | |
1484 | struct btrfs_extent_ref_v0); | |
1485 | num_refs = btrfs_ref_count_v0(leaf, ref0); | |
4b4e25f2 | 1486 | #endif |
5d4f98a2 YZ |
1487 | } else { |
1488 | WARN_ON(1); | |
1489 | } | |
1490 | return num_refs; | |
1491 | } | |
15916de8 | 1492 | |
5d4f98a2 | 1493 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1494 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1495 | struct btrfs_path *path, |
1496 | u64 bytenr, u64 parent, | |
1497 | u64 root_objectid) | |
1f3c79a2 | 1498 | { |
87bde3cd | 1499 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 | 1500 | struct btrfs_key key; |
1f3c79a2 | 1501 | int ret; |
1f3c79a2 | 1502 | |
5d4f98a2 YZ |
1503 | key.objectid = bytenr; |
1504 | if (parent) { | |
1505 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1506 | key.offset = parent; | |
1507 | } else { | |
1508 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1509 | key.offset = root_objectid; | |
1f3c79a2 LH |
1510 | } |
1511 | ||
5d4f98a2 YZ |
1512 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1513 | if (ret > 0) | |
1514 | ret = -ENOENT; | |
1515 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1516 | if (ret == -ENOENT && parent) { | |
b3b4aa74 | 1517 | btrfs_release_path(path); |
5d4f98a2 YZ |
1518 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
1519 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1520 | if (ret > 0) | |
1521 | ret = -ENOENT; | |
1522 | } | |
1f3c79a2 | 1523 | #endif |
5d4f98a2 | 1524 | return ret; |
1f3c79a2 LH |
1525 | } |
1526 | ||
5d4f98a2 | 1527 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
87bde3cd | 1528 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1529 | struct btrfs_path *path, |
1530 | u64 bytenr, u64 parent, | |
1531 | u64 root_objectid) | |
31840ae1 | 1532 | { |
5d4f98a2 | 1533 | struct btrfs_key key; |
31840ae1 | 1534 | int ret; |
31840ae1 | 1535 | |
5d4f98a2 YZ |
1536 | key.objectid = bytenr; |
1537 | if (parent) { | |
1538 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1539 | key.offset = parent; | |
1540 | } else { | |
1541 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1542 | key.offset = root_objectid; | |
1543 | } | |
1544 | ||
87bde3cd JM |
1545 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, |
1546 | path, &key, 0); | |
b3b4aa74 | 1547 | btrfs_release_path(path); |
31840ae1 ZY |
1548 | return ret; |
1549 | } | |
1550 | ||
5d4f98a2 | 1551 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1552 | { |
5d4f98a2 YZ |
1553 | int type; |
1554 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1555 | if (parent > 0) | |
1556 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1557 | else | |
1558 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1559 | } else { | |
1560 | if (parent > 0) | |
1561 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1562 | else | |
1563 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1564 | } | |
1565 | return type; | |
31840ae1 | 1566 | } |
56bec294 | 1567 | |
2c47e605 YZ |
1568 | static int find_next_key(struct btrfs_path *path, int level, |
1569 | struct btrfs_key *key) | |
56bec294 | 1570 | |
02217ed2 | 1571 | { |
2c47e605 | 1572 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1573 | if (!path->nodes[level]) |
1574 | break; | |
5d4f98a2 YZ |
1575 | if (path->slots[level] + 1 >= |
1576 | btrfs_header_nritems(path->nodes[level])) | |
1577 | continue; | |
1578 | if (level == 0) | |
1579 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1580 | path->slots[level] + 1); | |
1581 | else | |
1582 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1583 | path->slots[level] + 1); | |
1584 | return 0; | |
1585 | } | |
1586 | return 1; | |
1587 | } | |
037e6390 | 1588 | |
5d4f98a2 YZ |
1589 | /* |
1590 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1591 | * to the address of inline back ref, and 0 is returned. | |
1592 | * | |
1593 | * if back ref isn't found, *ref_ret is set to the address where it | |
1594 | * should be inserted, and -ENOENT is returned. | |
1595 | * | |
1596 | * if insert is true and there are too many inline back refs, the path | |
1597 | * points to the extent item, and -EAGAIN is returned. | |
1598 | * | |
1599 | * NOTE: inline back refs are ordered in the same way that back ref | |
1600 | * items in the tree are ordered. | |
1601 | */ | |
1602 | static noinline_for_stack | |
1603 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1604 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1605 | struct btrfs_path *path, |
1606 | struct btrfs_extent_inline_ref **ref_ret, | |
1607 | u64 bytenr, u64 num_bytes, | |
1608 | u64 parent, u64 root_objectid, | |
1609 | u64 owner, u64 offset, int insert) | |
1610 | { | |
87bde3cd | 1611 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1612 | struct btrfs_key key; |
1613 | struct extent_buffer *leaf; | |
1614 | struct btrfs_extent_item *ei; | |
1615 | struct btrfs_extent_inline_ref *iref; | |
1616 | u64 flags; | |
1617 | u64 item_size; | |
1618 | unsigned long ptr; | |
1619 | unsigned long end; | |
1620 | int extra_size; | |
1621 | int type; | |
1622 | int want; | |
1623 | int ret; | |
1624 | int err = 0; | |
0b246afa | 1625 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 1626 | int needed; |
26b8003f | 1627 | |
db94535d | 1628 | key.objectid = bytenr; |
31840ae1 | 1629 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1630 | key.offset = num_bytes; |
31840ae1 | 1631 | |
5d4f98a2 YZ |
1632 | want = extent_ref_type(parent, owner); |
1633 | if (insert) { | |
1634 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1635 | path->keep_locks = 1; |
5d4f98a2 YZ |
1636 | } else |
1637 | extra_size = -1; | |
3173a18f JB |
1638 | |
1639 | /* | |
1640 | * Owner is our parent level, so we can just add one to get the level | |
1641 | * for the block we are interested in. | |
1642 | */ | |
1643 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1644 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1645 | key.offset = owner; | |
1646 | } | |
1647 | ||
1648 | again: | |
5d4f98a2 | 1649 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1650 | if (ret < 0) { |
5d4f98a2 YZ |
1651 | err = ret; |
1652 | goto out; | |
1653 | } | |
3173a18f JB |
1654 | |
1655 | /* | |
1656 | * We may be a newly converted file system which still has the old fat | |
1657 | * extent entries for metadata, so try and see if we have one of those. | |
1658 | */ | |
1659 | if (ret > 0 && skinny_metadata) { | |
1660 | skinny_metadata = false; | |
1661 | if (path->slots[0]) { | |
1662 | path->slots[0]--; | |
1663 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1664 | path->slots[0]); | |
1665 | if (key.objectid == bytenr && | |
1666 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1667 | key.offset == num_bytes) | |
1668 | ret = 0; | |
1669 | } | |
1670 | if (ret) { | |
9ce49a0b | 1671 | key.objectid = bytenr; |
3173a18f JB |
1672 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1673 | key.offset = num_bytes; | |
1674 | btrfs_release_path(path); | |
1675 | goto again; | |
1676 | } | |
1677 | } | |
1678 | ||
79787eaa JM |
1679 | if (ret && !insert) { |
1680 | err = -ENOENT; | |
1681 | goto out; | |
fae7f21c | 1682 | } else if (WARN_ON(ret)) { |
492104c8 | 1683 | err = -EIO; |
492104c8 | 1684 | goto out; |
79787eaa | 1685 | } |
5d4f98a2 YZ |
1686 | |
1687 | leaf = path->nodes[0]; | |
1688 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1689 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
1690 | if (item_size < sizeof(*ei)) { | |
1691 | if (!insert) { | |
1692 | err = -ENOENT; | |
1693 | goto out; | |
1694 | } | |
87bde3cd | 1695 | ret = convert_extent_item_v0(trans, fs_info, path, owner, |
5d4f98a2 YZ |
1696 | extra_size); |
1697 | if (ret < 0) { | |
1698 | err = ret; | |
1699 | goto out; | |
1700 | } | |
1701 | leaf = path->nodes[0]; | |
1702 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1703 | } | |
1704 | #endif | |
1705 | BUG_ON(item_size < sizeof(*ei)); | |
1706 | ||
5d4f98a2 YZ |
1707 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1708 | flags = btrfs_extent_flags(leaf, ei); | |
1709 | ||
1710 | ptr = (unsigned long)(ei + 1); | |
1711 | end = (unsigned long)ei + item_size; | |
1712 | ||
3173a18f | 1713 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1714 | ptr += sizeof(struct btrfs_tree_block_info); |
1715 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1716 | } |
1717 | ||
3de28d57 LB |
1718 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
1719 | needed = BTRFS_REF_TYPE_DATA; | |
1720 | else | |
1721 | needed = BTRFS_REF_TYPE_BLOCK; | |
1722 | ||
5d4f98a2 YZ |
1723 | err = -ENOENT; |
1724 | while (1) { | |
1725 | if (ptr >= end) { | |
1726 | WARN_ON(ptr > end); | |
1727 | break; | |
1728 | } | |
1729 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
1730 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
1731 | if (type == BTRFS_REF_TYPE_INVALID) { | |
1732 | err = -EINVAL; | |
1733 | goto out; | |
1734 | } | |
1735 | ||
5d4f98a2 YZ |
1736 | if (want < type) |
1737 | break; | |
1738 | if (want > type) { | |
1739 | ptr += btrfs_extent_inline_ref_size(type); | |
1740 | continue; | |
1741 | } | |
1742 | ||
1743 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1744 | struct btrfs_extent_data_ref *dref; | |
1745 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1746 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1747 | owner, offset)) { | |
1748 | err = 0; | |
1749 | break; | |
1750 | } | |
1751 | if (hash_extent_data_ref_item(leaf, dref) < | |
1752 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1753 | break; | |
1754 | } else { | |
1755 | u64 ref_offset; | |
1756 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1757 | if (parent > 0) { | |
1758 | if (parent == ref_offset) { | |
1759 | err = 0; | |
1760 | break; | |
1761 | } | |
1762 | if (ref_offset < parent) | |
1763 | break; | |
1764 | } else { | |
1765 | if (root_objectid == ref_offset) { | |
1766 | err = 0; | |
1767 | break; | |
1768 | } | |
1769 | if (ref_offset < root_objectid) | |
1770 | break; | |
1771 | } | |
1772 | } | |
1773 | ptr += btrfs_extent_inline_ref_size(type); | |
1774 | } | |
1775 | if (err == -ENOENT && insert) { | |
1776 | if (item_size + extra_size >= | |
1777 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1778 | err = -EAGAIN; | |
1779 | goto out; | |
1780 | } | |
1781 | /* | |
1782 | * To add new inline back ref, we have to make sure | |
1783 | * there is no corresponding back ref item. | |
1784 | * For simplicity, we just do not add new inline back | |
1785 | * ref if there is any kind of item for this block | |
1786 | */ | |
2c47e605 YZ |
1787 | if (find_next_key(path, 0, &key) == 0 && |
1788 | key.objectid == bytenr && | |
85d4198e | 1789 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1790 | err = -EAGAIN; |
1791 | goto out; | |
1792 | } | |
1793 | } | |
1794 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1795 | out: | |
85d4198e | 1796 | if (insert) { |
5d4f98a2 YZ |
1797 | path->keep_locks = 0; |
1798 | btrfs_unlock_up_safe(path, 1); | |
1799 | } | |
1800 | return err; | |
1801 | } | |
1802 | ||
1803 | /* | |
1804 | * helper to add new inline back ref | |
1805 | */ | |
1806 | static noinline_for_stack | |
87bde3cd | 1807 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1808 | struct btrfs_path *path, |
1809 | struct btrfs_extent_inline_ref *iref, | |
1810 | u64 parent, u64 root_objectid, | |
1811 | u64 owner, u64 offset, int refs_to_add, | |
1812 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1813 | { |
1814 | struct extent_buffer *leaf; | |
1815 | struct btrfs_extent_item *ei; | |
1816 | unsigned long ptr; | |
1817 | unsigned long end; | |
1818 | unsigned long item_offset; | |
1819 | u64 refs; | |
1820 | int size; | |
1821 | int type; | |
5d4f98a2 YZ |
1822 | |
1823 | leaf = path->nodes[0]; | |
1824 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1825 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1826 | ||
1827 | type = extent_ref_type(parent, owner); | |
1828 | size = btrfs_extent_inline_ref_size(type); | |
1829 | ||
87bde3cd | 1830 | btrfs_extend_item(fs_info, path, size); |
5d4f98a2 YZ |
1831 | |
1832 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1833 | refs = btrfs_extent_refs(leaf, ei); | |
1834 | refs += refs_to_add; | |
1835 | btrfs_set_extent_refs(leaf, ei, refs); | |
1836 | if (extent_op) | |
1837 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1838 | ||
1839 | ptr = (unsigned long)ei + item_offset; | |
1840 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1841 | if (ptr < end - size) | |
1842 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1843 | end - size - ptr); | |
1844 | ||
1845 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1846 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1847 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1848 | struct btrfs_extent_data_ref *dref; | |
1849 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1850 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1851 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1852 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1853 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1854 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1855 | struct btrfs_shared_data_ref *sref; | |
1856 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1857 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1858 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1859 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1860 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1861 | } else { | |
1862 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1863 | } | |
1864 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1865 | } |
1866 | ||
1867 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1868 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1869 | struct btrfs_path *path, |
1870 | struct btrfs_extent_inline_ref **ref_ret, | |
1871 | u64 bytenr, u64 num_bytes, u64 parent, | |
1872 | u64 root_objectid, u64 owner, u64 offset) | |
1873 | { | |
1874 | int ret; | |
1875 | ||
87bde3cd | 1876 | ret = lookup_inline_extent_backref(trans, fs_info, path, ref_ret, |
5d4f98a2 YZ |
1877 | bytenr, num_bytes, parent, |
1878 | root_objectid, owner, offset, 0); | |
1879 | if (ret != -ENOENT) | |
54aa1f4d | 1880 | return ret; |
5d4f98a2 | 1881 | |
b3b4aa74 | 1882 | btrfs_release_path(path); |
5d4f98a2 YZ |
1883 | *ref_ret = NULL; |
1884 | ||
1885 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
87bde3cd JM |
1886 | ret = lookup_tree_block_ref(trans, fs_info, path, bytenr, |
1887 | parent, root_objectid); | |
5d4f98a2 | 1888 | } else { |
87bde3cd JM |
1889 | ret = lookup_extent_data_ref(trans, fs_info, path, bytenr, |
1890 | parent, root_objectid, owner, | |
1891 | offset); | |
b9473439 | 1892 | } |
5d4f98a2 YZ |
1893 | return ret; |
1894 | } | |
31840ae1 | 1895 | |
5d4f98a2 YZ |
1896 | /* |
1897 | * helper to update/remove inline back ref | |
1898 | */ | |
1899 | static noinline_for_stack | |
87bde3cd | 1900 | void update_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1901 | struct btrfs_path *path, |
1902 | struct btrfs_extent_inline_ref *iref, | |
1903 | int refs_to_mod, | |
fcebe456 JB |
1904 | struct btrfs_delayed_extent_op *extent_op, |
1905 | int *last_ref) | |
5d4f98a2 YZ |
1906 | { |
1907 | struct extent_buffer *leaf; | |
1908 | struct btrfs_extent_item *ei; | |
1909 | struct btrfs_extent_data_ref *dref = NULL; | |
1910 | struct btrfs_shared_data_ref *sref = NULL; | |
1911 | unsigned long ptr; | |
1912 | unsigned long end; | |
1913 | u32 item_size; | |
1914 | int size; | |
1915 | int type; | |
5d4f98a2 YZ |
1916 | u64 refs; |
1917 | ||
1918 | leaf = path->nodes[0]; | |
1919 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1920 | refs = btrfs_extent_refs(leaf, ei); | |
1921 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1922 | refs += refs_to_mod; | |
1923 | btrfs_set_extent_refs(leaf, ei, refs); | |
1924 | if (extent_op) | |
1925 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1926 | ||
3de28d57 LB |
1927 | /* |
1928 | * If type is invalid, we should have bailed out after | |
1929 | * lookup_inline_extent_backref(). | |
1930 | */ | |
1931 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
1932 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
5d4f98a2 YZ |
1933 | |
1934 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1935 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1936 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1937 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1938 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1939 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1940 | } else { | |
1941 | refs = 1; | |
1942 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1943 | } |
31840ae1 | 1944 | |
5d4f98a2 YZ |
1945 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1946 | refs += refs_to_mod; | |
1947 | ||
1948 | if (refs > 0) { | |
1949 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1950 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1951 | else | |
1952 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1953 | } else { | |
fcebe456 | 1954 | *last_ref = 1; |
5d4f98a2 YZ |
1955 | size = btrfs_extent_inline_ref_size(type); |
1956 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1957 | ptr = (unsigned long)iref; | |
1958 | end = (unsigned long)ei + item_size; | |
1959 | if (ptr + size < end) | |
1960 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1961 | end - ptr - size); | |
1962 | item_size -= size; | |
87bde3cd | 1963 | btrfs_truncate_item(fs_info, path, item_size, 1); |
5d4f98a2 YZ |
1964 | } |
1965 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1966 | } |
1967 | ||
1968 | static noinline_for_stack | |
1969 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
87bde3cd | 1970 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1971 | struct btrfs_path *path, |
1972 | u64 bytenr, u64 num_bytes, u64 parent, | |
1973 | u64 root_objectid, u64 owner, | |
1974 | u64 offset, int refs_to_add, | |
1975 | struct btrfs_delayed_extent_op *extent_op) | |
1976 | { | |
1977 | struct btrfs_extent_inline_ref *iref; | |
1978 | int ret; | |
1979 | ||
87bde3cd | 1980 | ret = lookup_inline_extent_backref(trans, fs_info, path, &iref, |
5d4f98a2 YZ |
1981 | bytenr, num_bytes, parent, |
1982 | root_objectid, owner, offset, 1); | |
1983 | if (ret == 0) { | |
1984 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
87bde3cd | 1985 | update_inline_extent_backref(fs_info, path, iref, |
fcebe456 | 1986 | refs_to_add, extent_op, NULL); |
5d4f98a2 | 1987 | } else if (ret == -ENOENT) { |
87bde3cd | 1988 | setup_inline_extent_backref(fs_info, path, iref, parent, |
143bede5 JM |
1989 | root_objectid, owner, offset, |
1990 | refs_to_add, extent_op); | |
1991 | ret = 0; | |
771ed689 | 1992 | } |
5d4f98a2 YZ |
1993 | return ret; |
1994 | } | |
31840ae1 | 1995 | |
5d4f98a2 | 1996 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
87bde3cd | 1997 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
1998 | struct btrfs_path *path, |
1999 | u64 bytenr, u64 parent, u64 root_objectid, | |
2000 | u64 owner, u64 offset, int refs_to_add) | |
2001 | { | |
2002 | int ret; | |
2003 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
2004 | BUG_ON(refs_to_add != 1); | |
87bde3cd | 2005 | ret = insert_tree_block_ref(trans, fs_info, path, bytenr, |
5d4f98a2 YZ |
2006 | parent, root_objectid); |
2007 | } else { | |
87bde3cd | 2008 | ret = insert_extent_data_ref(trans, fs_info, path, bytenr, |
5d4f98a2 YZ |
2009 | parent, root_objectid, |
2010 | owner, offset, refs_to_add); | |
2011 | } | |
2012 | return ret; | |
2013 | } | |
56bec294 | 2014 | |
5d4f98a2 | 2015 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
87bde3cd | 2016 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2017 | struct btrfs_path *path, |
2018 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 2019 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 2020 | { |
143bede5 | 2021 | int ret = 0; |
b9473439 | 2022 | |
5d4f98a2 YZ |
2023 | BUG_ON(!is_data && refs_to_drop != 1); |
2024 | if (iref) { | |
87bde3cd | 2025 | update_inline_extent_backref(fs_info, path, iref, |
fcebe456 | 2026 | -refs_to_drop, NULL, last_ref); |
5d4f98a2 | 2027 | } else if (is_data) { |
87bde3cd | 2028 | ret = remove_extent_data_ref(trans, fs_info, path, refs_to_drop, |
fcebe456 | 2029 | last_ref); |
5d4f98a2 | 2030 | } else { |
fcebe456 | 2031 | *last_ref = 1; |
87bde3cd | 2032 | ret = btrfs_del_item(trans, fs_info->extent_root, path); |
5d4f98a2 YZ |
2033 | } |
2034 | return ret; | |
2035 | } | |
2036 | ||
86557861 | 2037 | #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len)) |
d04c6b88 JM |
2038 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
2039 | u64 *discarded_bytes) | |
5d4f98a2 | 2040 | { |
86557861 JM |
2041 | int j, ret = 0; |
2042 | u64 bytes_left, end; | |
4d89d377 | 2043 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 2044 | |
4d89d377 JM |
2045 | if (WARN_ON(start != aligned_start)) { |
2046 | len -= aligned_start - start; | |
2047 | len = round_down(len, 1 << 9); | |
2048 | start = aligned_start; | |
2049 | } | |
d04c6b88 | 2050 | |
4d89d377 | 2051 | *discarded_bytes = 0; |
86557861 JM |
2052 | |
2053 | if (!len) | |
2054 | return 0; | |
2055 | ||
2056 | end = start + len; | |
2057 | bytes_left = len; | |
2058 | ||
2059 | /* Skip any superblocks on this device. */ | |
2060 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
2061 | u64 sb_start = btrfs_sb_offset(j); | |
2062 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
2063 | u64 size = sb_start - start; | |
2064 | ||
2065 | if (!in_range(sb_start, start, bytes_left) && | |
2066 | !in_range(sb_end, start, bytes_left) && | |
2067 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
2068 | continue; | |
2069 | ||
2070 | /* | |
2071 | * Superblock spans beginning of range. Adjust start and | |
2072 | * try again. | |
2073 | */ | |
2074 | if (sb_start <= start) { | |
2075 | start += sb_end - start; | |
2076 | if (start > end) { | |
2077 | bytes_left = 0; | |
2078 | break; | |
2079 | } | |
2080 | bytes_left = end - start; | |
2081 | continue; | |
2082 | } | |
2083 | ||
2084 | if (size) { | |
2085 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
2086 | GFP_NOFS, 0); | |
2087 | if (!ret) | |
2088 | *discarded_bytes += size; | |
2089 | else if (ret != -EOPNOTSUPP) | |
2090 | return ret; | |
2091 | } | |
2092 | ||
2093 | start = sb_end; | |
2094 | if (start > end) { | |
2095 | bytes_left = 0; | |
2096 | break; | |
2097 | } | |
2098 | bytes_left = end - start; | |
2099 | } | |
2100 | ||
2101 | if (bytes_left) { | |
2102 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
2103 | GFP_NOFS, 0); |
2104 | if (!ret) | |
86557861 | 2105 | *discarded_bytes += bytes_left; |
4d89d377 | 2106 | } |
d04c6b88 | 2107 | return ret; |
5d4f98a2 | 2108 | } |
5d4f98a2 | 2109 | |
2ff7e61e | 2110 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 2111 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 2112 | { |
5d4f98a2 | 2113 | int ret; |
5378e607 | 2114 | u64 discarded_bytes = 0; |
a1d3c478 | 2115 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 2116 | |
e244a0ae | 2117 | |
2999241d FM |
2118 | /* |
2119 | * Avoid races with device replace and make sure our bbio has devices | |
2120 | * associated to its stripes that don't go away while we are discarding. | |
2121 | */ | |
0b246afa | 2122 | btrfs_bio_counter_inc_blocked(fs_info); |
5d4f98a2 | 2123 | /* Tell the block device(s) that the sectors can be discarded */ |
0b246afa JM |
2124 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes, |
2125 | &bbio, 0); | |
79787eaa | 2126 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 2127 | if (!ret) { |
a1d3c478 | 2128 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
2129 | int i; |
2130 | ||
5d4f98a2 | 2131 | |
a1d3c478 | 2132 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 2133 | u64 bytes; |
38b5f68e AJ |
2134 | struct request_queue *req_q; |
2135 | ||
627e0873 FM |
2136 | if (!stripe->dev->bdev) { |
2137 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); | |
2138 | continue; | |
2139 | } | |
38b5f68e AJ |
2140 | req_q = bdev_get_queue(stripe->dev->bdev); |
2141 | if (!blk_queue_discard(req_q)) | |
d5e2003c JB |
2142 | continue; |
2143 | ||
5378e607 LD |
2144 | ret = btrfs_issue_discard(stripe->dev->bdev, |
2145 | stripe->physical, | |
d04c6b88 JM |
2146 | stripe->length, |
2147 | &bytes); | |
5378e607 | 2148 | if (!ret) |
d04c6b88 | 2149 | discarded_bytes += bytes; |
5378e607 | 2150 | else if (ret != -EOPNOTSUPP) |
79787eaa | 2151 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
2152 | |
2153 | /* | |
2154 | * Just in case we get back EOPNOTSUPP for some reason, | |
2155 | * just ignore the return value so we don't screw up | |
2156 | * people calling discard_extent. | |
2157 | */ | |
2158 | ret = 0; | |
5d4f98a2 | 2159 | } |
6e9606d2 | 2160 | btrfs_put_bbio(bbio); |
5d4f98a2 | 2161 | } |
0b246afa | 2162 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
2163 | |
2164 | if (actual_bytes) | |
2165 | *actual_bytes = discarded_bytes; | |
2166 | ||
5d4f98a2 | 2167 | |
53b381b3 DW |
2168 | if (ret == -EOPNOTSUPP) |
2169 | ret = 0; | |
5d4f98a2 | 2170 | return ret; |
5d4f98a2 YZ |
2171 | } |
2172 | ||
79787eaa | 2173 | /* Can return -ENOMEM */ |
5d4f98a2 | 2174 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
84f7d8e6 | 2175 | struct btrfs_root *root, |
5d4f98a2 | 2176 | u64 bytenr, u64 num_bytes, u64 parent, |
b06c4bf5 | 2177 | u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 | 2178 | { |
84f7d8e6 | 2179 | struct btrfs_fs_info *fs_info = root->fs_info; |
d7eae340 | 2180 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 2181 | int ret; |
66d7e7f0 | 2182 | |
5d4f98a2 YZ |
2183 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
2184 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
2185 | ||
fd708b81 JB |
2186 | btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid, |
2187 | owner, offset, BTRFS_ADD_DELAYED_REF); | |
2188 | ||
5d4f98a2 | 2189 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 | 2190 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
7be07912 OS |
2191 | num_bytes, parent, |
2192 | root_objectid, (int)owner, | |
2193 | BTRFS_ADD_DELAYED_REF, NULL, | |
d7eae340 | 2194 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 2195 | } else { |
66d7e7f0 | 2196 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
7be07912 OS |
2197 | num_bytes, parent, |
2198 | root_objectid, owner, offset, | |
d7eae340 OS |
2199 | 0, BTRFS_ADD_DELAYED_REF, |
2200 | &old_ref_mod, &new_ref_mod); | |
5d4f98a2 | 2201 | } |
d7eae340 | 2202 | |
29d2b84c NB |
2203 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) { |
2204 | bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID; | |
2205 | ||
2206 | add_pinned_bytes(fs_info, -num_bytes, metadata, root_objectid); | |
2207 | } | |
d7eae340 | 2208 | |
5d4f98a2 YZ |
2209 | return ret; |
2210 | } | |
2211 | ||
2212 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2213 | struct btrfs_fs_info *fs_info, |
c682f9b3 | 2214 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2215 | u64 parent, u64 root_objectid, |
2216 | u64 owner, u64 offset, int refs_to_add, | |
2217 | struct btrfs_delayed_extent_op *extent_op) | |
2218 | { | |
2219 | struct btrfs_path *path; | |
2220 | struct extent_buffer *leaf; | |
2221 | struct btrfs_extent_item *item; | |
fcebe456 | 2222 | struct btrfs_key key; |
c682f9b3 QW |
2223 | u64 bytenr = node->bytenr; |
2224 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2225 | u64 refs; |
2226 | int ret; | |
5d4f98a2 YZ |
2227 | |
2228 | path = btrfs_alloc_path(); | |
2229 | if (!path) | |
2230 | return -ENOMEM; | |
2231 | ||
e4058b54 | 2232 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2233 | path->leave_spinning = 1; |
2234 | /* this will setup the path even if it fails to insert the back ref */ | |
87bde3cd JM |
2235 | ret = insert_inline_extent_backref(trans, fs_info, path, bytenr, |
2236 | num_bytes, parent, root_objectid, | |
2237 | owner, offset, | |
5d4f98a2 | 2238 | refs_to_add, extent_op); |
0ed4792a | 2239 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2240 | goto out; |
fcebe456 JB |
2241 | |
2242 | /* | |
2243 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2244 | * inline extent ref, so just update the reference count and add a | |
2245 | * normal backref. | |
2246 | */ | |
5d4f98a2 | 2247 | leaf = path->nodes[0]; |
fcebe456 | 2248 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2249 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2250 | refs = btrfs_extent_refs(leaf, item); | |
2251 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2252 | if (extent_op) | |
2253 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2254 | |
5d4f98a2 | 2255 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2256 | btrfs_release_path(path); |
56bec294 | 2257 | |
e4058b54 | 2258 | path->reada = READA_FORWARD; |
b9473439 | 2259 | path->leave_spinning = 1; |
56bec294 | 2260 | /* now insert the actual backref */ |
87bde3cd JM |
2261 | ret = insert_extent_backref(trans, fs_info, path, bytenr, parent, |
2262 | root_objectid, owner, offset, refs_to_add); | |
79787eaa | 2263 | if (ret) |
66642832 | 2264 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 2265 | out: |
56bec294 | 2266 | btrfs_free_path(path); |
30d133fc | 2267 | return ret; |
56bec294 CM |
2268 | } |
2269 | ||
5d4f98a2 | 2270 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2271 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2272 | struct btrfs_delayed_ref_node *node, |
2273 | struct btrfs_delayed_extent_op *extent_op, | |
2274 | int insert_reserved) | |
56bec294 | 2275 | { |
5d4f98a2 YZ |
2276 | int ret = 0; |
2277 | struct btrfs_delayed_data_ref *ref; | |
2278 | struct btrfs_key ins; | |
2279 | u64 parent = 0; | |
2280 | u64 ref_root = 0; | |
2281 | u64 flags = 0; | |
2282 | ||
2283 | ins.objectid = node->bytenr; | |
2284 | ins.offset = node->num_bytes; | |
2285 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2286 | ||
2287 | ref = btrfs_delayed_node_to_data_ref(node); | |
0b246afa | 2288 | trace_run_delayed_data_ref(fs_info, node, ref, node->action); |
599c75ec | 2289 | |
5d4f98a2 YZ |
2290 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2291 | parent = ref->parent; | |
fcebe456 | 2292 | ref_root = ref->root; |
5d4f98a2 YZ |
2293 | |
2294 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2295 | if (extent_op) |
5d4f98a2 | 2296 | flags |= extent_op->flags_to_set; |
2ff7e61e | 2297 | ret = alloc_reserved_file_extent(trans, fs_info, |
5d4f98a2 YZ |
2298 | parent, ref_root, flags, |
2299 | ref->objectid, ref->offset, | |
2300 | &ins, node->ref_mod); | |
5d4f98a2 | 2301 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2ff7e61e | 2302 | ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent, |
5d4f98a2 YZ |
2303 | ref_root, ref->objectid, |
2304 | ref->offset, node->ref_mod, | |
c682f9b3 | 2305 | extent_op); |
5d4f98a2 | 2306 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
2ff7e61e | 2307 | ret = __btrfs_free_extent(trans, fs_info, node, parent, |
5d4f98a2 YZ |
2308 | ref_root, ref->objectid, |
2309 | ref->offset, node->ref_mod, | |
c682f9b3 | 2310 | extent_op); |
5d4f98a2 YZ |
2311 | } else { |
2312 | BUG(); | |
2313 | } | |
2314 | return ret; | |
2315 | } | |
2316 | ||
2317 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2318 | struct extent_buffer *leaf, | |
2319 | struct btrfs_extent_item *ei) | |
2320 | { | |
2321 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2322 | if (extent_op->update_flags) { | |
2323 | flags |= extent_op->flags_to_set; | |
2324 | btrfs_set_extent_flags(leaf, ei, flags); | |
2325 | } | |
2326 | ||
2327 | if (extent_op->update_key) { | |
2328 | struct btrfs_tree_block_info *bi; | |
2329 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2330 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2331 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2332 | } | |
2333 | } | |
2334 | ||
2335 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
2ff7e61e | 2336 | struct btrfs_fs_info *fs_info, |
d278850e | 2337 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
2338 | struct btrfs_delayed_extent_op *extent_op) |
2339 | { | |
2340 | struct btrfs_key key; | |
2341 | struct btrfs_path *path; | |
2342 | struct btrfs_extent_item *ei; | |
2343 | struct extent_buffer *leaf; | |
2344 | u32 item_size; | |
56bec294 | 2345 | int ret; |
5d4f98a2 | 2346 | int err = 0; |
b1c79e09 | 2347 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2348 | |
79787eaa JM |
2349 | if (trans->aborted) |
2350 | return 0; | |
2351 | ||
0b246afa | 2352 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
2353 | metadata = 0; |
2354 | ||
5d4f98a2 YZ |
2355 | path = btrfs_alloc_path(); |
2356 | if (!path) | |
2357 | return -ENOMEM; | |
2358 | ||
d278850e | 2359 | key.objectid = head->bytenr; |
5d4f98a2 | 2360 | |
3173a18f | 2361 | if (metadata) { |
3173a18f | 2362 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2363 | key.offset = extent_op->level; |
3173a18f JB |
2364 | } else { |
2365 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 2366 | key.offset = head->num_bytes; |
3173a18f JB |
2367 | } |
2368 | ||
2369 | again: | |
e4058b54 | 2370 | path->reada = READA_FORWARD; |
5d4f98a2 | 2371 | path->leave_spinning = 1; |
0b246afa | 2372 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
2373 | if (ret < 0) { |
2374 | err = ret; | |
2375 | goto out; | |
2376 | } | |
2377 | if (ret > 0) { | |
3173a18f | 2378 | if (metadata) { |
55994887 FDBM |
2379 | if (path->slots[0] > 0) { |
2380 | path->slots[0]--; | |
2381 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2382 | path->slots[0]); | |
d278850e | 2383 | if (key.objectid == head->bytenr && |
55994887 | 2384 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 2385 | key.offset == head->num_bytes) |
55994887 FDBM |
2386 | ret = 0; |
2387 | } | |
2388 | if (ret > 0) { | |
2389 | btrfs_release_path(path); | |
2390 | metadata = 0; | |
3173a18f | 2391 | |
d278850e JB |
2392 | key.objectid = head->bytenr; |
2393 | key.offset = head->num_bytes; | |
55994887 FDBM |
2394 | key.type = BTRFS_EXTENT_ITEM_KEY; |
2395 | goto again; | |
2396 | } | |
2397 | } else { | |
2398 | err = -EIO; | |
2399 | goto out; | |
3173a18f | 2400 | } |
5d4f98a2 YZ |
2401 | } |
2402 | ||
2403 | leaf = path->nodes[0]; | |
2404 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2405 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
2406 | if (item_size < sizeof(*ei)) { | |
87bde3cd | 2407 | ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0); |
5d4f98a2 YZ |
2408 | if (ret < 0) { |
2409 | err = ret; | |
2410 | goto out; | |
2411 | } | |
2412 | leaf = path->nodes[0]; | |
2413 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
2414 | } | |
2415 | #endif | |
2416 | BUG_ON(item_size < sizeof(*ei)); | |
2417 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
2418 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2419 | |
5d4f98a2 YZ |
2420 | btrfs_mark_buffer_dirty(leaf); |
2421 | out: | |
2422 | btrfs_free_path(path); | |
2423 | return err; | |
56bec294 CM |
2424 | } |
2425 | ||
5d4f98a2 | 2426 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2427 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2428 | struct btrfs_delayed_ref_node *node, |
2429 | struct btrfs_delayed_extent_op *extent_op, | |
2430 | int insert_reserved) | |
56bec294 CM |
2431 | { |
2432 | int ret = 0; | |
5d4f98a2 YZ |
2433 | struct btrfs_delayed_tree_ref *ref; |
2434 | struct btrfs_key ins; | |
2435 | u64 parent = 0; | |
2436 | u64 ref_root = 0; | |
0b246afa | 2437 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
56bec294 | 2438 | |
5d4f98a2 | 2439 | ref = btrfs_delayed_node_to_tree_ref(node); |
0b246afa | 2440 | trace_run_delayed_tree_ref(fs_info, node, ref, node->action); |
599c75ec | 2441 | |
5d4f98a2 YZ |
2442 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2443 | parent = ref->parent; | |
fcebe456 | 2444 | ref_root = ref->root; |
5d4f98a2 | 2445 | |
3173a18f JB |
2446 | ins.objectid = node->bytenr; |
2447 | if (skinny_metadata) { | |
2448 | ins.offset = ref->level; | |
2449 | ins.type = BTRFS_METADATA_ITEM_KEY; | |
2450 | } else { | |
2451 | ins.offset = node->num_bytes; | |
2452 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2453 | } | |
2454 | ||
02794222 | 2455 | if (node->ref_mod != 1) { |
2ff7e61e | 2456 | btrfs_err(fs_info, |
02794222 LB |
2457 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
2458 | node->bytenr, node->ref_mod, node->action, ref_root, | |
2459 | parent); | |
2460 | return -EIO; | |
2461 | } | |
5d4f98a2 | 2462 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 2463 | BUG_ON(!extent_op || !extent_op->update_flags); |
2ff7e61e | 2464 | ret = alloc_reserved_tree_block(trans, fs_info, |
5d4f98a2 YZ |
2465 | parent, ref_root, |
2466 | extent_op->flags_to_set, | |
2467 | &extent_op->key, | |
b06c4bf5 | 2468 | ref->level, &ins); |
5d4f98a2 | 2469 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2ff7e61e | 2470 | ret = __btrfs_inc_extent_ref(trans, fs_info, node, |
c682f9b3 QW |
2471 | parent, ref_root, |
2472 | ref->level, 0, 1, | |
fcebe456 | 2473 | extent_op); |
5d4f98a2 | 2474 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
2ff7e61e | 2475 | ret = __btrfs_free_extent(trans, fs_info, node, |
c682f9b3 QW |
2476 | parent, ref_root, |
2477 | ref->level, 0, 1, extent_op); | |
5d4f98a2 YZ |
2478 | } else { |
2479 | BUG(); | |
2480 | } | |
56bec294 CM |
2481 | return ret; |
2482 | } | |
2483 | ||
2484 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 2485 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
2ff7e61e | 2486 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
2487 | struct btrfs_delayed_ref_node *node, |
2488 | struct btrfs_delayed_extent_op *extent_op, | |
2489 | int insert_reserved) | |
56bec294 | 2490 | { |
79787eaa JM |
2491 | int ret = 0; |
2492 | ||
857cc2fc JB |
2493 | if (trans->aborted) { |
2494 | if (insert_reserved) | |
2ff7e61e | 2495 | btrfs_pin_extent(fs_info, node->bytenr, |
857cc2fc | 2496 | node->num_bytes, 1); |
79787eaa | 2497 | return 0; |
857cc2fc | 2498 | } |
79787eaa | 2499 | |
5d4f98a2 YZ |
2500 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2501 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
2ff7e61e | 2502 | ret = run_delayed_tree_ref(trans, fs_info, node, extent_op, |
5d4f98a2 YZ |
2503 | insert_reserved); |
2504 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2505 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2ff7e61e | 2506 | ret = run_delayed_data_ref(trans, fs_info, node, extent_op, |
5d4f98a2 YZ |
2507 | insert_reserved); |
2508 | else | |
2509 | BUG(); | |
2510 | return ret; | |
56bec294 CM |
2511 | } |
2512 | ||
c6fc2454 | 2513 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2514 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2515 | { | |
cffc3374 FM |
2516 | struct btrfs_delayed_ref_node *ref; |
2517 | ||
0e0adbcf | 2518 | if (RB_EMPTY_ROOT(&head->ref_tree)) |
c6fc2454 | 2519 | return NULL; |
d7df2c79 | 2520 | |
cffc3374 FM |
2521 | /* |
2522 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2523 | * This is to prevent a ref count from going down to zero, which deletes | |
2524 | * the extent item from the extent tree, when there still are references | |
2525 | * to add, which would fail because they would not find the extent item. | |
2526 | */ | |
1d57ee94 WX |
2527 | if (!list_empty(&head->ref_add_list)) |
2528 | return list_first_entry(&head->ref_add_list, | |
2529 | struct btrfs_delayed_ref_node, add_list); | |
2530 | ||
0e0adbcf JB |
2531 | ref = rb_entry(rb_first(&head->ref_tree), |
2532 | struct btrfs_delayed_ref_node, ref_node); | |
1d57ee94 WX |
2533 | ASSERT(list_empty(&ref->add_list)); |
2534 | return ref; | |
56bec294 CM |
2535 | } |
2536 | ||
2eadaa22 JB |
2537 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
2538 | struct btrfs_delayed_ref_head *head) | |
2539 | { | |
2540 | spin_lock(&delayed_refs->lock); | |
2541 | head->processing = 0; | |
2542 | delayed_refs->num_heads_ready++; | |
2543 | spin_unlock(&delayed_refs->lock); | |
2544 | btrfs_delayed_ref_unlock(head); | |
2545 | } | |
2546 | ||
b00e6250 JB |
2547 | static int cleanup_extent_op(struct btrfs_trans_handle *trans, |
2548 | struct btrfs_fs_info *fs_info, | |
2549 | struct btrfs_delayed_ref_head *head) | |
2550 | { | |
2551 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
2552 | int ret; | |
2553 | ||
2554 | if (!extent_op) | |
2555 | return 0; | |
2556 | head->extent_op = NULL; | |
2557 | if (head->must_insert_reserved) { | |
2558 | btrfs_free_delayed_extent_op(extent_op); | |
2559 | return 0; | |
2560 | } | |
2561 | spin_unlock(&head->lock); | |
d278850e | 2562 | ret = run_delayed_extent_op(trans, fs_info, head, extent_op); |
b00e6250 JB |
2563 | btrfs_free_delayed_extent_op(extent_op); |
2564 | return ret ? ret : 1; | |
2565 | } | |
2566 | ||
194ab0bc JB |
2567 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
2568 | struct btrfs_fs_info *fs_info, | |
2569 | struct btrfs_delayed_ref_head *head) | |
2570 | { | |
2571 | struct btrfs_delayed_ref_root *delayed_refs; | |
2572 | int ret; | |
2573 | ||
2574 | delayed_refs = &trans->transaction->delayed_refs; | |
2575 | ||
2576 | ret = cleanup_extent_op(trans, fs_info, head); | |
2577 | if (ret < 0) { | |
2578 | unselect_delayed_ref_head(delayed_refs, head); | |
2579 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
2580 | return ret; | |
2581 | } else if (ret) { | |
2582 | return ret; | |
2583 | } | |
2584 | ||
2585 | /* | |
2586 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
2587 | * and then re-check to make sure nobody got added. | |
2588 | */ | |
2589 | spin_unlock(&head->lock); | |
2590 | spin_lock(&delayed_refs->lock); | |
2591 | spin_lock(&head->lock); | |
0e0adbcf | 2592 | if (!RB_EMPTY_ROOT(&head->ref_tree) || head->extent_op) { |
194ab0bc JB |
2593 | spin_unlock(&head->lock); |
2594 | spin_unlock(&delayed_refs->lock); | |
2595 | return 1; | |
2596 | } | |
194ab0bc JB |
2597 | delayed_refs->num_heads--; |
2598 | rb_erase(&head->href_node, &delayed_refs->href_root); | |
d278850e | 2599 | RB_CLEAR_NODE(&head->href_node); |
c1103f7a | 2600 | spin_unlock(&head->lock); |
1e7a1421 | 2601 | spin_unlock(&delayed_refs->lock); |
c1103f7a JB |
2602 | atomic_dec(&delayed_refs->num_entries); |
2603 | ||
d278850e | 2604 | trace_run_delayed_ref_head(fs_info, head, 0); |
c1103f7a JB |
2605 | |
2606 | if (head->total_ref_mod < 0) { | |
5e388e95 NB |
2607 | struct btrfs_space_info *space_info; |
2608 | u64 flags; | |
c1103f7a | 2609 | |
5e388e95 NB |
2610 | if (head->is_data) |
2611 | flags = BTRFS_BLOCK_GROUP_DATA; | |
2612 | else if (head->is_system) | |
2613 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
2614 | else | |
2615 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
2616 | space_info = __find_space_info(fs_info, flags); | |
2617 | ASSERT(space_info); | |
2618 | percpu_counter_add(&space_info->total_bytes_pinned, | |
d278850e | 2619 | -head->num_bytes); |
c1103f7a JB |
2620 | |
2621 | if (head->is_data) { | |
2622 | spin_lock(&delayed_refs->lock); | |
d278850e | 2623 | delayed_refs->pending_csums -= head->num_bytes; |
c1103f7a JB |
2624 | spin_unlock(&delayed_refs->lock); |
2625 | } | |
2626 | } | |
2627 | ||
2628 | if (head->must_insert_reserved) { | |
d278850e JB |
2629 | btrfs_pin_extent(fs_info, head->bytenr, |
2630 | head->num_bytes, 1); | |
c1103f7a | 2631 | if (head->is_data) { |
d278850e JB |
2632 | ret = btrfs_del_csums(trans, fs_info, head->bytenr, |
2633 | head->num_bytes); | |
c1103f7a JB |
2634 | } |
2635 | } | |
2636 | ||
2637 | /* Also free its reserved qgroup space */ | |
2638 | btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root, | |
2639 | head->qgroup_reserved); | |
2640 | btrfs_delayed_ref_unlock(head); | |
d278850e | 2641 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
2642 | return 0; |
2643 | } | |
2644 | ||
79787eaa JM |
2645 | /* |
2646 | * Returns 0 on success or if called with an already aborted transaction. | |
2647 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2648 | */ | |
d7df2c79 | 2649 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
d7df2c79 | 2650 | unsigned long nr) |
56bec294 | 2651 | { |
0a1e458a | 2652 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 CM |
2653 | struct btrfs_delayed_ref_root *delayed_refs; |
2654 | struct btrfs_delayed_ref_node *ref; | |
2655 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2656 | struct btrfs_delayed_extent_op *extent_op; |
0a2b2a84 | 2657 | ktime_t start = ktime_get(); |
56bec294 | 2658 | int ret; |
d7df2c79 | 2659 | unsigned long count = 0; |
0a2b2a84 | 2660 | unsigned long actual_count = 0; |
56bec294 | 2661 | int must_insert_reserved = 0; |
56bec294 CM |
2662 | |
2663 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2664 | while (1) { |
2665 | if (!locked_ref) { | |
d7df2c79 | 2666 | if (count >= nr) |
56bec294 | 2667 | break; |
56bec294 | 2668 | |
d7df2c79 JB |
2669 | spin_lock(&delayed_refs->lock); |
2670 | locked_ref = btrfs_select_ref_head(trans); | |
2671 | if (!locked_ref) { | |
2672 | spin_unlock(&delayed_refs->lock); | |
2673 | break; | |
2674 | } | |
c3e69d58 CM |
2675 | |
2676 | /* grab the lock that says we are going to process | |
2677 | * all the refs for this head */ | |
2678 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
d7df2c79 | 2679 | spin_unlock(&delayed_refs->lock); |
c3e69d58 CM |
2680 | /* |
2681 | * we may have dropped the spin lock to get the head | |
2682 | * mutex lock, and that might have given someone else | |
2683 | * time to free the head. If that's true, it has been | |
2684 | * removed from our list and we can move on. | |
2685 | */ | |
2686 | if (ret == -EAGAIN) { | |
2687 | locked_ref = NULL; | |
2688 | count++; | |
2689 | continue; | |
56bec294 CM |
2690 | } |
2691 | } | |
a28ec197 | 2692 | |
2c3cf7d5 FM |
2693 | /* |
2694 | * We need to try and merge add/drops of the same ref since we | |
2695 | * can run into issues with relocate dropping the implicit ref | |
2696 | * and then it being added back again before the drop can | |
2697 | * finish. If we merged anything we need to re-loop so we can | |
2698 | * get a good ref. | |
2699 | * Or we can get node references of the same type that weren't | |
2700 | * merged when created due to bumps in the tree mod seq, and | |
2701 | * we need to merge them to prevent adding an inline extent | |
2702 | * backref before dropping it (triggering a BUG_ON at | |
2703 | * insert_inline_extent_backref()). | |
2704 | */ | |
d7df2c79 | 2705 | spin_lock(&locked_ref->lock); |
2c3cf7d5 FM |
2706 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, |
2707 | locked_ref); | |
ae1e206b | 2708 | |
d1270cd9 AJ |
2709 | /* |
2710 | * locked_ref is the head node, so we have to go one | |
2711 | * node back for any delayed ref updates | |
2712 | */ | |
2713 | ref = select_delayed_ref(locked_ref); | |
2714 | ||
2715 | if (ref && ref->seq && | |
41d0bd3b | 2716 | btrfs_check_delayed_seq(fs_info, ref->seq)) { |
d7df2c79 | 2717 | spin_unlock(&locked_ref->lock); |
2eadaa22 | 2718 | unselect_delayed_ref_head(delayed_refs, locked_ref); |
d7df2c79 | 2719 | locked_ref = NULL; |
d1270cd9 | 2720 | cond_resched(); |
27a377db | 2721 | count++; |
d1270cd9 AJ |
2722 | continue; |
2723 | } | |
2724 | ||
c1103f7a JB |
2725 | /* |
2726 | * We're done processing refs in this ref_head, clean everything | |
2727 | * up and move on to the next ref_head. | |
2728 | */ | |
56bec294 | 2729 | if (!ref) { |
194ab0bc JB |
2730 | ret = cleanup_ref_head(trans, fs_info, locked_ref); |
2731 | if (ret > 0 ) { | |
b00e6250 JB |
2732 | /* We dropped our lock, we need to loop. */ |
2733 | ret = 0; | |
d7df2c79 | 2734 | continue; |
194ab0bc JB |
2735 | } else if (ret) { |
2736 | return ret; | |
5d4f98a2 | 2737 | } |
c1103f7a JB |
2738 | locked_ref = NULL; |
2739 | count++; | |
2740 | continue; | |
2741 | } | |
02217ed2 | 2742 | |
c1103f7a JB |
2743 | actual_count++; |
2744 | ref->in_tree = 0; | |
0e0adbcf JB |
2745 | rb_erase(&ref->ref_node, &locked_ref->ref_tree); |
2746 | RB_CLEAR_NODE(&ref->ref_node); | |
c1103f7a JB |
2747 | if (!list_empty(&ref->add_list)) |
2748 | list_del(&ref->add_list); | |
2749 | /* | |
2750 | * When we play the delayed ref, also correct the ref_mod on | |
2751 | * head | |
2752 | */ | |
2753 | switch (ref->action) { | |
2754 | case BTRFS_ADD_DELAYED_REF: | |
2755 | case BTRFS_ADD_DELAYED_EXTENT: | |
d278850e | 2756 | locked_ref->ref_mod -= ref->ref_mod; |
c1103f7a JB |
2757 | break; |
2758 | case BTRFS_DROP_DELAYED_REF: | |
d278850e | 2759 | locked_ref->ref_mod += ref->ref_mod; |
c1103f7a JB |
2760 | break; |
2761 | default: | |
2762 | WARN_ON(1); | |
22cd2e7d | 2763 | } |
1ce7a5ec JB |
2764 | atomic_dec(&delayed_refs->num_entries); |
2765 | ||
b00e6250 JB |
2766 | /* |
2767 | * Record the must-insert_reserved flag before we drop the spin | |
2768 | * lock. | |
2769 | */ | |
2770 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2771 | locked_ref->must_insert_reserved = 0; | |
2772 | ||
2773 | extent_op = locked_ref->extent_op; | |
2774 | locked_ref->extent_op = NULL; | |
d7df2c79 | 2775 | spin_unlock(&locked_ref->lock); |
925baedd | 2776 | |
2ff7e61e | 2777 | ret = run_one_delayed_ref(trans, fs_info, ref, extent_op, |
56bec294 | 2778 | must_insert_reserved); |
eb099670 | 2779 | |
78a6184a | 2780 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2781 | if (ret) { |
2eadaa22 | 2782 | unselect_delayed_ref_head(delayed_refs, locked_ref); |
093486c4 | 2783 | btrfs_put_delayed_ref(ref); |
5d163e0e JM |
2784 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", |
2785 | ret); | |
79787eaa JM |
2786 | return ret; |
2787 | } | |
2788 | ||
093486c4 MX |
2789 | btrfs_put_delayed_ref(ref); |
2790 | count++; | |
c3e69d58 | 2791 | cond_resched(); |
c3e69d58 | 2792 | } |
0a2b2a84 JB |
2793 | |
2794 | /* | |
2795 | * We don't want to include ref heads since we can have empty ref heads | |
2796 | * and those will drastically skew our runtime down since we just do | |
2797 | * accounting, no actual extent tree updates. | |
2798 | */ | |
2799 | if (actual_count > 0) { | |
2800 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2801 | u64 avg; | |
2802 | ||
2803 | /* | |
2804 | * We weigh the current average higher than our current runtime | |
2805 | * to avoid large swings in the average. | |
2806 | */ | |
2807 | spin_lock(&delayed_refs->lock); | |
2808 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2809 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2810 | spin_unlock(&delayed_refs->lock); |
2811 | } | |
d7df2c79 | 2812 | return 0; |
c3e69d58 CM |
2813 | } |
2814 | ||
709c0486 AJ |
2815 | #ifdef SCRAMBLE_DELAYED_REFS |
2816 | /* | |
2817 | * Normally delayed refs get processed in ascending bytenr order. This | |
2818 | * correlates in most cases to the order added. To expose dependencies on this | |
2819 | * order, we start to process the tree in the middle instead of the beginning | |
2820 | */ | |
2821 | static u64 find_middle(struct rb_root *root) | |
2822 | { | |
2823 | struct rb_node *n = root->rb_node; | |
2824 | struct btrfs_delayed_ref_node *entry; | |
2825 | int alt = 1; | |
2826 | u64 middle; | |
2827 | u64 first = 0, last = 0; | |
2828 | ||
2829 | n = rb_first(root); | |
2830 | if (n) { | |
2831 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2832 | first = entry->bytenr; | |
2833 | } | |
2834 | n = rb_last(root); | |
2835 | if (n) { | |
2836 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2837 | last = entry->bytenr; | |
2838 | } | |
2839 | n = root->rb_node; | |
2840 | ||
2841 | while (n) { | |
2842 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2843 | WARN_ON(!entry->in_tree); | |
2844 | ||
2845 | middle = entry->bytenr; | |
2846 | ||
2847 | if (alt) | |
2848 | n = n->rb_left; | |
2849 | else | |
2850 | n = n->rb_right; | |
2851 | ||
2852 | alt = 1 - alt; | |
2853 | } | |
2854 | return middle; | |
2855 | } | |
2856 | #endif | |
2857 | ||
2ff7e61e | 2858 | static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads) |
1be41b78 JB |
2859 | { |
2860 | u64 num_bytes; | |
2861 | ||
2862 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2863 | sizeof(struct btrfs_extent_inline_ref)); | |
0b246afa | 2864 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1be41b78 JB |
2865 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
2866 | ||
2867 | /* | |
2868 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2869 | * closer to what we're really going to want to use. |
1be41b78 | 2870 | */ |
0b246afa | 2871 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info)); |
1be41b78 JB |
2872 | } |
2873 | ||
1262133b JB |
2874 | /* |
2875 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2876 | * would require to store the csums for that many bytes. | |
2877 | */ | |
2ff7e61e | 2878 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2879 | { |
2880 | u64 csum_size; | |
2881 | u64 num_csums_per_leaf; | |
2882 | u64 num_csums; | |
2883 | ||
0b246afa | 2884 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2885 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2886 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2887 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2888 | num_csums += num_csums_per_leaf - 1; |
2889 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2890 | return num_csums; | |
2891 | } | |
2892 | ||
0a2b2a84 | 2893 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2894 | struct btrfs_fs_info *fs_info) |
1be41b78 JB |
2895 | { |
2896 | struct btrfs_block_rsv *global_rsv; | |
2897 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
1262133b | 2898 | u64 csum_bytes = trans->transaction->delayed_refs.pending_csums; |
165c8b02 | 2899 | unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs; |
cb723e49 | 2900 | u64 num_bytes, num_dirty_bgs_bytes; |
1be41b78 JB |
2901 | int ret = 0; |
2902 | ||
0b246afa | 2903 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
2ff7e61e | 2904 | num_heads = heads_to_leaves(fs_info, num_heads); |
1be41b78 | 2905 | if (num_heads > 1) |
0b246afa | 2906 | num_bytes += (num_heads - 1) * fs_info->nodesize; |
1be41b78 | 2907 | num_bytes <<= 1; |
2ff7e61e JM |
2908 | num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) * |
2909 | fs_info->nodesize; | |
0b246afa | 2910 | num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info, |
cb723e49 | 2911 | num_dirty_bgs); |
0b246afa | 2912 | global_rsv = &fs_info->global_block_rsv; |
1be41b78 JB |
2913 | |
2914 | /* | |
2915 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2916 | * wiggle room since running delayed refs can create more delayed refs. | |
2917 | */ | |
cb723e49 JB |
2918 | if (global_rsv->space_info->full) { |
2919 | num_dirty_bgs_bytes <<= 1; | |
1be41b78 | 2920 | num_bytes <<= 1; |
cb723e49 | 2921 | } |
1be41b78 JB |
2922 | |
2923 | spin_lock(&global_rsv->lock); | |
cb723e49 | 2924 | if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes) |
1be41b78 JB |
2925 | ret = 1; |
2926 | spin_unlock(&global_rsv->lock); | |
2927 | return ret; | |
2928 | } | |
2929 | ||
0a2b2a84 | 2930 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2931 | struct btrfs_fs_info *fs_info) |
0a2b2a84 | 2932 | { |
0a2b2a84 JB |
2933 | u64 num_entries = |
2934 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2935 | u64 avg_runtime; | |
a79b7d4b | 2936 | u64 val; |
0a2b2a84 JB |
2937 | |
2938 | smp_mb(); | |
2939 | avg_runtime = fs_info->avg_delayed_ref_runtime; | |
a79b7d4b | 2940 | val = num_entries * avg_runtime; |
dc1a90c6 | 2941 | if (val >= NSEC_PER_SEC) |
0a2b2a84 | 2942 | return 1; |
a79b7d4b CM |
2943 | if (val >= NSEC_PER_SEC / 2) |
2944 | return 2; | |
0a2b2a84 | 2945 | |
2ff7e61e | 2946 | return btrfs_check_space_for_delayed_refs(trans, fs_info); |
0a2b2a84 JB |
2947 | } |
2948 | ||
a79b7d4b CM |
2949 | struct async_delayed_refs { |
2950 | struct btrfs_root *root; | |
31b9655f | 2951 | u64 transid; |
a79b7d4b CM |
2952 | int count; |
2953 | int error; | |
2954 | int sync; | |
2955 | struct completion wait; | |
2956 | struct btrfs_work work; | |
2957 | }; | |
2958 | ||
2ff7e61e JM |
2959 | static inline struct async_delayed_refs * |
2960 | to_async_delayed_refs(struct btrfs_work *work) | |
2961 | { | |
2962 | return container_of(work, struct async_delayed_refs, work); | |
2963 | } | |
2964 | ||
a79b7d4b CM |
2965 | static void delayed_ref_async_start(struct btrfs_work *work) |
2966 | { | |
2ff7e61e | 2967 | struct async_delayed_refs *async = to_async_delayed_refs(work); |
a79b7d4b | 2968 | struct btrfs_trans_handle *trans; |
2ff7e61e | 2969 | struct btrfs_fs_info *fs_info = async->root->fs_info; |
a79b7d4b CM |
2970 | int ret; |
2971 | ||
0f873eca | 2972 | /* if the commit is already started, we don't need to wait here */ |
2ff7e61e | 2973 | if (btrfs_transaction_blocked(fs_info)) |
31b9655f | 2974 | goto done; |
31b9655f | 2975 | |
0f873eca CM |
2976 | trans = btrfs_join_transaction(async->root); |
2977 | if (IS_ERR(trans)) { | |
2978 | async->error = PTR_ERR(trans); | |
a79b7d4b CM |
2979 | goto done; |
2980 | } | |
2981 | ||
2982 | /* | |
01327610 | 2983 | * trans->sync means that when we call end_transaction, we won't |
a79b7d4b CM |
2984 | * wait on delayed refs |
2985 | */ | |
2986 | trans->sync = true; | |
0f873eca CM |
2987 | |
2988 | /* Don't bother flushing if we got into a different transaction */ | |
2989 | if (trans->transid > async->transid) | |
2990 | goto end; | |
2991 | ||
c79a70b1 | 2992 | ret = btrfs_run_delayed_refs(trans, async->count); |
a79b7d4b CM |
2993 | if (ret) |
2994 | async->error = ret; | |
0f873eca | 2995 | end: |
3a45bb20 | 2996 | ret = btrfs_end_transaction(trans); |
a79b7d4b CM |
2997 | if (ret && !async->error) |
2998 | async->error = ret; | |
2999 | done: | |
3000 | if (async->sync) | |
3001 | complete(&async->wait); | |
3002 | else | |
3003 | kfree(async); | |
3004 | } | |
3005 | ||
2ff7e61e | 3006 | int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info, |
31b9655f | 3007 | unsigned long count, u64 transid, int wait) |
a79b7d4b CM |
3008 | { |
3009 | struct async_delayed_refs *async; | |
3010 | int ret; | |
3011 | ||
3012 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
3013 | if (!async) | |
3014 | return -ENOMEM; | |
3015 | ||
0b246afa | 3016 | async->root = fs_info->tree_root; |
a79b7d4b CM |
3017 | async->count = count; |
3018 | async->error = 0; | |
31b9655f | 3019 | async->transid = transid; |
a79b7d4b CM |
3020 | if (wait) |
3021 | async->sync = 1; | |
3022 | else | |
3023 | async->sync = 0; | |
3024 | init_completion(&async->wait); | |
3025 | ||
9e0af237 LB |
3026 | btrfs_init_work(&async->work, btrfs_extent_refs_helper, |
3027 | delayed_ref_async_start, NULL, NULL); | |
a79b7d4b | 3028 | |
0b246afa | 3029 | btrfs_queue_work(fs_info->extent_workers, &async->work); |
a79b7d4b CM |
3030 | |
3031 | if (wait) { | |
3032 | wait_for_completion(&async->wait); | |
3033 | ret = async->error; | |
3034 | kfree(async); | |
3035 | return ret; | |
3036 | } | |
3037 | return 0; | |
3038 | } | |
3039 | ||
c3e69d58 CM |
3040 | /* |
3041 | * this starts processing the delayed reference count updates and | |
3042 | * extent insertions we have queued up so far. count can be | |
3043 | * 0, which means to process everything in the tree at the start | |
3044 | * of the run (but not newly added entries), or it can be some target | |
3045 | * number you'd like to process. | |
79787eaa JM |
3046 | * |
3047 | * Returns 0 on success or if called with an aborted transaction | |
3048 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
3049 | */ |
3050 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
c79a70b1 | 3051 | unsigned long count) |
c3e69d58 | 3052 | { |
c79a70b1 | 3053 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 CM |
3054 | struct rb_node *node; |
3055 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 3056 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
3057 | int ret; |
3058 | int run_all = count == (unsigned long)-1; | |
d9a0540a | 3059 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
c3e69d58 | 3060 | |
79787eaa JM |
3061 | /* We'll clean this up in btrfs_cleanup_transaction */ |
3062 | if (trans->aborted) | |
3063 | return 0; | |
3064 | ||
0b246afa | 3065 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
3066 | return 0; |
3067 | ||
c3e69d58 | 3068 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 3069 | if (count == 0) |
d7df2c79 | 3070 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 3071 | |
c3e69d58 | 3072 | again: |
709c0486 AJ |
3073 | #ifdef SCRAMBLE_DELAYED_REFS |
3074 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
3075 | #endif | |
d9a0540a | 3076 | trans->can_flush_pending_bgs = false; |
0a1e458a | 3077 | ret = __btrfs_run_delayed_refs(trans, count); |
d7df2c79 | 3078 | if (ret < 0) { |
66642832 | 3079 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 3080 | return ret; |
eb099670 | 3081 | } |
c3e69d58 | 3082 | |
56bec294 | 3083 | if (run_all) { |
d7df2c79 | 3084 | if (!list_empty(&trans->new_bgs)) |
6c686b35 | 3085 | btrfs_create_pending_block_groups(trans); |
ea658bad | 3086 | |
d7df2c79 | 3087 | spin_lock(&delayed_refs->lock); |
c46effa6 | 3088 | node = rb_first(&delayed_refs->href_root); |
d7df2c79 JB |
3089 | if (!node) { |
3090 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 3091 | goto out; |
d7df2c79 | 3092 | } |
d278850e JB |
3093 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
3094 | href_node); | |
3095 | refcount_inc(&head->refs); | |
3096 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 3097 | |
d278850e JB |
3098 | /* Mutex was contended, block until it's released and retry. */ |
3099 | mutex_lock(&head->mutex); | |
3100 | mutex_unlock(&head->mutex); | |
56bec294 | 3101 | |
d278850e | 3102 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 3103 | cond_resched(); |
56bec294 | 3104 | goto again; |
5f39d397 | 3105 | } |
54aa1f4d | 3106 | out: |
d9a0540a | 3107 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
a28ec197 CM |
3108 | return 0; |
3109 | } | |
3110 | ||
5d4f98a2 | 3111 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2ff7e61e | 3112 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 3113 | u64 bytenr, u64 num_bytes, u64 flags, |
b1c79e09 | 3114 | int level, int is_data) |
5d4f98a2 YZ |
3115 | { |
3116 | struct btrfs_delayed_extent_op *extent_op; | |
3117 | int ret; | |
3118 | ||
78a6184a | 3119 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
3120 | if (!extent_op) |
3121 | return -ENOMEM; | |
3122 | ||
3123 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
3124 | extent_op->update_flags = true; |
3125 | extent_op->update_key = false; | |
3126 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 3127 | extent_op->level = level; |
5d4f98a2 | 3128 | |
0b246afa | 3129 | ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr, |
66d7e7f0 | 3130 | num_bytes, extent_op); |
5d4f98a2 | 3131 | if (ret) |
78a6184a | 3132 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
3133 | return ret; |
3134 | } | |
3135 | ||
e4c3b2dc | 3136 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3137 | struct btrfs_path *path, |
3138 | u64 objectid, u64 offset, u64 bytenr) | |
3139 | { | |
3140 | struct btrfs_delayed_ref_head *head; | |
3141 | struct btrfs_delayed_ref_node *ref; | |
3142 | struct btrfs_delayed_data_ref *data_ref; | |
3143 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 3144 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 3145 | struct rb_node *node; |
5d4f98a2 YZ |
3146 | int ret = 0; |
3147 | ||
998ac6d2 | 3148 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 3149 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 3150 | if (cur_trans) |
3151 | refcount_inc(&cur_trans->use_count); | |
3152 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
3153 | if (!cur_trans) |
3154 | return 0; | |
3155 | ||
3156 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 3157 | spin_lock(&delayed_refs->lock); |
f72ad18e | 3158 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
3159 | if (!head) { |
3160 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 3161 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
3162 | return 0; |
3163 | } | |
5d4f98a2 YZ |
3164 | |
3165 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 3166 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
3167 | spin_unlock(&delayed_refs->lock); |
3168 | ||
b3b4aa74 | 3169 | btrfs_release_path(path); |
5d4f98a2 | 3170 | |
8cc33e5c DS |
3171 | /* |
3172 | * Mutex was contended, block until it's released and let | |
3173 | * caller try again | |
3174 | */ | |
5d4f98a2 YZ |
3175 | mutex_lock(&head->mutex); |
3176 | mutex_unlock(&head->mutex); | |
d278850e | 3177 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 3178 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
3179 | return -EAGAIN; |
3180 | } | |
d7df2c79 | 3181 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 3182 | |
d7df2c79 | 3183 | spin_lock(&head->lock); |
0e0adbcf JB |
3184 | /* |
3185 | * XXX: We should replace this with a proper search function in the | |
3186 | * future. | |
3187 | */ | |
3188 | for (node = rb_first(&head->ref_tree); node; node = rb_next(node)) { | |
3189 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); | |
d7df2c79 JB |
3190 | /* If it's a shared ref we know a cross reference exists */ |
3191 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3192 | ret = 1; | |
3193 | break; | |
3194 | } | |
5d4f98a2 | 3195 | |
d7df2c79 | 3196 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 3197 | |
d7df2c79 JB |
3198 | /* |
3199 | * If our ref doesn't match the one we're currently looking at | |
3200 | * then we have a cross reference. | |
3201 | */ | |
3202 | if (data_ref->root != root->root_key.objectid || | |
3203 | data_ref->objectid != objectid || | |
3204 | data_ref->offset != offset) { | |
3205 | ret = 1; | |
3206 | break; | |
3207 | } | |
5d4f98a2 | 3208 | } |
d7df2c79 | 3209 | spin_unlock(&head->lock); |
5d4f98a2 | 3210 | mutex_unlock(&head->mutex); |
998ac6d2 | 3211 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
3212 | return ret; |
3213 | } | |
3214 | ||
e4c3b2dc | 3215 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3216 | struct btrfs_path *path, |
3217 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d | 3218 | { |
0b246afa JM |
3219 | struct btrfs_fs_info *fs_info = root->fs_info; |
3220 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 3221 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3222 | struct btrfs_extent_data_ref *ref; |
3223 | struct btrfs_extent_inline_ref *iref; | |
3224 | struct btrfs_extent_item *ei; | |
f321e491 | 3225 | struct btrfs_key key; |
5d4f98a2 | 3226 | u32 item_size; |
3de28d57 | 3227 | int type; |
be20aa9d | 3228 | int ret; |
925baedd | 3229 | |
be20aa9d | 3230 | key.objectid = bytenr; |
31840ae1 | 3231 | key.offset = (u64)-1; |
f321e491 | 3232 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 3233 | |
be20aa9d CM |
3234 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
3235 | if (ret < 0) | |
3236 | goto out; | |
79787eaa | 3237 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
3238 | |
3239 | ret = -ENOENT; | |
3240 | if (path->slots[0] == 0) | |
31840ae1 | 3241 | goto out; |
be20aa9d | 3242 | |
31840ae1 | 3243 | path->slots[0]--; |
f321e491 | 3244 | leaf = path->nodes[0]; |
5d4f98a2 | 3245 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3246 | |
5d4f98a2 | 3247 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3248 | goto out; |
f321e491 | 3249 | |
5d4f98a2 YZ |
3250 | ret = 1; |
3251 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
3252 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
3253 | if (item_size < sizeof(*ei)) { | |
3254 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); | |
3255 | goto out; | |
3256 | } | |
3257 | #endif | |
3258 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
bd09835d | 3259 | |
5d4f98a2 YZ |
3260 | if (item_size != sizeof(*ei) + |
3261 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3262 | goto out; | |
be20aa9d | 3263 | |
5d4f98a2 YZ |
3264 | if (btrfs_extent_generation(leaf, ei) <= |
3265 | btrfs_root_last_snapshot(&root->root_item)) | |
3266 | goto out; | |
3267 | ||
3268 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 LB |
3269 | |
3270 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
3271 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
3272 | goto out; |
3273 | ||
3274 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3275 | if (btrfs_extent_refs(leaf, ei) != | |
3276 | btrfs_extent_data_ref_count(leaf, ref) || | |
3277 | btrfs_extent_data_ref_root(leaf, ref) != | |
3278 | root->root_key.objectid || | |
3279 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3280 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3281 | goto out; | |
3282 | ||
3283 | ret = 0; | |
3284 | out: | |
3285 | return ret; | |
3286 | } | |
3287 | ||
e4c3b2dc LB |
3288 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
3289 | u64 bytenr) | |
5d4f98a2 YZ |
3290 | { |
3291 | struct btrfs_path *path; | |
3292 | int ret; | |
3293 | int ret2; | |
3294 | ||
3295 | path = btrfs_alloc_path(); | |
3296 | if (!path) | |
3297 | return -ENOENT; | |
3298 | ||
3299 | do { | |
e4c3b2dc | 3300 | ret = check_committed_ref(root, path, objectid, |
5d4f98a2 YZ |
3301 | offset, bytenr); |
3302 | if (ret && ret != -ENOENT) | |
f321e491 | 3303 | goto out; |
80ff3856 | 3304 | |
e4c3b2dc | 3305 | ret2 = check_delayed_ref(root, path, objectid, |
5d4f98a2 YZ |
3306 | offset, bytenr); |
3307 | } while (ret2 == -EAGAIN); | |
3308 | ||
3309 | if (ret2 && ret2 != -ENOENT) { | |
3310 | ret = ret2; | |
3311 | goto out; | |
f321e491 | 3312 | } |
5d4f98a2 YZ |
3313 | |
3314 | if (ret != -ENOENT || ret2 != -ENOENT) | |
3315 | ret = 0; | |
be20aa9d | 3316 | out: |
80ff3856 | 3317 | btrfs_free_path(path); |
f0486c68 YZ |
3318 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3319 | WARN_ON(ret > 0); | |
f321e491 | 3320 | return ret; |
be20aa9d | 3321 | } |
c5739bba | 3322 | |
5d4f98a2 | 3323 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3324 | struct btrfs_root *root, |
5d4f98a2 | 3325 | struct extent_buffer *buf, |
e339a6b0 | 3326 | int full_backref, int inc) |
31840ae1 | 3327 | { |
0b246afa | 3328 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 3329 | u64 bytenr; |
5d4f98a2 YZ |
3330 | u64 num_bytes; |
3331 | u64 parent; | |
31840ae1 | 3332 | u64 ref_root; |
31840ae1 | 3333 | u32 nritems; |
31840ae1 ZY |
3334 | struct btrfs_key key; |
3335 | struct btrfs_file_extent_item *fi; | |
3336 | int i; | |
3337 | int level; | |
3338 | int ret = 0; | |
2ff7e61e | 3339 | int (*process_func)(struct btrfs_trans_handle *, |
84f7d8e6 | 3340 | struct btrfs_root *, |
b06c4bf5 | 3341 | u64, u64, u64, u64, u64, u64); |
31840ae1 | 3342 | |
fccb84c9 | 3343 | |
0b246afa | 3344 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3345 | return 0; |
fccb84c9 | 3346 | |
31840ae1 | 3347 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3348 | nritems = btrfs_header_nritems(buf); |
3349 | level = btrfs_header_level(buf); | |
3350 | ||
27cdeb70 | 3351 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3352 | return 0; |
31840ae1 | 3353 | |
5d4f98a2 YZ |
3354 | if (inc) |
3355 | process_func = btrfs_inc_extent_ref; | |
3356 | else | |
3357 | process_func = btrfs_free_extent; | |
31840ae1 | 3358 | |
5d4f98a2 YZ |
3359 | if (full_backref) |
3360 | parent = buf->start; | |
3361 | else | |
3362 | parent = 0; | |
3363 | ||
3364 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3365 | if (level == 0) { |
5d4f98a2 | 3366 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3367 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3368 | continue; |
5d4f98a2 | 3369 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3370 | struct btrfs_file_extent_item); |
3371 | if (btrfs_file_extent_type(buf, fi) == | |
3372 | BTRFS_FILE_EXTENT_INLINE) | |
3373 | continue; | |
3374 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3375 | if (bytenr == 0) | |
3376 | continue; | |
5d4f98a2 YZ |
3377 | |
3378 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3379 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
84f7d8e6 | 3380 | ret = process_func(trans, root, bytenr, num_bytes, |
5d4f98a2 | 3381 | parent, ref_root, key.objectid, |
b06c4bf5 | 3382 | key.offset); |
31840ae1 ZY |
3383 | if (ret) |
3384 | goto fail; | |
3385 | } else { | |
5d4f98a2 | 3386 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 3387 | num_bytes = fs_info->nodesize; |
84f7d8e6 | 3388 | ret = process_func(trans, root, bytenr, num_bytes, |
b06c4bf5 | 3389 | parent, ref_root, level - 1, 0); |
31840ae1 ZY |
3390 | if (ret) |
3391 | goto fail; | |
3392 | } | |
3393 | } | |
3394 | return 0; | |
3395 | fail: | |
5d4f98a2 YZ |
3396 | return ret; |
3397 | } | |
3398 | ||
3399 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3400 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3401 | { |
e339a6b0 | 3402 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3403 | } |
3404 | ||
3405 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3406 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3407 | { |
e339a6b0 | 3408 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3409 | } |
3410 | ||
9078a3e1 | 3411 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 3412 | struct btrfs_fs_info *fs_info, |
9078a3e1 CM |
3413 | struct btrfs_path *path, |
3414 | struct btrfs_block_group_cache *cache) | |
3415 | { | |
3416 | int ret; | |
0b246afa | 3417 | struct btrfs_root *extent_root = fs_info->extent_root; |
5f39d397 CM |
3418 | unsigned long bi; |
3419 | struct extent_buffer *leaf; | |
9078a3e1 | 3420 | |
9078a3e1 | 3421 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3422 | if (ret) { |
3423 | if (ret > 0) | |
3424 | ret = -ENOENT; | |
54aa1f4d | 3425 | goto fail; |
df95e7f0 | 3426 | } |
5f39d397 CM |
3427 | |
3428 | leaf = path->nodes[0]; | |
3429 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3430 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3431 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3432 | fail: |
24b89d08 | 3433 | btrfs_release_path(path); |
df95e7f0 | 3434 | return ret; |
9078a3e1 CM |
3435 | |
3436 | } | |
3437 | ||
4a8c9a62 | 3438 | static struct btrfs_block_group_cache * |
2ff7e61e | 3439 | next_block_group(struct btrfs_fs_info *fs_info, |
4a8c9a62 YZ |
3440 | struct btrfs_block_group_cache *cache) |
3441 | { | |
3442 | struct rb_node *node; | |
292cbd51 | 3443 | |
0b246afa | 3444 | spin_lock(&fs_info->block_group_cache_lock); |
292cbd51 FM |
3445 | |
3446 | /* If our block group was removed, we need a full search. */ | |
3447 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3448 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3449 | ||
0b246afa | 3450 | spin_unlock(&fs_info->block_group_cache_lock); |
292cbd51 | 3451 | btrfs_put_block_group(cache); |
0b246afa | 3452 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; |
292cbd51 | 3453 | } |
4a8c9a62 YZ |
3454 | node = rb_next(&cache->cache_node); |
3455 | btrfs_put_block_group(cache); | |
3456 | if (node) { | |
3457 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3458 | cache_node); | |
11dfe35a | 3459 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3460 | } else |
3461 | cache = NULL; | |
0b246afa | 3462 | spin_unlock(&fs_info->block_group_cache_lock); |
4a8c9a62 YZ |
3463 | return cache; |
3464 | } | |
3465 | ||
0af3d00b JB |
3466 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3467 | struct btrfs_trans_handle *trans, | |
3468 | struct btrfs_path *path) | |
3469 | { | |
0b246afa JM |
3470 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
3471 | struct btrfs_root *root = fs_info->tree_root; | |
0af3d00b | 3472 | struct inode *inode = NULL; |
364ecf36 | 3473 | struct extent_changeset *data_reserved = NULL; |
0af3d00b | 3474 | u64 alloc_hint = 0; |
2b20982e | 3475 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3476 | u64 num_pages = 0; |
0af3d00b JB |
3477 | int retries = 0; |
3478 | int ret = 0; | |
3479 | ||
3480 | /* | |
3481 | * If this block group is smaller than 100 megs don't bother caching the | |
3482 | * block group. | |
3483 | */ | |
ee22184b | 3484 | if (block_group->key.offset < (100 * SZ_1M)) { |
0af3d00b JB |
3485 | spin_lock(&block_group->lock); |
3486 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3487 | spin_unlock(&block_group->lock); | |
3488 | return 0; | |
3489 | } | |
3490 | ||
0c0ef4bc JB |
3491 | if (trans->aborted) |
3492 | return 0; | |
0af3d00b | 3493 | again: |
77ab86bf | 3494 | inode = lookup_free_space_inode(fs_info, block_group, path); |
0af3d00b JB |
3495 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { |
3496 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3497 | btrfs_release_path(path); |
0af3d00b JB |
3498 | goto out; |
3499 | } | |
3500 | ||
3501 | if (IS_ERR(inode)) { | |
3502 | BUG_ON(retries); | |
3503 | retries++; | |
3504 | ||
3505 | if (block_group->ro) | |
3506 | goto out_free; | |
3507 | ||
77ab86bf JM |
3508 | ret = create_free_space_inode(fs_info, trans, block_group, |
3509 | path); | |
0af3d00b JB |
3510 | if (ret) |
3511 | goto out_free; | |
3512 | goto again; | |
3513 | } | |
3514 | ||
3515 | /* | |
3516 | * We want to set the generation to 0, that way if anything goes wrong | |
3517 | * from here on out we know not to trust this cache when we load up next | |
3518 | * time. | |
3519 | */ | |
3520 | BTRFS_I(inode)->generation = 0; | |
3521 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3522 | if (ret) { |
3523 | /* | |
3524 | * So theoretically we could recover from this, simply set the | |
3525 | * super cache generation to 0 so we know to invalidate the | |
3526 | * cache, but then we'd have to keep track of the block groups | |
3527 | * that fail this way so we know we _have_ to reset this cache | |
3528 | * before the next commit or risk reading stale cache. So to | |
3529 | * limit our exposure to horrible edge cases lets just abort the | |
3530 | * transaction, this only happens in really bad situations | |
3531 | * anyway. | |
3532 | */ | |
66642832 | 3533 | btrfs_abort_transaction(trans, ret); |
0c0ef4bc JB |
3534 | goto out_put; |
3535 | } | |
0af3d00b JB |
3536 | WARN_ON(ret); |
3537 | ||
8e138e0d JB |
3538 | /* We've already setup this transaction, go ahead and exit */ |
3539 | if (block_group->cache_generation == trans->transid && | |
3540 | i_size_read(inode)) { | |
3541 | dcs = BTRFS_DC_SETUP; | |
3542 | goto out_put; | |
3543 | } | |
3544 | ||
0af3d00b | 3545 | if (i_size_read(inode) > 0) { |
2ff7e61e | 3546 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
0b246afa | 3547 | &fs_info->global_block_rsv); |
7b61cd92 MX |
3548 | if (ret) |
3549 | goto out_put; | |
3550 | ||
77ab86bf | 3551 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); |
0af3d00b JB |
3552 | if (ret) |
3553 | goto out_put; | |
3554 | } | |
3555 | ||
3556 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3557 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
0b246afa | 3558 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cf7c1ef6 LB |
3559 | /* |
3560 | * don't bother trying to write stuff out _if_ | |
3561 | * a) we're not cached, | |
1a79c1f2 LB |
3562 | * b) we're with nospace_cache mount option, |
3563 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
cf7c1ef6 | 3564 | */ |
2b20982e | 3565 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3566 | spin_unlock(&block_group->lock); |
3567 | goto out_put; | |
3568 | } | |
3569 | spin_unlock(&block_group->lock); | |
3570 | ||
2968b1f4 JB |
3571 | /* |
3572 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3573 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3574 | */ | |
3575 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3576 | ret = -ENOSPC; | |
3577 | goto out_put; | |
3578 | } | |
3579 | ||
6fc823b1 JB |
3580 | /* |
3581 | * Try to preallocate enough space based on how big the block group is. | |
3582 | * Keep in mind this has to include any pinned space which could end up | |
3583 | * taking up quite a bit since it's not folded into the other space | |
3584 | * cache. | |
3585 | */ | |
ee22184b | 3586 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
0af3d00b JB |
3587 | if (!num_pages) |
3588 | num_pages = 1; | |
3589 | ||
0af3d00b | 3590 | num_pages *= 16; |
09cbfeaf | 3591 | num_pages *= PAGE_SIZE; |
0af3d00b | 3592 | |
364ecf36 | 3593 | ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages); |
0af3d00b JB |
3594 | if (ret) |
3595 | goto out_put; | |
3596 | ||
3597 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3598 | num_pages, num_pages, | |
3599 | &alloc_hint); | |
2968b1f4 JB |
3600 | /* |
3601 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3602 | * of metadata or split extents when writing the cache out, which means | |
3603 | * we can enospc if we are heavily fragmented in addition to just normal | |
3604 | * out of space conditions. So if we hit this just skip setting up any | |
3605 | * other block groups for this transaction, maybe we'll unpin enough | |
3606 | * space the next time around. | |
3607 | */ | |
2b20982e JB |
3608 | if (!ret) |
3609 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3610 | else if (ret == -ENOSPC) |
3611 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
c09544e0 | 3612 | |
0af3d00b JB |
3613 | out_put: |
3614 | iput(inode); | |
3615 | out_free: | |
b3b4aa74 | 3616 | btrfs_release_path(path); |
0af3d00b JB |
3617 | out: |
3618 | spin_lock(&block_group->lock); | |
e65cbb94 | 3619 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3620 | block_group->cache_generation = trans->transid; |
2b20982e | 3621 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3622 | spin_unlock(&block_group->lock); |
3623 | ||
364ecf36 | 3624 | extent_changeset_free(data_reserved); |
0af3d00b JB |
3625 | return ret; |
3626 | } | |
3627 | ||
dcdf7f6d | 3628 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans, |
2ff7e61e | 3629 | struct btrfs_fs_info *fs_info) |
dcdf7f6d JB |
3630 | { |
3631 | struct btrfs_block_group_cache *cache, *tmp; | |
3632 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3633 | struct btrfs_path *path; | |
3634 | ||
3635 | if (list_empty(&cur_trans->dirty_bgs) || | |
0b246afa | 3636 | !btrfs_test_opt(fs_info, SPACE_CACHE)) |
dcdf7f6d JB |
3637 | return 0; |
3638 | ||
3639 | path = btrfs_alloc_path(); | |
3640 | if (!path) | |
3641 | return -ENOMEM; | |
3642 | ||
3643 | /* Could add new block groups, use _safe just in case */ | |
3644 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3645 | dirty_list) { | |
3646 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3647 | cache_save_setup(cache, trans, path); | |
3648 | } | |
3649 | ||
3650 | btrfs_free_path(path); | |
3651 | return 0; | |
3652 | } | |
3653 | ||
1bbc621e CM |
3654 | /* |
3655 | * transaction commit does final block group cache writeback during a | |
3656 | * critical section where nothing is allowed to change the FS. This is | |
3657 | * required in order for the cache to actually match the block group, | |
3658 | * but can introduce a lot of latency into the commit. | |
3659 | * | |
3660 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3661 | * cache IO. There's a chance we'll have to redo some of it if the | |
3662 | * block group changes again during the commit, but it greatly reduces | |
3663 | * the commit latency by getting rid of the easy block groups while | |
3664 | * we're still allowing others to join the commit. | |
3665 | */ | |
21217054 | 3666 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) |
9078a3e1 | 3667 | { |
21217054 | 3668 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4a8c9a62 | 3669 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3670 | struct btrfs_transaction *cur_trans = trans->transaction; |
3671 | int ret = 0; | |
c9dc4c65 | 3672 | int should_put; |
1bbc621e CM |
3673 | struct btrfs_path *path = NULL; |
3674 | LIST_HEAD(dirty); | |
3675 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3676 | int num_started = 0; |
1bbc621e CM |
3677 | int loops = 0; |
3678 | ||
3679 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3680 | if (list_empty(&cur_trans->dirty_bgs)) { |
3681 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3682 | return 0; | |
1bbc621e | 3683 | } |
b58d1a9e | 3684 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3685 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3686 | |
1bbc621e | 3687 | again: |
1bbc621e CM |
3688 | /* |
3689 | * make sure all the block groups on our dirty list actually | |
3690 | * exist | |
3691 | */ | |
6c686b35 | 3692 | btrfs_create_pending_block_groups(trans); |
1bbc621e CM |
3693 | |
3694 | if (!path) { | |
3695 | path = btrfs_alloc_path(); | |
3696 | if (!path) | |
3697 | return -ENOMEM; | |
3698 | } | |
3699 | ||
b58d1a9e FM |
3700 | /* |
3701 | * cache_write_mutex is here only to save us from balance or automatic | |
3702 | * removal of empty block groups deleting this block group while we are | |
3703 | * writing out the cache | |
3704 | */ | |
3705 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e CM |
3706 | while (!list_empty(&dirty)) { |
3707 | cache = list_first_entry(&dirty, | |
3708 | struct btrfs_block_group_cache, | |
3709 | dirty_list); | |
1bbc621e CM |
3710 | /* |
3711 | * this can happen if something re-dirties a block | |
3712 | * group that is already under IO. Just wait for it to | |
3713 | * finish and then do it all again | |
3714 | */ | |
3715 | if (!list_empty(&cache->io_list)) { | |
3716 | list_del_init(&cache->io_list); | |
afdb5718 | 3717 | btrfs_wait_cache_io(trans, cache, path); |
1bbc621e CM |
3718 | btrfs_put_block_group(cache); |
3719 | } | |
3720 | ||
3721 | ||
3722 | /* | |
3723 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3724 | * if it should update the cache_state. Don't delete | |
3725 | * until after we wait. | |
3726 | * | |
3727 | * Since we're not running in the commit critical section | |
3728 | * we need the dirty_bgs_lock to protect from update_block_group | |
3729 | */ | |
3730 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3731 | list_del_init(&cache->dirty_list); | |
3732 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3733 | ||
3734 | should_put = 1; | |
3735 | ||
3736 | cache_save_setup(cache, trans, path); | |
3737 | ||
3738 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3739 | cache->io_ctl.inode = NULL; | |
0b246afa | 3740 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3741 | cache, path); |
1bbc621e CM |
3742 | if (ret == 0 && cache->io_ctl.inode) { |
3743 | num_started++; | |
3744 | should_put = 0; | |
3745 | ||
3746 | /* | |
45ae2c18 NB |
3747 | * The cache_write_mutex is protecting the |
3748 | * io_list, also refer to the definition of | |
3749 | * btrfs_transaction::io_bgs for more details | |
1bbc621e CM |
3750 | */ |
3751 | list_add_tail(&cache->io_list, io); | |
3752 | } else { | |
3753 | /* | |
3754 | * if we failed to write the cache, the | |
3755 | * generation will be bad and life goes on | |
3756 | */ | |
3757 | ret = 0; | |
3758 | } | |
3759 | } | |
ff1f8250 | 3760 | if (!ret) { |
2ff7e61e JM |
3761 | ret = write_one_cache_group(trans, fs_info, |
3762 | path, cache); | |
ff1f8250 FM |
3763 | /* |
3764 | * Our block group might still be attached to the list | |
3765 | * of new block groups in the transaction handle of some | |
3766 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3767 | * means its block group item isn't yet in the extent | |
3768 | * tree. If this happens ignore the error, as we will | |
3769 | * try again later in the critical section of the | |
3770 | * transaction commit. | |
3771 | */ | |
3772 | if (ret == -ENOENT) { | |
3773 | ret = 0; | |
3774 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3775 | if (list_empty(&cache->dirty_list)) { | |
3776 | list_add_tail(&cache->dirty_list, | |
3777 | &cur_trans->dirty_bgs); | |
3778 | btrfs_get_block_group(cache); | |
3779 | } | |
3780 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3781 | } else if (ret) { | |
66642832 | 3782 | btrfs_abort_transaction(trans, ret); |
ff1f8250 FM |
3783 | } |
3784 | } | |
1bbc621e CM |
3785 | |
3786 | /* if its not on the io list, we need to put the block group */ | |
3787 | if (should_put) | |
3788 | btrfs_put_block_group(cache); | |
3789 | ||
3790 | if (ret) | |
3791 | break; | |
b58d1a9e FM |
3792 | |
3793 | /* | |
3794 | * Avoid blocking other tasks for too long. It might even save | |
3795 | * us from writing caches for block groups that are going to be | |
3796 | * removed. | |
3797 | */ | |
3798 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3799 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3800 | } |
b58d1a9e | 3801 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3802 | |
3803 | /* | |
3804 | * go through delayed refs for all the stuff we've just kicked off | |
3805 | * and then loop back (just once) | |
3806 | */ | |
c79a70b1 | 3807 | ret = btrfs_run_delayed_refs(trans, 0); |
1bbc621e CM |
3808 | if (!ret && loops == 0) { |
3809 | loops++; | |
3810 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3811 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3812 | /* |
3813 | * dirty_bgs_lock protects us from concurrent block group | |
3814 | * deletes too (not just cache_write_mutex). | |
3815 | */ | |
3816 | if (!list_empty(&dirty)) { | |
3817 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3818 | goto again; | |
3819 | } | |
1bbc621e | 3820 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c79a1751 | 3821 | } else if (ret < 0) { |
2ff7e61e | 3822 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
1bbc621e CM |
3823 | } |
3824 | ||
3825 | btrfs_free_path(path); | |
3826 | return ret; | |
3827 | } | |
3828 | ||
3829 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3830 | struct btrfs_fs_info *fs_info) |
1bbc621e CM |
3831 | { |
3832 | struct btrfs_block_group_cache *cache; | |
3833 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3834 | int ret = 0; | |
3835 | int should_put; | |
3836 | struct btrfs_path *path; | |
3837 | struct list_head *io = &cur_trans->io_bgs; | |
3838 | int num_started = 0; | |
9078a3e1 CM |
3839 | |
3840 | path = btrfs_alloc_path(); | |
3841 | if (!path) | |
3842 | return -ENOMEM; | |
3843 | ||
ce93ec54 | 3844 | /* |
e44081ef FM |
3845 | * Even though we are in the critical section of the transaction commit, |
3846 | * we can still have concurrent tasks adding elements to this | |
3847 | * transaction's list of dirty block groups. These tasks correspond to | |
3848 | * endio free space workers started when writeback finishes for a | |
3849 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3850 | * allocate new block groups as a result of COWing nodes of the root | |
3851 | * tree when updating the free space inode. The writeback for the space | |
3852 | * caches is triggered by an earlier call to | |
3853 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3854 | * loop. | |
3855 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3856 | * delayed refs to make sure we have the best chance at doing this all |
3857 | * in one shot. | |
3858 | */ | |
e44081ef | 3859 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3860 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3861 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3862 | struct btrfs_block_group_cache, | |
3863 | dirty_list); | |
c9dc4c65 CM |
3864 | |
3865 | /* | |
3866 | * this can happen if cache_save_setup re-dirties a block | |
3867 | * group that is already under IO. Just wait for it to | |
3868 | * finish and then do it all again | |
3869 | */ | |
3870 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3871 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3872 | list_del_init(&cache->io_list); |
afdb5718 | 3873 | btrfs_wait_cache_io(trans, cache, path); |
c9dc4c65 | 3874 | btrfs_put_block_group(cache); |
e44081ef | 3875 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3876 | } |
3877 | ||
1bbc621e CM |
3878 | /* |
3879 | * don't remove from the dirty list until after we've waited | |
3880 | * on any pending IO | |
3881 | */ | |
ce93ec54 | 3882 | list_del_init(&cache->dirty_list); |
e44081ef | 3883 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3884 | should_put = 1; |
3885 | ||
1bbc621e | 3886 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3887 | |
ce93ec54 | 3888 | if (!ret) |
c79a70b1 | 3889 | ret = btrfs_run_delayed_refs(trans, |
2ff7e61e | 3890 | (unsigned long) -1); |
c9dc4c65 CM |
3891 | |
3892 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3893 | cache->io_ctl.inode = NULL; | |
0b246afa | 3894 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3895 | cache, path); |
c9dc4c65 CM |
3896 | if (ret == 0 && cache->io_ctl.inode) { |
3897 | num_started++; | |
3898 | should_put = 0; | |
1bbc621e | 3899 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3900 | } else { |
3901 | /* | |
3902 | * if we failed to write the cache, the | |
3903 | * generation will be bad and life goes on | |
3904 | */ | |
3905 | ret = 0; | |
3906 | } | |
3907 | } | |
ff1f8250 | 3908 | if (!ret) { |
2ff7e61e JM |
3909 | ret = write_one_cache_group(trans, fs_info, |
3910 | path, cache); | |
2bc0bb5f FM |
3911 | /* |
3912 | * One of the free space endio workers might have | |
3913 | * created a new block group while updating a free space | |
3914 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3915 | * and hasn't released its transaction handle yet, in | |
3916 | * which case the new block group is still attached to | |
3917 | * its transaction handle and its creation has not | |
3918 | * finished yet (no block group item in the extent tree | |
3919 | * yet, etc). If this is the case, wait for all free | |
3920 | * space endio workers to finish and retry. This is a | |
3921 | * a very rare case so no need for a more efficient and | |
3922 | * complex approach. | |
3923 | */ | |
3924 | if (ret == -ENOENT) { | |
3925 | wait_event(cur_trans->writer_wait, | |
3926 | atomic_read(&cur_trans->num_writers) == 1); | |
2ff7e61e JM |
3927 | ret = write_one_cache_group(trans, fs_info, |
3928 | path, cache); | |
2bc0bb5f | 3929 | } |
ff1f8250 | 3930 | if (ret) |
66642832 | 3931 | btrfs_abort_transaction(trans, ret); |
ff1f8250 | 3932 | } |
c9dc4c65 CM |
3933 | |
3934 | /* if its not on the io list, we need to put the block group */ | |
3935 | if (should_put) | |
3936 | btrfs_put_block_group(cache); | |
e44081ef | 3937 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3938 | } |
e44081ef | 3939 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3940 | |
45ae2c18 NB |
3941 | /* |
3942 | * Refer to the definition of io_bgs member for details why it's safe | |
3943 | * to use it without any locking | |
3944 | */ | |
1bbc621e CM |
3945 | while (!list_empty(io)) { |
3946 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3947 | io_list); |
3948 | list_del_init(&cache->io_list); | |
afdb5718 | 3949 | btrfs_wait_cache_io(trans, cache, path); |
0cb59c99 JB |
3950 | btrfs_put_block_group(cache); |
3951 | } | |
3952 | ||
9078a3e1 | 3953 | btrfs_free_path(path); |
ce93ec54 | 3954 | return ret; |
9078a3e1 CM |
3955 | } |
3956 | ||
2ff7e61e | 3957 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 YZ |
3958 | { |
3959 | struct btrfs_block_group_cache *block_group; | |
3960 | int readonly = 0; | |
3961 | ||
0b246afa | 3962 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
3963 | if (!block_group || block_group->ro) |
3964 | readonly = 1; | |
3965 | if (block_group) | |
fa9c0d79 | 3966 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3967 | return readonly; |
3968 | } | |
3969 | ||
f78c436c FM |
3970 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
3971 | { | |
3972 | struct btrfs_block_group_cache *bg; | |
3973 | bool ret = true; | |
3974 | ||
3975 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3976 | if (!bg) | |
3977 | return false; | |
3978 | ||
3979 | spin_lock(&bg->lock); | |
3980 | if (bg->ro) | |
3981 | ret = false; | |
3982 | else | |
3983 | atomic_inc(&bg->nocow_writers); | |
3984 | spin_unlock(&bg->lock); | |
3985 | ||
3986 | /* no put on block group, done by btrfs_dec_nocow_writers */ | |
3987 | if (!ret) | |
3988 | btrfs_put_block_group(bg); | |
3989 | ||
3990 | return ret; | |
3991 | ||
3992 | } | |
3993 | ||
3994 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
3995 | { | |
3996 | struct btrfs_block_group_cache *bg; | |
3997 | ||
3998 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3999 | ASSERT(bg); | |
4000 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
4625956a | 4001 | wake_up_var(&bg->nocow_writers); |
f78c436c FM |
4002 | /* |
4003 | * Once for our lookup and once for the lookup done by a previous call | |
4004 | * to btrfs_inc_nocow_writers() | |
4005 | */ | |
4006 | btrfs_put_block_group(bg); | |
4007 | btrfs_put_block_group(bg); | |
4008 | } | |
4009 | ||
f78c436c FM |
4010 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) |
4011 | { | |
4625956a | 4012 | wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers)); |
f78c436c FM |
4013 | } |
4014 | ||
6ab0a202 JM |
4015 | static const char *alloc_name(u64 flags) |
4016 | { | |
4017 | switch (flags) { | |
4018 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
4019 | return "mixed"; | |
4020 | case BTRFS_BLOCK_GROUP_METADATA: | |
4021 | return "metadata"; | |
4022 | case BTRFS_BLOCK_GROUP_DATA: | |
4023 | return "data"; | |
4024 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
4025 | return "system"; | |
4026 | default: | |
4027 | WARN_ON(1); | |
4028 | return "invalid-combination"; | |
4029 | }; | |
4030 | } | |
4031 | ||
2be12ef7 NB |
4032 | static int create_space_info(struct btrfs_fs_info *info, u64 flags, |
4033 | struct btrfs_space_info **new) | |
4034 | { | |
4035 | ||
4036 | struct btrfs_space_info *space_info; | |
4037 | int i; | |
4038 | int ret; | |
4039 | ||
4040 | space_info = kzalloc(sizeof(*space_info), GFP_NOFS); | |
4041 | if (!space_info) | |
4042 | return -ENOMEM; | |
4043 | ||
4044 | ret = percpu_counter_init(&space_info->total_bytes_pinned, 0, | |
4045 | GFP_KERNEL); | |
4046 | if (ret) { | |
4047 | kfree(space_info); | |
4048 | return ret; | |
4049 | } | |
4050 | ||
4051 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
4052 | INIT_LIST_HEAD(&space_info->block_groups[i]); | |
4053 | init_rwsem(&space_info->groups_sem); | |
4054 | spin_lock_init(&space_info->lock); | |
4055 | space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; | |
4056 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
4057 | init_waitqueue_head(&space_info->wait); | |
4058 | INIT_LIST_HEAD(&space_info->ro_bgs); | |
4059 | INIT_LIST_HEAD(&space_info->tickets); | |
4060 | INIT_LIST_HEAD(&space_info->priority_tickets); | |
4061 | ||
4062 | ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype, | |
4063 | info->space_info_kobj, "%s", | |
4064 | alloc_name(space_info->flags)); | |
4065 | if (ret) { | |
4066 | percpu_counter_destroy(&space_info->total_bytes_pinned); | |
4067 | kfree(space_info); | |
4068 | return ret; | |
4069 | } | |
4070 | ||
4071 | *new = space_info; | |
4072 | list_add_rcu(&space_info->list, &info->space_info); | |
4073 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
4074 | info->data_sinfo = space_info; | |
4075 | ||
4076 | return ret; | |
4077 | } | |
4078 | ||
d2006e6d | 4079 | static void update_space_info(struct btrfs_fs_info *info, u64 flags, |
593060d7 | 4080 | u64 total_bytes, u64 bytes_used, |
e40edf2d | 4081 | u64 bytes_readonly, |
593060d7 CM |
4082 | struct btrfs_space_info **space_info) |
4083 | { | |
4084 | struct btrfs_space_info *found; | |
b742bb82 YZ |
4085 | int factor; |
4086 | ||
4087 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
4088 | BTRFS_BLOCK_GROUP_RAID10)) | |
4089 | factor = 2; | |
4090 | else | |
4091 | factor = 1; | |
593060d7 CM |
4092 | |
4093 | found = __find_space_info(info, flags); | |
d2006e6d NB |
4094 | ASSERT(found); |
4095 | spin_lock(&found->lock); | |
4096 | found->total_bytes += total_bytes; | |
4097 | found->disk_total += total_bytes * factor; | |
4098 | found->bytes_used += bytes_used; | |
4099 | found->disk_used += bytes_used * factor; | |
4100 | found->bytes_readonly += bytes_readonly; | |
4101 | if (total_bytes > 0) | |
4102 | found->full = 0; | |
4103 | space_info_add_new_bytes(info, found, total_bytes - | |
4104 | bytes_used - bytes_readonly); | |
4105 | spin_unlock(&found->lock); | |
4106 | *space_info = found; | |
593060d7 CM |
4107 | } |
4108 | ||
8790d502 CM |
4109 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
4110 | { | |
899c81ea ID |
4111 | u64 extra_flags = chunk_to_extended(flags) & |
4112 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 4113 | |
de98ced9 | 4114 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
4115 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
4116 | fs_info->avail_data_alloc_bits |= extra_flags; | |
4117 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
4118 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
4119 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
4120 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 4121 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 4122 | } |
593060d7 | 4123 | |
fc67c450 ID |
4124 | /* |
4125 | * returns target flags in extended format or 0 if restripe for this | |
4126 | * chunk_type is not in progress | |
c6664b42 ID |
4127 | * |
4128 | * should be called with either volume_mutex or balance_lock held | |
fc67c450 ID |
4129 | */ |
4130 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
4131 | { | |
4132 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
4133 | u64 target = 0; | |
4134 | ||
fc67c450 ID |
4135 | if (!bctl) |
4136 | return 0; | |
4137 | ||
4138 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
4139 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4140 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
4141 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
4142 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4143 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
4144 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
4145 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
4146 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
4147 | } | |
4148 | ||
4149 | return target; | |
4150 | } | |
4151 | ||
a46d11a8 ID |
4152 | /* |
4153 | * @flags: available profiles in extended format (see ctree.h) | |
4154 | * | |
e4d8ec0f ID |
4155 | * Returns reduced profile in chunk format. If profile changing is in |
4156 | * progress (either running or paused) picks the target profile (if it's | |
4157 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 4158 | */ |
2ff7e61e | 4159 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c | 4160 | { |
0b246afa | 4161 | u64 num_devices = fs_info->fs_devices->rw_devices; |
fc67c450 | 4162 | u64 target; |
9c170b26 ZL |
4163 | u64 raid_type; |
4164 | u64 allowed = 0; | |
a061fc8d | 4165 | |
fc67c450 ID |
4166 | /* |
4167 | * see if restripe for this chunk_type is in progress, if so | |
4168 | * try to reduce to the target profile | |
4169 | */ | |
0b246afa JM |
4170 | spin_lock(&fs_info->balance_lock); |
4171 | target = get_restripe_target(fs_info, flags); | |
fc67c450 ID |
4172 | if (target) { |
4173 | /* pick target profile only if it's already available */ | |
4174 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
0b246afa | 4175 | spin_unlock(&fs_info->balance_lock); |
fc67c450 | 4176 | return extended_to_chunk(target); |
e4d8ec0f ID |
4177 | } |
4178 | } | |
0b246afa | 4179 | spin_unlock(&fs_info->balance_lock); |
e4d8ec0f | 4180 | |
53b381b3 | 4181 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
4182 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
4183 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
4184 | allowed |= btrfs_raid_group[raid_type]; | |
4185 | } | |
4186 | allowed &= flags; | |
4187 | ||
4188 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
4189 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
4190 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
4191 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
4192 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
4193 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
4194 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
4195 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
4196 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
4197 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
4198 | ||
4199 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
4200 | ||
4201 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
4202 | } |
4203 | ||
2ff7e61e | 4204 | static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
6a63209f | 4205 | { |
de98ced9 | 4206 | unsigned seq; |
f8213bdc | 4207 | u64 flags; |
de98ced9 MX |
4208 | |
4209 | do { | |
f8213bdc | 4210 | flags = orig_flags; |
0b246afa | 4211 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 MX |
4212 | |
4213 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
0b246afa | 4214 | flags |= fs_info->avail_data_alloc_bits; |
de98ced9 | 4215 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
0b246afa | 4216 | flags |= fs_info->avail_system_alloc_bits; |
de98ced9 | 4217 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
0b246afa JM |
4218 | flags |= fs_info->avail_metadata_alloc_bits; |
4219 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
6fef8df1 | 4220 | |
2ff7e61e | 4221 | return btrfs_reduce_alloc_profile(fs_info, flags); |
6a63209f JB |
4222 | } |
4223 | ||
1b86826d | 4224 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 4225 | { |
0b246afa | 4226 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 4227 | u64 flags; |
53b381b3 | 4228 | u64 ret; |
9ed74f2d | 4229 | |
b742bb82 YZ |
4230 | if (data) |
4231 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 4232 | else if (root == fs_info->chunk_root) |
b742bb82 | 4233 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 4234 | else |
b742bb82 | 4235 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 4236 | |
2ff7e61e | 4237 | ret = get_alloc_profile(fs_info, flags); |
53b381b3 | 4238 | return ret; |
6a63209f | 4239 | } |
9ed74f2d | 4240 | |
1b86826d JM |
4241 | u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info) |
4242 | { | |
4243 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA); | |
4244 | } | |
4245 | ||
4246 | u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info) | |
4247 | { | |
4248 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4249 | } | |
4250 | ||
4251 | u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info) | |
4252 | { | |
4253 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
4254 | } | |
4255 | ||
4136135b LB |
4256 | static u64 btrfs_space_info_used(struct btrfs_space_info *s_info, |
4257 | bool may_use_included) | |
4258 | { | |
4259 | ASSERT(s_info); | |
4260 | return s_info->bytes_used + s_info->bytes_reserved + | |
4261 | s_info->bytes_pinned + s_info->bytes_readonly + | |
4262 | (may_use_included ? s_info->bytes_may_use : 0); | |
4263 | } | |
4264 | ||
04f4f916 | 4265 | int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes) |
6a63209f | 4266 | { |
04f4f916 | 4267 | struct btrfs_root *root = inode->root; |
b4d7c3c9 | 4268 | struct btrfs_fs_info *fs_info = root->fs_info; |
1174cade | 4269 | struct btrfs_space_info *data_sinfo = fs_info->data_sinfo; |
ab6e2410 | 4270 | u64 used; |
94b947b2 | 4271 | int ret = 0; |
c99f1b0c ZL |
4272 | int need_commit = 2; |
4273 | int have_pinned_space; | |
6a63209f | 4274 | |
6a63209f | 4275 | /* make sure bytes are sectorsize aligned */ |
0b246afa | 4276 | bytes = ALIGN(bytes, fs_info->sectorsize); |
6a63209f | 4277 | |
9dced186 | 4278 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 4279 | need_commit = 0; |
9dced186 | 4280 | ASSERT(current->journal_info); |
0af3d00b JB |
4281 | } |
4282 | ||
6a63209f JB |
4283 | again: |
4284 | /* make sure we have enough space to handle the data first */ | |
4285 | spin_lock(&data_sinfo->lock); | |
4136135b | 4286 | used = btrfs_space_info_used(data_sinfo, true); |
ab6e2410 JB |
4287 | |
4288 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 4289 | struct btrfs_trans_handle *trans; |
9ed74f2d | 4290 | |
6a63209f JB |
4291 | /* |
4292 | * if we don't have enough free bytes in this space then we need | |
4293 | * to alloc a new chunk. | |
4294 | */ | |
b9fd47cd | 4295 | if (!data_sinfo->full) { |
6a63209f | 4296 | u64 alloc_target; |
9ed74f2d | 4297 | |
0e4f8f88 | 4298 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 4299 | spin_unlock(&data_sinfo->lock); |
1174cade | 4300 | |
1b86826d | 4301 | alloc_target = btrfs_data_alloc_profile(fs_info); |
9dced186 MX |
4302 | /* |
4303 | * It is ugly that we don't call nolock join | |
4304 | * transaction for the free space inode case here. | |
4305 | * But it is safe because we only do the data space | |
4306 | * reservation for the free space cache in the | |
4307 | * transaction context, the common join transaction | |
4308 | * just increase the counter of the current transaction | |
4309 | * handler, doesn't try to acquire the trans_lock of | |
4310 | * the fs. | |
4311 | */ | |
7a7eaa40 | 4312 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4313 | if (IS_ERR(trans)) |
4314 | return PTR_ERR(trans); | |
9ed74f2d | 4315 | |
2ff7e61e | 4316 | ret = do_chunk_alloc(trans, fs_info, alloc_target, |
0e4f8f88 | 4317 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 4318 | btrfs_end_transaction(trans); |
d52a5b5f MX |
4319 | if (ret < 0) { |
4320 | if (ret != -ENOSPC) | |
4321 | return ret; | |
c99f1b0c ZL |
4322 | else { |
4323 | have_pinned_space = 1; | |
d52a5b5f | 4324 | goto commit_trans; |
c99f1b0c | 4325 | } |
d52a5b5f | 4326 | } |
9ed74f2d | 4327 | |
6a63209f JB |
4328 | goto again; |
4329 | } | |
f2bb8f5c JB |
4330 | |
4331 | /* | |
b150a4f1 | 4332 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
4333 | * allocation, and no removed chunk in current transaction, |
4334 | * don't bother committing the transaction. | |
f2bb8f5c | 4335 | */ |
c99f1b0c ZL |
4336 | have_pinned_space = percpu_counter_compare( |
4337 | &data_sinfo->total_bytes_pinned, | |
4338 | used + bytes - data_sinfo->total_bytes); | |
6a63209f | 4339 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4340 | |
4e06bdd6 | 4341 | /* commit the current transaction and try again */ |
d52a5b5f | 4342 | commit_trans: |
92e2f7e3 | 4343 | if (need_commit) { |
c99f1b0c | 4344 | need_commit--; |
b150a4f1 | 4345 | |
e1746e83 ZL |
4346 | if (need_commit > 0) { |
4347 | btrfs_start_delalloc_roots(fs_info, 0, -1); | |
6374e57a | 4348 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, |
0b246afa | 4349 | (u64)-1); |
e1746e83 | 4350 | } |
9a4e7276 | 4351 | |
7a7eaa40 | 4352 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4353 | if (IS_ERR(trans)) |
4354 | return PTR_ERR(trans); | |
c99f1b0c | 4355 | if (have_pinned_space >= 0 || |
3204d33c JB |
4356 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4357 | &trans->transaction->flags) || | |
c99f1b0c | 4358 | need_commit > 0) { |
3a45bb20 | 4359 | ret = btrfs_commit_transaction(trans); |
94b947b2 ZL |
4360 | if (ret) |
4361 | return ret; | |
d7c15171 | 4362 | /* |
c2d6cb16 FM |
4363 | * The cleaner kthread might still be doing iput |
4364 | * operations. Wait for it to finish so that | |
4365 | * more space is released. | |
d7c15171 | 4366 | */ |
0b246afa JM |
4367 | mutex_lock(&fs_info->cleaner_delayed_iput_mutex); |
4368 | mutex_unlock(&fs_info->cleaner_delayed_iput_mutex); | |
94b947b2 ZL |
4369 | goto again; |
4370 | } else { | |
3a45bb20 | 4371 | btrfs_end_transaction(trans); |
94b947b2 | 4372 | } |
4e06bdd6 | 4373 | } |
9ed74f2d | 4374 | |
0b246afa | 4375 | trace_btrfs_space_reservation(fs_info, |
cab45e22 JM |
4376 | "space_info:enospc", |
4377 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4378 | return -ENOSPC; |
4379 | } | |
4380 | data_sinfo->bytes_may_use += bytes; | |
0b246afa | 4381 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4382 | data_sinfo->flags, bytes, 1); |
6a63209f | 4383 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4384 | |
237c0e9f | 4385 | return ret; |
9ed74f2d | 4386 | } |
6a63209f | 4387 | |
364ecf36 QW |
4388 | int btrfs_check_data_free_space(struct inode *inode, |
4389 | struct extent_changeset **reserved, u64 start, u64 len) | |
4ceff079 | 4390 | { |
0b246afa | 4391 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4392 | int ret; |
4393 | ||
4394 | /* align the range */ | |
0b246afa JM |
4395 | len = round_up(start + len, fs_info->sectorsize) - |
4396 | round_down(start, fs_info->sectorsize); | |
4397 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4398 | |
04f4f916 | 4399 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len); |
4ceff079 QW |
4400 | if (ret < 0) |
4401 | return ret; | |
4402 | ||
1e5ec2e7 | 4403 | /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */ |
364ecf36 | 4404 | ret = btrfs_qgroup_reserve_data(inode, reserved, start, len); |
7bc329c1 | 4405 | if (ret < 0) |
1e5ec2e7 | 4406 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
364ecf36 QW |
4407 | else |
4408 | ret = 0; | |
4ceff079 QW |
4409 | return ret; |
4410 | } | |
4411 | ||
4ceff079 QW |
4412 | /* |
4413 | * Called if we need to clear a data reservation for this inode | |
4414 | * Normally in a error case. | |
4415 | * | |
51773bec QW |
4416 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4417 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4418 | * Like clear_bit_hook(). | |
4ceff079 | 4419 | */ |
51773bec QW |
4420 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4421 | u64 len) | |
4ceff079 | 4422 | { |
0b246afa | 4423 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4424 | struct btrfs_space_info *data_sinfo; |
4425 | ||
4426 | /* Make sure the range is aligned to sectorsize */ | |
0b246afa JM |
4427 | len = round_up(start + len, fs_info->sectorsize) - |
4428 | round_down(start, fs_info->sectorsize); | |
4429 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4430 | |
0b246afa | 4431 | data_sinfo = fs_info->data_sinfo; |
4ceff079 QW |
4432 | spin_lock(&data_sinfo->lock); |
4433 | if (WARN_ON(data_sinfo->bytes_may_use < len)) | |
4434 | data_sinfo->bytes_may_use = 0; | |
4435 | else | |
4436 | data_sinfo->bytes_may_use -= len; | |
0b246afa | 4437 | trace_btrfs_space_reservation(fs_info, "space_info", |
4ceff079 QW |
4438 | data_sinfo->flags, len, 0); |
4439 | spin_unlock(&data_sinfo->lock); | |
4440 | } | |
4441 | ||
51773bec QW |
4442 | /* |
4443 | * Called if we need to clear a data reservation for this inode | |
4444 | * Normally in a error case. | |
4445 | * | |
01327610 | 4446 | * This one will handle the per-inode data rsv map for accurate reserved |
51773bec QW |
4447 | * space framework. |
4448 | */ | |
bc42bda2 QW |
4449 | void btrfs_free_reserved_data_space(struct inode *inode, |
4450 | struct extent_changeset *reserved, u64 start, u64 len) | |
51773bec | 4451 | { |
0c476a5d JM |
4452 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4453 | ||
4454 | /* Make sure the range is aligned to sectorsize */ | |
da17066c JM |
4455 | len = round_up(start + len, root->fs_info->sectorsize) - |
4456 | round_down(start, root->fs_info->sectorsize); | |
4457 | start = round_down(start, root->fs_info->sectorsize); | |
0c476a5d | 4458 | |
51773bec | 4459 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
bc42bda2 | 4460 | btrfs_qgroup_free_data(inode, reserved, start, len); |
51773bec QW |
4461 | } |
4462 | ||
97e728d4 | 4463 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4464 | { |
97e728d4 JB |
4465 | struct list_head *head = &info->space_info; |
4466 | struct btrfs_space_info *found; | |
e3ccfa98 | 4467 | |
97e728d4 JB |
4468 | rcu_read_lock(); |
4469 | list_for_each_entry_rcu(found, head, list) { | |
4470 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4471 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4472 | } |
97e728d4 | 4473 | rcu_read_unlock(); |
e3ccfa98 JB |
4474 | } |
4475 | ||
3c76cd84 MX |
4476 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
4477 | { | |
4478 | return (global->size << 1); | |
4479 | } | |
4480 | ||
2ff7e61e | 4481 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, |
698d0082 | 4482 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4483 | { |
0b246afa | 4484 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
8d8aafee | 4485 | u64 bytes_used = btrfs_space_info_used(sinfo, false); |
e5bc2458 | 4486 | u64 thresh; |
e3ccfa98 | 4487 | |
0e4f8f88 CM |
4488 | if (force == CHUNK_ALLOC_FORCE) |
4489 | return 1; | |
4490 | ||
fb25e914 JB |
4491 | /* |
4492 | * We need to take into account the global rsv because for all intents | |
4493 | * and purposes it's used space. Don't worry about locking the | |
4494 | * global_rsv, it doesn't change except when the transaction commits. | |
4495 | */ | |
54338b5c | 4496 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
8d8aafee | 4497 | bytes_used += calc_global_rsv_need_space(global_rsv); |
fb25e914 | 4498 | |
0e4f8f88 CM |
4499 | /* |
4500 | * in limited mode, we want to have some free space up to | |
4501 | * about 1% of the FS size. | |
4502 | */ | |
4503 | if (force == CHUNK_ALLOC_LIMITED) { | |
0b246afa | 4504 | thresh = btrfs_super_total_bytes(fs_info->super_copy); |
ee22184b | 4505 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
0e4f8f88 | 4506 | |
8d8aafee | 4507 | if (sinfo->total_bytes - bytes_used < thresh) |
0e4f8f88 CM |
4508 | return 1; |
4509 | } | |
0e4f8f88 | 4510 | |
8d8aafee | 4511 | if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) |
14ed0ca6 | 4512 | return 0; |
424499db | 4513 | return 1; |
32c00aff JB |
4514 | } |
4515 | ||
2ff7e61e | 4516 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4517 | { |
4518 | u64 num_dev; | |
4519 | ||
53b381b3 DW |
4520 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
4521 | BTRFS_BLOCK_GROUP_RAID0 | | |
4522 | BTRFS_BLOCK_GROUP_RAID5 | | |
4523 | BTRFS_BLOCK_GROUP_RAID6)) | |
0b246afa | 4524 | num_dev = fs_info->fs_devices->rw_devices; |
15d1ff81 LB |
4525 | else if (type & BTRFS_BLOCK_GROUP_RAID1) |
4526 | num_dev = 2; | |
4527 | else | |
4528 | num_dev = 1; /* DUP or single */ | |
4529 | ||
39c2d7fa | 4530 | return num_dev; |
15d1ff81 LB |
4531 | } |
4532 | ||
39c2d7fa FM |
4533 | /* |
4534 | * If @is_allocation is true, reserve space in the system space info necessary | |
4535 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4536 | * removing a chunk. | |
4537 | */ | |
4538 | void check_system_chunk(struct btrfs_trans_handle *trans, | |
2ff7e61e | 4539 | struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4540 | { |
4541 | struct btrfs_space_info *info; | |
4542 | u64 left; | |
4543 | u64 thresh; | |
4fbcdf66 | 4544 | int ret = 0; |
39c2d7fa | 4545 | u64 num_devs; |
4fbcdf66 FM |
4546 | |
4547 | /* | |
4548 | * Needed because we can end up allocating a system chunk and for an | |
4549 | * atomic and race free space reservation in the chunk block reserve. | |
4550 | */ | |
a32bf9a3 | 4551 | lockdep_assert_held(&fs_info->chunk_mutex); |
15d1ff81 | 4552 | |
0b246afa | 4553 | info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
15d1ff81 | 4554 | spin_lock(&info->lock); |
4136135b | 4555 | left = info->total_bytes - btrfs_space_info_used(info, true); |
15d1ff81 LB |
4556 | spin_unlock(&info->lock); |
4557 | ||
2ff7e61e | 4558 | num_devs = get_profile_num_devs(fs_info, type); |
39c2d7fa FM |
4559 | |
4560 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
0b246afa JM |
4561 | thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) + |
4562 | btrfs_calc_trans_metadata_size(fs_info, 1); | |
39c2d7fa | 4563 | |
0b246afa JM |
4564 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
4565 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
4566 | left, thresh, type); | |
4567 | dump_space_info(fs_info, info, 0, 0); | |
15d1ff81 LB |
4568 | } |
4569 | ||
4570 | if (left < thresh) { | |
1b86826d | 4571 | u64 flags = btrfs_system_alloc_profile(fs_info); |
15d1ff81 | 4572 | |
4fbcdf66 FM |
4573 | /* |
4574 | * Ignore failure to create system chunk. We might end up not | |
4575 | * needing it, as we might not need to COW all nodes/leafs from | |
4576 | * the paths we visit in the chunk tree (they were already COWed | |
4577 | * or created in the current transaction for example). | |
4578 | */ | |
2ff7e61e | 4579 | ret = btrfs_alloc_chunk(trans, fs_info, flags); |
4fbcdf66 FM |
4580 | } |
4581 | ||
4582 | if (!ret) { | |
0b246afa JM |
4583 | ret = btrfs_block_rsv_add(fs_info->chunk_root, |
4584 | &fs_info->chunk_block_rsv, | |
4fbcdf66 FM |
4585 | thresh, BTRFS_RESERVE_NO_FLUSH); |
4586 | if (!ret) | |
4587 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4588 | } |
4589 | } | |
4590 | ||
28b737f6 LB |
4591 | /* |
4592 | * If force is CHUNK_ALLOC_FORCE: | |
4593 | * - return 1 if it successfully allocates a chunk, | |
4594 | * - return errors including -ENOSPC otherwise. | |
4595 | * If force is NOT CHUNK_ALLOC_FORCE: | |
4596 | * - return 0 if it doesn't need to allocate a new chunk, | |
4597 | * - return 1 if it successfully allocates a chunk, | |
4598 | * - return errors including -ENOSPC otherwise. | |
4599 | */ | |
6324fbf3 | 4600 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 4601 | struct btrfs_fs_info *fs_info, u64 flags, int force) |
9ed74f2d | 4602 | { |
6324fbf3 | 4603 | struct btrfs_space_info *space_info; |
6d74119f | 4604 | int wait_for_alloc = 0; |
9ed74f2d | 4605 | int ret = 0; |
9ed74f2d | 4606 | |
c6b305a8 JB |
4607 | /* Don't re-enter if we're already allocating a chunk */ |
4608 | if (trans->allocating_chunk) | |
4609 | return -ENOSPC; | |
4610 | ||
0b246afa | 4611 | space_info = __find_space_info(fs_info, flags); |
dc2d3005 | 4612 | ASSERT(space_info); |
9ed74f2d | 4613 | |
6d74119f | 4614 | again: |
25179201 | 4615 | spin_lock(&space_info->lock); |
9e622d6b | 4616 | if (force < space_info->force_alloc) |
0e4f8f88 | 4617 | force = space_info->force_alloc; |
25179201 | 4618 | if (space_info->full) { |
2ff7e61e | 4619 | if (should_alloc_chunk(fs_info, space_info, force)) |
09fb99a6 FDBM |
4620 | ret = -ENOSPC; |
4621 | else | |
4622 | ret = 0; | |
25179201 | 4623 | spin_unlock(&space_info->lock); |
09fb99a6 | 4624 | return ret; |
9ed74f2d JB |
4625 | } |
4626 | ||
2ff7e61e | 4627 | if (!should_alloc_chunk(fs_info, space_info, force)) { |
25179201 | 4628 | spin_unlock(&space_info->lock); |
6d74119f JB |
4629 | return 0; |
4630 | } else if (space_info->chunk_alloc) { | |
4631 | wait_for_alloc = 1; | |
4632 | } else { | |
4633 | space_info->chunk_alloc = 1; | |
9ed74f2d | 4634 | } |
0e4f8f88 | 4635 | |
25179201 | 4636 | spin_unlock(&space_info->lock); |
9ed74f2d | 4637 | |
6d74119f JB |
4638 | mutex_lock(&fs_info->chunk_mutex); |
4639 | ||
4640 | /* | |
4641 | * The chunk_mutex is held throughout the entirety of a chunk | |
4642 | * allocation, so once we've acquired the chunk_mutex we know that the | |
4643 | * other guy is done and we need to recheck and see if we should | |
4644 | * allocate. | |
4645 | */ | |
4646 | if (wait_for_alloc) { | |
4647 | mutex_unlock(&fs_info->chunk_mutex); | |
4648 | wait_for_alloc = 0; | |
1e1c50a9 | 4649 | cond_resched(); |
6d74119f JB |
4650 | goto again; |
4651 | } | |
4652 | ||
c6b305a8 JB |
4653 | trans->allocating_chunk = true; |
4654 | ||
67377734 JB |
4655 | /* |
4656 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4657 | * allocating mixed chunks instead of individual chunks. | |
4658 | */ | |
4659 | if (btrfs_mixed_space_info(space_info)) | |
4660 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4661 | ||
97e728d4 JB |
4662 | /* |
4663 | * if we're doing a data chunk, go ahead and make sure that | |
4664 | * we keep a reasonable number of metadata chunks allocated in the | |
4665 | * FS as well. | |
4666 | */ | |
9ed74f2d | 4667 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4668 | fs_info->data_chunk_allocations++; |
4669 | if (!(fs_info->data_chunk_allocations % | |
4670 | fs_info->metadata_ratio)) | |
4671 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4672 | } |
4673 | ||
15d1ff81 LB |
4674 | /* |
4675 | * Check if we have enough space in SYSTEM chunk because we may need | |
4676 | * to update devices. | |
4677 | */ | |
2ff7e61e | 4678 | check_system_chunk(trans, fs_info, flags); |
15d1ff81 | 4679 | |
2ff7e61e | 4680 | ret = btrfs_alloc_chunk(trans, fs_info, flags); |
c6b305a8 | 4681 | trans->allocating_chunk = false; |
92b8e897 | 4682 | |
9ed74f2d | 4683 | spin_lock(&space_info->lock); |
57f1642e NB |
4684 | if (ret < 0) { |
4685 | if (ret == -ENOSPC) | |
4686 | space_info->full = 1; | |
4687 | else | |
4688 | goto out; | |
4689 | } else { | |
424499db | 4690 | ret = 1; |
57f1642e | 4691 | } |
6d74119f | 4692 | |
0e4f8f88 | 4693 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4694 | out: |
6d74119f | 4695 | space_info->chunk_alloc = 0; |
9ed74f2d | 4696 | spin_unlock(&space_info->lock); |
a25c75d5 | 4697 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4698 | /* |
4699 | * When we allocate a new chunk we reserve space in the chunk block | |
4700 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4701 | * add new nodes/leafs to it if we end up needing to do it when | |
4702 | * inserting the chunk item and updating device items as part of the | |
4703 | * second phase of chunk allocation, performed by | |
4704 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4705 | * large number of new block groups to create in our transaction | |
4706 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4707 | * in extreme cases - like having a single transaction create many new | |
4708 | * block groups when starting to write out the free space caches of all | |
4709 | * the block groups that were made dirty during the lifetime of the | |
4710 | * transaction. | |
4711 | */ | |
d9a0540a | 4712 | if (trans->can_flush_pending_bgs && |
ee22184b | 4713 | trans->chunk_bytes_reserved >= (u64)SZ_2M) { |
6c686b35 | 4714 | btrfs_create_pending_block_groups(trans); |
00d80e34 FM |
4715 | btrfs_trans_release_chunk_metadata(trans); |
4716 | } | |
0f9dd46c | 4717 | return ret; |
6324fbf3 | 4718 | } |
9ed74f2d | 4719 | |
c1c4919b | 4720 | static int can_overcommit(struct btrfs_fs_info *fs_info, |
a80c8dcf | 4721 | struct btrfs_space_info *space_info, u64 bytes, |
c1c4919b JM |
4722 | enum btrfs_reserve_flush_enum flush, |
4723 | bool system_chunk) | |
a80c8dcf | 4724 | { |
0b246afa | 4725 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
957780eb | 4726 | u64 profile; |
3c76cd84 | 4727 | u64 space_size; |
a80c8dcf JB |
4728 | u64 avail; |
4729 | u64 used; | |
4730 | ||
957780eb JB |
4731 | /* Don't overcommit when in mixed mode. */ |
4732 | if (space_info->flags & BTRFS_BLOCK_GROUP_DATA) | |
4733 | return 0; | |
4734 | ||
c1c4919b JM |
4735 | if (system_chunk) |
4736 | profile = btrfs_system_alloc_profile(fs_info); | |
4737 | else | |
4738 | profile = btrfs_metadata_alloc_profile(fs_info); | |
4739 | ||
4136135b | 4740 | used = btrfs_space_info_used(space_info, false); |
96f1bb57 | 4741 | |
96f1bb57 JB |
4742 | /* |
4743 | * We only want to allow over committing if we have lots of actual space | |
4744 | * free, but if we don't have enough space to handle the global reserve | |
4745 | * space then we could end up having a real enospc problem when trying | |
4746 | * to allocate a chunk or some other such important allocation. | |
4747 | */ | |
3c76cd84 MX |
4748 | spin_lock(&global_rsv->lock); |
4749 | space_size = calc_global_rsv_need_space(global_rsv); | |
4750 | spin_unlock(&global_rsv->lock); | |
4751 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
4752 | return 0; |
4753 | ||
4754 | used += space_info->bytes_may_use; | |
a80c8dcf | 4755 | |
a5ed45f8 | 4756 | avail = atomic64_read(&fs_info->free_chunk_space); |
a80c8dcf JB |
4757 | |
4758 | /* | |
4759 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
4760 | * space is actually useable. For raid56, the space info used |
4761 | * doesn't include the parity drive, so we don't have to | |
4762 | * change the math | |
a80c8dcf JB |
4763 | */ |
4764 | if (profile & (BTRFS_BLOCK_GROUP_DUP | | |
4765 | BTRFS_BLOCK_GROUP_RAID1 | | |
4766 | BTRFS_BLOCK_GROUP_RAID10)) | |
4767 | avail >>= 1; | |
4768 | ||
4769 | /* | |
561c294d MX |
4770 | * If we aren't flushing all things, let us overcommit up to |
4771 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4772 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 4773 | */ |
08e007d2 | 4774 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 4775 | avail >>= 3; |
a80c8dcf | 4776 | else |
14575aef | 4777 | avail >>= 1; |
a80c8dcf | 4778 | |
14575aef | 4779 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
4780 | return 1; |
4781 | return 0; | |
4782 | } | |
4783 | ||
2ff7e61e | 4784 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info, |
6c255e67 | 4785 | unsigned long nr_pages, int nr_items) |
da633a42 | 4786 | { |
0b246afa | 4787 | struct super_block *sb = fs_info->sb; |
da633a42 | 4788 | |
925a6efb JB |
4789 | if (down_read_trylock(&sb->s_umount)) { |
4790 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4791 | up_read(&sb->s_umount); | |
4792 | } else { | |
da633a42 MX |
4793 | /* |
4794 | * We needn't worry the filesystem going from r/w to r/o though | |
4795 | * we don't acquire ->s_umount mutex, because the filesystem | |
4796 | * should guarantee the delalloc inodes list be empty after | |
4797 | * the filesystem is readonly(all dirty pages are written to | |
4798 | * the disk). | |
4799 | */ | |
0b246afa | 4800 | btrfs_start_delalloc_roots(fs_info, 0, nr_items); |
98ad69cf | 4801 | if (!current->journal_info) |
0b246afa | 4802 | btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1); |
da633a42 MX |
4803 | } |
4804 | } | |
4805 | ||
6374e57a | 4806 | static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info, |
2ff7e61e | 4807 | u64 to_reclaim) |
18cd8ea6 MX |
4808 | { |
4809 | u64 bytes; | |
6374e57a | 4810 | u64 nr; |
18cd8ea6 | 4811 | |
2ff7e61e | 4812 | bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
6374e57a | 4813 | nr = div64_u64(to_reclaim, bytes); |
18cd8ea6 MX |
4814 | if (!nr) |
4815 | nr = 1; | |
4816 | return nr; | |
4817 | } | |
4818 | ||
ee22184b | 4819 | #define EXTENT_SIZE_PER_ITEM SZ_256K |
c61a16a7 | 4820 | |
9ed74f2d | 4821 | /* |
5da9d01b | 4822 | * shrink metadata reservation for delalloc |
9ed74f2d | 4823 | */ |
c1c4919b JM |
4824 | static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim, |
4825 | u64 orig, bool wait_ordered) | |
5da9d01b | 4826 | { |
0019f10d | 4827 | struct btrfs_space_info *space_info; |
663350ac | 4828 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4829 | u64 delalloc_bytes; |
5da9d01b | 4830 | u64 max_reclaim; |
6374e57a | 4831 | u64 items; |
b1953bce | 4832 | long time_left; |
d3ee29e3 MX |
4833 | unsigned long nr_pages; |
4834 | int loops; | |
5da9d01b | 4835 | |
c61a16a7 | 4836 | /* Calc the number of the pages we need flush for space reservation */ |
2ff7e61e | 4837 | items = calc_reclaim_items_nr(fs_info, to_reclaim); |
6374e57a | 4838 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; |
c61a16a7 | 4839 | |
663350ac | 4840 | trans = (struct btrfs_trans_handle *)current->journal_info; |
69fe2d75 | 4841 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
bf9022e0 | 4842 | |
963d678b | 4843 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4844 | &fs_info->delalloc_bytes); |
f4c738c2 | 4845 | if (delalloc_bytes == 0) { |
fdb5effd | 4846 | if (trans) |
f4c738c2 | 4847 | return; |
38c135af | 4848 | if (wait_ordered) |
0b246afa | 4849 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f4c738c2 | 4850 | return; |
fdb5effd JB |
4851 | } |
4852 | ||
d3ee29e3 | 4853 | loops = 0; |
f4c738c2 JB |
4854 | while (delalloc_bytes && loops < 3) { |
4855 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
09cbfeaf | 4856 | nr_pages = max_reclaim >> PAGE_SHIFT; |
2ff7e61e | 4857 | btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items); |
dea31f52 JB |
4858 | /* |
4859 | * We need to wait for the async pages to actually start before | |
4860 | * we do anything. | |
4861 | */ | |
0b246afa | 4862 | max_reclaim = atomic_read(&fs_info->async_delalloc_pages); |
9f3a074d MX |
4863 | if (!max_reclaim) |
4864 | goto skip_async; | |
4865 | ||
4866 | if (max_reclaim <= nr_pages) | |
4867 | max_reclaim = 0; | |
4868 | else | |
4869 | max_reclaim -= nr_pages; | |
dea31f52 | 4870 | |
0b246afa JM |
4871 | wait_event(fs_info->async_submit_wait, |
4872 | atomic_read(&fs_info->async_delalloc_pages) <= | |
9f3a074d MX |
4873 | (int)max_reclaim); |
4874 | skip_async: | |
0019f10d | 4875 | spin_lock(&space_info->lock); |
957780eb JB |
4876 | if (list_empty(&space_info->tickets) && |
4877 | list_empty(&space_info->priority_tickets)) { | |
4878 | spin_unlock(&space_info->lock); | |
4879 | break; | |
4880 | } | |
0019f10d | 4881 | spin_unlock(&space_info->lock); |
5da9d01b | 4882 | |
36e39c40 | 4883 | loops++; |
f104d044 | 4884 | if (wait_ordered && !trans) { |
0b246afa | 4885 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f104d044 | 4886 | } else { |
f4c738c2 | 4887 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4888 | if (time_left) |
4889 | break; | |
4890 | } | |
963d678b | 4891 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4892 | &fs_info->delalloc_bytes); |
5da9d01b | 4893 | } |
5da9d01b YZ |
4894 | } |
4895 | ||
996478ca JB |
4896 | struct reserve_ticket { |
4897 | u64 bytes; | |
4898 | int error; | |
4899 | struct list_head list; | |
4900 | wait_queue_head_t wait; | |
4901 | }; | |
4902 | ||
663350ac JB |
4903 | /** |
4904 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4905 | * @root - the root we're allocating for | |
4906 | * @bytes - the number of bytes we want to reserve | |
4907 | * @force - force the commit | |
8bb8ab2e | 4908 | * |
663350ac JB |
4909 | * This will check to make sure that committing the transaction will actually |
4910 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4911 | * will return -ENOSPC. | |
8bb8ab2e | 4912 | */ |
0c9ab349 | 4913 | static int may_commit_transaction(struct btrfs_fs_info *fs_info, |
996478ca | 4914 | struct btrfs_space_info *space_info) |
663350ac | 4915 | { |
996478ca | 4916 | struct reserve_ticket *ticket = NULL; |
0b246afa | 4917 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv; |
663350ac | 4918 | struct btrfs_trans_handle *trans; |
996478ca | 4919 | u64 bytes; |
663350ac JB |
4920 | |
4921 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4922 | if (trans) | |
4923 | return -EAGAIN; | |
4924 | ||
996478ca JB |
4925 | spin_lock(&space_info->lock); |
4926 | if (!list_empty(&space_info->priority_tickets)) | |
4927 | ticket = list_first_entry(&space_info->priority_tickets, | |
4928 | struct reserve_ticket, list); | |
4929 | else if (!list_empty(&space_info->tickets)) | |
4930 | ticket = list_first_entry(&space_info->tickets, | |
4931 | struct reserve_ticket, list); | |
4932 | bytes = (ticket) ? ticket->bytes : 0; | |
4933 | spin_unlock(&space_info->lock); | |
4934 | ||
4935 | if (!bytes) | |
4936 | return 0; | |
663350ac JB |
4937 | |
4938 | /* See if there is enough pinned space to make this reservation */ | |
b150a4f1 | 4939 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
0424c548 | 4940 | bytes) >= 0) |
663350ac | 4941 | goto commit; |
663350ac JB |
4942 | |
4943 | /* | |
4944 | * See if there is some space in the delayed insertion reservation for | |
4945 | * this reservation. | |
4946 | */ | |
4947 | if (space_info != delayed_rsv->space_info) | |
4948 | return -ENOSPC; | |
4949 | ||
4950 | spin_lock(&delayed_rsv->lock); | |
996478ca JB |
4951 | if (delayed_rsv->size > bytes) |
4952 | bytes = 0; | |
4953 | else | |
4954 | bytes -= delayed_rsv->size; | |
057aac3e NB |
4955 | spin_unlock(&delayed_rsv->lock); |
4956 | ||
b150a4f1 | 4957 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
996478ca | 4958 | bytes) < 0) { |
663350ac JB |
4959 | return -ENOSPC; |
4960 | } | |
663350ac JB |
4961 | |
4962 | commit: | |
a9b3311e | 4963 | trans = btrfs_join_transaction(fs_info->extent_root); |
663350ac JB |
4964 | if (IS_ERR(trans)) |
4965 | return -ENOSPC; | |
4966 | ||
3a45bb20 | 4967 | return btrfs_commit_transaction(trans); |
663350ac JB |
4968 | } |
4969 | ||
e38ae7a0 NB |
4970 | /* |
4971 | * Try to flush some data based on policy set by @state. This is only advisory | |
4972 | * and may fail for various reasons. The caller is supposed to examine the | |
4973 | * state of @space_info to detect the outcome. | |
4974 | */ | |
4975 | static void flush_space(struct btrfs_fs_info *fs_info, | |
96c3f433 | 4976 | struct btrfs_space_info *space_info, u64 num_bytes, |
7bdd6277 | 4977 | int state) |
96c3f433 | 4978 | { |
a9b3311e | 4979 | struct btrfs_root *root = fs_info->extent_root; |
96c3f433 JB |
4980 | struct btrfs_trans_handle *trans; |
4981 | int nr; | |
f4c738c2 | 4982 | int ret = 0; |
96c3f433 JB |
4983 | |
4984 | switch (state) { | |
96c3f433 JB |
4985 | case FLUSH_DELAYED_ITEMS_NR: |
4986 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 | 4987 | if (state == FLUSH_DELAYED_ITEMS_NR) |
2ff7e61e | 4988 | nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2; |
18cd8ea6 | 4989 | else |
96c3f433 | 4990 | nr = -1; |
18cd8ea6 | 4991 | |
96c3f433 JB |
4992 | trans = btrfs_join_transaction(root); |
4993 | if (IS_ERR(trans)) { | |
4994 | ret = PTR_ERR(trans); | |
4995 | break; | |
4996 | } | |
e5c304e6 | 4997 | ret = btrfs_run_delayed_items_nr(trans, nr); |
3a45bb20 | 4998 | btrfs_end_transaction(trans); |
96c3f433 | 4999 | break; |
67b0fd63 JB |
5000 | case FLUSH_DELALLOC: |
5001 | case FLUSH_DELALLOC_WAIT: | |
7bdd6277 | 5002 | shrink_delalloc(fs_info, num_bytes * 2, num_bytes, |
67b0fd63 JB |
5003 | state == FLUSH_DELALLOC_WAIT); |
5004 | break; | |
ea658bad JB |
5005 | case ALLOC_CHUNK: |
5006 | trans = btrfs_join_transaction(root); | |
5007 | if (IS_ERR(trans)) { | |
5008 | ret = PTR_ERR(trans); | |
5009 | break; | |
5010 | } | |
2ff7e61e | 5011 | ret = do_chunk_alloc(trans, fs_info, |
1b86826d | 5012 | btrfs_metadata_alloc_profile(fs_info), |
ea658bad | 5013 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 5014 | btrfs_end_transaction(trans); |
eecba891 | 5015 | if (ret > 0 || ret == -ENOSPC) |
ea658bad JB |
5016 | ret = 0; |
5017 | break; | |
96c3f433 | 5018 | case COMMIT_TRANS: |
996478ca | 5019 | ret = may_commit_transaction(fs_info, space_info); |
96c3f433 JB |
5020 | break; |
5021 | default: | |
5022 | ret = -ENOSPC; | |
5023 | break; | |
5024 | } | |
5025 | ||
7bdd6277 NB |
5026 | trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state, |
5027 | ret); | |
e38ae7a0 | 5028 | return; |
96c3f433 | 5029 | } |
21c7e756 MX |
5030 | |
5031 | static inline u64 | |
c1c4919b JM |
5032 | btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info, |
5033 | struct btrfs_space_info *space_info, | |
5034 | bool system_chunk) | |
21c7e756 | 5035 | { |
957780eb | 5036 | struct reserve_ticket *ticket; |
21c7e756 MX |
5037 | u64 used; |
5038 | u64 expected; | |
957780eb | 5039 | u64 to_reclaim = 0; |
21c7e756 | 5040 | |
957780eb JB |
5041 | list_for_each_entry(ticket, &space_info->tickets, list) |
5042 | to_reclaim += ticket->bytes; | |
5043 | list_for_each_entry(ticket, &space_info->priority_tickets, list) | |
5044 | to_reclaim += ticket->bytes; | |
5045 | if (to_reclaim) | |
5046 | return to_reclaim; | |
21c7e756 | 5047 | |
e0af2484 | 5048 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); |
c1c4919b JM |
5049 | if (can_overcommit(fs_info, space_info, to_reclaim, |
5050 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
e0af2484 WX |
5051 | return 0; |
5052 | ||
0eee8a49 NB |
5053 | used = btrfs_space_info_used(space_info, true); |
5054 | ||
c1c4919b JM |
5055 | if (can_overcommit(fs_info, space_info, SZ_1M, |
5056 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
21c7e756 MX |
5057 | expected = div_factor_fine(space_info->total_bytes, 95); |
5058 | else | |
5059 | expected = div_factor_fine(space_info->total_bytes, 90); | |
5060 | ||
5061 | if (used > expected) | |
5062 | to_reclaim = used - expected; | |
5063 | else | |
5064 | to_reclaim = 0; | |
5065 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
5066 | space_info->bytes_reserved); | |
21c7e756 MX |
5067 | return to_reclaim; |
5068 | } | |
5069 | ||
c1c4919b JM |
5070 | static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info, |
5071 | struct btrfs_space_info *space_info, | |
5072 | u64 used, bool system_chunk) | |
21c7e756 | 5073 | { |
365c5313 JB |
5074 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
5075 | ||
5076 | /* If we're just plain full then async reclaim just slows us down. */ | |
baee8790 | 5077 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) |
365c5313 JB |
5078 | return 0; |
5079 | ||
c1c4919b JM |
5080 | if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5081 | system_chunk)) | |
d38b349c JB |
5082 | return 0; |
5083 | ||
0b246afa JM |
5084 | return (used >= thresh && !btrfs_fs_closing(fs_info) && |
5085 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); | |
21c7e756 MX |
5086 | } |
5087 | ||
957780eb | 5088 | static void wake_all_tickets(struct list_head *head) |
21c7e756 | 5089 | { |
957780eb | 5090 | struct reserve_ticket *ticket; |
25ce459c | 5091 | |
957780eb JB |
5092 | while (!list_empty(head)) { |
5093 | ticket = list_first_entry(head, struct reserve_ticket, list); | |
5094 | list_del_init(&ticket->list); | |
5095 | ticket->error = -ENOSPC; | |
5096 | wake_up(&ticket->wait); | |
21c7e756 | 5097 | } |
21c7e756 MX |
5098 | } |
5099 | ||
957780eb JB |
5100 | /* |
5101 | * This is for normal flushers, we can wait all goddamned day if we want to. We | |
5102 | * will loop and continuously try to flush as long as we are making progress. | |
5103 | * We count progress as clearing off tickets each time we have to loop. | |
5104 | */ | |
21c7e756 MX |
5105 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) |
5106 | { | |
5107 | struct btrfs_fs_info *fs_info; | |
5108 | struct btrfs_space_info *space_info; | |
5109 | u64 to_reclaim; | |
5110 | int flush_state; | |
957780eb | 5111 | int commit_cycles = 0; |
ce129655 | 5112 | u64 last_tickets_id; |
21c7e756 MX |
5113 | |
5114 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
5115 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
5116 | ||
957780eb | 5117 | spin_lock(&space_info->lock); |
c1c4919b JM |
5118 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5119 | false); | |
957780eb JB |
5120 | if (!to_reclaim) { |
5121 | space_info->flush = 0; | |
5122 | spin_unlock(&space_info->lock); | |
21c7e756 | 5123 | return; |
957780eb | 5124 | } |
ce129655 | 5125 | last_tickets_id = space_info->tickets_id; |
957780eb | 5126 | spin_unlock(&space_info->lock); |
21c7e756 MX |
5127 | |
5128 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
957780eb | 5129 | do { |
e38ae7a0 | 5130 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
957780eb JB |
5131 | spin_lock(&space_info->lock); |
5132 | if (list_empty(&space_info->tickets)) { | |
5133 | space_info->flush = 0; | |
5134 | spin_unlock(&space_info->lock); | |
5135 | return; | |
5136 | } | |
c1c4919b JM |
5137 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, |
5138 | space_info, | |
5139 | false); | |
ce129655 | 5140 | if (last_tickets_id == space_info->tickets_id) { |
957780eb JB |
5141 | flush_state++; |
5142 | } else { | |
ce129655 | 5143 | last_tickets_id = space_info->tickets_id; |
957780eb JB |
5144 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
5145 | if (commit_cycles) | |
5146 | commit_cycles--; | |
5147 | } | |
5148 | ||
5149 | if (flush_state > COMMIT_TRANS) { | |
5150 | commit_cycles++; | |
5151 | if (commit_cycles > 2) { | |
5152 | wake_all_tickets(&space_info->tickets); | |
5153 | space_info->flush = 0; | |
5154 | } else { | |
5155 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5156 | } | |
5157 | } | |
5158 | spin_unlock(&space_info->lock); | |
5159 | } while (flush_state <= COMMIT_TRANS); | |
5160 | } | |
5161 | ||
5162 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
5163 | { | |
5164 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
5165 | } | |
5166 | ||
5167 | static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, | |
5168 | struct btrfs_space_info *space_info, | |
5169 | struct reserve_ticket *ticket) | |
5170 | { | |
5171 | u64 to_reclaim; | |
5172 | int flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5173 | ||
5174 | spin_lock(&space_info->lock); | |
c1c4919b JM |
5175 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5176 | false); | |
957780eb JB |
5177 | if (!to_reclaim) { |
5178 | spin_unlock(&space_info->lock); | |
5179 | return; | |
5180 | } | |
5181 | spin_unlock(&space_info->lock); | |
5182 | ||
21c7e756 | 5183 | do { |
7bdd6277 | 5184 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
21c7e756 | 5185 | flush_state++; |
957780eb JB |
5186 | spin_lock(&space_info->lock); |
5187 | if (ticket->bytes == 0) { | |
5188 | spin_unlock(&space_info->lock); | |
21c7e756 | 5189 | return; |
957780eb JB |
5190 | } |
5191 | spin_unlock(&space_info->lock); | |
5192 | ||
5193 | /* | |
5194 | * Priority flushers can't wait on delalloc without | |
5195 | * deadlocking. | |
5196 | */ | |
5197 | if (flush_state == FLUSH_DELALLOC || | |
5198 | flush_state == FLUSH_DELALLOC_WAIT) | |
5199 | flush_state = ALLOC_CHUNK; | |
365c5313 | 5200 | } while (flush_state < COMMIT_TRANS); |
21c7e756 MX |
5201 | } |
5202 | ||
957780eb JB |
5203 | static int wait_reserve_ticket(struct btrfs_fs_info *fs_info, |
5204 | struct btrfs_space_info *space_info, | |
5205 | struct reserve_ticket *ticket, u64 orig_bytes) | |
5206 | ||
21c7e756 | 5207 | { |
957780eb JB |
5208 | DEFINE_WAIT(wait); |
5209 | int ret = 0; | |
5210 | ||
5211 | spin_lock(&space_info->lock); | |
5212 | while (ticket->bytes > 0 && ticket->error == 0) { | |
5213 | ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE); | |
5214 | if (ret) { | |
5215 | ret = -EINTR; | |
5216 | break; | |
5217 | } | |
5218 | spin_unlock(&space_info->lock); | |
5219 | ||
5220 | schedule(); | |
5221 | ||
5222 | finish_wait(&ticket->wait, &wait); | |
5223 | spin_lock(&space_info->lock); | |
5224 | } | |
5225 | if (!ret) | |
5226 | ret = ticket->error; | |
5227 | if (!list_empty(&ticket->list)) | |
5228 | list_del_init(&ticket->list); | |
5229 | if (ticket->bytes && ticket->bytes < orig_bytes) { | |
5230 | u64 num_bytes = orig_bytes - ticket->bytes; | |
5231 | space_info->bytes_may_use -= num_bytes; | |
5232 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5233 | space_info->flags, num_bytes, 0); | |
5234 | } | |
5235 | spin_unlock(&space_info->lock); | |
5236 | ||
5237 | return ret; | |
21c7e756 MX |
5238 | } |
5239 | ||
4a92b1b8 JB |
5240 | /** |
5241 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5242 | * @root - the root we're allocating for | |
957780eb | 5243 | * @space_info - the space info we want to allocate from |
4a92b1b8 | 5244 | * @orig_bytes - the number of bytes we want |
48fc7f7e | 5245 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 5246 | * |
01327610 | 5247 | * This will reserve orig_bytes number of bytes from the space info associated |
4a92b1b8 JB |
5248 | * with the block_rsv. If there is not enough space it will make an attempt to |
5249 | * flush out space to make room. It will do this by flushing delalloc if | |
5250 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5251 | * regain reservations will be made and this will fail if there is not enough | |
5252 | * space already. | |
8bb8ab2e | 5253 | */ |
c1c4919b | 5254 | static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info, |
957780eb JB |
5255 | struct btrfs_space_info *space_info, |
5256 | u64 orig_bytes, | |
c1c4919b JM |
5257 | enum btrfs_reserve_flush_enum flush, |
5258 | bool system_chunk) | |
9ed74f2d | 5259 | { |
957780eb | 5260 | struct reserve_ticket ticket; |
2bf64758 | 5261 | u64 used; |
8bb8ab2e | 5262 | int ret = 0; |
9ed74f2d | 5263 | |
957780eb | 5264 | ASSERT(orig_bytes); |
8ca17f0f | 5265 | ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL); |
fdb5effd | 5266 | |
8bb8ab2e | 5267 | spin_lock(&space_info->lock); |
fdb5effd | 5268 | ret = -ENOSPC; |
4136135b | 5269 | used = btrfs_space_info_used(space_info, true); |
9ed74f2d | 5270 | |
8bb8ab2e | 5271 | /* |
957780eb JB |
5272 | * If we have enough space then hooray, make our reservation and carry |
5273 | * on. If not see if we can overcommit, and if we can, hooray carry on. | |
5274 | * If not things get more complicated. | |
8bb8ab2e | 5275 | */ |
957780eb JB |
5276 | if (used + orig_bytes <= space_info->total_bytes) { |
5277 | space_info->bytes_may_use += orig_bytes; | |
0b246afa JM |
5278 | trace_btrfs_space_reservation(fs_info, "space_info", |
5279 | space_info->flags, orig_bytes, 1); | |
957780eb | 5280 | ret = 0; |
c1c4919b JM |
5281 | } else if (can_overcommit(fs_info, space_info, orig_bytes, flush, |
5282 | system_chunk)) { | |
44734ed1 | 5283 | space_info->bytes_may_use += orig_bytes; |
0b246afa JM |
5284 | trace_btrfs_space_reservation(fs_info, "space_info", |
5285 | space_info->flags, orig_bytes, 1); | |
44734ed1 | 5286 | ret = 0; |
2bf64758 JB |
5287 | } |
5288 | ||
8bb8ab2e | 5289 | /* |
957780eb JB |
5290 | * If we couldn't make a reservation then setup our reservation ticket |
5291 | * and kick the async worker if it's not already running. | |
08e007d2 | 5292 | * |
957780eb JB |
5293 | * If we are a priority flusher then we just need to add our ticket to |
5294 | * the list and we will do our own flushing further down. | |
8bb8ab2e | 5295 | */ |
72bcd99d | 5296 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
957780eb JB |
5297 | ticket.bytes = orig_bytes; |
5298 | ticket.error = 0; | |
5299 | init_waitqueue_head(&ticket.wait); | |
5300 | if (flush == BTRFS_RESERVE_FLUSH_ALL) { | |
5301 | list_add_tail(&ticket.list, &space_info->tickets); | |
5302 | if (!space_info->flush) { | |
5303 | space_info->flush = 1; | |
0b246afa | 5304 | trace_btrfs_trigger_flush(fs_info, |
f376df2b JB |
5305 | space_info->flags, |
5306 | orig_bytes, flush, | |
5307 | "enospc"); | |
957780eb | 5308 | queue_work(system_unbound_wq, |
c1c4919b | 5309 | &fs_info->async_reclaim_work); |
957780eb JB |
5310 | } |
5311 | } else { | |
5312 | list_add_tail(&ticket.list, | |
5313 | &space_info->priority_tickets); | |
5314 | } | |
21c7e756 MX |
5315 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
5316 | used += orig_bytes; | |
f6acfd50 JB |
5317 | /* |
5318 | * We will do the space reservation dance during log replay, | |
5319 | * which means we won't have fs_info->fs_root set, so don't do | |
5320 | * the async reclaim as we will panic. | |
5321 | */ | |
0b246afa | 5322 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) && |
c1c4919b JM |
5323 | need_do_async_reclaim(fs_info, space_info, |
5324 | used, system_chunk) && | |
0b246afa JM |
5325 | !work_busy(&fs_info->async_reclaim_work)) { |
5326 | trace_btrfs_trigger_flush(fs_info, space_info->flags, | |
5327 | orig_bytes, flush, "preempt"); | |
21c7e756 | 5328 | queue_work(system_unbound_wq, |
0b246afa | 5329 | &fs_info->async_reclaim_work); |
f376df2b | 5330 | } |
8bb8ab2e | 5331 | } |
f0486c68 | 5332 | spin_unlock(&space_info->lock); |
08e007d2 | 5333 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
957780eb | 5334 | return ret; |
f0486c68 | 5335 | |
957780eb | 5336 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
0b246afa | 5337 | return wait_reserve_ticket(fs_info, space_info, &ticket, |
957780eb | 5338 | orig_bytes); |
08e007d2 | 5339 | |
957780eb | 5340 | ret = 0; |
0b246afa | 5341 | priority_reclaim_metadata_space(fs_info, space_info, &ticket); |
957780eb JB |
5342 | spin_lock(&space_info->lock); |
5343 | if (ticket.bytes) { | |
5344 | if (ticket.bytes < orig_bytes) { | |
5345 | u64 num_bytes = orig_bytes - ticket.bytes; | |
5346 | space_info->bytes_may_use -= num_bytes; | |
0b246afa JM |
5347 | trace_btrfs_space_reservation(fs_info, "space_info", |
5348 | space_info->flags, | |
5349 | num_bytes, 0); | |
08e007d2 | 5350 | |
957780eb JB |
5351 | } |
5352 | list_del_init(&ticket.list); | |
5353 | ret = -ENOSPC; | |
5354 | } | |
5355 | spin_unlock(&space_info->lock); | |
5356 | ASSERT(list_empty(&ticket.list)); | |
5357 | return ret; | |
5358 | } | |
8bb8ab2e | 5359 | |
957780eb JB |
5360 | /** |
5361 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5362 | * @root - the root we're allocating for | |
5363 | * @block_rsv - the block_rsv we're allocating for | |
5364 | * @orig_bytes - the number of bytes we want | |
5365 | * @flush - whether or not we can flush to make our reservation | |
5366 | * | |
5367 | * This will reserve orgi_bytes number of bytes from the space info associated | |
5368 | * with the block_rsv. If there is not enough space it will make an attempt to | |
5369 | * flush out space to make room. It will do this by flushing delalloc if | |
5370 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5371 | * regain reservations will be made and this will fail if there is not enough | |
5372 | * space already. | |
5373 | */ | |
5374 | static int reserve_metadata_bytes(struct btrfs_root *root, | |
5375 | struct btrfs_block_rsv *block_rsv, | |
5376 | u64 orig_bytes, | |
5377 | enum btrfs_reserve_flush_enum flush) | |
5378 | { | |
0b246afa JM |
5379 | struct btrfs_fs_info *fs_info = root->fs_info; |
5380 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
957780eb | 5381 | int ret; |
c1c4919b | 5382 | bool system_chunk = (root == fs_info->chunk_root); |
957780eb | 5383 | |
c1c4919b JM |
5384 | ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info, |
5385 | orig_bytes, flush, system_chunk); | |
5d80366e JB |
5386 | if (ret == -ENOSPC && |
5387 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5d80366e JB |
5388 | if (block_rsv != global_rsv && |
5389 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5390 | ret = 0; | |
5391 | } | |
9a3daff3 | 5392 | if (ret == -ENOSPC) { |
0b246afa | 5393 | trace_btrfs_space_reservation(fs_info, "space_info:enospc", |
957780eb JB |
5394 | block_rsv->space_info->flags, |
5395 | orig_bytes, 1); | |
9a3daff3 NB |
5396 | |
5397 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) | |
5398 | dump_space_info(fs_info, block_rsv->space_info, | |
5399 | orig_bytes, 0); | |
5400 | } | |
f0486c68 YZ |
5401 | return ret; |
5402 | } | |
5403 | ||
79787eaa JM |
5404 | static struct btrfs_block_rsv *get_block_rsv( |
5405 | const struct btrfs_trans_handle *trans, | |
5406 | const struct btrfs_root *root) | |
f0486c68 | 5407 | { |
0b246afa | 5408 | struct btrfs_fs_info *fs_info = root->fs_info; |
4c13d758 JB |
5409 | struct btrfs_block_rsv *block_rsv = NULL; |
5410 | ||
e9cf439f | 5411 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa JM |
5412 | (root == fs_info->csum_root && trans->adding_csums) || |
5413 | (root == fs_info->uuid_root)) | |
f7a81ea4 SB |
5414 | block_rsv = trans->block_rsv; |
5415 | ||
4c13d758 | 5416 | if (!block_rsv) |
f0486c68 YZ |
5417 | block_rsv = root->block_rsv; |
5418 | ||
5419 | if (!block_rsv) | |
0b246afa | 5420 | block_rsv = &fs_info->empty_block_rsv; |
f0486c68 YZ |
5421 | |
5422 | return block_rsv; | |
5423 | } | |
5424 | ||
5425 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
5426 | u64 num_bytes) | |
5427 | { | |
5428 | int ret = -ENOSPC; | |
5429 | spin_lock(&block_rsv->lock); | |
5430 | if (block_rsv->reserved >= num_bytes) { | |
5431 | block_rsv->reserved -= num_bytes; | |
5432 | if (block_rsv->reserved < block_rsv->size) | |
5433 | block_rsv->full = 0; | |
5434 | ret = 0; | |
5435 | } | |
5436 | spin_unlock(&block_rsv->lock); | |
5437 | return ret; | |
5438 | } | |
5439 | ||
5440 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
5441 | u64 num_bytes, int update_size) | |
5442 | { | |
5443 | spin_lock(&block_rsv->lock); | |
5444 | block_rsv->reserved += num_bytes; | |
5445 | if (update_size) | |
5446 | block_rsv->size += num_bytes; | |
5447 | else if (block_rsv->reserved >= block_rsv->size) | |
5448 | block_rsv->full = 1; | |
5449 | spin_unlock(&block_rsv->lock); | |
5450 | } | |
5451 | ||
d52be818 JB |
5452 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
5453 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5454 | int min_factor) | |
5455 | { | |
5456 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5457 | u64 min_bytes; | |
5458 | ||
5459 | if (global_rsv->space_info != dest->space_info) | |
5460 | return -ENOSPC; | |
5461 | ||
5462 | spin_lock(&global_rsv->lock); | |
5463 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5464 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5465 | spin_unlock(&global_rsv->lock); | |
5466 | return -ENOSPC; | |
5467 | } | |
5468 | global_rsv->reserved -= num_bytes; | |
5469 | if (global_rsv->reserved < global_rsv->size) | |
5470 | global_rsv->full = 0; | |
5471 | spin_unlock(&global_rsv->lock); | |
5472 | ||
5473 | block_rsv_add_bytes(dest, num_bytes, 1); | |
5474 | return 0; | |
5475 | } | |
5476 | ||
957780eb JB |
5477 | /* |
5478 | * This is for space we already have accounted in space_info->bytes_may_use, so | |
5479 | * basically when we're returning space from block_rsv's. | |
5480 | */ | |
5481 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
5482 | struct btrfs_space_info *space_info, | |
5483 | u64 num_bytes) | |
5484 | { | |
5485 | struct reserve_ticket *ticket; | |
5486 | struct list_head *head; | |
5487 | u64 used; | |
5488 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH; | |
5489 | bool check_overcommit = false; | |
5490 | ||
5491 | spin_lock(&space_info->lock); | |
5492 | head = &space_info->priority_tickets; | |
5493 | ||
5494 | /* | |
5495 | * If we are over our limit then we need to check and see if we can | |
5496 | * overcommit, and if we can't then we just need to free up our space | |
5497 | * and not satisfy any requests. | |
5498 | */ | |
0eee8a49 | 5499 | used = btrfs_space_info_used(space_info, true); |
957780eb JB |
5500 | if (used - num_bytes >= space_info->total_bytes) |
5501 | check_overcommit = true; | |
5502 | again: | |
5503 | while (!list_empty(head) && num_bytes) { | |
5504 | ticket = list_first_entry(head, struct reserve_ticket, | |
5505 | list); | |
5506 | /* | |
5507 | * We use 0 bytes because this space is already reserved, so | |
5508 | * adding the ticket space would be a double count. | |
5509 | */ | |
5510 | if (check_overcommit && | |
c1c4919b | 5511 | !can_overcommit(fs_info, space_info, 0, flush, false)) |
957780eb JB |
5512 | break; |
5513 | if (num_bytes >= ticket->bytes) { | |
5514 | list_del_init(&ticket->list); | |
5515 | num_bytes -= ticket->bytes; | |
5516 | ticket->bytes = 0; | |
ce129655 | 5517 | space_info->tickets_id++; |
957780eb JB |
5518 | wake_up(&ticket->wait); |
5519 | } else { | |
5520 | ticket->bytes -= num_bytes; | |
5521 | num_bytes = 0; | |
5522 | } | |
5523 | } | |
5524 | ||
5525 | if (num_bytes && head == &space_info->priority_tickets) { | |
5526 | head = &space_info->tickets; | |
5527 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
5528 | goto again; | |
5529 | } | |
5530 | space_info->bytes_may_use -= num_bytes; | |
5531 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5532 | space_info->flags, num_bytes, 0); | |
5533 | spin_unlock(&space_info->lock); | |
5534 | } | |
5535 | ||
5536 | /* | |
5537 | * This is for newly allocated space that isn't accounted in | |
5538 | * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent | |
5539 | * we use this helper. | |
5540 | */ | |
5541 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, | |
5542 | struct btrfs_space_info *space_info, | |
5543 | u64 num_bytes) | |
5544 | { | |
5545 | struct reserve_ticket *ticket; | |
5546 | struct list_head *head = &space_info->priority_tickets; | |
5547 | ||
5548 | again: | |
5549 | while (!list_empty(head) && num_bytes) { | |
5550 | ticket = list_first_entry(head, struct reserve_ticket, | |
5551 | list); | |
5552 | if (num_bytes >= ticket->bytes) { | |
5553 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5554 | space_info->flags, | |
5555 | ticket->bytes, 1); | |
5556 | list_del_init(&ticket->list); | |
5557 | num_bytes -= ticket->bytes; | |
5558 | space_info->bytes_may_use += ticket->bytes; | |
5559 | ticket->bytes = 0; | |
ce129655 | 5560 | space_info->tickets_id++; |
957780eb JB |
5561 | wake_up(&ticket->wait); |
5562 | } else { | |
5563 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5564 | space_info->flags, | |
5565 | num_bytes, 1); | |
5566 | space_info->bytes_may_use += num_bytes; | |
5567 | ticket->bytes -= num_bytes; | |
5568 | num_bytes = 0; | |
5569 | } | |
5570 | } | |
5571 | ||
5572 | if (num_bytes && head == &space_info->priority_tickets) { | |
5573 | head = &space_info->tickets; | |
5574 | goto again; | |
5575 | } | |
5576 | } | |
5577 | ||
69fe2d75 | 5578 | static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
8c2a3ca2 | 5579 | struct btrfs_block_rsv *block_rsv, |
ff6bc37e QW |
5580 | struct btrfs_block_rsv *dest, u64 num_bytes, |
5581 | u64 *qgroup_to_release_ret) | |
f0486c68 YZ |
5582 | { |
5583 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
ff6bc37e | 5584 | u64 qgroup_to_release = 0; |
69fe2d75 | 5585 | u64 ret; |
f0486c68 YZ |
5586 | |
5587 | spin_lock(&block_rsv->lock); | |
ff6bc37e | 5588 | if (num_bytes == (u64)-1) { |
f0486c68 | 5589 | num_bytes = block_rsv->size; |
ff6bc37e QW |
5590 | qgroup_to_release = block_rsv->qgroup_rsv_size; |
5591 | } | |
f0486c68 YZ |
5592 | block_rsv->size -= num_bytes; |
5593 | if (block_rsv->reserved >= block_rsv->size) { | |
5594 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5595 | block_rsv->reserved = block_rsv->size; | |
5596 | block_rsv->full = 1; | |
5597 | } else { | |
5598 | num_bytes = 0; | |
5599 | } | |
ff6bc37e QW |
5600 | if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) { |
5601 | qgroup_to_release = block_rsv->qgroup_rsv_reserved - | |
5602 | block_rsv->qgroup_rsv_size; | |
5603 | block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size; | |
5604 | } else { | |
5605 | qgroup_to_release = 0; | |
5606 | } | |
f0486c68 YZ |
5607 | spin_unlock(&block_rsv->lock); |
5608 | ||
69fe2d75 | 5609 | ret = num_bytes; |
f0486c68 YZ |
5610 | if (num_bytes > 0) { |
5611 | if (dest) { | |
e9e22899 JB |
5612 | spin_lock(&dest->lock); |
5613 | if (!dest->full) { | |
5614 | u64 bytes_to_add; | |
5615 | ||
5616 | bytes_to_add = dest->size - dest->reserved; | |
5617 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5618 | dest->reserved += bytes_to_add; | |
5619 | if (dest->reserved >= dest->size) | |
5620 | dest->full = 1; | |
5621 | num_bytes -= bytes_to_add; | |
5622 | } | |
5623 | spin_unlock(&dest->lock); | |
5624 | } | |
957780eb JB |
5625 | if (num_bytes) |
5626 | space_info_add_old_bytes(fs_info, space_info, | |
5627 | num_bytes); | |
9ed74f2d | 5628 | } |
ff6bc37e QW |
5629 | if (qgroup_to_release_ret) |
5630 | *qgroup_to_release_ret = qgroup_to_release; | |
69fe2d75 | 5631 | return ret; |
f0486c68 | 5632 | } |
4e06bdd6 | 5633 | |
25d609f8 JB |
5634 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src, |
5635 | struct btrfs_block_rsv *dst, u64 num_bytes, | |
5636 | int update_size) | |
f0486c68 YZ |
5637 | { |
5638 | int ret; | |
9ed74f2d | 5639 | |
f0486c68 YZ |
5640 | ret = block_rsv_use_bytes(src, num_bytes); |
5641 | if (ret) | |
5642 | return ret; | |
9ed74f2d | 5643 | |
25d609f8 | 5644 | block_rsv_add_bytes(dst, num_bytes, update_size); |
9ed74f2d JB |
5645 | return 0; |
5646 | } | |
5647 | ||
66d8f3dd | 5648 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 5649 | { |
f0486c68 YZ |
5650 | memset(rsv, 0, sizeof(*rsv)); |
5651 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 5652 | rsv->type = type; |
f0486c68 YZ |
5653 | } |
5654 | ||
69fe2d75 JB |
5655 | void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info, |
5656 | struct btrfs_block_rsv *rsv, | |
5657 | unsigned short type) | |
5658 | { | |
5659 | btrfs_init_block_rsv(rsv, type); | |
5660 | rsv->space_info = __find_space_info(fs_info, | |
5661 | BTRFS_BLOCK_GROUP_METADATA); | |
5662 | } | |
5663 | ||
2ff7e61e | 5664 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info, |
66d8f3dd | 5665 | unsigned short type) |
f0486c68 YZ |
5666 | { |
5667 | struct btrfs_block_rsv *block_rsv; | |
9ed74f2d | 5668 | |
f0486c68 YZ |
5669 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
5670 | if (!block_rsv) | |
5671 | return NULL; | |
9ed74f2d | 5672 | |
69fe2d75 | 5673 | btrfs_init_metadata_block_rsv(fs_info, block_rsv, type); |
f0486c68 YZ |
5674 | return block_rsv; |
5675 | } | |
9ed74f2d | 5676 | |
2ff7e61e | 5677 | void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5678 | struct btrfs_block_rsv *rsv) |
5679 | { | |
2aaa6655 JB |
5680 | if (!rsv) |
5681 | return; | |
2ff7e61e | 5682 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
dabdb640 | 5683 | kfree(rsv); |
9ed74f2d JB |
5684 | } |
5685 | ||
cdfb080e CM |
5686 | void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv) |
5687 | { | |
5688 | kfree(rsv); | |
5689 | } | |
5690 | ||
08e007d2 MX |
5691 | int btrfs_block_rsv_add(struct btrfs_root *root, |
5692 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5693 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5694 | { |
f0486c68 | 5695 | int ret; |
9ed74f2d | 5696 | |
f0486c68 YZ |
5697 | if (num_bytes == 0) |
5698 | return 0; | |
8bb8ab2e | 5699 | |
61b520a9 | 5700 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
f0486c68 YZ |
5701 | if (!ret) { |
5702 | block_rsv_add_bytes(block_rsv, num_bytes, 1); | |
5703 | return 0; | |
5704 | } | |
9ed74f2d | 5705 | |
f0486c68 | 5706 | return ret; |
f0486c68 | 5707 | } |
9ed74f2d | 5708 | |
2ff7e61e | 5709 | int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
5710 | { |
5711 | u64 num_bytes = 0; | |
f0486c68 | 5712 | int ret = -ENOSPC; |
9ed74f2d | 5713 | |
f0486c68 YZ |
5714 | if (!block_rsv) |
5715 | return 0; | |
9ed74f2d | 5716 | |
f0486c68 | 5717 | spin_lock(&block_rsv->lock); |
36ba022a JB |
5718 | num_bytes = div_factor(block_rsv->size, min_factor); |
5719 | if (block_rsv->reserved >= num_bytes) | |
5720 | ret = 0; | |
5721 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 5722 | |
36ba022a JB |
5723 | return ret; |
5724 | } | |
5725 | ||
08e007d2 MX |
5726 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
5727 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5728 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
5729 | { |
5730 | u64 num_bytes = 0; | |
5731 | int ret = -ENOSPC; | |
5732 | ||
5733 | if (!block_rsv) | |
5734 | return 0; | |
5735 | ||
5736 | spin_lock(&block_rsv->lock); | |
5737 | num_bytes = min_reserved; | |
13553e52 | 5738 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 5739 | ret = 0; |
13553e52 | 5740 | else |
f0486c68 | 5741 | num_bytes -= block_rsv->reserved; |
f0486c68 | 5742 | spin_unlock(&block_rsv->lock); |
13553e52 | 5743 | |
f0486c68 YZ |
5744 | if (!ret) |
5745 | return 0; | |
5746 | ||
aa38a711 | 5747 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 JB |
5748 | if (!ret) { |
5749 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
f0486c68 | 5750 | return 0; |
6a63209f | 5751 | } |
9ed74f2d | 5752 | |
13553e52 | 5753 | return ret; |
f0486c68 YZ |
5754 | } |
5755 | ||
69fe2d75 JB |
5756 | /** |
5757 | * btrfs_inode_rsv_refill - refill the inode block rsv. | |
5758 | * @inode - the inode we are refilling. | |
5759 | * @flush - the flusing restriction. | |
5760 | * | |
5761 | * Essentially the same as btrfs_block_rsv_refill, except it uses the | |
5762 | * block_rsv->size as the minimum size. We'll either refill the missing amount | |
5763 | * or return if we already have enough space. This will also handle the resreve | |
5764 | * tracepoint for the reserved amount. | |
5765 | */ | |
3f2dd7a0 QW |
5766 | static int btrfs_inode_rsv_refill(struct btrfs_inode *inode, |
5767 | enum btrfs_reserve_flush_enum flush) | |
69fe2d75 JB |
5768 | { |
5769 | struct btrfs_root *root = inode->root; | |
5770 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; | |
5771 | u64 num_bytes = 0; | |
ff6bc37e | 5772 | u64 qgroup_num_bytes = 0; |
69fe2d75 JB |
5773 | int ret = -ENOSPC; |
5774 | ||
5775 | spin_lock(&block_rsv->lock); | |
5776 | if (block_rsv->reserved < block_rsv->size) | |
5777 | num_bytes = block_rsv->size - block_rsv->reserved; | |
ff6bc37e QW |
5778 | if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size) |
5779 | qgroup_num_bytes = block_rsv->qgroup_rsv_size - | |
5780 | block_rsv->qgroup_rsv_reserved; | |
69fe2d75 JB |
5781 | spin_unlock(&block_rsv->lock); |
5782 | ||
5783 | if (num_bytes == 0) | |
5784 | return 0; | |
5785 | ||
ff6bc37e | 5786 | ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes, true); |
43b18595 QW |
5787 | if (ret) |
5788 | return ret; | |
69fe2d75 JB |
5789 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
5790 | if (!ret) { | |
5791 | block_rsv_add_bytes(block_rsv, num_bytes, 0); | |
5792 | trace_btrfs_space_reservation(root->fs_info, "delalloc", | |
5793 | btrfs_ino(inode), num_bytes, 1); | |
ff6bc37e QW |
5794 | |
5795 | /* Don't forget to increase qgroup_rsv_reserved */ | |
5796 | spin_lock(&block_rsv->lock); | |
5797 | block_rsv->qgroup_rsv_reserved += qgroup_num_bytes; | |
5798 | spin_unlock(&block_rsv->lock); | |
5799 | } else | |
5800 | btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes); | |
69fe2d75 JB |
5801 | return ret; |
5802 | } | |
5803 | ||
5804 | /** | |
5805 | * btrfs_inode_rsv_release - release any excessive reservation. | |
5806 | * @inode - the inode we need to release from. | |
43b18595 QW |
5807 | * @qgroup_free - free or convert qgroup meta. |
5808 | * Unlike normal operation, qgroup meta reservation needs to know if we are | |
5809 | * freeing qgroup reservation or just converting it into per-trans. Normally | |
5810 | * @qgroup_free is true for error handling, and false for normal release. | |
69fe2d75 JB |
5811 | * |
5812 | * This is the same as btrfs_block_rsv_release, except that it handles the | |
5813 | * tracepoint for the reservation. | |
5814 | */ | |
43b18595 | 5815 | static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free) |
69fe2d75 JB |
5816 | { |
5817 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
5818 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5819 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; | |
5820 | u64 released = 0; | |
ff6bc37e | 5821 | u64 qgroup_to_release = 0; |
69fe2d75 JB |
5822 | |
5823 | /* | |
5824 | * Since we statically set the block_rsv->size we just want to say we | |
5825 | * are releasing 0 bytes, and then we'll just get the reservation over | |
5826 | * the size free'd. | |
5827 | */ | |
ff6bc37e QW |
5828 | released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0, |
5829 | &qgroup_to_release); | |
69fe2d75 JB |
5830 | if (released > 0) |
5831 | trace_btrfs_space_reservation(fs_info, "delalloc", | |
5832 | btrfs_ino(inode), released, 0); | |
43b18595 | 5833 | if (qgroup_free) |
ff6bc37e | 5834 | btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release); |
43b18595 | 5835 | else |
ff6bc37e QW |
5836 | btrfs_qgroup_convert_reserved_meta(inode->root, |
5837 | qgroup_to_release); | |
69fe2d75 JB |
5838 | } |
5839 | ||
2ff7e61e | 5840 | void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5841 | struct btrfs_block_rsv *block_rsv, |
5842 | u64 num_bytes) | |
5843 | { | |
0b246afa JM |
5844 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
5845 | ||
17504584 | 5846 | if (global_rsv == block_rsv || |
f0486c68 YZ |
5847 | block_rsv->space_info != global_rsv->space_info) |
5848 | global_rsv = NULL; | |
ff6bc37e | 5849 | block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes, NULL); |
6a63209f JB |
5850 | } |
5851 | ||
8929ecfa YZ |
5852 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5853 | { | |
5854 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5855 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5856 | u64 num_bytes; | |
6a63209f | 5857 | |
ae2e4728 JB |
5858 | /* |
5859 | * The global block rsv is based on the size of the extent tree, the | |
5860 | * checksum tree and the root tree. If the fs is empty we want to set | |
5861 | * it to a minimal amount for safety. | |
5862 | */ | |
5863 | num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) + | |
5864 | btrfs_root_used(&fs_info->csum_root->root_item) + | |
5865 | btrfs_root_used(&fs_info->tree_root->root_item); | |
5866 | num_bytes = max_t(u64, num_bytes, SZ_16M); | |
33b4d47f | 5867 | |
8929ecfa | 5868 | spin_lock(&sinfo->lock); |
1f699d38 | 5869 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5870 | |
ee22184b | 5871 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); |
4e06bdd6 | 5872 | |
fb4b10e5 | 5873 | if (block_rsv->reserved < block_rsv->size) { |
4136135b | 5874 | num_bytes = btrfs_space_info_used(sinfo, true); |
fb4b10e5 JB |
5875 | if (sinfo->total_bytes > num_bytes) { |
5876 | num_bytes = sinfo->total_bytes - num_bytes; | |
5877 | num_bytes = min(num_bytes, | |
5878 | block_rsv->size - block_rsv->reserved); | |
5879 | block_rsv->reserved += num_bytes; | |
5880 | sinfo->bytes_may_use += num_bytes; | |
5881 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5882 | sinfo->flags, num_bytes, | |
5883 | 1); | |
5884 | } | |
5885 | } else if (block_rsv->reserved > block_rsv->size) { | |
8929ecfa | 5886 | num_bytes = block_rsv->reserved - block_rsv->size; |
fb25e914 | 5887 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 5888 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5889 | sinfo->flags, num_bytes, 0); |
8929ecfa | 5890 | block_rsv->reserved = block_rsv->size; |
8929ecfa | 5891 | } |
182608c8 | 5892 | |
fb4b10e5 JB |
5893 | if (block_rsv->reserved == block_rsv->size) |
5894 | block_rsv->full = 1; | |
5895 | else | |
5896 | block_rsv->full = 0; | |
5897 | ||
8929ecfa | 5898 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5899 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5900 | } |
5901 | ||
f0486c68 | 5902 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5903 | { |
f0486c68 | 5904 | struct btrfs_space_info *space_info; |
6a63209f | 5905 | |
f0486c68 YZ |
5906 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
5907 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 5908 | |
f0486c68 | 5909 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5910 | fs_info->global_block_rsv.space_info = space_info; |
f0486c68 YZ |
5911 | fs_info->trans_block_rsv.space_info = space_info; |
5912 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5913 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 5914 | |
8929ecfa YZ |
5915 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
5916 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
5917 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
5918 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5919 | if (fs_info->quota_root) |
5920 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5921 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5922 | |
8929ecfa | 5923 | update_global_block_rsv(fs_info); |
6a63209f JB |
5924 | } |
5925 | ||
8929ecfa | 5926 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5927 | { |
8c2a3ca2 | 5928 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
ff6bc37e | 5929 | (u64)-1, NULL); |
8929ecfa YZ |
5930 | WARN_ON(fs_info->trans_block_rsv.size > 0); |
5931 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5932 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5933 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5934 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5935 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
5936 | } |
5937 | ||
6a63209f | 5938 | |
4fbcdf66 FM |
5939 | /* |
5940 | * To be called after all the new block groups attached to the transaction | |
5941 | * handle have been created (btrfs_create_pending_block_groups()). | |
5942 | */ | |
5943 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5944 | { | |
64b63580 | 5945 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4fbcdf66 FM |
5946 | |
5947 | if (!trans->chunk_bytes_reserved) | |
5948 | return; | |
5949 | ||
5950 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5951 | ||
5952 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
ff6bc37e | 5953 | trans->chunk_bytes_reserved, NULL); |
4fbcdf66 FM |
5954 | trans->chunk_bytes_reserved = 0; |
5955 | } | |
5956 | ||
79787eaa | 5957 | /* Can only return 0 or -ENOSPC */ |
d68fc57b | 5958 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
8ed7a2a0 | 5959 | struct btrfs_inode *inode) |
d68fc57b | 5960 | { |
8ed7a2a0 NB |
5961 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
5962 | struct btrfs_root *root = inode->root; | |
40acc3ee JB |
5963 | /* |
5964 | * We always use trans->block_rsv here as we will have reserved space | |
5965 | * for our orphan when starting the transaction, using get_block_rsv() | |
5966 | * here will sometimes make us choose the wrong block rsv as we could be | |
5967 | * doing a reloc inode for a non refcounted root. | |
5968 | */ | |
5969 | struct btrfs_block_rsv *src_rsv = trans->block_rsv; | |
d68fc57b YZ |
5970 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; |
5971 | ||
5972 | /* | |
fcb80c2a JB |
5973 | * We need to hold space in order to delete our orphan item once we've |
5974 | * added it, so this takes the reservation so we can release it later | |
5975 | * when we are truly done with the orphan item. | |
d68fc57b | 5976 | */ |
0b246afa JM |
5977 | u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
5978 | ||
9678c543 | 5979 | trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode), |
8ed7a2a0 | 5980 | num_bytes, 1); |
25d609f8 | 5981 | return btrfs_block_rsv_migrate(src_rsv, dst_rsv, num_bytes, 1); |
6a63209f JB |
5982 | } |
5983 | ||
703b391a | 5984 | void btrfs_orphan_release_metadata(struct btrfs_inode *inode) |
97e728d4 | 5985 | { |
703b391a NB |
5986 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
5987 | struct btrfs_root *root = inode->root; | |
0b246afa JM |
5988 | u64 num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
5989 | ||
703b391a NB |
5990 | trace_btrfs_space_reservation(fs_info, "orphan", btrfs_ino(inode), |
5991 | num_bytes, 0); | |
2ff7e61e | 5992 | btrfs_block_rsv_release(fs_info, root->orphan_block_rsv, num_bytes); |
d68fc57b | 5993 | } |
97e728d4 | 5994 | |
d5c12070 MX |
5995 | /* |
5996 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5997 | * root: the root of the parent directory | |
5998 | * rsv: block reservation | |
5999 | * items: the number of items that we need do reservation | |
6000 | * qgroup_reserved: used to return the reserved size in qgroup | |
6001 | * | |
6002 | * This function is used to reserve the space for snapshot/subvolume | |
6003 | * creation and deletion. Those operations are different with the | |
6004 | * common file/directory operations, they change two fs/file trees | |
6005 | * and root tree, the number of items that the qgroup reserves is | |
6006 | * different with the free space reservation. So we can not use | |
01327610 | 6007 | * the space reservation mechanism in start_transaction(). |
d5c12070 MX |
6008 | */ |
6009 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
6010 | struct btrfs_block_rsv *rsv, | |
6011 | int items, | |
ee3441b4 JM |
6012 | u64 *qgroup_reserved, |
6013 | bool use_global_rsv) | |
a22285a6 | 6014 | { |
d5c12070 MX |
6015 | u64 num_bytes; |
6016 | int ret; | |
0b246afa JM |
6017 | struct btrfs_fs_info *fs_info = root->fs_info; |
6018 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d5c12070 | 6019 | |
0b246afa | 6020 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
d5c12070 | 6021 | /* One for parent inode, two for dir entries */ |
0b246afa | 6022 | num_bytes = 3 * fs_info->nodesize; |
733e03a0 | 6023 | ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes, true); |
d5c12070 MX |
6024 | if (ret) |
6025 | return ret; | |
6026 | } else { | |
6027 | num_bytes = 0; | |
6028 | } | |
6029 | ||
6030 | *qgroup_reserved = num_bytes; | |
6031 | ||
0b246afa JM |
6032 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, items); |
6033 | rsv->space_info = __find_space_info(fs_info, | |
d5c12070 MX |
6034 | BTRFS_BLOCK_GROUP_METADATA); |
6035 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
6036 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
6037 | |
6038 | if (ret == -ENOSPC && use_global_rsv) | |
25d609f8 | 6039 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1); |
ee3441b4 | 6040 | |
7174109c | 6041 | if (ret && *qgroup_reserved) |
733e03a0 | 6042 | btrfs_qgroup_free_meta_prealloc(root, *qgroup_reserved); |
d5c12070 MX |
6043 | |
6044 | return ret; | |
6045 | } | |
6046 | ||
2ff7e61e | 6047 | void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info, |
7775c818 | 6048 | struct btrfs_block_rsv *rsv) |
d5c12070 | 6049 | { |
2ff7e61e | 6050 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
97e728d4 JB |
6051 | } |
6052 | ||
69fe2d75 JB |
6053 | static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info, |
6054 | struct btrfs_inode *inode) | |
9e0baf60 | 6055 | { |
69fe2d75 JB |
6056 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
6057 | u64 reserve_size = 0; | |
ff6bc37e | 6058 | u64 qgroup_rsv_size = 0; |
69fe2d75 JB |
6059 | u64 csum_leaves; |
6060 | unsigned outstanding_extents; | |
9e0baf60 | 6061 | |
69fe2d75 JB |
6062 | lockdep_assert_held(&inode->lock); |
6063 | outstanding_extents = inode->outstanding_extents; | |
6064 | if (outstanding_extents) | |
6065 | reserve_size = btrfs_calc_trans_metadata_size(fs_info, | |
6066 | outstanding_extents + 1); | |
6067 | csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, | |
6068 | inode->csum_bytes); | |
6069 | reserve_size += btrfs_calc_trans_metadata_size(fs_info, | |
6070 | csum_leaves); | |
ff6bc37e QW |
6071 | /* |
6072 | * For qgroup rsv, the calculation is very simple: | |
6073 | * account one nodesize for each outstanding extent | |
6074 | * | |
6075 | * This is overestimating in most cases. | |
6076 | */ | |
6077 | qgroup_rsv_size = outstanding_extents * fs_info->nodesize; | |
9e0baf60 | 6078 | |
69fe2d75 JB |
6079 | spin_lock(&block_rsv->lock); |
6080 | block_rsv->size = reserve_size; | |
ff6bc37e | 6081 | block_rsv->qgroup_rsv_size = qgroup_rsv_size; |
69fe2d75 | 6082 | spin_unlock(&block_rsv->lock); |
0ca1f7ce | 6083 | } |
c146afad | 6084 | |
9f3db423 | 6085 | int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 6086 | { |
9f3db423 | 6087 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
69fe2d75 | 6088 | unsigned nr_extents; |
08e007d2 | 6089 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 6090 | int ret = 0; |
c64c2bd8 | 6091 | bool delalloc_lock = true; |
6324fbf3 | 6092 | |
c64c2bd8 JB |
6093 | /* If we are a free space inode we need to not flush since we will be in |
6094 | * the middle of a transaction commit. We also don't need the delalloc | |
6095 | * mutex since we won't race with anybody. We need this mostly to make | |
6096 | * lockdep shut its filthy mouth. | |
bac357dc JB |
6097 | * |
6098 | * If we have a transaction open (can happen if we call truncate_block | |
6099 | * from truncate), then we need FLUSH_LIMIT so we don't deadlock. | |
c64c2bd8 JB |
6100 | */ |
6101 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 6102 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 | 6103 | delalloc_lock = false; |
da07d4ab NB |
6104 | } else { |
6105 | if (current->journal_info) | |
6106 | flush = BTRFS_RESERVE_FLUSH_LIMIT; | |
c09544e0 | 6107 | |
da07d4ab NB |
6108 | if (btrfs_transaction_in_commit(fs_info)) |
6109 | schedule_timeout(1); | |
6110 | } | |
ec44a35c | 6111 | |
c64c2bd8 | 6112 | if (delalloc_lock) |
9f3db423 | 6113 | mutex_lock(&inode->delalloc_mutex); |
c64c2bd8 | 6114 | |
0b246afa | 6115 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
69fe2d75 JB |
6116 | |
6117 | /* Add our new extents and calculate the new rsv size. */ | |
9f3db423 | 6118 | spin_lock(&inode->lock); |
69fe2d75 | 6119 | nr_extents = count_max_extents(num_bytes); |
8b62f87b | 6120 | btrfs_mod_outstanding_extents(inode, nr_extents); |
69fe2d75 JB |
6121 | inode->csum_bytes += num_bytes; |
6122 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 6123 | spin_unlock(&inode->lock); |
57a45ced | 6124 | |
69fe2d75 | 6125 | ret = btrfs_inode_rsv_refill(inode, flush); |
43b18595 | 6126 | if (unlikely(ret)) |
88e081bf | 6127 | goto out_fail; |
25179201 | 6128 | |
c64c2bd8 | 6129 | if (delalloc_lock) |
9f3db423 | 6130 | mutex_unlock(&inode->delalloc_mutex); |
0ca1f7ce | 6131 | return 0; |
88e081bf WS |
6132 | |
6133 | out_fail: | |
9f3db423 | 6134 | spin_lock(&inode->lock); |
8b62f87b JB |
6135 | nr_extents = count_max_extents(num_bytes); |
6136 | btrfs_mod_outstanding_extents(inode, -nr_extents); | |
69fe2d75 JB |
6137 | inode->csum_bytes -= num_bytes; |
6138 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 6139 | spin_unlock(&inode->lock); |
88e081bf | 6140 | |
43b18595 | 6141 | btrfs_inode_rsv_release(inode, true); |
88e081bf | 6142 | if (delalloc_lock) |
9f3db423 | 6143 | mutex_unlock(&inode->delalloc_mutex); |
88e081bf | 6144 | return ret; |
0ca1f7ce YZ |
6145 | } |
6146 | ||
7709cde3 JB |
6147 | /** |
6148 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
8b62f87b JB |
6149 | * @inode: the inode to release the reservation for. |
6150 | * @num_bytes: the number of bytes we are releasing. | |
43b18595 | 6151 | * @qgroup_free: free qgroup reservation or convert it to per-trans reservation |
7709cde3 JB |
6152 | * |
6153 | * This will release the metadata reservation for an inode. This can be called | |
6154 | * once we complete IO for a given set of bytes to release their metadata | |
8b62f87b | 6155 | * reservations, or on error for the same reason. |
7709cde3 | 6156 | */ |
43b18595 QW |
6157 | void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes, |
6158 | bool qgroup_free) | |
0ca1f7ce | 6159 | { |
691fa059 | 6160 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
0ca1f7ce | 6161 | |
0b246afa | 6162 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
691fa059 | 6163 | spin_lock(&inode->lock); |
69fe2d75 JB |
6164 | inode->csum_bytes -= num_bytes; |
6165 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
691fa059 | 6166 | spin_unlock(&inode->lock); |
0ca1f7ce | 6167 | |
0b246afa | 6168 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
6169 | return; |
6170 | ||
43b18595 | 6171 | btrfs_inode_rsv_release(inode, qgroup_free); |
0ca1f7ce YZ |
6172 | } |
6173 | ||
8b62f87b JB |
6174 | /** |
6175 | * btrfs_delalloc_release_extents - release our outstanding_extents | |
6176 | * @inode: the inode to balance the reservation for. | |
6177 | * @num_bytes: the number of bytes we originally reserved with | |
43b18595 | 6178 | * @qgroup_free: do we need to free qgroup meta reservation or convert them. |
8b62f87b JB |
6179 | * |
6180 | * When we reserve space we increase outstanding_extents for the extents we may | |
6181 | * add. Once we've set the range as delalloc or created our ordered extents we | |
6182 | * have outstanding_extents to track the real usage, so we use this to free our | |
6183 | * temporarily tracked outstanding_extents. This _must_ be used in conjunction | |
6184 | * with btrfs_delalloc_reserve_metadata. | |
6185 | */ | |
43b18595 QW |
6186 | void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes, |
6187 | bool qgroup_free) | |
8b62f87b JB |
6188 | { |
6189 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); | |
6190 | unsigned num_extents; | |
8b62f87b JB |
6191 | |
6192 | spin_lock(&inode->lock); | |
6193 | num_extents = count_max_extents(num_bytes); | |
6194 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
69fe2d75 | 6195 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); |
8b62f87b JB |
6196 | spin_unlock(&inode->lock); |
6197 | ||
8b62f87b JB |
6198 | if (btrfs_is_testing(fs_info)) |
6199 | return; | |
6200 | ||
43b18595 | 6201 | btrfs_inode_rsv_release(inode, qgroup_free); |
8b62f87b JB |
6202 | } |
6203 | ||
1ada3a62 | 6204 | /** |
7cf5b976 | 6205 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
6206 | * delalloc |
6207 | * @inode: inode we're writing to | |
6208 | * @start: start range we are writing to | |
6209 | * @len: how long the range we are writing to | |
364ecf36 QW |
6210 | * @reserved: mandatory parameter, record actually reserved qgroup ranges of |
6211 | * current reservation. | |
1ada3a62 | 6212 | * |
1ada3a62 QW |
6213 | * This will do the following things |
6214 | * | |
6215 | * o reserve space in data space info for num bytes | |
6216 | * and reserve precious corresponding qgroup space | |
6217 | * (Done in check_data_free_space) | |
6218 | * | |
6219 | * o reserve space for metadata space, based on the number of outstanding | |
6220 | * extents and how much csums will be needed | |
6221 | * also reserve metadata space in a per root over-reserve method. | |
6222 | * o add to the inodes->delalloc_bytes | |
6223 | * o add it to the fs_info's delalloc inodes list. | |
6224 | * (Above 3 all done in delalloc_reserve_metadata) | |
6225 | * | |
6226 | * Return 0 for success | |
6227 | * Return <0 for error(-ENOSPC or -EQUOT) | |
6228 | */ | |
364ecf36 QW |
6229 | int btrfs_delalloc_reserve_space(struct inode *inode, |
6230 | struct extent_changeset **reserved, u64 start, u64 len) | |
1ada3a62 QW |
6231 | { |
6232 | int ret; | |
6233 | ||
364ecf36 | 6234 | ret = btrfs_check_data_free_space(inode, reserved, start, len); |
1ada3a62 QW |
6235 | if (ret < 0) |
6236 | return ret; | |
9f3db423 | 6237 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len); |
1ada3a62 | 6238 | if (ret < 0) |
bc42bda2 | 6239 | btrfs_free_reserved_data_space(inode, *reserved, start, len); |
1ada3a62 QW |
6240 | return ret; |
6241 | } | |
6242 | ||
7709cde3 | 6243 | /** |
7cf5b976 | 6244 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
6245 | * @inode: inode we're releasing space for |
6246 | * @start: start position of the space already reserved | |
6247 | * @len: the len of the space already reserved | |
8b62f87b | 6248 | * @release_bytes: the len of the space we consumed or didn't use |
1ada3a62 QW |
6249 | * |
6250 | * This function will release the metadata space that was not used and will | |
6251 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
6252 | * list if there are no delalloc bytes left. | |
6253 | * Also it will handle the qgroup reserved space. | |
6254 | */ | |
bc42bda2 | 6255 | void btrfs_delalloc_release_space(struct inode *inode, |
8b62f87b | 6256 | struct extent_changeset *reserved, |
43b18595 | 6257 | u64 start, u64 len, bool qgroup_free) |
1ada3a62 | 6258 | { |
43b18595 | 6259 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free); |
bc42bda2 | 6260 | btrfs_free_reserved_data_space(inode, reserved, start, len); |
6324fbf3 CM |
6261 | } |
6262 | ||
ce93ec54 | 6263 | static int update_block_group(struct btrfs_trans_handle *trans, |
6202df69 | 6264 | struct btrfs_fs_info *info, u64 bytenr, |
ce93ec54 | 6265 | u64 num_bytes, int alloc) |
9078a3e1 | 6266 | { |
0af3d00b | 6267 | struct btrfs_block_group_cache *cache = NULL; |
db94535d | 6268 | u64 total = num_bytes; |
9078a3e1 | 6269 | u64 old_val; |
db94535d | 6270 | u64 byte_in_group; |
0af3d00b | 6271 | int factor; |
3e1ad54f | 6272 | |
5d4f98a2 | 6273 | /* block accounting for super block */ |
eb73c1b7 | 6274 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 6275 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
6276 | if (alloc) |
6277 | old_val += num_bytes; | |
6278 | else | |
6279 | old_val -= num_bytes; | |
6c41761f | 6280 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 6281 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 6282 | |
d397712b | 6283 | while (total) { |
db94535d | 6284 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 6285 | if (!cache) |
79787eaa | 6286 | return -ENOENT; |
b742bb82 YZ |
6287 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
6288 | BTRFS_BLOCK_GROUP_RAID1 | | |
6289 | BTRFS_BLOCK_GROUP_RAID10)) | |
6290 | factor = 2; | |
6291 | else | |
6292 | factor = 1; | |
9d66e233 JB |
6293 | /* |
6294 | * If this block group has free space cache written out, we | |
6295 | * need to make sure to load it if we are removing space. This | |
6296 | * is because we need the unpinning stage to actually add the | |
6297 | * space back to the block group, otherwise we will leak space. | |
6298 | */ | |
6299 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 6300 | cache_block_group(cache, 1); |
0af3d00b | 6301 | |
db94535d CM |
6302 | byte_in_group = bytenr - cache->key.objectid; |
6303 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 6304 | |
25179201 | 6305 | spin_lock(&cache->space_info->lock); |
c286ac48 | 6306 | spin_lock(&cache->lock); |
0af3d00b | 6307 | |
6202df69 | 6308 | if (btrfs_test_opt(info, SPACE_CACHE) && |
0af3d00b JB |
6309 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
6310 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
6311 | ||
9078a3e1 | 6312 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 6313 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 6314 | if (alloc) { |
db94535d | 6315 | old_val += num_bytes; |
11833d66 YZ |
6316 | btrfs_set_block_group_used(&cache->item, old_val); |
6317 | cache->reserved -= num_bytes; | |
11833d66 | 6318 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
6319 | cache->space_info->bytes_used += num_bytes; |
6320 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 6321 | spin_unlock(&cache->lock); |
25179201 | 6322 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 6323 | } else { |
db94535d | 6324 | old_val -= num_bytes; |
ae0ab003 FM |
6325 | btrfs_set_block_group_used(&cache->item, old_val); |
6326 | cache->pinned += num_bytes; | |
6327 | cache->space_info->bytes_pinned += num_bytes; | |
6328 | cache->space_info->bytes_used -= num_bytes; | |
6329 | cache->space_info->disk_used -= num_bytes * factor; | |
6330 | spin_unlock(&cache->lock); | |
6331 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 6332 | |
0b246afa | 6333 | trace_btrfs_space_reservation(info, "pinned", |
c51e7bb1 JB |
6334 | cache->space_info->flags, |
6335 | num_bytes, 1); | |
d7eae340 OS |
6336 | percpu_counter_add(&cache->space_info->total_bytes_pinned, |
6337 | num_bytes); | |
ae0ab003 FM |
6338 | set_extent_dirty(info->pinned_extents, |
6339 | bytenr, bytenr + num_bytes - 1, | |
6340 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 6341 | } |
1bbc621e CM |
6342 | |
6343 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
6344 | if (list_empty(&cache->dirty_list)) { | |
6345 | list_add_tail(&cache->dirty_list, | |
6346 | &trans->transaction->dirty_bgs); | |
6347 | trans->transaction->num_dirty_bgs++; | |
6348 | btrfs_get_block_group(cache); | |
6349 | } | |
6350 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
6351 | ||
036a9348 FM |
6352 | /* |
6353 | * No longer have used bytes in this block group, queue it for | |
6354 | * deletion. We do this after adding the block group to the | |
6355 | * dirty list to avoid races between cleaner kthread and space | |
6356 | * cache writeout. | |
6357 | */ | |
6358 | if (!alloc && old_val == 0) { | |
6359 | spin_lock(&info->unused_bgs_lock); | |
6360 | if (list_empty(&cache->bg_list)) { | |
6361 | btrfs_get_block_group(cache); | |
6362 | list_add_tail(&cache->bg_list, | |
6363 | &info->unused_bgs); | |
6364 | } | |
6365 | spin_unlock(&info->unused_bgs_lock); | |
6366 | } | |
6367 | ||
fa9c0d79 | 6368 | btrfs_put_block_group(cache); |
db94535d CM |
6369 | total -= num_bytes; |
6370 | bytenr += num_bytes; | |
9078a3e1 CM |
6371 | } |
6372 | return 0; | |
6373 | } | |
6324fbf3 | 6374 | |
2ff7e61e | 6375 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 6376 | { |
0f9dd46c | 6377 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 6378 | u64 bytenr; |
0f9dd46c | 6379 | |
0b246afa JM |
6380 | spin_lock(&fs_info->block_group_cache_lock); |
6381 | bytenr = fs_info->first_logical_byte; | |
6382 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
6383 | |
6384 | if (bytenr < (u64)-1) | |
6385 | return bytenr; | |
6386 | ||
0b246afa | 6387 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 6388 | if (!cache) |
a061fc8d | 6389 | return 0; |
0f9dd46c | 6390 | |
d2fb3437 | 6391 | bytenr = cache->key.objectid; |
fa9c0d79 | 6392 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
6393 | |
6394 | return bytenr; | |
a061fc8d CM |
6395 | } |
6396 | ||
2ff7e61e | 6397 | static int pin_down_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6398 | struct btrfs_block_group_cache *cache, |
6399 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 6400 | { |
11833d66 YZ |
6401 | spin_lock(&cache->space_info->lock); |
6402 | spin_lock(&cache->lock); | |
6403 | cache->pinned += num_bytes; | |
6404 | cache->space_info->bytes_pinned += num_bytes; | |
6405 | if (reserved) { | |
6406 | cache->reserved -= num_bytes; | |
6407 | cache->space_info->bytes_reserved -= num_bytes; | |
6408 | } | |
6409 | spin_unlock(&cache->lock); | |
6410 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 6411 | |
0b246afa | 6412 | trace_btrfs_space_reservation(fs_info, "pinned", |
c51e7bb1 | 6413 | cache->space_info->flags, num_bytes, 1); |
4da8b76d | 6414 | percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes); |
0b246afa | 6415 | set_extent_dirty(fs_info->pinned_extents, bytenr, |
f0486c68 YZ |
6416 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
6417 | return 0; | |
6418 | } | |
68b38550 | 6419 | |
f0486c68 YZ |
6420 | /* |
6421 | * this function must be called within transaction | |
6422 | */ | |
2ff7e61e | 6423 | int btrfs_pin_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6424 | u64 bytenr, u64 num_bytes, int reserved) |
6425 | { | |
6426 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6427 | |
0b246afa | 6428 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
79787eaa | 6429 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 6430 | |
2ff7e61e | 6431 | pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
6432 | |
6433 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6434 | return 0; |
6435 | } | |
6436 | ||
f0486c68 | 6437 | /* |
e688b725 CM |
6438 | * this function must be called within transaction |
6439 | */ | |
2ff7e61e | 6440 | int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info, |
e688b725 CM |
6441 | u64 bytenr, u64 num_bytes) |
6442 | { | |
6443 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6444 | int ret; |
e688b725 | 6445 | |
0b246afa | 6446 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
b50c6e25 JB |
6447 | if (!cache) |
6448 | return -EINVAL; | |
e688b725 CM |
6449 | |
6450 | /* | |
6451 | * pull in the free space cache (if any) so that our pin | |
6452 | * removes the free space from the cache. We have load_only set | |
6453 | * to one because the slow code to read in the free extents does check | |
6454 | * the pinned extents. | |
6455 | */ | |
f6373bf3 | 6456 | cache_block_group(cache, 1); |
e688b725 | 6457 | |
2ff7e61e | 6458 | pin_down_extent(fs_info, cache, bytenr, num_bytes, 0); |
e688b725 CM |
6459 | |
6460 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6461 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6462 | btrfs_put_block_group(cache); |
b50c6e25 | 6463 | return ret; |
e688b725 CM |
6464 | } |
6465 | ||
2ff7e61e JM |
6466 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
6467 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
6468 | { |
6469 | int ret; | |
6470 | struct btrfs_block_group_cache *block_group; | |
6471 | struct btrfs_caching_control *caching_ctl; | |
6472 | ||
0b246afa | 6473 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
6474 | if (!block_group) |
6475 | return -EINVAL; | |
6476 | ||
6477 | cache_block_group(block_group, 0); | |
6478 | caching_ctl = get_caching_control(block_group); | |
6479 | ||
6480 | if (!caching_ctl) { | |
6481 | /* Logic error */ | |
6482 | BUG_ON(!block_group_cache_done(block_group)); | |
6483 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6484 | } else { | |
6485 | mutex_lock(&caching_ctl->mutex); | |
6486 | ||
6487 | if (start >= caching_ctl->progress) { | |
2ff7e61e | 6488 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6489 | } else if (start + num_bytes <= caching_ctl->progress) { |
6490 | ret = btrfs_remove_free_space(block_group, | |
6491 | start, num_bytes); | |
6492 | } else { | |
6493 | num_bytes = caching_ctl->progress - start; | |
6494 | ret = btrfs_remove_free_space(block_group, | |
6495 | start, num_bytes); | |
6496 | if (ret) | |
6497 | goto out_lock; | |
6498 | ||
6499 | num_bytes = (start + num_bytes) - | |
6500 | caching_ctl->progress; | |
6501 | start = caching_ctl->progress; | |
2ff7e61e | 6502 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6503 | } |
6504 | out_lock: | |
6505 | mutex_unlock(&caching_ctl->mutex); | |
6506 | put_caching_control(caching_ctl); | |
6507 | } | |
6508 | btrfs_put_block_group(block_group); | |
6509 | return ret; | |
6510 | } | |
6511 | ||
2ff7e61e | 6512 | int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info, |
8c2a1a30 JB |
6513 | struct extent_buffer *eb) |
6514 | { | |
6515 | struct btrfs_file_extent_item *item; | |
6516 | struct btrfs_key key; | |
6517 | int found_type; | |
6518 | int i; | |
6519 | ||
2ff7e61e | 6520 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
6521 | return 0; |
6522 | ||
6523 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6524 | btrfs_item_key_to_cpu(eb, &key, i); | |
6525 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6526 | continue; | |
6527 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6528 | found_type = btrfs_file_extent_type(eb, item); | |
6529 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6530 | continue; | |
6531 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6532 | continue; | |
6533 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6534 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
2ff7e61e | 6535 | __exclude_logged_extent(fs_info, key.objectid, key.offset); |
8c2a1a30 JB |
6536 | } |
6537 | ||
6538 | return 0; | |
6539 | } | |
6540 | ||
9cfa3e34 FM |
6541 | static void |
6542 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6543 | { | |
6544 | atomic_inc(&bg->reservations); | |
6545 | } | |
6546 | ||
6547 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
6548 | const u64 start) | |
6549 | { | |
6550 | struct btrfs_block_group_cache *bg; | |
6551 | ||
6552 | bg = btrfs_lookup_block_group(fs_info, start); | |
6553 | ASSERT(bg); | |
6554 | if (atomic_dec_and_test(&bg->reservations)) | |
4625956a | 6555 | wake_up_var(&bg->reservations); |
9cfa3e34 FM |
6556 | btrfs_put_block_group(bg); |
6557 | } | |
6558 | ||
9cfa3e34 FM |
6559 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) |
6560 | { | |
6561 | struct btrfs_space_info *space_info = bg->space_info; | |
6562 | ||
6563 | ASSERT(bg->ro); | |
6564 | ||
6565 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
6566 | return; | |
6567 | ||
6568 | /* | |
6569 | * Our block group is read only but before we set it to read only, | |
6570 | * some task might have had allocated an extent from it already, but it | |
6571 | * has not yet created a respective ordered extent (and added it to a | |
6572 | * root's list of ordered extents). | |
6573 | * Therefore wait for any task currently allocating extents, since the | |
6574 | * block group's reservations counter is incremented while a read lock | |
6575 | * on the groups' semaphore is held and decremented after releasing | |
6576 | * the read access on that semaphore and creating the ordered extent. | |
6577 | */ | |
6578 | down_write(&space_info->groups_sem); | |
6579 | up_write(&space_info->groups_sem); | |
6580 | ||
4625956a | 6581 | wait_var_event(&bg->reservations, !atomic_read(&bg->reservations)); |
9cfa3e34 FM |
6582 | } |
6583 | ||
fb25e914 | 6584 | /** |
4824f1f4 | 6585 | * btrfs_add_reserved_bytes - update the block_group and space info counters |
fb25e914 | 6586 | * @cache: The cache we are manipulating |
18513091 WX |
6587 | * @ram_bytes: The number of bytes of file content, and will be same to |
6588 | * @num_bytes except for the compress path. | |
fb25e914 | 6589 | * @num_bytes: The number of bytes in question |
e570fd27 | 6590 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 | 6591 | * |
745699ef XW |
6592 | * This is called by the allocator when it reserves space. If this is a |
6593 | * reservation and the block group has become read only we cannot make the | |
6594 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
f0486c68 | 6595 | */ |
4824f1f4 | 6596 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 6597 | u64 ram_bytes, u64 num_bytes, int delalloc) |
11833d66 | 6598 | { |
fb25e914 | 6599 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6600 | int ret = 0; |
79787eaa | 6601 | |
fb25e914 JB |
6602 | spin_lock(&space_info->lock); |
6603 | spin_lock(&cache->lock); | |
4824f1f4 WX |
6604 | if (cache->ro) { |
6605 | ret = -EAGAIN; | |
fb25e914 | 6606 | } else { |
4824f1f4 WX |
6607 | cache->reserved += num_bytes; |
6608 | space_info->bytes_reserved += num_bytes; | |
e570fd27 | 6609 | |
18513091 WX |
6610 | trace_btrfs_space_reservation(cache->fs_info, |
6611 | "space_info", space_info->flags, | |
6612 | ram_bytes, 0); | |
6613 | space_info->bytes_may_use -= ram_bytes; | |
e570fd27 | 6614 | if (delalloc) |
4824f1f4 | 6615 | cache->delalloc_bytes += num_bytes; |
324ae4df | 6616 | } |
fb25e914 JB |
6617 | spin_unlock(&cache->lock); |
6618 | spin_unlock(&space_info->lock); | |
f0486c68 | 6619 | return ret; |
324ae4df | 6620 | } |
9078a3e1 | 6621 | |
4824f1f4 WX |
6622 | /** |
6623 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
6624 | * @cache: The cache we are manipulating | |
6625 | * @num_bytes: The number of bytes in question | |
6626 | * @delalloc: The blocks are allocated for the delalloc write | |
6627 | * | |
6628 | * This is called by somebody who is freeing space that was never actually used | |
6629 | * on disk. For example if you reserve some space for a new leaf in transaction | |
6630 | * A and before transaction A commits you free that leaf, you call this with | |
6631 | * reserve set to 0 in order to clear the reservation. | |
6632 | */ | |
6633 | ||
6634 | static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, | |
6635 | u64 num_bytes, int delalloc) | |
6636 | { | |
6637 | struct btrfs_space_info *space_info = cache->space_info; | |
6638 | int ret = 0; | |
6639 | ||
6640 | spin_lock(&space_info->lock); | |
6641 | spin_lock(&cache->lock); | |
6642 | if (cache->ro) | |
6643 | space_info->bytes_readonly += num_bytes; | |
6644 | cache->reserved -= num_bytes; | |
6645 | space_info->bytes_reserved -= num_bytes; | |
6646 | ||
6647 | if (delalloc) | |
6648 | cache->delalloc_bytes -= num_bytes; | |
6649 | spin_unlock(&cache->lock); | |
6650 | spin_unlock(&space_info->lock); | |
6651 | return ret; | |
6652 | } | |
8b74c03e | 6653 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 6654 | { |
11833d66 YZ |
6655 | struct btrfs_caching_control *next; |
6656 | struct btrfs_caching_control *caching_ctl; | |
6657 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6658 | |
9e351cc8 | 6659 | down_write(&fs_info->commit_root_sem); |
25179201 | 6660 | |
11833d66 YZ |
6661 | list_for_each_entry_safe(caching_ctl, next, |
6662 | &fs_info->caching_block_groups, list) { | |
6663 | cache = caching_ctl->block_group; | |
6664 | if (block_group_cache_done(cache)) { | |
6665 | cache->last_byte_to_unpin = (u64)-1; | |
6666 | list_del_init(&caching_ctl->list); | |
6667 | put_caching_control(caching_ctl); | |
e8569813 | 6668 | } else { |
11833d66 | 6669 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6670 | } |
e8569813 | 6671 | } |
11833d66 YZ |
6672 | |
6673 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6674 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6675 | else | |
6676 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6677 | ||
9e351cc8 | 6678 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6679 | |
6680 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6681 | } |
6682 | ||
c759c4e1 JB |
6683 | /* |
6684 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6685 | * what it should be based on the mount options. | |
6686 | */ | |
6687 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
6688 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
6689 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
6690 | { |
6691 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
6692 | |
6693 | *empty_cluster = 0; | |
6694 | if (btrfs_mixed_space_info(space_info)) | |
6695 | return ret; | |
6696 | ||
c759c4e1 | 6697 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 6698 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
6699 | if (btrfs_test_opt(fs_info, SSD)) |
6700 | *empty_cluster = SZ_2M; | |
6701 | else | |
ee22184b | 6702 | *empty_cluster = SZ_64K; |
583b7231 HK |
6703 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
6704 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
6705 | *empty_cluster = SZ_2M; | |
0b246afa | 6706 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
6707 | } |
6708 | ||
6709 | return ret; | |
6710 | } | |
6711 | ||
2ff7e61e JM |
6712 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
6713 | u64 start, u64 end, | |
678886bd | 6714 | const bool return_free_space) |
ccd467d6 | 6715 | { |
11833d66 | 6716 | struct btrfs_block_group_cache *cache = NULL; |
7b398f8e JB |
6717 | struct btrfs_space_info *space_info; |
6718 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6719 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6720 | u64 len; |
c759c4e1 JB |
6721 | u64 total_unpinned = 0; |
6722 | u64 empty_cluster = 0; | |
7b398f8e | 6723 | bool readonly; |
ccd467d6 | 6724 | |
11833d66 | 6725 | while (start <= end) { |
7b398f8e | 6726 | readonly = false; |
11833d66 YZ |
6727 | if (!cache || |
6728 | start >= cache->key.objectid + cache->key.offset) { | |
6729 | if (cache) | |
6730 | btrfs_put_block_group(cache); | |
c759c4e1 | 6731 | total_unpinned = 0; |
11833d66 | 6732 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6733 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 6734 | |
2ff7e61e | 6735 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
6736 | cache->space_info, |
6737 | &empty_cluster); | |
6738 | empty_cluster <<= 1; | |
11833d66 YZ |
6739 | } |
6740 | ||
6741 | len = cache->key.objectid + cache->key.offset - start; | |
6742 | len = min(len, end + 1 - start); | |
6743 | ||
6744 | if (start < cache->last_byte_to_unpin) { | |
6745 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6746 | if (return_free_space) |
6747 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6748 | } |
6749 | ||
f0486c68 | 6750 | start += len; |
c759c4e1 | 6751 | total_unpinned += len; |
7b398f8e | 6752 | space_info = cache->space_info; |
f0486c68 | 6753 | |
c759c4e1 JB |
6754 | /* |
6755 | * If this space cluster has been marked as fragmented and we've | |
6756 | * unpinned enough in this block group to potentially allow a | |
6757 | * cluster to be created inside of it go ahead and clear the | |
6758 | * fragmented check. | |
6759 | */ | |
6760 | if (cluster && cluster->fragmented && | |
6761 | total_unpinned > empty_cluster) { | |
6762 | spin_lock(&cluster->lock); | |
6763 | cluster->fragmented = 0; | |
6764 | spin_unlock(&cluster->lock); | |
6765 | } | |
6766 | ||
7b398f8e | 6767 | spin_lock(&space_info->lock); |
11833d66 YZ |
6768 | spin_lock(&cache->lock); |
6769 | cache->pinned -= len; | |
7b398f8e | 6770 | space_info->bytes_pinned -= len; |
c51e7bb1 JB |
6771 | |
6772 | trace_btrfs_space_reservation(fs_info, "pinned", | |
6773 | space_info->flags, len, 0); | |
4f4db217 | 6774 | space_info->max_extent_size = 0; |
d288db5d | 6775 | percpu_counter_add(&space_info->total_bytes_pinned, -len); |
7b398f8e JB |
6776 | if (cache->ro) { |
6777 | space_info->bytes_readonly += len; | |
6778 | readonly = true; | |
6779 | } | |
11833d66 | 6780 | spin_unlock(&cache->lock); |
957780eb JB |
6781 | if (!readonly && return_free_space && |
6782 | global_rsv->space_info == space_info) { | |
6783 | u64 to_add = len; | |
92ac58ec | 6784 | |
7b398f8e JB |
6785 | spin_lock(&global_rsv->lock); |
6786 | if (!global_rsv->full) { | |
957780eb JB |
6787 | to_add = min(len, global_rsv->size - |
6788 | global_rsv->reserved); | |
6789 | global_rsv->reserved += to_add; | |
6790 | space_info->bytes_may_use += to_add; | |
7b398f8e JB |
6791 | if (global_rsv->reserved >= global_rsv->size) |
6792 | global_rsv->full = 1; | |
957780eb JB |
6793 | trace_btrfs_space_reservation(fs_info, |
6794 | "space_info", | |
6795 | space_info->flags, | |
6796 | to_add, 1); | |
6797 | len -= to_add; | |
7b398f8e JB |
6798 | } |
6799 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
6800 | /* Add to any tickets we may have */ |
6801 | if (len) | |
6802 | space_info_add_new_bytes(fs_info, space_info, | |
6803 | len); | |
7b398f8e JB |
6804 | } |
6805 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6806 | } |
11833d66 YZ |
6807 | |
6808 | if (cache) | |
6809 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6810 | return 0; |
6811 | } | |
6812 | ||
5ead2dd0 | 6813 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 6814 | { |
5ead2dd0 | 6815 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e33e17ee JM |
6816 | struct btrfs_block_group_cache *block_group, *tmp; |
6817 | struct list_head *deleted_bgs; | |
11833d66 | 6818 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6819 | u64 start; |
6820 | u64 end; | |
a28ec197 | 6821 | int ret; |
a28ec197 | 6822 | |
11833d66 YZ |
6823 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6824 | unpin = &fs_info->freed_extents[1]; | |
6825 | else | |
6826 | unpin = &fs_info->freed_extents[0]; | |
6827 | ||
e33e17ee | 6828 | while (!trans->aborted) { |
d4b450cd | 6829 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6830 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 6831 | EXTENT_DIRTY, NULL); |
d4b450cd FM |
6832 | if (ret) { |
6833 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6834 | break; |
d4b450cd | 6835 | } |
1f3c79a2 | 6836 | |
0b246afa | 6837 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 6838 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 6839 | end + 1 - start, NULL); |
1f3c79a2 | 6840 | |
af6f8f60 | 6841 | clear_extent_dirty(unpin, start, end); |
2ff7e61e | 6842 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 6843 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
b9473439 | 6844 | cond_resched(); |
a28ec197 | 6845 | } |
817d52f8 | 6846 | |
e33e17ee JM |
6847 | /* |
6848 | * Transaction is finished. We don't need the lock anymore. We | |
6849 | * do need to clean up the block groups in case of a transaction | |
6850 | * abort. | |
6851 | */ | |
6852 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6853 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6854 | u64 trimmed = 0; | |
6855 | ||
6856 | ret = -EROFS; | |
6857 | if (!trans->aborted) | |
2ff7e61e | 6858 | ret = btrfs_discard_extent(fs_info, |
e33e17ee JM |
6859 | block_group->key.objectid, |
6860 | block_group->key.offset, | |
6861 | &trimmed); | |
6862 | ||
6863 | list_del_init(&block_group->bg_list); | |
6864 | btrfs_put_block_group_trimming(block_group); | |
6865 | btrfs_put_block_group(block_group); | |
6866 | ||
6867 | if (ret) { | |
6868 | const char *errstr = btrfs_decode_error(ret); | |
6869 | btrfs_warn(fs_info, | |
913e1535 | 6870 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
6871 | ret, errstr); |
6872 | } | |
6873 | } | |
6874 | ||
e20d96d6 CM |
6875 | return 0; |
6876 | } | |
6877 | ||
5d4f98a2 | 6878 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 6879 | struct btrfs_fs_info *info, |
c682f9b3 | 6880 | struct btrfs_delayed_ref_node *node, u64 parent, |
5d4f98a2 YZ |
6881 | u64 root_objectid, u64 owner_objectid, |
6882 | u64 owner_offset, int refs_to_drop, | |
c682f9b3 | 6883 | struct btrfs_delayed_extent_op *extent_op) |
a28ec197 | 6884 | { |
e2fa7227 | 6885 | struct btrfs_key key; |
5d4f98a2 | 6886 | struct btrfs_path *path; |
1261ec42 | 6887 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 6888 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6889 | struct btrfs_extent_item *ei; |
6890 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6891 | int ret; |
5d4f98a2 | 6892 | int is_data; |
952fccac CM |
6893 | int extent_slot = 0; |
6894 | int found_extent = 0; | |
6895 | int num_to_del = 1; | |
5d4f98a2 YZ |
6896 | u32 item_size; |
6897 | u64 refs; | |
c682f9b3 QW |
6898 | u64 bytenr = node->bytenr; |
6899 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6900 | int last_ref = 0; |
0b246afa | 6901 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 6902 | |
5caf2a00 | 6903 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6904 | if (!path) |
6905 | return -ENOMEM; | |
5f26f772 | 6906 | |
e4058b54 | 6907 | path->reada = READA_FORWARD; |
b9473439 | 6908 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6909 | |
6910 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6911 | BUG_ON(!is_data && refs_to_drop != 1); | |
6912 | ||
3173a18f | 6913 | if (is_data) |
897ca819 | 6914 | skinny_metadata = false; |
3173a18f | 6915 | |
87bde3cd | 6916 | ret = lookup_extent_backref(trans, info, path, &iref, |
5d4f98a2 YZ |
6917 | bytenr, num_bytes, parent, |
6918 | root_objectid, owner_objectid, | |
6919 | owner_offset); | |
7bb86316 | 6920 | if (ret == 0) { |
952fccac | 6921 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6922 | while (extent_slot >= 0) { |
6923 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6924 | extent_slot); |
5d4f98a2 | 6925 | if (key.objectid != bytenr) |
952fccac | 6926 | break; |
5d4f98a2 YZ |
6927 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6928 | key.offset == num_bytes) { | |
952fccac CM |
6929 | found_extent = 1; |
6930 | break; | |
6931 | } | |
3173a18f JB |
6932 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
6933 | key.offset == owner_objectid) { | |
6934 | found_extent = 1; | |
6935 | break; | |
6936 | } | |
952fccac CM |
6937 | if (path->slots[0] - extent_slot > 5) |
6938 | break; | |
5d4f98a2 | 6939 | extent_slot--; |
952fccac | 6940 | } |
5d4f98a2 YZ |
6941 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
6942 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); | |
6943 | if (found_extent && item_size < sizeof(*ei)) | |
6944 | found_extent = 0; | |
6945 | #endif | |
31840ae1 | 6946 | if (!found_extent) { |
5d4f98a2 | 6947 | BUG_ON(iref); |
87bde3cd JM |
6948 | ret = remove_extent_backref(trans, info, path, NULL, |
6949 | refs_to_drop, | |
fcebe456 | 6950 | is_data, &last_ref); |
005d6427 | 6951 | if (ret) { |
66642832 | 6952 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6953 | goto out; |
6954 | } | |
b3b4aa74 | 6955 | btrfs_release_path(path); |
b9473439 | 6956 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6957 | |
6958 | key.objectid = bytenr; | |
6959 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6960 | key.offset = num_bytes; | |
6961 | ||
3173a18f JB |
6962 | if (!is_data && skinny_metadata) { |
6963 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6964 | key.offset = owner_objectid; | |
6965 | } | |
6966 | ||
31840ae1 ZY |
6967 | ret = btrfs_search_slot(trans, extent_root, |
6968 | &key, path, -1, 1); | |
3173a18f JB |
6969 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
6970 | /* | |
6971 | * Couldn't find our skinny metadata item, | |
6972 | * see if we have ye olde extent item. | |
6973 | */ | |
6974 | path->slots[0]--; | |
6975 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6976 | path->slots[0]); | |
6977 | if (key.objectid == bytenr && | |
6978 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6979 | key.offset == num_bytes) | |
6980 | ret = 0; | |
6981 | } | |
6982 | ||
6983 | if (ret > 0 && skinny_metadata) { | |
6984 | skinny_metadata = false; | |
9ce49a0b | 6985 | key.objectid = bytenr; |
3173a18f JB |
6986 | key.type = BTRFS_EXTENT_ITEM_KEY; |
6987 | key.offset = num_bytes; | |
6988 | btrfs_release_path(path); | |
6989 | ret = btrfs_search_slot(trans, extent_root, | |
6990 | &key, path, -1, 1); | |
6991 | } | |
6992 | ||
f3465ca4 | 6993 | if (ret) { |
5d163e0e JM |
6994 | btrfs_err(info, |
6995 | "umm, got %d back from search, was looking for %llu", | |
6996 | ret, bytenr); | |
b783e62d | 6997 | if (ret > 0) |
a4f78750 | 6998 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 6999 | } |
005d6427 | 7000 | if (ret < 0) { |
66642832 | 7001 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7002 | goto out; |
7003 | } | |
31840ae1 ZY |
7004 | extent_slot = path->slots[0]; |
7005 | } | |
fae7f21c | 7006 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 7007 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
7008 | btrfs_err(info, |
7009 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
7010 | bytenr, parent, root_objectid, owner_objectid, |
7011 | owner_offset); | |
66642832 | 7012 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 7013 | goto out; |
79787eaa | 7014 | } else { |
66642832 | 7015 | btrfs_abort_transaction(trans, ret); |
005d6427 | 7016 | goto out; |
7bb86316 | 7017 | } |
5f39d397 CM |
7018 | |
7019 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
7020 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
7021 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 | |
7022 | if (item_size < sizeof(*ei)) { | |
7023 | BUG_ON(found_extent || extent_slot != path->slots[0]); | |
87bde3cd JM |
7024 | ret = convert_extent_item_v0(trans, info, path, owner_objectid, |
7025 | 0); | |
005d6427 | 7026 | if (ret < 0) { |
66642832 | 7027 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7028 | goto out; |
7029 | } | |
5d4f98a2 | 7030 | |
b3b4aa74 | 7031 | btrfs_release_path(path); |
5d4f98a2 YZ |
7032 | path->leave_spinning = 1; |
7033 | ||
7034 | key.objectid = bytenr; | |
7035 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
7036 | key.offset = num_bytes; | |
7037 | ||
7038 | ret = btrfs_search_slot(trans, extent_root, &key, path, | |
7039 | -1, 1); | |
7040 | if (ret) { | |
5d163e0e JM |
7041 | btrfs_err(info, |
7042 | "umm, got %d back from search, was looking for %llu", | |
c1c9ff7c | 7043 | ret, bytenr); |
a4f78750 | 7044 | btrfs_print_leaf(path->nodes[0]); |
5d4f98a2 | 7045 | } |
005d6427 | 7046 | if (ret < 0) { |
66642832 | 7047 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7048 | goto out; |
7049 | } | |
7050 | ||
5d4f98a2 YZ |
7051 | extent_slot = path->slots[0]; |
7052 | leaf = path->nodes[0]; | |
7053 | item_size = btrfs_item_size_nr(leaf, extent_slot); | |
7054 | } | |
7055 | #endif | |
7056 | BUG_ON(item_size < sizeof(*ei)); | |
952fccac | 7057 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 7058 | struct btrfs_extent_item); |
3173a18f JB |
7059 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
7060 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
7061 | struct btrfs_tree_block_info *bi; |
7062 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
7063 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
7064 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
7065 | } | |
56bec294 | 7066 | |
5d4f98a2 | 7067 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 7068 | if (refs < refs_to_drop) { |
5d163e0e JM |
7069 | btrfs_err(info, |
7070 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
7071 | refs_to_drop, refs, bytenr); | |
32b02538 | 7072 | ret = -EINVAL; |
66642832 | 7073 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
7074 | goto out; |
7075 | } | |
56bec294 | 7076 | refs -= refs_to_drop; |
5f39d397 | 7077 | |
5d4f98a2 YZ |
7078 | if (refs > 0) { |
7079 | if (extent_op) | |
7080 | __run_delayed_extent_op(extent_op, leaf, ei); | |
7081 | /* | |
7082 | * In the case of inline back ref, reference count will | |
7083 | * be updated by remove_extent_backref | |
952fccac | 7084 | */ |
5d4f98a2 YZ |
7085 | if (iref) { |
7086 | BUG_ON(!found_extent); | |
7087 | } else { | |
7088 | btrfs_set_extent_refs(leaf, ei, refs); | |
7089 | btrfs_mark_buffer_dirty(leaf); | |
7090 | } | |
7091 | if (found_extent) { | |
87bde3cd | 7092 | ret = remove_extent_backref(trans, info, path, |
5d4f98a2 | 7093 | iref, refs_to_drop, |
fcebe456 | 7094 | is_data, &last_ref); |
005d6427 | 7095 | if (ret) { |
66642832 | 7096 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7097 | goto out; |
7098 | } | |
952fccac | 7099 | } |
5d4f98a2 | 7100 | } else { |
5d4f98a2 YZ |
7101 | if (found_extent) { |
7102 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 7103 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
7104 | if (iref) { |
7105 | BUG_ON(path->slots[0] != extent_slot); | |
7106 | } else { | |
7107 | BUG_ON(path->slots[0] != extent_slot + 1); | |
7108 | path->slots[0] = extent_slot; | |
7109 | num_to_del = 2; | |
7110 | } | |
78fae27e | 7111 | } |
b9473439 | 7112 | |
fcebe456 | 7113 | last_ref = 1; |
952fccac CM |
7114 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
7115 | num_to_del); | |
005d6427 | 7116 | if (ret) { |
66642832 | 7117 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7118 | goto out; |
7119 | } | |
b3b4aa74 | 7120 | btrfs_release_path(path); |
21af804c | 7121 | |
5d4f98a2 | 7122 | if (is_data) { |
5b4aacef | 7123 | ret = btrfs_del_csums(trans, info, bytenr, num_bytes); |
005d6427 | 7124 | if (ret) { |
66642832 | 7125 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7126 | goto out; |
7127 | } | |
459931ec CM |
7128 | } |
7129 | ||
0b246afa | 7130 | ret = add_to_free_space_tree(trans, info, bytenr, num_bytes); |
1e144fb8 | 7131 | if (ret) { |
66642832 | 7132 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
7133 | goto out; |
7134 | } | |
7135 | ||
0b246afa | 7136 | ret = update_block_group(trans, info, bytenr, num_bytes, 0); |
005d6427 | 7137 | if (ret) { |
66642832 | 7138 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
7139 | goto out; |
7140 | } | |
a28ec197 | 7141 | } |
fcebe456 JB |
7142 | btrfs_release_path(path); |
7143 | ||
79787eaa | 7144 | out: |
5caf2a00 | 7145 | btrfs_free_path(path); |
a28ec197 CM |
7146 | return ret; |
7147 | } | |
7148 | ||
1887be66 | 7149 | /* |
f0486c68 | 7150 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
7151 | * delayed ref for that extent as well. This searches the delayed ref tree for |
7152 | * a given extent, and if there are no other delayed refs to be processed, it | |
7153 | * removes it from the tree. | |
7154 | */ | |
7155 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 7156 | u64 bytenr) |
1887be66 CM |
7157 | { |
7158 | struct btrfs_delayed_ref_head *head; | |
7159 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 7160 | int ret = 0; |
1887be66 CM |
7161 | |
7162 | delayed_refs = &trans->transaction->delayed_refs; | |
7163 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 7164 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 7165 | if (!head) |
cf93da7b | 7166 | goto out_delayed_unlock; |
1887be66 | 7167 | |
d7df2c79 | 7168 | spin_lock(&head->lock); |
0e0adbcf | 7169 | if (!RB_EMPTY_ROOT(&head->ref_tree)) |
1887be66 CM |
7170 | goto out; |
7171 | ||
5d4f98a2 YZ |
7172 | if (head->extent_op) { |
7173 | if (!head->must_insert_reserved) | |
7174 | goto out; | |
78a6184a | 7175 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
7176 | head->extent_op = NULL; |
7177 | } | |
7178 | ||
1887be66 CM |
7179 | /* |
7180 | * waiting for the lock here would deadlock. If someone else has it | |
7181 | * locked they are already in the process of dropping it anyway | |
7182 | */ | |
7183 | if (!mutex_trylock(&head->mutex)) | |
7184 | goto out; | |
7185 | ||
7186 | /* | |
7187 | * at this point we have a head with no other entries. Go | |
7188 | * ahead and process it. | |
7189 | */ | |
c46effa6 | 7190 | rb_erase(&head->href_node, &delayed_refs->href_root); |
d278850e | 7191 | RB_CLEAR_NODE(&head->href_node); |
d7df2c79 | 7192 | atomic_dec(&delayed_refs->num_entries); |
1887be66 CM |
7193 | |
7194 | /* | |
7195 | * we don't take a ref on the node because we're removing it from the | |
7196 | * tree, so we just steal the ref the tree was holding. | |
7197 | */ | |
c3e69d58 | 7198 | delayed_refs->num_heads--; |
d7df2c79 | 7199 | if (head->processing == 0) |
c3e69d58 | 7200 | delayed_refs->num_heads_ready--; |
d7df2c79 JB |
7201 | head->processing = 0; |
7202 | spin_unlock(&head->lock); | |
1887be66 CM |
7203 | spin_unlock(&delayed_refs->lock); |
7204 | ||
f0486c68 YZ |
7205 | BUG_ON(head->extent_op); |
7206 | if (head->must_insert_reserved) | |
7207 | ret = 1; | |
7208 | ||
7209 | mutex_unlock(&head->mutex); | |
d278850e | 7210 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 7211 | return ret; |
1887be66 | 7212 | out: |
d7df2c79 | 7213 | spin_unlock(&head->lock); |
cf93da7b CM |
7214 | |
7215 | out_delayed_unlock: | |
1887be66 CM |
7216 | spin_unlock(&delayed_refs->lock); |
7217 | return 0; | |
7218 | } | |
7219 | ||
f0486c68 YZ |
7220 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
7221 | struct btrfs_root *root, | |
7222 | struct extent_buffer *buf, | |
5581a51a | 7223 | u64 parent, int last_ref) |
f0486c68 | 7224 | { |
0b246afa | 7225 | struct btrfs_fs_info *fs_info = root->fs_info; |
b150a4f1 | 7226 | int pin = 1; |
f0486c68 YZ |
7227 | int ret; |
7228 | ||
7229 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
d7eae340 OS |
7230 | int old_ref_mod, new_ref_mod; |
7231 | ||
fd708b81 JB |
7232 | btrfs_ref_tree_mod(root, buf->start, buf->len, parent, |
7233 | root->root_key.objectid, | |
7234 | btrfs_header_level(buf), 0, | |
7235 | BTRFS_DROP_DELAYED_REF); | |
7be07912 OS |
7236 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start, |
7237 | buf->len, parent, | |
0b246afa JM |
7238 | root->root_key.objectid, |
7239 | btrfs_header_level(buf), | |
7be07912 | 7240 | BTRFS_DROP_DELAYED_REF, NULL, |
d7eae340 | 7241 | &old_ref_mod, &new_ref_mod); |
79787eaa | 7242 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 7243 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
7244 | } |
7245 | ||
0a16c7d7 | 7246 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
7247 | struct btrfs_block_group_cache *cache; |
7248 | ||
f0486c68 | 7249 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 7250 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 7251 | if (!ret) |
37be25bc | 7252 | goto out; |
f0486c68 YZ |
7253 | } |
7254 | ||
4da8b76d | 7255 | pin = 0; |
0b246afa | 7256 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 7257 | |
f0486c68 | 7258 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
2ff7e61e JM |
7259 | pin_down_extent(fs_info, cache, buf->start, |
7260 | buf->len, 1); | |
6219872d | 7261 | btrfs_put_block_group(cache); |
37be25bc | 7262 | goto out; |
f0486c68 YZ |
7263 | } |
7264 | ||
7265 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
7266 | ||
7267 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 7268 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 7269 | btrfs_put_block_group(cache); |
71ff6437 | 7270 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
7271 | } |
7272 | out: | |
b150a4f1 | 7273 | if (pin) |
29d2b84c | 7274 | add_pinned_bytes(fs_info, buf->len, true, |
b150a4f1 JB |
7275 | root->root_key.objectid); |
7276 | ||
0a16c7d7 OS |
7277 | if (last_ref) { |
7278 | /* | |
7279 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
7280 | * matter anymore. | |
7281 | */ | |
7282 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
7283 | } | |
f0486c68 YZ |
7284 | } |
7285 | ||
79787eaa | 7286 | /* Can return -ENOMEM */ |
2ff7e61e | 7287 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
84f7d8e6 | 7288 | struct btrfs_root *root, |
66d7e7f0 | 7289 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, |
b06c4bf5 | 7290 | u64 owner, u64 offset) |
925baedd | 7291 | { |
84f7d8e6 | 7292 | struct btrfs_fs_info *fs_info = root->fs_info; |
d7eae340 | 7293 | int old_ref_mod, new_ref_mod; |
925baedd CM |
7294 | int ret; |
7295 | ||
f5ee5c9a | 7296 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 7297 | return 0; |
fccb84c9 | 7298 | |
fd708b81 JB |
7299 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) |
7300 | btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, | |
7301 | root_objectid, owner, offset, | |
7302 | BTRFS_DROP_DELAYED_REF); | |
7303 | ||
56bec294 CM |
7304 | /* |
7305 | * tree log blocks never actually go into the extent allocation | |
7306 | * tree, just update pinning info and exit early. | |
56bec294 | 7307 | */ |
5d4f98a2 YZ |
7308 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
7309 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 7310 | /* unlocks the pinned mutex */ |
2ff7e61e | 7311 | btrfs_pin_extent(fs_info, bytenr, num_bytes, 1); |
d7eae340 | 7312 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 7313 | ret = 0; |
5d4f98a2 | 7314 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
66d7e7f0 | 7315 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
7be07912 OS |
7316 | num_bytes, parent, |
7317 | root_objectid, (int)owner, | |
7318 | BTRFS_DROP_DELAYED_REF, NULL, | |
d7eae340 | 7319 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 7320 | } else { |
66d7e7f0 | 7321 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
7be07912 OS |
7322 | num_bytes, parent, |
7323 | root_objectid, owner, offset, | |
7324 | 0, BTRFS_DROP_DELAYED_REF, | |
d7eae340 | 7325 | &old_ref_mod, &new_ref_mod); |
56bec294 | 7326 | } |
d7eae340 | 7327 | |
29d2b84c NB |
7328 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) { |
7329 | bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID; | |
7330 | ||
7331 | add_pinned_bytes(fs_info, num_bytes, metadata, root_objectid); | |
7332 | } | |
d7eae340 | 7333 | |
925baedd CM |
7334 | return ret; |
7335 | } | |
7336 | ||
817d52f8 JB |
7337 | /* |
7338 | * when we wait for progress in the block group caching, its because | |
7339 | * our allocation attempt failed at least once. So, we must sleep | |
7340 | * and let some progress happen before we try again. | |
7341 | * | |
7342 | * This function will sleep at least once waiting for new free space to | |
7343 | * show up, and then it will check the block group free space numbers | |
7344 | * for our min num_bytes. Another option is to have it go ahead | |
7345 | * and look in the rbtree for a free extent of a given size, but this | |
7346 | * is a good start. | |
36cce922 JB |
7347 | * |
7348 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
7349 | * any of the information in this block group. | |
817d52f8 | 7350 | */ |
36cce922 | 7351 | static noinline void |
817d52f8 JB |
7352 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
7353 | u64 num_bytes) | |
7354 | { | |
11833d66 | 7355 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 7356 | |
11833d66 YZ |
7357 | caching_ctl = get_caching_control(cache); |
7358 | if (!caching_ctl) | |
36cce922 | 7359 | return; |
817d52f8 | 7360 | |
11833d66 | 7361 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 7362 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
7363 | |
7364 | put_caching_control(caching_ctl); | |
11833d66 YZ |
7365 | } |
7366 | ||
7367 | static noinline int | |
7368 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
7369 | { | |
7370 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 7371 | int ret = 0; |
11833d66 YZ |
7372 | |
7373 | caching_ctl = get_caching_control(cache); | |
7374 | if (!caching_ctl) | |
36cce922 | 7375 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
7376 | |
7377 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
7378 | if (cache->cached == BTRFS_CACHE_ERROR) |
7379 | ret = -EIO; | |
11833d66 | 7380 | put_caching_control(caching_ctl); |
36cce922 | 7381 | return ret; |
817d52f8 JB |
7382 | } |
7383 | ||
6ab0a202 JM |
7384 | static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = { |
7385 | [BTRFS_RAID_RAID10] = "raid10", | |
7386 | [BTRFS_RAID_RAID1] = "raid1", | |
7387 | [BTRFS_RAID_DUP] = "dup", | |
7388 | [BTRFS_RAID_RAID0] = "raid0", | |
7389 | [BTRFS_RAID_SINGLE] = "single", | |
7390 | [BTRFS_RAID_RAID5] = "raid5", | |
7391 | [BTRFS_RAID_RAID6] = "raid6", | |
7392 | }; | |
7393 | ||
1b8e5df6 | 7394 | static const char *get_raid_name(enum btrfs_raid_types type) |
6ab0a202 JM |
7395 | { |
7396 | if (type >= BTRFS_NR_RAID_TYPES) | |
7397 | return NULL; | |
7398 | ||
7399 | return btrfs_raid_type_names[type]; | |
7400 | } | |
7401 | ||
817d52f8 | 7402 | enum btrfs_loop_type { |
285ff5af JB |
7403 | LOOP_CACHING_NOWAIT = 0, |
7404 | LOOP_CACHING_WAIT = 1, | |
7405 | LOOP_ALLOC_CHUNK = 2, | |
7406 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
7407 | }; |
7408 | ||
e570fd27 MX |
7409 | static inline void |
7410 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
7411 | int delalloc) | |
7412 | { | |
7413 | if (delalloc) | |
7414 | down_read(&cache->data_rwsem); | |
7415 | } | |
7416 | ||
7417 | static inline void | |
7418 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
7419 | int delalloc) | |
7420 | { | |
7421 | btrfs_get_block_group(cache); | |
7422 | if (delalloc) | |
7423 | down_read(&cache->data_rwsem); | |
7424 | } | |
7425 | ||
7426 | static struct btrfs_block_group_cache * | |
7427 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
7428 | struct btrfs_free_cluster *cluster, | |
7429 | int delalloc) | |
7430 | { | |
89771cc9 | 7431 | struct btrfs_block_group_cache *used_bg = NULL; |
6719afdc | 7432 | |
e570fd27 | 7433 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
7434 | while (1) { |
7435 | used_bg = cluster->block_group; | |
7436 | if (!used_bg) | |
7437 | return NULL; | |
7438 | ||
7439 | if (used_bg == block_group) | |
e570fd27 MX |
7440 | return used_bg; |
7441 | ||
6719afdc | 7442 | btrfs_get_block_group(used_bg); |
e570fd27 | 7443 | |
6719afdc GU |
7444 | if (!delalloc) |
7445 | return used_bg; | |
e570fd27 | 7446 | |
6719afdc GU |
7447 | if (down_read_trylock(&used_bg->data_rwsem)) |
7448 | return used_bg; | |
e570fd27 | 7449 | |
6719afdc | 7450 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 7451 | |
e321f8a8 LB |
7452 | /* We should only have one-level nested. */ |
7453 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 7454 | |
6719afdc GU |
7455 | spin_lock(&cluster->refill_lock); |
7456 | if (used_bg == cluster->block_group) | |
7457 | return used_bg; | |
e570fd27 | 7458 | |
6719afdc GU |
7459 | up_read(&used_bg->data_rwsem); |
7460 | btrfs_put_block_group(used_bg); | |
7461 | } | |
e570fd27 MX |
7462 | } |
7463 | ||
7464 | static inline void | |
7465 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7466 | int delalloc) | |
7467 | { | |
7468 | if (delalloc) | |
7469 | up_read(&cache->data_rwsem); | |
7470 | btrfs_put_block_group(cache); | |
7471 | } | |
7472 | ||
fec577fb CM |
7473 | /* |
7474 | * walks the btree of allocated extents and find a hole of a given size. | |
7475 | * The key ins is changed to record the hole: | |
a4820398 | 7476 | * ins->objectid == start position |
62e2749e | 7477 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7478 | * ins->offset == the size of the hole. |
fec577fb | 7479 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7480 | * |
7481 | * If there is no suitable free space, we will record the max size of | |
7482 | * the free space extent currently. | |
fec577fb | 7483 | */ |
87bde3cd | 7484 | static noinline int find_free_extent(struct btrfs_fs_info *fs_info, |
18513091 WX |
7485 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
7486 | u64 hint_byte, struct btrfs_key *ins, | |
7487 | u64 flags, int delalloc) | |
fec577fb | 7488 | { |
80eb234a | 7489 | int ret = 0; |
0b246afa | 7490 | struct btrfs_root *root = fs_info->extent_root; |
fa9c0d79 | 7491 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7492 | struct btrfs_block_group_cache *block_group = NULL; |
81c9ad23 | 7493 | u64 search_start = 0; |
a4820398 | 7494 | u64 max_extent_size = 0; |
c759c4e1 | 7495 | u64 empty_cluster = 0; |
80eb234a | 7496 | struct btrfs_space_info *space_info; |
fa9c0d79 | 7497 | int loop = 0; |
3e72ee88 | 7498 | int index = btrfs_bg_flags_to_raid_index(flags); |
0a24325e | 7499 | bool failed_cluster_refill = false; |
1cdda9b8 | 7500 | bool failed_alloc = false; |
67377734 | 7501 | bool use_cluster = true; |
60d2adbb | 7502 | bool have_caching_bg = false; |
13a0db5a | 7503 | bool orig_have_caching_bg = false; |
a5e681d9 | 7504 | bool full_search = false; |
fec577fb | 7505 | |
0b246afa | 7506 | WARN_ON(num_bytes < fs_info->sectorsize); |
962a298f | 7507 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7508 | ins->objectid = 0; |
7509 | ins->offset = 0; | |
b1a4d965 | 7510 | |
71ff6437 | 7511 | trace_find_free_extent(fs_info, num_bytes, empty_size, flags); |
3f7de037 | 7512 | |
0b246afa | 7513 | space_info = __find_space_info(fs_info, flags); |
1b1d1f66 | 7514 | if (!space_info) { |
0b246afa | 7515 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7516 | return -ENOSPC; |
7517 | } | |
2552d17e | 7518 | |
67377734 | 7519 | /* |
4f4db217 JB |
7520 | * If our free space is heavily fragmented we may not be able to make |
7521 | * big contiguous allocations, so instead of doing the expensive search | |
7522 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7523 | * can go ahead and search for a more manageable chunk. | |
7524 | * | |
7525 | * If our max_extent_size is large enough for our allocation simply | |
7526 | * disable clustering since we will likely not be able to find enough | |
7527 | * space to create a cluster and induce latency trying. | |
67377734 | 7528 | */ |
4f4db217 JB |
7529 | if (unlikely(space_info->max_extent_size)) { |
7530 | spin_lock(&space_info->lock); | |
7531 | if (space_info->max_extent_size && | |
7532 | num_bytes > space_info->max_extent_size) { | |
7533 | ins->offset = space_info->max_extent_size; | |
7534 | spin_unlock(&space_info->lock); | |
7535 | return -ENOSPC; | |
7536 | } else if (space_info->max_extent_size) { | |
7537 | use_cluster = false; | |
7538 | } | |
7539 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7540 | } |
0f9dd46c | 7541 | |
2ff7e61e | 7542 | last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster); |
239b14b3 | 7543 | if (last_ptr) { |
fa9c0d79 CM |
7544 | spin_lock(&last_ptr->lock); |
7545 | if (last_ptr->block_group) | |
7546 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7547 | if (last_ptr->fragmented) { |
7548 | /* | |
7549 | * We still set window_start so we can keep track of the | |
7550 | * last place we found an allocation to try and save | |
7551 | * some time. | |
7552 | */ | |
7553 | hint_byte = last_ptr->window_start; | |
7554 | use_cluster = false; | |
7555 | } | |
fa9c0d79 | 7556 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7557 | } |
fa9c0d79 | 7558 | |
2ff7e61e | 7559 | search_start = max(search_start, first_logical_byte(fs_info, 0)); |
239b14b3 | 7560 | search_start = max(search_start, hint_byte); |
2552d17e | 7561 | if (search_start == hint_byte) { |
0b246afa | 7562 | block_group = btrfs_lookup_block_group(fs_info, search_start); |
817d52f8 JB |
7563 | /* |
7564 | * we don't want to use the block group if it doesn't match our | |
7565 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7566 | * |
7567 | * However if we are re-searching with an ideal block group | |
7568 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7569 | */ |
b6919a58 | 7570 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7571 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7572 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7573 | if (list_empty(&block_group->list) || |
7574 | block_group->ro) { | |
7575 | /* | |
7576 | * someone is removing this block group, | |
7577 | * we can't jump into the have_block_group | |
7578 | * target because our list pointers are not | |
7579 | * valid | |
7580 | */ | |
7581 | btrfs_put_block_group(block_group); | |
7582 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7583 | } else { |
3e72ee88 QW |
7584 | index = btrfs_bg_flags_to_raid_index( |
7585 | block_group->flags); | |
e570fd27 | 7586 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7587 | goto have_block_group; |
ccf0e725 | 7588 | } |
2552d17e | 7589 | } else if (block_group) { |
fa9c0d79 | 7590 | btrfs_put_block_group(block_group); |
2552d17e | 7591 | } |
42e70e7a | 7592 | } |
2552d17e | 7593 | search: |
60d2adbb | 7594 | have_caching_bg = false; |
3e72ee88 | 7595 | if (index == 0 || index == btrfs_bg_flags_to_raid_index(flags)) |
a5e681d9 | 7596 | full_search = true; |
80eb234a | 7597 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
7598 | list_for_each_entry(block_group, &space_info->block_groups[index], |
7599 | list) { | |
6226cb0a | 7600 | u64 offset; |
817d52f8 | 7601 | int cached; |
8a1413a2 | 7602 | |
14443937 JM |
7603 | /* If the block group is read-only, we can skip it entirely. */ |
7604 | if (unlikely(block_group->ro)) | |
7605 | continue; | |
7606 | ||
e570fd27 | 7607 | btrfs_grab_block_group(block_group, delalloc); |
2552d17e | 7608 | search_start = block_group->key.objectid; |
42e70e7a | 7609 | |
83a50de9 CM |
7610 | /* |
7611 | * this can happen if we end up cycling through all the | |
7612 | * raid types, but we want to make sure we only allocate | |
7613 | * for the proper type. | |
7614 | */ | |
b6919a58 | 7615 | if (!block_group_bits(block_group, flags)) { |
83a50de9 CM |
7616 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
7617 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
7618 | BTRFS_BLOCK_GROUP_RAID5 | |
7619 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
7620 | BTRFS_BLOCK_GROUP_RAID10; |
7621 | ||
7622 | /* | |
7623 | * if they asked for extra copies and this block group | |
7624 | * doesn't provide them, bail. This does allow us to | |
7625 | * fill raid0 from raid1. | |
7626 | */ | |
b6919a58 | 7627 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
7628 | goto loop; |
7629 | } | |
7630 | ||
2552d17e | 7631 | have_block_group: |
291c7d2f JB |
7632 | cached = block_group_cache_done(block_group); |
7633 | if (unlikely(!cached)) { | |
a5e681d9 | 7634 | have_caching_bg = true; |
f6373bf3 | 7635 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7636 | BUG_ON(ret < 0); |
7637 | ret = 0; | |
817d52f8 JB |
7638 | } |
7639 | ||
36cce922 JB |
7640 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7641 | goto loop; | |
0f9dd46c | 7642 | |
0a24325e | 7643 | /* |
062c05c4 AO |
7644 | * Ok we want to try and use the cluster allocator, so |
7645 | * lets look there | |
0a24325e | 7646 | */ |
c759c4e1 | 7647 | if (last_ptr && use_cluster) { |
215a63d1 | 7648 | struct btrfs_block_group_cache *used_block_group; |
8de972b4 | 7649 | unsigned long aligned_cluster; |
fa9c0d79 CM |
7650 | /* |
7651 | * the refill lock keeps out other | |
7652 | * people trying to start a new cluster | |
7653 | */ | |
e570fd27 MX |
7654 | used_block_group = btrfs_lock_cluster(block_group, |
7655 | last_ptr, | |
7656 | delalloc); | |
7657 | if (!used_block_group) | |
44fb5511 | 7658 | goto refill_cluster; |
274bd4fb | 7659 | |
e570fd27 MX |
7660 | if (used_block_group != block_group && |
7661 | (used_block_group->ro || | |
7662 | !block_group_bits(used_block_group, flags))) | |
7663 | goto release_cluster; | |
44fb5511 | 7664 | |
274bd4fb | 7665 | offset = btrfs_alloc_from_cluster(used_block_group, |
a4820398 MX |
7666 | last_ptr, |
7667 | num_bytes, | |
7668 | used_block_group->key.objectid, | |
7669 | &max_extent_size); | |
fa9c0d79 CM |
7670 | if (offset) { |
7671 | /* we have a block, we're done */ | |
7672 | spin_unlock(&last_ptr->refill_lock); | |
71ff6437 | 7673 | trace_btrfs_reserve_extent_cluster(fs_info, |
89d4346a MX |
7674 | used_block_group, |
7675 | search_start, num_bytes); | |
215a63d1 | 7676 | if (used_block_group != block_group) { |
e570fd27 MX |
7677 | btrfs_release_block_group(block_group, |
7678 | delalloc); | |
215a63d1 MX |
7679 | block_group = used_block_group; |
7680 | } | |
fa9c0d79 CM |
7681 | goto checks; |
7682 | } | |
7683 | ||
274bd4fb | 7684 | WARN_ON(last_ptr->block_group != used_block_group); |
e570fd27 | 7685 | release_cluster: |
062c05c4 AO |
7686 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
7687 | * set up a new clusters, so lets just skip it | |
7688 | * and let the allocator find whatever block | |
7689 | * it can find. If we reach this point, we | |
7690 | * will have tried the cluster allocator | |
7691 | * plenty of times and not have found | |
7692 | * anything, so we are likely way too | |
7693 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
7694 | * anything. |
7695 | * | |
7696 | * However, if the cluster is taken from the | |
7697 | * current block group, release the cluster | |
7698 | * first, so that we stand a better chance of | |
7699 | * succeeding in the unclustered | |
7700 | * allocation. */ | |
7701 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
e570fd27 | 7702 | used_block_group != block_group) { |
062c05c4 | 7703 | spin_unlock(&last_ptr->refill_lock); |
e570fd27 MX |
7704 | btrfs_release_block_group(used_block_group, |
7705 | delalloc); | |
062c05c4 AO |
7706 | goto unclustered_alloc; |
7707 | } | |
7708 | ||
fa9c0d79 CM |
7709 | /* |
7710 | * this cluster didn't work out, free it and | |
7711 | * start over | |
7712 | */ | |
7713 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7714 | ||
e570fd27 MX |
7715 | if (used_block_group != block_group) |
7716 | btrfs_release_block_group(used_block_group, | |
7717 | delalloc); | |
7718 | refill_cluster: | |
a5f6f719 AO |
7719 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
7720 | spin_unlock(&last_ptr->refill_lock); | |
7721 | goto unclustered_alloc; | |
7722 | } | |
7723 | ||
8de972b4 CM |
7724 | aligned_cluster = max_t(unsigned long, |
7725 | empty_cluster + empty_size, | |
7726 | block_group->full_stripe_len); | |
7727 | ||
fa9c0d79 | 7728 | /* allocate a cluster in this block group */ |
2ff7e61e | 7729 | ret = btrfs_find_space_cluster(fs_info, block_group, |
00361589 JB |
7730 | last_ptr, search_start, |
7731 | num_bytes, | |
7732 | aligned_cluster); | |
fa9c0d79 CM |
7733 | if (ret == 0) { |
7734 | /* | |
7735 | * now pull our allocation out of this | |
7736 | * cluster | |
7737 | */ | |
7738 | offset = btrfs_alloc_from_cluster(block_group, | |
a4820398 MX |
7739 | last_ptr, |
7740 | num_bytes, | |
7741 | search_start, | |
7742 | &max_extent_size); | |
fa9c0d79 CM |
7743 | if (offset) { |
7744 | /* we found one, proceed */ | |
7745 | spin_unlock(&last_ptr->refill_lock); | |
71ff6437 | 7746 | trace_btrfs_reserve_extent_cluster(fs_info, |
3f7de037 JB |
7747 | block_group, search_start, |
7748 | num_bytes); | |
fa9c0d79 CM |
7749 | goto checks; |
7750 | } | |
0a24325e JB |
7751 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
7752 | && !failed_cluster_refill) { | |
817d52f8 JB |
7753 | spin_unlock(&last_ptr->refill_lock); |
7754 | ||
0a24325e | 7755 | failed_cluster_refill = true; |
817d52f8 JB |
7756 | wait_block_group_cache_progress(block_group, |
7757 | num_bytes + empty_cluster + empty_size); | |
7758 | goto have_block_group; | |
fa9c0d79 | 7759 | } |
817d52f8 | 7760 | |
fa9c0d79 CM |
7761 | /* |
7762 | * at this point we either didn't find a cluster | |
7763 | * or we weren't able to allocate a block from our | |
7764 | * cluster. Free the cluster we've been trying | |
7765 | * to use, and go to the next block group | |
7766 | */ | |
0a24325e | 7767 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 7768 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 7769 | goto loop; |
fa9c0d79 CM |
7770 | } |
7771 | ||
062c05c4 | 7772 | unclustered_alloc: |
c759c4e1 JB |
7773 | /* |
7774 | * We are doing an unclustered alloc, set the fragmented flag so | |
7775 | * we don't bother trying to setup a cluster again until we get | |
7776 | * more space. | |
7777 | */ | |
7778 | if (unlikely(last_ptr)) { | |
7779 | spin_lock(&last_ptr->lock); | |
7780 | last_ptr->fragmented = 1; | |
7781 | spin_unlock(&last_ptr->lock); | |
7782 | } | |
0c9b36e0 LB |
7783 | if (cached) { |
7784 | struct btrfs_free_space_ctl *ctl = | |
7785 | block_group->free_space_ctl; | |
7786 | ||
7787 | spin_lock(&ctl->tree_lock); | |
7788 | if (ctl->free_space < | |
7789 | num_bytes + empty_cluster + empty_size) { | |
7790 | if (ctl->free_space > max_extent_size) | |
7791 | max_extent_size = ctl->free_space; | |
7792 | spin_unlock(&ctl->tree_lock); | |
7793 | goto loop; | |
7794 | } | |
7795 | spin_unlock(&ctl->tree_lock); | |
a5f6f719 | 7796 | } |
a5f6f719 | 7797 | |
6226cb0a | 7798 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
a4820398 MX |
7799 | num_bytes, empty_size, |
7800 | &max_extent_size); | |
1cdda9b8 JB |
7801 | /* |
7802 | * If we didn't find a chunk, and we haven't failed on this | |
7803 | * block group before, and this block group is in the middle of | |
7804 | * caching and we are ok with waiting, then go ahead and wait | |
7805 | * for progress to be made, and set failed_alloc to true. | |
7806 | * | |
7807 | * If failed_alloc is true then we've already waited on this | |
7808 | * block group once and should move on to the next block group. | |
7809 | */ | |
7810 | if (!offset && !failed_alloc && !cached && | |
7811 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 7812 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
7813 | num_bytes + empty_size); |
7814 | failed_alloc = true; | |
817d52f8 | 7815 | goto have_block_group; |
1cdda9b8 JB |
7816 | } else if (!offset) { |
7817 | goto loop; | |
817d52f8 | 7818 | } |
fa9c0d79 | 7819 | checks: |
0b246afa | 7820 | search_start = ALIGN(offset, fs_info->stripesize); |
25179201 | 7821 | |
2552d17e JB |
7822 | /* move on to the next group */ |
7823 | if (search_start + num_bytes > | |
215a63d1 MX |
7824 | block_group->key.objectid + block_group->key.offset) { |
7825 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 7826 | goto loop; |
6226cb0a | 7827 | } |
f5a31e16 | 7828 | |
f0486c68 | 7829 | if (offset < search_start) |
215a63d1 | 7830 | btrfs_add_free_space(block_group, offset, |
f0486c68 YZ |
7831 | search_start - offset); |
7832 | BUG_ON(offset > search_start); | |
2552d17e | 7833 | |
18513091 WX |
7834 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
7835 | num_bytes, delalloc); | |
f0486c68 | 7836 | if (ret == -EAGAIN) { |
215a63d1 | 7837 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 7838 | goto loop; |
0f9dd46c | 7839 | } |
9cfa3e34 | 7840 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 7841 | |
f0486c68 | 7842 | /* we are all good, lets return */ |
2552d17e JB |
7843 | ins->objectid = search_start; |
7844 | ins->offset = num_bytes; | |
d2fb3437 | 7845 | |
71ff6437 | 7846 | trace_btrfs_reserve_extent(fs_info, block_group, |
3f7de037 | 7847 | search_start, num_bytes); |
e570fd27 | 7848 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7849 | break; |
7850 | loop: | |
0a24325e | 7851 | failed_cluster_refill = false; |
1cdda9b8 | 7852 | failed_alloc = false; |
3e72ee88 QW |
7853 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != |
7854 | index); | |
e570fd27 | 7855 | btrfs_release_block_group(block_group, delalloc); |
14443937 | 7856 | cond_resched(); |
2552d17e JB |
7857 | } |
7858 | up_read(&space_info->groups_sem); | |
7859 | ||
13a0db5a | 7860 | if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg |
7861 | && !orig_have_caching_bg) | |
7862 | orig_have_caching_bg = true; | |
7863 | ||
60d2adbb MX |
7864 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
7865 | goto search; | |
7866 | ||
b742bb82 YZ |
7867 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
7868 | goto search; | |
7869 | ||
285ff5af | 7870 | /* |
ccf0e725 JB |
7871 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
7872 | * caching kthreads as we move along | |
817d52f8 JB |
7873 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
7874 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7875 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7876 | * again | |
fa9c0d79 | 7877 | */ |
723bda20 | 7878 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 7879 | index = 0; |
a5e681d9 JB |
7880 | if (loop == LOOP_CACHING_NOWAIT) { |
7881 | /* | |
7882 | * We want to skip the LOOP_CACHING_WAIT step if we | |
01327610 | 7883 | * don't have any uncached bgs and we've already done a |
a5e681d9 JB |
7884 | * full search through. |
7885 | */ | |
13a0db5a | 7886 | if (orig_have_caching_bg || !full_search) |
a5e681d9 JB |
7887 | loop = LOOP_CACHING_WAIT; |
7888 | else | |
7889 | loop = LOOP_ALLOC_CHUNK; | |
7890 | } else { | |
7891 | loop++; | |
7892 | } | |
7893 | ||
817d52f8 | 7894 | if (loop == LOOP_ALLOC_CHUNK) { |
00361589 | 7895 | struct btrfs_trans_handle *trans; |
f017f15f WS |
7896 | int exist = 0; |
7897 | ||
7898 | trans = current->journal_info; | |
7899 | if (trans) | |
7900 | exist = 1; | |
7901 | else | |
7902 | trans = btrfs_join_transaction(root); | |
00361589 | 7903 | |
00361589 JB |
7904 | if (IS_ERR(trans)) { |
7905 | ret = PTR_ERR(trans); | |
7906 | goto out; | |
7907 | } | |
7908 | ||
2ff7e61e | 7909 | ret = do_chunk_alloc(trans, fs_info, flags, |
ea658bad | 7910 | CHUNK_ALLOC_FORCE); |
a5e681d9 JB |
7911 | |
7912 | /* | |
7913 | * If we can't allocate a new chunk we've already looped | |
7914 | * through at least once, move on to the NO_EMPTY_SIZE | |
7915 | * case. | |
7916 | */ | |
7917 | if (ret == -ENOSPC) | |
7918 | loop = LOOP_NO_EMPTY_SIZE; | |
7919 | ||
ea658bad JB |
7920 | /* |
7921 | * Do not bail out on ENOSPC since we | |
7922 | * can do more things. | |
7923 | */ | |
00361589 | 7924 | if (ret < 0 && ret != -ENOSPC) |
66642832 | 7925 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
7926 | else |
7927 | ret = 0; | |
f017f15f | 7928 | if (!exist) |
3a45bb20 | 7929 | btrfs_end_transaction(trans); |
00361589 | 7930 | if (ret) |
ea658bad | 7931 | goto out; |
2552d17e JB |
7932 | } |
7933 | ||
723bda20 | 7934 | if (loop == LOOP_NO_EMPTY_SIZE) { |
a5e681d9 JB |
7935 | /* |
7936 | * Don't loop again if we already have no empty_size and | |
7937 | * no empty_cluster. | |
7938 | */ | |
7939 | if (empty_size == 0 && | |
7940 | empty_cluster == 0) { | |
7941 | ret = -ENOSPC; | |
7942 | goto out; | |
7943 | } | |
723bda20 JB |
7944 | empty_size = 0; |
7945 | empty_cluster = 0; | |
fa9c0d79 | 7946 | } |
723bda20 JB |
7947 | |
7948 | goto search; | |
2552d17e JB |
7949 | } else if (!ins->objectid) { |
7950 | ret = -ENOSPC; | |
d82a6f1d | 7951 | } else if (ins->objectid) { |
c759c4e1 JB |
7952 | if (!use_cluster && last_ptr) { |
7953 | spin_lock(&last_ptr->lock); | |
7954 | last_ptr->window_start = ins->objectid; | |
7955 | spin_unlock(&last_ptr->lock); | |
7956 | } | |
80eb234a | 7957 | ret = 0; |
be744175 | 7958 | } |
79787eaa | 7959 | out: |
4f4db217 JB |
7960 | if (ret == -ENOSPC) { |
7961 | spin_lock(&space_info->lock); | |
7962 | space_info->max_extent_size = max_extent_size; | |
7963 | spin_unlock(&space_info->lock); | |
a4820398 | 7964 | ins->offset = max_extent_size; |
4f4db217 | 7965 | } |
0f70abe2 | 7966 | return ret; |
fec577fb | 7967 | } |
ec44a35c | 7968 | |
ab8d0fc4 JM |
7969 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
7970 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 7971 | int dump_block_groups) |
0f9dd46c JB |
7972 | { |
7973 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 7974 | int index = 0; |
0f9dd46c | 7975 | |
9ed74f2d | 7976 | spin_lock(&info->lock); |
ab8d0fc4 JM |
7977 | btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull", |
7978 | info->flags, | |
4136135b LB |
7979 | info->total_bytes - btrfs_space_info_used(info, true), |
7980 | info->full ? "" : "not "); | |
ab8d0fc4 JM |
7981 | btrfs_info(fs_info, |
7982 | "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu", | |
7983 | info->total_bytes, info->bytes_used, info->bytes_pinned, | |
7984 | info->bytes_reserved, info->bytes_may_use, | |
7985 | info->bytes_readonly); | |
9ed74f2d JB |
7986 | spin_unlock(&info->lock); |
7987 | ||
7988 | if (!dump_block_groups) | |
7989 | return; | |
0f9dd46c | 7990 | |
80eb234a | 7991 | down_read(&info->groups_sem); |
b742bb82 YZ |
7992 | again: |
7993 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 7994 | spin_lock(&cache->lock); |
ab8d0fc4 JM |
7995 | btrfs_info(fs_info, |
7996 | "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s", | |
7997 | cache->key.objectid, cache->key.offset, | |
7998 | btrfs_block_group_used(&cache->item), cache->pinned, | |
7999 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
8000 | btrfs_dump_free_space(cache, bytes); |
8001 | spin_unlock(&cache->lock); | |
8002 | } | |
b742bb82 YZ |
8003 | if (++index < BTRFS_NR_RAID_TYPES) |
8004 | goto again; | |
80eb234a | 8005 | up_read(&info->groups_sem); |
0f9dd46c | 8006 | } |
e8569813 | 8007 | |
6f47c706 NB |
8008 | /* |
8009 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a | |
8010 | * hole that is at least as big as @num_bytes. | |
8011 | * | |
8012 | * @root - The root that will contain this extent | |
8013 | * | |
8014 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
8015 | * is used for accounting purposes. This value differs | |
8016 | * from @num_bytes only in the case of compressed extents. | |
8017 | * | |
8018 | * @num_bytes - Number of bytes to allocate on-disk. | |
8019 | * | |
8020 | * @min_alloc_size - Indicates the minimum amount of space that the | |
8021 | * allocator should try to satisfy. In some cases | |
8022 | * @num_bytes may be larger than what is required and if | |
8023 | * the filesystem is fragmented then allocation fails. | |
8024 | * However, the presence of @min_alloc_size gives a | |
8025 | * chance to try and satisfy the smaller allocation. | |
8026 | * | |
8027 | * @empty_size - A hint that you plan on doing more COW. This is the | |
8028 | * size in bytes the allocator should try to find free | |
8029 | * next to the block it returns. This is just a hint and | |
8030 | * may be ignored by the allocator. | |
8031 | * | |
8032 | * @hint_byte - Hint to the allocator to start searching above the byte | |
8033 | * address passed. It might be ignored. | |
8034 | * | |
8035 | * @ins - This key is modified to record the found hole. It will | |
8036 | * have the following values: | |
8037 | * ins->objectid == start position | |
8038 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
8039 | * ins->offset == the size of the hole. | |
8040 | * | |
8041 | * @is_data - Boolean flag indicating whether an extent is | |
8042 | * allocated for data (true) or metadata (false) | |
8043 | * | |
8044 | * @delalloc - Boolean flag indicating whether this allocation is for | |
8045 | * delalloc or not. If 'true' data_rwsem of block groups | |
8046 | * is going to be acquired. | |
8047 | * | |
8048 | * | |
8049 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
8050 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
8051 | * largest available hole the allocator managed to find. | |
8052 | */ | |
18513091 | 8053 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
8054 | u64 num_bytes, u64 min_alloc_size, |
8055 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 8056 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 8057 | { |
ab8d0fc4 | 8058 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 8059 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 8060 | u64 flags; |
fec577fb | 8061 | int ret; |
925baedd | 8062 | |
1b86826d | 8063 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 8064 | again: |
0b246afa | 8065 | WARN_ON(num_bytes < fs_info->sectorsize); |
87bde3cd | 8066 | ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size, |
18513091 | 8067 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 8068 | if (!ret && !is_data) { |
ab8d0fc4 | 8069 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 8070 | } else if (ret == -ENOSPC) { |
a4820398 MX |
8071 | if (!final_tried && ins->offset) { |
8072 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 8073 | num_bytes = round_down(num_bytes, |
0b246afa | 8074 | fs_info->sectorsize); |
9e622d6b | 8075 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 8076 | ram_bytes = num_bytes; |
9e622d6b MX |
8077 | if (num_bytes == min_alloc_size) |
8078 | final_tried = true; | |
8079 | goto again; | |
ab8d0fc4 | 8080 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
8081 | struct btrfs_space_info *sinfo; |
8082 | ||
ab8d0fc4 | 8083 | sinfo = __find_space_info(fs_info, flags); |
0b246afa | 8084 | btrfs_err(fs_info, |
5d163e0e JM |
8085 | "allocation failed flags %llu, wanted %llu", |
8086 | flags, num_bytes); | |
53804280 | 8087 | if (sinfo) |
ab8d0fc4 | 8088 | dump_space_info(fs_info, sinfo, num_bytes, 1); |
9e622d6b | 8089 | } |
925baedd | 8090 | } |
0f9dd46c JB |
8091 | |
8092 | return ret; | |
e6dcd2dc CM |
8093 | } |
8094 | ||
2ff7e61e | 8095 | static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
8096 | u64 start, u64 len, |
8097 | int pin, int delalloc) | |
65b51a00 | 8098 | { |
0f9dd46c | 8099 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 8100 | int ret = 0; |
0f9dd46c | 8101 | |
0b246afa | 8102 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 8103 | if (!cache) { |
0b246afa JM |
8104 | btrfs_err(fs_info, "Unable to find block group for %llu", |
8105 | start); | |
0f9dd46c JB |
8106 | return -ENOSPC; |
8107 | } | |
1f3c79a2 | 8108 | |
e688b725 | 8109 | if (pin) |
2ff7e61e | 8110 | pin_down_extent(fs_info, cache, start, len, 1); |
e688b725 | 8111 | else { |
0b246afa | 8112 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 8113 | ret = btrfs_discard_extent(fs_info, start, len, NULL); |
e688b725 | 8114 | btrfs_add_free_space(cache, start, len); |
4824f1f4 | 8115 | btrfs_free_reserved_bytes(cache, len, delalloc); |
71ff6437 | 8116 | trace_btrfs_reserved_extent_free(fs_info, start, len); |
e688b725 | 8117 | } |
31193213 | 8118 | |
fa9c0d79 | 8119 | btrfs_put_block_group(cache); |
e6dcd2dc CM |
8120 | return ret; |
8121 | } | |
8122 | ||
2ff7e61e | 8123 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 | 8124 | u64 start, u64 len, int delalloc) |
e688b725 | 8125 | { |
2ff7e61e | 8126 | return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc); |
e688b725 CM |
8127 | } |
8128 | ||
2ff7e61e | 8129 | int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info, |
e688b725 CM |
8130 | u64 start, u64 len) |
8131 | { | |
2ff7e61e | 8132 | return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0); |
e688b725 CM |
8133 | } |
8134 | ||
5d4f98a2 | 8135 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 8136 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8137 | u64 parent, u64 root_objectid, |
8138 | u64 flags, u64 owner, u64 offset, | |
8139 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc CM |
8140 | { |
8141 | int ret; | |
e6dcd2dc | 8142 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 8143 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 8144 | struct btrfs_path *path; |
5d4f98a2 YZ |
8145 | struct extent_buffer *leaf; |
8146 | int type; | |
8147 | u32 size; | |
26b8003f | 8148 | |
5d4f98a2 YZ |
8149 | if (parent > 0) |
8150 | type = BTRFS_SHARED_DATA_REF_KEY; | |
8151 | else | |
8152 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 8153 | |
5d4f98a2 | 8154 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
8155 | |
8156 | path = btrfs_alloc_path(); | |
db5b493a TI |
8157 | if (!path) |
8158 | return -ENOMEM; | |
47e4bb98 | 8159 | |
b9473439 | 8160 | path->leave_spinning = 1; |
5d4f98a2 YZ |
8161 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
8162 | ins, size); | |
79787eaa JM |
8163 | if (ret) { |
8164 | btrfs_free_path(path); | |
8165 | return ret; | |
8166 | } | |
0f9dd46c | 8167 | |
5d4f98a2 YZ |
8168 | leaf = path->nodes[0]; |
8169 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 8170 | struct btrfs_extent_item); |
5d4f98a2 YZ |
8171 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
8172 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8173 | btrfs_set_extent_flags(leaf, extent_item, | |
8174 | flags | BTRFS_EXTENT_FLAG_DATA); | |
8175 | ||
8176 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8177 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
8178 | if (parent > 0) { | |
8179 | struct btrfs_shared_data_ref *ref; | |
8180 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
8181 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8182 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
8183 | } else { | |
8184 | struct btrfs_extent_data_ref *ref; | |
8185 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
8186 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
8187 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
8188 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
8189 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
8190 | } | |
47e4bb98 CM |
8191 | |
8192 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 8193 | btrfs_free_path(path); |
f510cfec | 8194 | |
1e144fb8 OS |
8195 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
8196 | ins->offset); | |
8197 | if (ret) | |
8198 | return ret; | |
8199 | ||
6202df69 | 8200 | ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1); |
79787eaa | 8201 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8202 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8203 | ins->objectid, ins->offset); |
f5947066 CM |
8204 | BUG(); |
8205 | } | |
71ff6437 | 8206 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
8207 | return ret; |
8208 | } | |
8209 | ||
5d4f98a2 | 8210 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
2ff7e61e | 8211 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8212 | u64 parent, u64 root_objectid, |
8213 | u64 flags, struct btrfs_disk_key *key, | |
b06c4bf5 | 8214 | int level, struct btrfs_key *ins) |
e6dcd2dc CM |
8215 | { |
8216 | int ret; | |
5d4f98a2 YZ |
8217 | struct btrfs_extent_item *extent_item; |
8218 | struct btrfs_tree_block_info *block_info; | |
8219 | struct btrfs_extent_inline_ref *iref; | |
8220 | struct btrfs_path *path; | |
8221 | struct extent_buffer *leaf; | |
3173a18f | 8222 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
fcebe456 | 8223 | u64 num_bytes = ins->offset; |
0b246afa | 8224 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f JB |
8225 | |
8226 | if (!skinny_metadata) | |
8227 | size += sizeof(*block_info); | |
1c2308f8 | 8228 | |
5d4f98a2 | 8229 | path = btrfs_alloc_path(); |
857cc2fc | 8230 | if (!path) { |
2ff7e61e | 8231 | btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid, |
0b246afa | 8232 | fs_info->nodesize); |
d8926bb3 | 8233 | return -ENOMEM; |
857cc2fc | 8234 | } |
56bec294 | 8235 | |
5d4f98a2 YZ |
8236 | path->leave_spinning = 1; |
8237 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
8238 | ins, size); | |
79787eaa | 8239 | if (ret) { |
dd825259 | 8240 | btrfs_free_path(path); |
2ff7e61e | 8241 | btrfs_free_and_pin_reserved_extent(fs_info, ins->objectid, |
0b246afa | 8242 | fs_info->nodesize); |
79787eaa JM |
8243 | return ret; |
8244 | } | |
5d4f98a2 YZ |
8245 | |
8246 | leaf = path->nodes[0]; | |
8247 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
8248 | struct btrfs_extent_item); | |
8249 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
8250 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
8251 | btrfs_set_extent_flags(leaf, extent_item, | |
8252 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 8253 | |
3173a18f JB |
8254 | if (skinny_metadata) { |
8255 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
0b246afa | 8256 | num_bytes = fs_info->nodesize; |
3173a18f JB |
8257 | } else { |
8258 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
8259 | btrfs_set_tree_block_key(leaf, block_info, key); | |
8260 | btrfs_set_tree_block_level(leaf, block_info, level); | |
8261 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); | |
8262 | } | |
5d4f98a2 | 8263 | |
5d4f98a2 YZ |
8264 | if (parent > 0) { |
8265 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); | |
8266 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8267 | BTRFS_SHARED_BLOCK_REF_KEY); | |
8268 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
8269 | } else { | |
8270 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8271 | BTRFS_TREE_BLOCK_REF_KEY); | |
8272 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
8273 | } | |
8274 | ||
8275 | btrfs_mark_buffer_dirty(leaf); | |
8276 | btrfs_free_path(path); | |
8277 | ||
1e144fb8 OS |
8278 | ret = remove_from_free_space_tree(trans, fs_info, ins->objectid, |
8279 | num_bytes); | |
8280 | if (ret) | |
8281 | return ret; | |
8282 | ||
6202df69 JM |
8283 | ret = update_block_group(trans, fs_info, ins->objectid, |
8284 | fs_info->nodesize, 1); | |
79787eaa | 8285 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8286 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 8287 | ins->objectid, ins->offset); |
5d4f98a2 YZ |
8288 | BUG(); |
8289 | } | |
0be5dc67 | 8290 | |
71ff6437 | 8291 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, |
0b246afa | 8292 | fs_info->nodesize); |
5d4f98a2 YZ |
8293 | return ret; |
8294 | } | |
8295 | ||
8296 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 8297 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
8298 | u64 offset, u64 ram_bytes, |
8299 | struct btrfs_key *ins) | |
5d4f98a2 | 8300 | { |
84f7d8e6 | 8301 | struct btrfs_fs_info *fs_info = root->fs_info; |
5d4f98a2 YZ |
8302 | int ret; |
8303 | ||
84f7d8e6 | 8304 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 8305 | |
fd708b81 JB |
8306 | btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0, |
8307 | root->root_key.objectid, owner, offset, | |
8308 | BTRFS_ADD_DELAYED_EXTENT); | |
8309 | ||
0b246afa | 8310 | ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid, |
84f7d8e6 JB |
8311 | ins->offset, 0, |
8312 | root->root_key.objectid, owner, | |
7be07912 OS |
8313 | offset, ram_bytes, |
8314 | BTRFS_ADD_DELAYED_EXTENT, NULL, NULL); | |
e6dcd2dc CM |
8315 | return ret; |
8316 | } | |
e02119d5 CM |
8317 | |
8318 | /* | |
8319 | * this is used by the tree logging recovery code. It records that | |
8320 | * an extent has been allocated and makes sure to clear the free | |
8321 | * space cache bits as well | |
8322 | */ | |
5d4f98a2 | 8323 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
2ff7e61e | 8324 | struct btrfs_fs_info *fs_info, |
5d4f98a2 YZ |
8325 | u64 root_objectid, u64 owner, u64 offset, |
8326 | struct btrfs_key *ins) | |
e02119d5 CM |
8327 | { |
8328 | int ret; | |
8329 | struct btrfs_block_group_cache *block_group; | |
ed7a6948 | 8330 | struct btrfs_space_info *space_info; |
11833d66 | 8331 | |
8c2a1a30 JB |
8332 | /* |
8333 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 8334 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 8335 | */ |
0b246afa | 8336 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
8337 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
8338 | ins->offset); | |
b50c6e25 | 8339 | if (ret) |
8c2a1a30 | 8340 | return ret; |
11833d66 YZ |
8341 | } |
8342 | ||
0b246afa | 8343 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
8344 | if (!block_group) |
8345 | return -EINVAL; | |
8346 | ||
ed7a6948 WX |
8347 | space_info = block_group->space_info; |
8348 | spin_lock(&space_info->lock); | |
8349 | spin_lock(&block_group->lock); | |
8350 | space_info->bytes_reserved += ins->offset; | |
8351 | block_group->reserved += ins->offset; | |
8352 | spin_unlock(&block_group->lock); | |
8353 | spin_unlock(&space_info->lock); | |
8354 | ||
2ff7e61e | 8355 | ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid, |
5d4f98a2 | 8356 | 0, owner, offset, ins, 1); |
b50c6e25 | 8357 | btrfs_put_block_group(block_group); |
e02119d5 CM |
8358 | return ret; |
8359 | } | |
8360 | ||
48a3b636 ES |
8361 | static struct extent_buffer * |
8362 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
fe864576 | 8363 | u64 bytenr, int level) |
65b51a00 | 8364 | { |
0b246afa | 8365 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
8366 | struct extent_buffer *buf; |
8367 | ||
2ff7e61e | 8368 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8369 | if (IS_ERR(buf)) |
8370 | return buf; | |
8371 | ||
65b51a00 | 8372 | btrfs_set_header_generation(buf, trans->transid); |
85d4e461 | 8373 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 8374 | btrfs_tree_lock(buf); |
7c302b49 | 8375 | clean_tree_block(fs_info, buf); |
3083ee2e | 8376 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
8377 | |
8378 | btrfs_set_lock_blocking(buf); | |
4db8c528 | 8379 | set_extent_buffer_uptodate(buf); |
b4ce94de | 8380 | |
d0c803c4 | 8381 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 8382 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
8383 | /* |
8384 | * we allow two log transactions at a time, use different | |
8385 | * EXENT bit to differentiate dirty pages. | |
8386 | */ | |
656f30db | 8387 | if (buf->log_index == 0) |
8cef4e16 YZ |
8388 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
8389 | buf->start + buf->len - 1, GFP_NOFS); | |
8390 | else | |
8391 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 8392 | buf->start + buf->len - 1); |
d0c803c4 | 8393 | } else { |
656f30db | 8394 | buf->log_index = -1; |
d0c803c4 | 8395 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 8396 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 8397 | } |
64c12921 | 8398 | trans->dirty = true; |
b4ce94de | 8399 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
8400 | return buf; |
8401 | } | |
8402 | ||
f0486c68 YZ |
8403 | static struct btrfs_block_rsv * |
8404 | use_block_rsv(struct btrfs_trans_handle *trans, | |
8405 | struct btrfs_root *root, u32 blocksize) | |
8406 | { | |
0b246afa | 8407 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8408 | struct btrfs_block_rsv *block_rsv; |
0b246afa | 8409 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
f0486c68 | 8410 | int ret; |
d88033db | 8411 | bool global_updated = false; |
f0486c68 YZ |
8412 | |
8413 | block_rsv = get_block_rsv(trans, root); | |
8414 | ||
b586b323 MX |
8415 | if (unlikely(block_rsv->size == 0)) |
8416 | goto try_reserve; | |
d88033db | 8417 | again: |
f0486c68 YZ |
8418 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
8419 | if (!ret) | |
8420 | return block_rsv; | |
8421 | ||
b586b323 MX |
8422 | if (block_rsv->failfast) |
8423 | return ERR_PTR(ret); | |
8424 | ||
d88033db MX |
8425 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
8426 | global_updated = true; | |
0b246afa | 8427 | update_global_block_rsv(fs_info); |
d88033db MX |
8428 | goto again; |
8429 | } | |
8430 | ||
0b246afa | 8431 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
b586b323 MX |
8432 | static DEFINE_RATELIMIT_STATE(_rs, |
8433 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
8434 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
8435 | if (__ratelimit(&_rs)) | |
8436 | WARN(1, KERN_DEBUG | |
efe120a0 | 8437 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
8438 | } |
8439 | try_reserve: | |
8440 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
8441 | BTRFS_RESERVE_NO_FLUSH); | |
8442 | if (!ret) | |
8443 | return block_rsv; | |
8444 | /* | |
8445 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
8446 | * the global reserve if its space type is the same as the global |
8447 | * reservation. | |
b586b323 | 8448 | */ |
5881cfc9 MX |
8449 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
8450 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
8451 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
8452 | if (!ret) | |
8453 | return global_rsv; | |
8454 | } | |
8455 | return ERR_PTR(ret); | |
f0486c68 YZ |
8456 | } |
8457 | ||
8c2a3ca2 JB |
8458 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
8459 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 YZ |
8460 | { |
8461 | block_rsv_add_bytes(block_rsv, blocksize, 0); | |
ff6bc37e | 8462 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL); |
f0486c68 YZ |
8463 | } |
8464 | ||
fec577fb | 8465 | /* |
f0486c68 | 8466 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 8467 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 8468 | */ |
4d75f8a9 | 8469 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
8470 | struct btrfs_root *root, |
8471 | u64 parent, u64 root_objectid, | |
8472 | const struct btrfs_disk_key *key, | |
8473 | int level, u64 hint, | |
8474 | u64 empty_size) | |
fec577fb | 8475 | { |
0b246afa | 8476 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 8477 | struct btrfs_key ins; |
f0486c68 | 8478 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 8479 | struct extent_buffer *buf; |
67b7859e | 8480 | struct btrfs_delayed_extent_op *extent_op; |
f0486c68 YZ |
8481 | u64 flags = 0; |
8482 | int ret; | |
0b246afa JM |
8483 | u32 blocksize = fs_info->nodesize; |
8484 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 8485 | |
05653ef3 | 8486 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 8487 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 8488 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
fe864576 | 8489 | level); |
faa2dbf0 JB |
8490 | if (!IS_ERR(buf)) |
8491 | root->alloc_bytenr += blocksize; | |
8492 | return buf; | |
8493 | } | |
05653ef3 | 8494 | #endif |
fccb84c9 | 8495 | |
f0486c68 YZ |
8496 | block_rsv = use_block_rsv(trans, root, blocksize); |
8497 | if (IS_ERR(block_rsv)) | |
8498 | return ERR_CAST(block_rsv); | |
8499 | ||
18513091 | 8500 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 8501 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
8502 | if (ret) |
8503 | goto out_unuse; | |
55c69072 | 8504 | |
fe864576 | 8505 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level); |
67b7859e OS |
8506 | if (IS_ERR(buf)) { |
8507 | ret = PTR_ERR(buf); | |
8508 | goto out_free_reserved; | |
8509 | } | |
f0486c68 YZ |
8510 | |
8511 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8512 | if (parent == 0) | |
8513 | parent = ins.objectid; | |
8514 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8515 | } else | |
8516 | BUG_ON(parent > 0); | |
8517 | ||
8518 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 8519 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
8520 | if (!extent_op) { |
8521 | ret = -ENOMEM; | |
8522 | goto out_free_buf; | |
8523 | } | |
f0486c68 YZ |
8524 | if (key) |
8525 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8526 | else | |
8527 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8528 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
8529 | extent_op->update_key = skinny_metadata ? false : true; |
8530 | extent_op->update_flags = true; | |
8531 | extent_op->is_data = false; | |
b1c79e09 | 8532 | extent_op->level = level; |
f0486c68 | 8533 | |
fd708b81 JB |
8534 | btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent, |
8535 | root_objectid, level, 0, | |
8536 | BTRFS_ADD_DELAYED_EXTENT); | |
7be07912 OS |
8537 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid, |
8538 | ins.offset, parent, | |
8539 | root_objectid, level, | |
67b7859e | 8540 | BTRFS_ADD_DELAYED_EXTENT, |
7be07912 | 8541 | extent_op, NULL, NULL); |
67b7859e OS |
8542 | if (ret) |
8543 | goto out_free_delayed; | |
f0486c68 | 8544 | } |
fec577fb | 8545 | return buf; |
67b7859e OS |
8546 | |
8547 | out_free_delayed: | |
8548 | btrfs_free_delayed_extent_op(extent_op); | |
8549 | out_free_buf: | |
8550 | free_extent_buffer(buf); | |
8551 | out_free_reserved: | |
2ff7e61e | 8552 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 8553 | out_unuse: |
0b246afa | 8554 | unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 8555 | return ERR_PTR(ret); |
fec577fb | 8556 | } |
a28ec197 | 8557 | |
2c47e605 YZ |
8558 | struct walk_control { |
8559 | u64 refs[BTRFS_MAX_LEVEL]; | |
8560 | u64 flags[BTRFS_MAX_LEVEL]; | |
8561 | struct btrfs_key update_progress; | |
8562 | int stage; | |
8563 | int level; | |
8564 | int shared_level; | |
8565 | int update_ref; | |
8566 | int keep_locks; | |
1c4850e2 YZ |
8567 | int reada_slot; |
8568 | int reada_count; | |
66d7e7f0 | 8569 | int for_reloc; |
2c47e605 YZ |
8570 | }; |
8571 | ||
8572 | #define DROP_REFERENCE 1 | |
8573 | #define UPDATE_BACKREF 2 | |
8574 | ||
1c4850e2 YZ |
8575 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8576 | struct btrfs_root *root, | |
8577 | struct walk_control *wc, | |
8578 | struct btrfs_path *path) | |
6407bf6d | 8579 | { |
0b246afa | 8580 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8581 | u64 bytenr; |
8582 | u64 generation; | |
8583 | u64 refs; | |
94fcca9f | 8584 | u64 flags; |
5d4f98a2 | 8585 | u32 nritems; |
1c4850e2 YZ |
8586 | struct btrfs_key key; |
8587 | struct extent_buffer *eb; | |
6407bf6d | 8588 | int ret; |
1c4850e2 YZ |
8589 | int slot; |
8590 | int nread = 0; | |
6407bf6d | 8591 | |
1c4850e2 YZ |
8592 | if (path->slots[wc->level] < wc->reada_slot) { |
8593 | wc->reada_count = wc->reada_count * 2 / 3; | |
8594 | wc->reada_count = max(wc->reada_count, 2); | |
8595 | } else { | |
8596 | wc->reada_count = wc->reada_count * 3 / 2; | |
8597 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 8598 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 8599 | } |
7bb86316 | 8600 | |
1c4850e2 YZ |
8601 | eb = path->nodes[wc->level]; |
8602 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 8603 | |
1c4850e2 YZ |
8604 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8605 | if (nread >= wc->reada_count) | |
8606 | break; | |
bd56b302 | 8607 | |
2dd3e67b | 8608 | cond_resched(); |
1c4850e2 YZ |
8609 | bytenr = btrfs_node_blockptr(eb, slot); |
8610 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8611 | |
1c4850e2 YZ |
8612 | if (slot == path->slots[wc->level]) |
8613 | goto reada; | |
5d4f98a2 | 8614 | |
1c4850e2 YZ |
8615 | if (wc->stage == UPDATE_BACKREF && |
8616 | generation <= root->root_key.offset) | |
bd56b302 CM |
8617 | continue; |
8618 | ||
94fcca9f | 8619 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 8620 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
8621 | wc->level - 1, 1, &refs, |
8622 | &flags); | |
79787eaa JM |
8623 | /* We don't care about errors in readahead. */ |
8624 | if (ret < 0) | |
8625 | continue; | |
94fcca9f YZ |
8626 | BUG_ON(refs == 0); |
8627 | ||
1c4850e2 | 8628 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8629 | if (refs == 1) |
8630 | goto reada; | |
bd56b302 | 8631 | |
94fcca9f YZ |
8632 | if (wc->level == 1 && |
8633 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8634 | continue; | |
1c4850e2 YZ |
8635 | if (!wc->update_ref || |
8636 | generation <= root->root_key.offset) | |
8637 | continue; | |
8638 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8639 | ret = btrfs_comp_cpu_keys(&key, | |
8640 | &wc->update_progress); | |
8641 | if (ret < 0) | |
8642 | continue; | |
94fcca9f YZ |
8643 | } else { |
8644 | if (wc->level == 1 && | |
8645 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8646 | continue; | |
6407bf6d | 8647 | } |
1c4850e2 | 8648 | reada: |
2ff7e61e | 8649 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 8650 | nread++; |
20524f02 | 8651 | } |
1c4850e2 | 8652 | wc->reada_slot = slot; |
20524f02 | 8653 | } |
2c47e605 | 8654 | |
f82d02d9 | 8655 | /* |
2c016dc2 | 8656 | * helper to process tree block while walking down the tree. |
2c47e605 | 8657 | * |
2c47e605 YZ |
8658 | * when wc->stage == UPDATE_BACKREF, this function updates |
8659 | * back refs for pointers in the block. | |
8660 | * | |
8661 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8662 | */ |
2c47e605 | 8663 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8664 | struct btrfs_root *root, |
2c47e605 | 8665 | struct btrfs_path *path, |
94fcca9f | 8666 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8667 | { |
2ff7e61e | 8668 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
8669 | int level = wc->level; |
8670 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 8671 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
8672 | int ret; |
8673 | ||
2c47e605 YZ |
8674 | if (wc->stage == UPDATE_BACKREF && |
8675 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8676 | return 1; | |
f82d02d9 | 8677 | |
2c47e605 YZ |
8678 | /* |
8679 | * when reference count of tree block is 1, it won't increase | |
8680 | * again. once full backref flag is set, we never clear it. | |
8681 | */ | |
94fcca9f YZ |
8682 | if (lookup_info && |
8683 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8684 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 8685 | BUG_ON(!path->locks[level]); |
2ff7e61e | 8686 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8687 | eb->start, level, 1, |
2c47e605 YZ |
8688 | &wc->refs[level], |
8689 | &wc->flags[level]); | |
79787eaa JM |
8690 | BUG_ON(ret == -ENOMEM); |
8691 | if (ret) | |
8692 | return ret; | |
2c47e605 YZ |
8693 | BUG_ON(wc->refs[level] == 0); |
8694 | } | |
5d4f98a2 | 8695 | |
2c47e605 YZ |
8696 | if (wc->stage == DROP_REFERENCE) { |
8697 | if (wc->refs[level] > 1) | |
8698 | return 1; | |
f82d02d9 | 8699 | |
2c47e605 | 8700 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 8701 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8702 | path->locks[level] = 0; |
8703 | } | |
8704 | return 0; | |
8705 | } | |
f82d02d9 | 8706 | |
2c47e605 YZ |
8707 | /* wc->stage == UPDATE_BACKREF */ |
8708 | if (!(wc->flags[level] & flag)) { | |
8709 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 8710 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 8711 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 8712 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8713 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8714 | ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start, |
b1c79e09 JB |
8715 | eb->len, flag, |
8716 | btrfs_header_level(eb), 0); | |
79787eaa | 8717 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
8718 | wc->flags[level] |= flag; |
8719 | } | |
8720 | ||
8721 | /* | |
8722 | * the block is shared by multiple trees, so it's not good to | |
8723 | * keep the tree lock | |
8724 | */ | |
8725 | if (path->locks[level] && level > 0) { | |
bd681513 | 8726 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8727 | path->locks[level] = 0; |
8728 | } | |
8729 | return 0; | |
8730 | } | |
8731 | ||
1c4850e2 | 8732 | /* |
2c016dc2 | 8733 | * helper to process tree block pointer. |
1c4850e2 YZ |
8734 | * |
8735 | * when wc->stage == DROP_REFERENCE, this function checks | |
8736 | * reference count of the block pointed to. if the block | |
8737 | * is shared and we need update back refs for the subtree | |
8738 | * rooted at the block, this function changes wc->stage to | |
8739 | * UPDATE_BACKREF. if the block is shared and there is no | |
8740 | * need to update back, this function drops the reference | |
8741 | * to the block. | |
8742 | * | |
8743 | * NOTE: return value 1 means we should stop walking down. | |
8744 | */ | |
8745 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8746 | struct btrfs_root *root, | |
8747 | struct btrfs_path *path, | |
94fcca9f | 8748 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 8749 | { |
0b246afa | 8750 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8751 | u64 bytenr; |
8752 | u64 generation; | |
8753 | u64 parent; | |
8754 | u32 blocksize; | |
8755 | struct btrfs_key key; | |
581c1760 | 8756 | struct btrfs_key first_key; |
1c4850e2 YZ |
8757 | struct extent_buffer *next; |
8758 | int level = wc->level; | |
8759 | int reada = 0; | |
8760 | int ret = 0; | |
1152651a | 8761 | bool need_account = false; |
1c4850e2 YZ |
8762 | |
8763 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8764 | path->slots[level]); | |
8765 | /* | |
8766 | * if the lower level block was created before the snapshot | |
8767 | * was created, we know there is no need to update back refs | |
8768 | * for the subtree | |
8769 | */ | |
8770 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8771 | generation <= root->root_key.offset) { |
8772 | *lookup_info = 1; | |
1c4850e2 | 8773 | return 1; |
94fcca9f | 8774 | } |
1c4850e2 YZ |
8775 | |
8776 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
581c1760 QW |
8777 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
8778 | path->slots[level]); | |
0b246afa | 8779 | blocksize = fs_info->nodesize; |
1c4850e2 | 8780 | |
0b246afa | 8781 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 8782 | if (!next) { |
2ff7e61e | 8783 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8784 | if (IS_ERR(next)) |
8785 | return PTR_ERR(next); | |
8786 | ||
b2aaaa3b JB |
8787 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8788 | level - 1); | |
1c4850e2 YZ |
8789 | reada = 1; |
8790 | } | |
8791 | btrfs_tree_lock(next); | |
8792 | btrfs_set_lock_blocking(next); | |
8793 | ||
2ff7e61e | 8794 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
8795 | &wc->refs[level - 1], |
8796 | &wc->flags[level - 1]); | |
4867268c JB |
8797 | if (ret < 0) |
8798 | goto out_unlock; | |
79787eaa | 8799 | |
c2cf52eb | 8800 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 8801 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
8802 | ret = -EIO; |
8803 | goto out_unlock; | |
c2cf52eb | 8804 | } |
94fcca9f | 8805 | *lookup_info = 0; |
1c4850e2 | 8806 | |
94fcca9f | 8807 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8808 | if (wc->refs[level - 1] > 1) { |
1152651a | 8809 | need_account = true; |
94fcca9f YZ |
8810 | if (level == 1 && |
8811 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8812 | goto skip; | |
8813 | ||
1c4850e2 YZ |
8814 | if (!wc->update_ref || |
8815 | generation <= root->root_key.offset) | |
8816 | goto skip; | |
8817 | ||
8818 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8819 | path->slots[level]); | |
8820 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8821 | if (ret < 0) | |
8822 | goto skip; | |
8823 | ||
8824 | wc->stage = UPDATE_BACKREF; | |
8825 | wc->shared_level = level - 1; | |
8826 | } | |
94fcca9f YZ |
8827 | } else { |
8828 | if (level == 1 && | |
8829 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8830 | goto skip; | |
1c4850e2 YZ |
8831 | } |
8832 | ||
b9fab919 | 8833 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8834 | btrfs_tree_unlock(next); |
8835 | free_extent_buffer(next); | |
8836 | next = NULL; | |
94fcca9f | 8837 | *lookup_info = 1; |
1c4850e2 YZ |
8838 | } |
8839 | ||
8840 | if (!next) { | |
8841 | if (reada && level == 1) | |
8842 | reada_walk_down(trans, root, wc, path); | |
581c1760 QW |
8843 | next = read_tree_block(fs_info, bytenr, generation, level - 1, |
8844 | &first_key); | |
64c043de LB |
8845 | if (IS_ERR(next)) { |
8846 | return PTR_ERR(next); | |
8847 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8848 | free_extent_buffer(next); |
97d9a8a4 | 8849 | return -EIO; |
416bc658 | 8850 | } |
1c4850e2 YZ |
8851 | btrfs_tree_lock(next); |
8852 | btrfs_set_lock_blocking(next); | |
8853 | } | |
8854 | ||
8855 | level--; | |
4867268c JB |
8856 | ASSERT(level == btrfs_header_level(next)); |
8857 | if (level != btrfs_header_level(next)) { | |
8858 | btrfs_err(root->fs_info, "mismatched level"); | |
8859 | ret = -EIO; | |
8860 | goto out_unlock; | |
8861 | } | |
1c4850e2 YZ |
8862 | path->nodes[level] = next; |
8863 | path->slots[level] = 0; | |
bd681513 | 8864 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8865 | wc->level = level; |
8866 | if (wc->level == 1) | |
8867 | wc->reada_slot = 0; | |
8868 | return 0; | |
8869 | skip: | |
8870 | wc->refs[level - 1] = 0; | |
8871 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
8872 | if (wc->stage == DROP_REFERENCE) { |
8873 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8874 | parent = path->nodes[level]->start; | |
8875 | } else { | |
4867268c | 8876 | ASSERT(root->root_key.objectid == |
94fcca9f | 8877 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
8878 | if (root->root_key.objectid != |
8879 | btrfs_header_owner(path->nodes[level])) { | |
8880 | btrfs_err(root->fs_info, | |
8881 | "mismatched block owner"); | |
8882 | ret = -EIO; | |
8883 | goto out_unlock; | |
8884 | } | |
94fcca9f YZ |
8885 | parent = 0; |
8886 | } | |
1c4850e2 | 8887 | |
1152651a | 8888 | if (need_account) { |
33d1f05c QW |
8889 | ret = btrfs_qgroup_trace_subtree(trans, root, next, |
8890 | generation, level - 1); | |
1152651a | 8891 | if (ret) { |
0b246afa | 8892 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8893 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
8894 | ret); | |
1152651a MF |
8895 | } |
8896 | } | |
84f7d8e6 | 8897 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, |
2ff7e61e JM |
8898 | parent, root->root_key.objectid, |
8899 | level - 1, 0); | |
4867268c JB |
8900 | if (ret) |
8901 | goto out_unlock; | |
1c4850e2 | 8902 | } |
4867268c JB |
8903 | |
8904 | *lookup_info = 1; | |
8905 | ret = 1; | |
8906 | ||
8907 | out_unlock: | |
1c4850e2 YZ |
8908 | btrfs_tree_unlock(next); |
8909 | free_extent_buffer(next); | |
4867268c JB |
8910 | |
8911 | return ret; | |
1c4850e2 YZ |
8912 | } |
8913 | ||
2c47e605 | 8914 | /* |
2c016dc2 | 8915 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
8916 | * |
8917 | * when wc->stage == DROP_REFERENCE, this function drops | |
8918 | * reference count on the block. | |
8919 | * | |
8920 | * when wc->stage == UPDATE_BACKREF, this function changes | |
8921 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
8922 | * to UPDATE_BACKREF previously while processing the block. | |
8923 | * | |
8924 | * NOTE: return value 1 means we should stop walking up. | |
8925 | */ | |
8926 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
8927 | struct btrfs_root *root, | |
8928 | struct btrfs_path *path, | |
8929 | struct walk_control *wc) | |
8930 | { | |
0b246afa | 8931 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8932 | int ret; |
2c47e605 YZ |
8933 | int level = wc->level; |
8934 | struct extent_buffer *eb = path->nodes[level]; | |
8935 | u64 parent = 0; | |
8936 | ||
8937 | if (wc->stage == UPDATE_BACKREF) { | |
8938 | BUG_ON(wc->shared_level < level); | |
8939 | if (level < wc->shared_level) | |
8940 | goto out; | |
8941 | ||
2c47e605 YZ |
8942 | ret = find_next_key(path, level + 1, &wc->update_progress); |
8943 | if (ret > 0) | |
8944 | wc->update_ref = 0; | |
8945 | ||
8946 | wc->stage = DROP_REFERENCE; | |
8947 | wc->shared_level = -1; | |
8948 | path->slots[level] = 0; | |
8949 | ||
8950 | /* | |
8951 | * check reference count again if the block isn't locked. | |
8952 | * we should start walking down the tree again if reference | |
8953 | * count is one. | |
8954 | */ | |
8955 | if (!path->locks[level]) { | |
8956 | BUG_ON(level == 0); | |
8957 | btrfs_tree_lock(eb); | |
8958 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8959 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8960 | |
2ff7e61e | 8961 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8962 | eb->start, level, 1, |
2c47e605 YZ |
8963 | &wc->refs[level], |
8964 | &wc->flags[level]); | |
79787eaa JM |
8965 | if (ret < 0) { |
8966 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 8967 | path->locks[level] = 0; |
79787eaa JM |
8968 | return ret; |
8969 | } | |
2c47e605 YZ |
8970 | BUG_ON(wc->refs[level] == 0); |
8971 | if (wc->refs[level] == 1) { | |
bd681513 | 8972 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 8973 | path->locks[level] = 0; |
2c47e605 YZ |
8974 | return 1; |
8975 | } | |
f82d02d9 | 8976 | } |
2c47e605 | 8977 | } |
f82d02d9 | 8978 | |
2c47e605 YZ |
8979 | /* wc->stage == DROP_REFERENCE */ |
8980 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 8981 | |
2c47e605 YZ |
8982 | if (wc->refs[level] == 1) { |
8983 | if (level == 0) { | |
8984 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 8985 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 8986 | else |
e339a6b0 | 8987 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8988 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8989 | ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb); |
1152651a | 8990 | if (ret) { |
0b246afa | 8991 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8992 | "error %d accounting leaf items. Quota is out of sync, rescan required.", |
8993 | ret); | |
1152651a | 8994 | } |
2c47e605 YZ |
8995 | } |
8996 | /* make block locked assertion in clean_tree_block happy */ | |
8997 | if (!path->locks[level] && | |
8998 | btrfs_header_generation(eb) == trans->transid) { | |
8999 | btrfs_tree_lock(eb); | |
9000 | btrfs_set_lock_blocking(eb); | |
bd681513 | 9001 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9002 | } |
7c302b49 | 9003 | clean_tree_block(fs_info, eb); |
2c47e605 YZ |
9004 | } |
9005 | ||
9006 | if (eb == root->node) { | |
9007 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
9008 | parent = eb->start; | |
9009 | else | |
9010 | BUG_ON(root->root_key.objectid != | |
9011 | btrfs_header_owner(eb)); | |
9012 | } else { | |
9013 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
9014 | parent = path->nodes[level + 1]->start; | |
9015 | else | |
9016 | BUG_ON(root->root_key.objectid != | |
9017 | btrfs_header_owner(path->nodes[level + 1])); | |
f82d02d9 | 9018 | } |
f82d02d9 | 9019 | |
5581a51a | 9020 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
9021 | out: |
9022 | wc->refs[level] = 0; | |
9023 | wc->flags[level] = 0; | |
f0486c68 | 9024 | return 0; |
2c47e605 YZ |
9025 | } |
9026 | ||
9027 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
9028 | struct btrfs_root *root, | |
9029 | struct btrfs_path *path, | |
9030 | struct walk_control *wc) | |
9031 | { | |
2c47e605 | 9032 | int level = wc->level; |
94fcca9f | 9033 | int lookup_info = 1; |
2c47e605 YZ |
9034 | int ret; |
9035 | ||
9036 | while (level >= 0) { | |
94fcca9f | 9037 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
9038 | if (ret > 0) |
9039 | break; | |
9040 | ||
9041 | if (level == 0) | |
9042 | break; | |
9043 | ||
7a7965f8 YZ |
9044 | if (path->slots[level] >= |
9045 | btrfs_header_nritems(path->nodes[level])) | |
9046 | break; | |
9047 | ||
94fcca9f | 9048 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
9049 | if (ret > 0) { |
9050 | path->slots[level]++; | |
9051 | continue; | |
90d2c51d MX |
9052 | } else if (ret < 0) |
9053 | return ret; | |
1c4850e2 | 9054 | level = wc->level; |
f82d02d9 | 9055 | } |
f82d02d9 YZ |
9056 | return 0; |
9057 | } | |
9058 | ||
d397712b | 9059 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 9060 | struct btrfs_root *root, |
f82d02d9 | 9061 | struct btrfs_path *path, |
2c47e605 | 9062 | struct walk_control *wc, int max_level) |
20524f02 | 9063 | { |
2c47e605 | 9064 | int level = wc->level; |
20524f02 | 9065 | int ret; |
9f3a7427 | 9066 | |
2c47e605 YZ |
9067 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
9068 | while (level < max_level && path->nodes[level]) { | |
9069 | wc->level = level; | |
9070 | if (path->slots[level] + 1 < | |
9071 | btrfs_header_nritems(path->nodes[level])) { | |
9072 | path->slots[level]++; | |
20524f02 CM |
9073 | return 0; |
9074 | } else { | |
2c47e605 YZ |
9075 | ret = walk_up_proc(trans, root, path, wc); |
9076 | if (ret > 0) | |
9077 | return 0; | |
bd56b302 | 9078 | |
2c47e605 | 9079 | if (path->locks[level]) { |
bd681513 CM |
9080 | btrfs_tree_unlock_rw(path->nodes[level], |
9081 | path->locks[level]); | |
2c47e605 | 9082 | path->locks[level] = 0; |
f82d02d9 | 9083 | } |
2c47e605 YZ |
9084 | free_extent_buffer(path->nodes[level]); |
9085 | path->nodes[level] = NULL; | |
9086 | level++; | |
20524f02 CM |
9087 | } |
9088 | } | |
9089 | return 1; | |
9090 | } | |
9091 | ||
9aca1d51 | 9092 | /* |
2c47e605 YZ |
9093 | * drop a subvolume tree. |
9094 | * | |
9095 | * this function traverses the tree freeing any blocks that only | |
9096 | * referenced by the tree. | |
9097 | * | |
9098 | * when a shared tree block is found. this function decreases its | |
9099 | * reference count by one. if update_ref is true, this function | |
9100 | * also make sure backrefs for the shared block and all lower level | |
9101 | * blocks are properly updated. | |
9d1a2a3a DS |
9102 | * |
9103 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 9104 | */ |
2c536799 | 9105 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
9106 | struct btrfs_block_rsv *block_rsv, int update_ref, |
9107 | int for_reloc) | |
20524f02 | 9108 | { |
ab8d0fc4 | 9109 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 9110 | struct btrfs_path *path; |
2c47e605 | 9111 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 9112 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 9113 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
9114 | struct walk_control *wc; |
9115 | struct btrfs_key key; | |
9116 | int err = 0; | |
9117 | int ret; | |
9118 | int level; | |
d29a9f62 | 9119 | bool root_dropped = false; |
20524f02 | 9120 | |
ab8d0fc4 | 9121 | btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid); |
1152651a | 9122 | |
5caf2a00 | 9123 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
9124 | if (!path) { |
9125 | err = -ENOMEM; | |
9126 | goto out; | |
9127 | } | |
20524f02 | 9128 | |
2c47e605 | 9129 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
9130 | if (!wc) { |
9131 | btrfs_free_path(path); | |
cb1b69f4 TI |
9132 | err = -ENOMEM; |
9133 | goto out; | |
38a1a919 | 9134 | } |
2c47e605 | 9135 | |
a22285a6 | 9136 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9137 | if (IS_ERR(trans)) { |
9138 | err = PTR_ERR(trans); | |
9139 | goto out_free; | |
9140 | } | |
98d5dc13 | 9141 | |
3fd0a558 YZ |
9142 | if (block_rsv) |
9143 | trans->block_rsv = block_rsv; | |
2c47e605 | 9144 | |
9f3a7427 | 9145 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 9146 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
9147 | path->nodes[level] = btrfs_lock_root_node(root); |
9148 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 9149 | path->slots[level] = 0; |
bd681513 | 9150 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9151 | memset(&wc->update_progress, 0, |
9152 | sizeof(wc->update_progress)); | |
9f3a7427 | 9153 | } else { |
9f3a7427 | 9154 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
9155 | memcpy(&wc->update_progress, &key, |
9156 | sizeof(wc->update_progress)); | |
9157 | ||
6702ed49 | 9158 | level = root_item->drop_level; |
2c47e605 | 9159 | BUG_ON(level == 0); |
6702ed49 | 9160 | path->lowest_level = level; |
2c47e605 YZ |
9161 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
9162 | path->lowest_level = 0; | |
9163 | if (ret < 0) { | |
9164 | err = ret; | |
79787eaa | 9165 | goto out_end_trans; |
9f3a7427 | 9166 | } |
1c4850e2 | 9167 | WARN_ON(ret > 0); |
2c47e605 | 9168 | |
7d9eb12c CM |
9169 | /* |
9170 | * unlock our path, this is safe because only this | |
9171 | * function is allowed to delete this snapshot | |
9172 | */ | |
5d4f98a2 | 9173 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
9174 | |
9175 | level = btrfs_header_level(root->node); | |
9176 | while (1) { | |
9177 | btrfs_tree_lock(path->nodes[level]); | |
9178 | btrfs_set_lock_blocking(path->nodes[level]); | |
fec386ac | 9179 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 9180 | |
2ff7e61e | 9181 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 9182 | path->nodes[level]->start, |
3173a18f | 9183 | level, 1, &wc->refs[level], |
2c47e605 | 9184 | &wc->flags[level]); |
79787eaa JM |
9185 | if (ret < 0) { |
9186 | err = ret; | |
9187 | goto out_end_trans; | |
9188 | } | |
2c47e605 YZ |
9189 | BUG_ON(wc->refs[level] == 0); |
9190 | ||
9191 | if (level == root_item->drop_level) | |
9192 | break; | |
9193 | ||
9194 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 9195 | path->locks[level] = 0; |
2c47e605 YZ |
9196 | WARN_ON(wc->refs[level] != 1); |
9197 | level--; | |
9198 | } | |
9f3a7427 | 9199 | } |
2c47e605 YZ |
9200 | |
9201 | wc->level = level; | |
9202 | wc->shared_level = -1; | |
9203 | wc->stage = DROP_REFERENCE; | |
9204 | wc->update_ref = update_ref; | |
9205 | wc->keep_locks = 0; | |
66d7e7f0 | 9206 | wc->for_reloc = for_reloc; |
0b246afa | 9207 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 9208 | |
d397712b | 9209 | while (1) { |
9d1a2a3a | 9210 | |
2c47e605 YZ |
9211 | ret = walk_down_tree(trans, root, path, wc); |
9212 | if (ret < 0) { | |
9213 | err = ret; | |
20524f02 | 9214 | break; |
2c47e605 | 9215 | } |
9aca1d51 | 9216 | |
2c47e605 YZ |
9217 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
9218 | if (ret < 0) { | |
9219 | err = ret; | |
20524f02 | 9220 | break; |
2c47e605 YZ |
9221 | } |
9222 | ||
9223 | if (ret > 0) { | |
9224 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
9225 | break; |
9226 | } | |
2c47e605 YZ |
9227 | |
9228 | if (wc->stage == DROP_REFERENCE) { | |
9229 | level = wc->level; | |
9230 | btrfs_node_key(path->nodes[level], | |
9231 | &root_item->drop_progress, | |
9232 | path->slots[level]); | |
9233 | root_item->drop_level = level; | |
9234 | } | |
9235 | ||
9236 | BUG_ON(wc->level == 0); | |
3a45bb20 | 9237 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 9238 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
9239 | ret = btrfs_update_root(trans, tree_root, |
9240 | &root->root_key, | |
9241 | root_item); | |
79787eaa | 9242 | if (ret) { |
66642832 | 9243 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9244 | err = ret; |
9245 | goto out_end_trans; | |
9246 | } | |
2c47e605 | 9247 | |
3a45bb20 | 9248 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 9249 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
9250 | btrfs_debug(fs_info, |
9251 | "drop snapshot early exit"); | |
3c8f2422 JB |
9252 | err = -EAGAIN; |
9253 | goto out_free; | |
9254 | } | |
9255 | ||
a22285a6 | 9256 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9257 | if (IS_ERR(trans)) { |
9258 | err = PTR_ERR(trans); | |
9259 | goto out_free; | |
9260 | } | |
3fd0a558 YZ |
9261 | if (block_rsv) |
9262 | trans->block_rsv = block_rsv; | |
c3e69d58 | 9263 | } |
20524f02 | 9264 | } |
b3b4aa74 | 9265 | btrfs_release_path(path); |
79787eaa JM |
9266 | if (err) |
9267 | goto out_end_trans; | |
2c47e605 | 9268 | |
1cd5447e | 9269 | ret = btrfs_del_root(trans, fs_info, &root->root_key); |
79787eaa | 9270 | if (ret) { |
66642832 | 9271 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 9272 | err = ret; |
79787eaa JM |
9273 | goto out_end_trans; |
9274 | } | |
2c47e605 | 9275 | |
76dda93c | 9276 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
9277 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
9278 | NULL, NULL); | |
79787eaa | 9279 | if (ret < 0) { |
66642832 | 9280 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9281 | err = ret; |
9282 | goto out_end_trans; | |
9283 | } else if (ret > 0) { | |
84cd948c JB |
9284 | /* if we fail to delete the orphan item this time |
9285 | * around, it'll get picked up the next time. | |
9286 | * | |
9287 | * The most common failure here is just -ENOENT. | |
9288 | */ | |
9289 | btrfs_del_orphan_item(trans, tree_root, | |
9290 | root->root_key.objectid); | |
76dda93c YZ |
9291 | } |
9292 | } | |
9293 | ||
27cdeb70 | 9294 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 9295 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
9296 | } else { |
9297 | free_extent_buffer(root->node); | |
9298 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 9299 | btrfs_put_fs_root(root); |
76dda93c | 9300 | } |
d29a9f62 | 9301 | root_dropped = true; |
79787eaa | 9302 | out_end_trans: |
3a45bb20 | 9303 | btrfs_end_transaction_throttle(trans); |
79787eaa | 9304 | out_free: |
2c47e605 | 9305 | kfree(wc); |
5caf2a00 | 9306 | btrfs_free_path(path); |
cb1b69f4 | 9307 | out: |
d29a9f62 JB |
9308 | /* |
9309 | * So if we need to stop dropping the snapshot for whatever reason we | |
9310 | * need to make sure to add it back to the dead root list so that we | |
9311 | * keep trying to do the work later. This also cleans up roots if we | |
9312 | * don't have it in the radix (like when we recover after a power fail | |
9313 | * or unmount) so we don't leak memory. | |
9314 | */ | |
897ca819 | 9315 | if (!for_reloc && !root_dropped) |
d29a9f62 | 9316 | btrfs_add_dead_root(root); |
90515e7f | 9317 | if (err && err != -EAGAIN) |
ab8d0fc4 | 9318 | btrfs_handle_fs_error(fs_info, err, NULL); |
2c536799 | 9319 | return err; |
20524f02 | 9320 | } |
9078a3e1 | 9321 | |
2c47e605 YZ |
9322 | /* |
9323 | * drop subtree rooted at tree block 'node'. | |
9324 | * | |
9325 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 9326 | * only used by relocation code |
2c47e605 | 9327 | */ |
f82d02d9 YZ |
9328 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
9329 | struct btrfs_root *root, | |
9330 | struct extent_buffer *node, | |
9331 | struct extent_buffer *parent) | |
9332 | { | |
0b246afa | 9333 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 9334 | struct btrfs_path *path; |
2c47e605 | 9335 | struct walk_control *wc; |
f82d02d9 YZ |
9336 | int level; |
9337 | int parent_level; | |
9338 | int ret = 0; | |
9339 | int wret; | |
9340 | ||
2c47e605 YZ |
9341 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
9342 | ||
f82d02d9 | 9343 | path = btrfs_alloc_path(); |
db5b493a TI |
9344 | if (!path) |
9345 | return -ENOMEM; | |
f82d02d9 | 9346 | |
2c47e605 | 9347 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
9348 | if (!wc) { |
9349 | btrfs_free_path(path); | |
9350 | return -ENOMEM; | |
9351 | } | |
2c47e605 | 9352 | |
b9447ef8 | 9353 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
9354 | parent_level = btrfs_header_level(parent); |
9355 | extent_buffer_get(parent); | |
9356 | path->nodes[parent_level] = parent; | |
9357 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9358 | ||
b9447ef8 | 9359 | btrfs_assert_tree_locked(node); |
f82d02d9 | 9360 | level = btrfs_header_level(node); |
f82d02d9 YZ |
9361 | path->nodes[level] = node; |
9362 | path->slots[level] = 0; | |
bd681513 | 9363 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9364 | |
9365 | wc->refs[parent_level] = 1; | |
9366 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9367 | wc->level = level; | |
9368 | wc->shared_level = -1; | |
9369 | wc->stage = DROP_REFERENCE; | |
9370 | wc->update_ref = 0; | |
9371 | wc->keep_locks = 1; | |
66d7e7f0 | 9372 | wc->for_reloc = 1; |
0b246afa | 9373 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
9374 | |
9375 | while (1) { | |
2c47e605 YZ |
9376 | wret = walk_down_tree(trans, root, path, wc); |
9377 | if (wret < 0) { | |
f82d02d9 | 9378 | ret = wret; |
f82d02d9 | 9379 | break; |
2c47e605 | 9380 | } |
f82d02d9 | 9381 | |
2c47e605 | 9382 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9383 | if (wret < 0) |
9384 | ret = wret; | |
9385 | if (wret != 0) | |
9386 | break; | |
9387 | } | |
9388 | ||
2c47e605 | 9389 | kfree(wc); |
f82d02d9 YZ |
9390 | btrfs_free_path(path); |
9391 | return ret; | |
9392 | } | |
9393 | ||
6202df69 | 9394 | static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c CM |
9395 | { |
9396 | u64 num_devices; | |
fc67c450 | 9397 | u64 stripped; |
e4d8ec0f | 9398 | |
fc67c450 ID |
9399 | /* |
9400 | * if restripe for this chunk_type is on pick target profile and | |
9401 | * return, otherwise do the usual balance | |
9402 | */ | |
6202df69 | 9403 | stripped = get_restripe_target(fs_info, flags); |
fc67c450 ID |
9404 | if (stripped) |
9405 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9406 | |
6202df69 | 9407 | num_devices = fs_info->fs_devices->rw_devices; |
cd02dca5 | 9408 | |
fc67c450 | 9409 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 9410 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
9411 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
9412 | ||
ec44a35c CM |
9413 | if (num_devices == 1) { |
9414 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9415 | stripped = flags & ~stripped; | |
9416 | ||
9417 | /* turn raid0 into single device chunks */ | |
9418 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9419 | return stripped; | |
9420 | ||
9421 | /* turn mirroring into duplication */ | |
9422 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9423 | BTRFS_BLOCK_GROUP_RAID10)) | |
9424 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9425 | } else { |
9426 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9427 | if (flags & stripped) |
9428 | return flags; | |
9429 | ||
9430 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9431 | stripped = flags & ~stripped; | |
9432 | ||
9433 | /* switch duplicated blocks with raid1 */ | |
9434 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9435 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9436 | ||
e3176ca2 | 9437 | /* this is drive concat, leave it alone */ |
ec44a35c | 9438 | } |
e3176ca2 | 9439 | |
ec44a35c CM |
9440 | return flags; |
9441 | } | |
9442 | ||
868f401a | 9443 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9444 | { |
f0486c68 YZ |
9445 | struct btrfs_space_info *sinfo = cache->space_info; |
9446 | u64 num_bytes; | |
199c36ea | 9447 | u64 min_allocable_bytes; |
f0486c68 | 9448 | int ret = -ENOSPC; |
0ef3e66b | 9449 | |
199c36ea MX |
9450 | /* |
9451 | * We need some metadata space and system metadata space for | |
9452 | * allocating chunks in some corner cases until we force to set | |
9453 | * it to be readonly. | |
9454 | */ | |
9455 | if ((sinfo->flags & | |
9456 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9457 | !force) | |
ee22184b | 9458 | min_allocable_bytes = SZ_1M; |
199c36ea MX |
9459 | else |
9460 | min_allocable_bytes = 0; | |
9461 | ||
f0486c68 YZ |
9462 | spin_lock(&sinfo->lock); |
9463 | spin_lock(&cache->lock); | |
61cfea9b W |
9464 | |
9465 | if (cache->ro) { | |
868f401a | 9466 | cache->ro++; |
61cfea9b W |
9467 | ret = 0; |
9468 | goto out; | |
9469 | } | |
9470 | ||
f0486c68 YZ |
9471 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9472 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
9473 | ||
4136135b | 9474 | if (btrfs_space_info_used(sinfo, true) + num_bytes + |
37be25bc | 9475 | min_allocable_bytes <= sinfo->total_bytes) { |
f0486c68 | 9476 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9477 | cache->ro++; |
633c0aad | 9478 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9479 | ret = 0; |
9480 | } | |
61cfea9b | 9481 | out: |
f0486c68 YZ |
9482 | spin_unlock(&cache->lock); |
9483 | spin_unlock(&sinfo->lock); | |
9484 | return ret; | |
9485 | } | |
7d9eb12c | 9486 | |
5e00f193 | 9487 | int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info, |
f0486c68 | 9488 | struct btrfs_block_group_cache *cache) |
c286ac48 | 9489 | |
f0486c68 YZ |
9490 | { |
9491 | struct btrfs_trans_handle *trans; | |
9492 | u64 alloc_flags; | |
9493 | int ret; | |
7d9eb12c | 9494 | |
1bbc621e | 9495 | again: |
5e00f193 | 9496 | trans = btrfs_join_transaction(fs_info->extent_root); |
79787eaa JM |
9497 | if (IS_ERR(trans)) |
9498 | return PTR_ERR(trans); | |
5d4f98a2 | 9499 | |
1bbc621e CM |
9500 | /* |
9501 | * we're not allowed to set block groups readonly after the dirty | |
9502 | * block groups cache has started writing. If it already started, | |
9503 | * back off and let this transaction commit | |
9504 | */ | |
0b246afa | 9505 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c | 9506 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9507 | u64 transid = trans->transid; |
9508 | ||
0b246afa | 9509 | mutex_unlock(&fs_info->ro_block_group_mutex); |
3a45bb20 | 9510 | btrfs_end_transaction(trans); |
1bbc621e | 9511 | |
2ff7e61e | 9512 | ret = btrfs_wait_for_commit(fs_info, transid); |
1bbc621e CM |
9513 | if (ret) |
9514 | return ret; | |
9515 | goto again; | |
9516 | } | |
9517 | ||
153c35b6 CM |
9518 | /* |
9519 | * if we are changing raid levels, try to allocate a corresponding | |
9520 | * block group with the new raid level. | |
9521 | */ | |
0b246afa | 9522 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
153c35b6 | 9523 | if (alloc_flags != cache->flags) { |
2ff7e61e | 9524 | ret = do_chunk_alloc(trans, fs_info, alloc_flags, |
153c35b6 CM |
9525 | CHUNK_ALLOC_FORCE); |
9526 | /* | |
9527 | * ENOSPC is allowed here, we may have enough space | |
9528 | * already allocated at the new raid level to | |
9529 | * carry on | |
9530 | */ | |
9531 | if (ret == -ENOSPC) | |
9532 | ret = 0; | |
9533 | if (ret < 0) | |
9534 | goto out; | |
9535 | } | |
1bbc621e | 9536 | |
868f401a | 9537 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9538 | if (!ret) |
9539 | goto out; | |
2ff7e61e JM |
9540 | alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags); |
9541 | ret = do_chunk_alloc(trans, fs_info, alloc_flags, | |
0e4f8f88 | 9542 | CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9543 | if (ret < 0) |
9544 | goto out; | |
868f401a | 9545 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9546 | out: |
2f081088 | 9547 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 9548 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
34441361 | 9549 | mutex_lock(&fs_info->chunk_mutex); |
2ff7e61e | 9550 | check_system_chunk(trans, fs_info, alloc_flags); |
34441361 | 9551 | mutex_unlock(&fs_info->chunk_mutex); |
2f081088 | 9552 | } |
0b246afa | 9553 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2f081088 | 9554 | |
3a45bb20 | 9555 | btrfs_end_transaction(trans); |
f0486c68 YZ |
9556 | return ret; |
9557 | } | |
5d4f98a2 | 9558 | |
c87f08ca | 9559 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
2ff7e61e | 9560 | struct btrfs_fs_info *fs_info, u64 type) |
c87f08ca | 9561 | { |
2ff7e61e JM |
9562 | u64 alloc_flags = get_alloc_profile(fs_info, type); |
9563 | ||
9564 | return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE); | |
c87f08ca CM |
9565 | } |
9566 | ||
6d07bcec MX |
9567 | /* |
9568 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9569 | * space_info. takes mirrors into account. |
6d07bcec | 9570 | */ |
633c0aad | 9571 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9572 | { |
9573 | struct btrfs_block_group_cache *block_group; | |
9574 | u64 free_bytes = 0; | |
9575 | int factor; | |
9576 | ||
01327610 | 9577 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
9578 | if (list_empty(&sinfo->ro_bgs)) |
9579 | return 0; | |
9580 | ||
9581 | spin_lock(&sinfo->lock); | |
9582 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9583 | spin_lock(&block_group->lock); |
9584 | ||
9585 | if (!block_group->ro) { | |
9586 | spin_unlock(&block_group->lock); | |
9587 | continue; | |
9588 | } | |
9589 | ||
9590 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9591 | BTRFS_BLOCK_GROUP_RAID10 | | |
9592 | BTRFS_BLOCK_GROUP_DUP)) | |
9593 | factor = 2; | |
9594 | else | |
9595 | factor = 1; | |
9596 | ||
9597 | free_bytes += (block_group->key.offset - | |
9598 | btrfs_block_group_used(&block_group->item)) * | |
9599 | factor; | |
9600 | ||
9601 | spin_unlock(&block_group->lock); | |
9602 | } | |
6d07bcec MX |
9603 | spin_unlock(&sinfo->lock); |
9604 | ||
9605 | return free_bytes; | |
9606 | } | |
9607 | ||
2ff7e61e | 9608 | void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9609 | { |
f0486c68 YZ |
9610 | struct btrfs_space_info *sinfo = cache->space_info; |
9611 | u64 num_bytes; | |
9612 | ||
9613 | BUG_ON(!cache->ro); | |
9614 | ||
9615 | spin_lock(&sinfo->lock); | |
9616 | spin_lock(&cache->lock); | |
868f401a Z |
9617 | if (!--cache->ro) { |
9618 | num_bytes = cache->key.offset - cache->reserved - | |
9619 | cache->pinned - cache->bytes_super - | |
9620 | btrfs_block_group_used(&cache->item); | |
9621 | sinfo->bytes_readonly -= num_bytes; | |
9622 | list_del_init(&cache->ro_list); | |
9623 | } | |
f0486c68 YZ |
9624 | spin_unlock(&cache->lock); |
9625 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9626 | } |
9627 | ||
ba1bf481 JB |
9628 | /* |
9629 | * checks to see if its even possible to relocate this block group. | |
9630 | * | |
9631 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9632 | * ok to go ahead and try. | |
9633 | */ | |
6bccf3ab | 9634 | int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr) |
1a40e23b | 9635 | { |
6bccf3ab | 9636 | struct btrfs_root *root = fs_info->extent_root; |
ba1bf481 JB |
9637 | struct btrfs_block_group_cache *block_group; |
9638 | struct btrfs_space_info *space_info; | |
0b246afa | 9639 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
ba1bf481 | 9640 | struct btrfs_device *device; |
6df9a95e | 9641 | struct btrfs_trans_handle *trans; |
cdcb725c | 9642 | u64 min_free; |
6719db6a JB |
9643 | u64 dev_min = 1; |
9644 | u64 dev_nr = 0; | |
4a5e98f5 | 9645 | u64 target; |
0305bc27 | 9646 | int debug; |
cdcb725c | 9647 | int index; |
ba1bf481 JB |
9648 | int full = 0; |
9649 | int ret = 0; | |
1a40e23b | 9650 | |
0b246afa | 9651 | debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG); |
0305bc27 | 9652 | |
0b246afa | 9653 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
1a40e23b | 9654 | |
ba1bf481 | 9655 | /* odd, couldn't find the block group, leave it alone */ |
0305bc27 QW |
9656 | if (!block_group) { |
9657 | if (debug) | |
0b246afa | 9658 | btrfs_warn(fs_info, |
0305bc27 QW |
9659 | "can't find block group for bytenr %llu", |
9660 | bytenr); | |
ba1bf481 | 9661 | return -1; |
0305bc27 | 9662 | } |
1a40e23b | 9663 | |
cdcb725c | 9664 | min_free = btrfs_block_group_used(&block_group->item); |
9665 | ||
ba1bf481 | 9666 | /* no bytes used, we're good */ |
cdcb725c | 9667 | if (!min_free) |
1a40e23b ZY |
9668 | goto out; |
9669 | ||
ba1bf481 JB |
9670 | space_info = block_group->space_info; |
9671 | spin_lock(&space_info->lock); | |
17d217fe | 9672 | |
ba1bf481 | 9673 | full = space_info->full; |
17d217fe | 9674 | |
ba1bf481 JB |
9675 | /* |
9676 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9677 | * relocate it unless we're able to allocate a new chunk below. |
9678 | * | |
9679 | * Otherwise, we need to make sure we have room in the space to handle | |
9680 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9681 | */ |
7ce618db | 9682 | if ((space_info->total_bytes != block_group->key.offset) && |
4136135b LB |
9683 | (btrfs_space_info_used(space_info, false) + min_free < |
9684 | space_info->total_bytes)) { | |
ba1bf481 JB |
9685 | spin_unlock(&space_info->lock); |
9686 | goto out; | |
17d217fe | 9687 | } |
ba1bf481 | 9688 | spin_unlock(&space_info->lock); |
ea8c2819 | 9689 | |
ba1bf481 JB |
9690 | /* |
9691 | * ok we don't have enough space, but maybe we have free space on our | |
9692 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9693 | * alloc devices and guess if we have enough space. if this block |
9694 | * group is going to be restriped, run checks against the target | |
9695 | * profile instead of the current one. | |
ba1bf481 JB |
9696 | */ |
9697 | ret = -1; | |
ea8c2819 | 9698 | |
cdcb725c | 9699 | /* |
9700 | * index: | |
9701 | * 0: raid10 | |
9702 | * 1: raid1 | |
9703 | * 2: dup | |
9704 | * 3: raid0 | |
9705 | * 4: single | |
9706 | */ | |
0b246afa | 9707 | target = get_restripe_target(fs_info, block_group->flags); |
4a5e98f5 | 9708 | if (target) { |
3e72ee88 | 9709 | index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9710 | } else { |
9711 | /* | |
9712 | * this is just a balance, so if we were marked as full | |
9713 | * we know there is no space for a new chunk | |
9714 | */ | |
0305bc27 QW |
9715 | if (full) { |
9716 | if (debug) | |
0b246afa JM |
9717 | btrfs_warn(fs_info, |
9718 | "no space to alloc new chunk for block group %llu", | |
9719 | block_group->key.objectid); | |
4a5e98f5 | 9720 | goto out; |
0305bc27 | 9721 | } |
4a5e98f5 | 9722 | |
3e72ee88 | 9723 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
4a5e98f5 ID |
9724 | } |
9725 | ||
e6ec716f | 9726 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9727 | dev_min = 4; |
6719db6a JB |
9728 | /* Divide by 2 */ |
9729 | min_free >>= 1; | |
e6ec716f | 9730 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9731 | dev_min = 2; |
e6ec716f | 9732 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9733 | /* Multiply by 2 */ |
9734 | min_free <<= 1; | |
e6ec716f | 9735 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9736 | dev_min = fs_devices->rw_devices; |
47c5713f | 9737 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9738 | } |
9739 | ||
6df9a95e JB |
9740 | /* We need to do this so that we can look at pending chunks */ |
9741 | trans = btrfs_join_transaction(root); | |
9742 | if (IS_ERR(trans)) { | |
9743 | ret = PTR_ERR(trans); | |
9744 | goto out; | |
9745 | } | |
9746 | ||
0b246afa | 9747 | mutex_lock(&fs_info->chunk_mutex); |
ba1bf481 | 9748 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
7bfc837d | 9749 | u64 dev_offset; |
56bec294 | 9750 | |
ba1bf481 JB |
9751 | /* |
9752 | * check to make sure we can actually find a chunk with enough | |
9753 | * space to fit our block group in. | |
9754 | */ | |
63a212ab | 9755 | if (device->total_bytes > device->bytes_used + min_free && |
401e29c1 | 9756 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
6df9a95e | 9757 | ret = find_free_dev_extent(trans, device, min_free, |
7bfc837d | 9758 | &dev_offset, NULL); |
ba1bf481 | 9759 | if (!ret) |
cdcb725c | 9760 | dev_nr++; |
9761 | ||
9762 | if (dev_nr >= dev_min) | |
73e48b27 | 9763 | break; |
cdcb725c | 9764 | |
ba1bf481 | 9765 | ret = -1; |
725c8463 | 9766 | } |
edbd8d4e | 9767 | } |
0305bc27 | 9768 | if (debug && ret == -1) |
0b246afa JM |
9769 | btrfs_warn(fs_info, |
9770 | "no space to allocate a new chunk for block group %llu", | |
9771 | block_group->key.objectid); | |
9772 | mutex_unlock(&fs_info->chunk_mutex); | |
3a45bb20 | 9773 | btrfs_end_transaction(trans); |
edbd8d4e | 9774 | out: |
ba1bf481 | 9775 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9776 | return ret; |
9777 | } | |
9778 | ||
6bccf3ab JM |
9779 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
9780 | struct btrfs_path *path, | |
9781 | struct btrfs_key *key) | |
0b86a832 | 9782 | { |
6bccf3ab | 9783 | struct btrfs_root *root = fs_info->extent_root; |
925baedd | 9784 | int ret = 0; |
0b86a832 CM |
9785 | struct btrfs_key found_key; |
9786 | struct extent_buffer *leaf; | |
9787 | int slot; | |
edbd8d4e | 9788 | |
0b86a832 CM |
9789 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9790 | if (ret < 0) | |
925baedd CM |
9791 | goto out; |
9792 | ||
d397712b | 9793 | while (1) { |
0b86a832 | 9794 | slot = path->slots[0]; |
edbd8d4e | 9795 | leaf = path->nodes[0]; |
0b86a832 CM |
9796 | if (slot >= btrfs_header_nritems(leaf)) { |
9797 | ret = btrfs_next_leaf(root, path); | |
9798 | if (ret == 0) | |
9799 | continue; | |
9800 | if (ret < 0) | |
925baedd | 9801 | goto out; |
0b86a832 | 9802 | break; |
edbd8d4e | 9803 | } |
0b86a832 | 9804 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9805 | |
0b86a832 | 9806 | if (found_key.objectid >= key->objectid && |
925baedd | 9807 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6fb37b75 LB |
9808 | struct extent_map_tree *em_tree; |
9809 | struct extent_map *em; | |
9810 | ||
9811 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
9812 | read_lock(&em_tree->lock); | |
9813 | em = lookup_extent_mapping(em_tree, found_key.objectid, | |
9814 | found_key.offset); | |
9815 | read_unlock(&em_tree->lock); | |
9816 | if (!em) { | |
0b246afa | 9817 | btrfs_err(fs_info, |
6fb37b75 LB |
9818 | "logical %llu len %llu found bg but no related chunk", |
9819 | found_key.objectid, found_key.offset); | |
9820 | ret = -ENOENT; | |
9821 | } else { | |
9822 | ret = 0; | |
9823 | } | |
187ee58c | 9824 | free_extent_map(em); |
925baedd CM |
9825 | goto out; |
9826 | } | |
0b86a832 | 9827 | path->slots[0]++; |
edbd8d4e | 9828 | } |
925baedd | 9829 | out: |
0b86a832 | 9830 | return ret; |
edbd8d4e CM |
9831 | } |
9832 | ||
0af3d00b JB |
9833 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
9834 | { | |
9835 | struct btrfs_block_group_cache *block_group; | |
9836 | u64 last = 0; | |
9837 | ||
9838 | while (1) { | |
9839 | struct inode *inode; | |
9840 | ||
9841 | block_group = btrfs_lookup_first_block_group(info, last); | |
9842 | while (block_group) { | |
9843 | spin_lock(&block_group->lock); | |
9844 | if (block_group->iref) | |
9845 | break; | |
9846 | spin_unlock(&block_group->lock); | |
2ff7e61e | 9847 | block_group = next_block_group(info, block_group); |
0af3d00b JB |
9848 | } |
9849 | if (!block_group) { | |
9850 | if (last == 0) | |
9851 | break; | |
9852 | last = 0; | |
9853 | continue; | |
9854 | } | |
9855 | ||
9856 | inode = block_group->inode; | |
9857 | block_group->iref = 0; | |
9858 | block_group->inode = NULL; | |
9859 | spin_unlock(&block_group->lock); | |
f3bca802 | 9860 | ASSERT(block_group->io_ctl.inode == NULL); |
0af3d00b JB |
9861 | iput(inode); |
9862 | last = block_group->key.objectid + block_group->key.offset; | |
9863 | btrfs_put_block_group(block_group); | |
9864 | } | |
9865 | } | |
9866 | ||
5cdd7db6 FM |
9867 | /* |
9868 | * Must be called only after stopping all workers, since we could have block | |
9869 | * group caching kthreads running, and therefore they could race with us if we | |
9870 | * freed the block groups before stopping them. | |
9871 | */ | |
1a40e23b ZY |
9872 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9873 | { | |
9874 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 9875 | struct btrfs_space_info *space_info; |
11833d66 | 9876 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
9877 | struct rb_node *n; |
9878 | ||
9e351cc8 | 9879 | down_write(&info->commit_root_sem); |
11833d66 YZ |
9880 | while (!list_empty(&info->caching_block_groups)) { |
9881 | caching_ctl = list_entry(info->caching_block_groups.next, | |
9882 | struct btrfs_caching_control, list); | |
9883 | list_del(&caching_ctl->list); | |
9884 | put_caching_control(caching_ctl); | |
9885 | } | |
9e351cc8 | 9886 | up_write(&info->commit_root_sem); |
11833d66 | 9887 | |
47ab2a6c JB |
9888 | spin_lock(&info->unused_bgs_lock); |
9889 | while (!list_empty(&info->unused_bgs)) { | |
9890 | block_group = list_first_entry(&info->unused_bgs, | |
9891 | struct btrfs_block_group_cache, | |
9892 | bg_list); | |
9893 | list_del_init(&block_group->bg_list); | |
9894 | btrfs_put_block_group(block_group); | |
9895 | } | |
9896 | spin_unlock(&info->unused_bgs_lock); | |
9897 | ||
1a40e23b ZY |
9898 | spin_lock(&info->block_group_cache_lock); |
9899 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
9900 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
9901 | cache_node); | |
1a40e23b ZY |
9902 | rb_erase(&block_group->cache_node, |
9903 | &info->block_group_cache_tree); | |
01eacb27 | 9904 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
9905 | spin_unlock(&info->block_group_cache_lock); |
9906 | ||
80eb234a | 9907 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 9908 | list_del(&block_group->list); |
80eb234a | 9909 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 9910 | |
3c14874a JB |
9911 | /* |
9912 | * We haven't cached this block group, which means we could | |
9913 | * possibly have excluded extents on this block group. | |
9914 | */ | |
36cce922 JB |
9915 | if (block_group->cached == BTRFS_CACHE_NO || |
9916 | block_group->cached == BTRFS_CACHE_ERROR) | |
2ff7e61e | 9917 | free_excluded_extents(info, block_group); |
3c14874a | 9918 | |
817d52f8 | 9919 | btrfs_remove_free_space_cache(block_group); |
5cdd7db6 | 9920 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); |
f3bca802 LB |
9921 | ASSERT(list_empty(&block_group->dirty_list)); |
9922 | ASSERT(list_empty(&block_group->io_list)); | |
9923 | ASSERT(list_empty(&block_group->bg_list)); | |
9924 | ASSERT(atomic_read(&block_group->count) == 1); | |
11dfe35a | 9925 | btrfs_put_block_group(block_group); |
d899e052 YZ |
9926 | |
9927 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
9928 | } |
9929 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
9930 | |
9931 | /* now that all the block groups are freed, go through and | |
9932 | * free all the space_info structs. This is only called during | |
9933 | * the final stages of unmount, and so we know nobody is | |
9934 | * using them. We call synchronize_rcu() once before we start, | |
9935 | * just to be on the safe side. | |
9936 | */ | |
9937 | synchronize_rcu(); | |
9938 | ||
8929ecfa YZ |
9939 | release_global_block_rsv(info); |
9940 | ||
67871254 | 9941 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
9942 | int i; |
9943 | ||
4184ea7f CM |
9944 | space_info = list_entry(info->space_info.next, |
9945 | struct btrfs_space_info, | |
9946 | list); | |
d555b6c3 JB |
9947 | |
9948 | /* | |
9949 | * Do not hide this behind enospc_debug, this is actually | |
9950 | * important and indicates a real bug if this happens. | |
9951 | */ | |
9952 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
b069e0c3 | 9953 | space_info->bytes_reserved > 0 || |
d555b6c3 | 9954 | space_info->bytes_may_use > 0)) |
ab8d0fc4 | 9955 | dump_space_info(info, space_info, 0, 0); |
4184ea7f | 9956 | list_del(&space_info->list); |
6ab0a202 JM |
9957 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
9958 | struct kobject *kobj; | |
c1895442 JM |
9959 | kobj = space_info->block_group_kobjs[i]; |
9960 | space_info->block_group_kobjs[i] = NULL; | |
9961 | if (kobj) { | |
6ab0a202 JM |
9962 | kobject_del(kobj); |
9963 | kobject_put(kobj); | |
9964 | } | |
9965 | } | |
9966 | kobject_del(&space_info->kobj); | |
9967 | kobject_put(&space_info->kobj); | |
4184ea7f | 9968 | } |
1a40e23b ZY |
9969 | return 0; |
9970 | } | |
9971 | ||
75cb379d JM |
9972 | /* link_block_group will queue up kobjects to add when we're reclaim-safe */ |
9973 | void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info) | |
9974 | { | |
9975 | struct btrfs_space_info *space_info; | |
9976 | struct raid_kobject *rkobj; | |
9977 | LIST_HEAD(list); | |
9978 | int index; | |
9979 | int ret = 0; | |
9980 | ||
9981 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
9982 | list_splice_init(&fs_info->pending_raid_kobjs, &list); | |
9983 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
9984 | ||
9985 | list_for_each_entry(rkobj, &list, list) { | |
9986 | space_info = __find_space_info(fs_info, rkobj->flags); | |
9987 | index = btrfs_bg_flags_to_raid_index(rkobj->flags); | |
9988 | ||
9989 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
9990 | "%s", get_raid_name(index)); | |
9991 | if (ret) { | |
9992 | kobject_put(&rkobj->kobj); | |
9993 | break; | |
9994 | } | |
9995 | } | |
9996 | if (ret) | |
9997 | btrfs_warn(fs_info, | |
9998 | "failed to add kobject for block cache, ignoring"); | |
9999 | } | |
10000 | ||
c434d21c | 10001 | static void link_block_group(struct btrfs_block_group_cache *cache) |
b742bb82 | 10002 | { |
c434d21c | 10003 | struct btrfs_space_info *space_info = cache->space_info; |
75cb379d | 10004 | struct btrfs_fs_info *fs_info = cache->fs_info; |
3e72ee88 | 10005 | int index = btrfs_bg_flags_to_raid_index(cache->flags); |
ed55b6ac | 10006 | bool first = false; |
b742bb82 YZ |
10007 | |
10008 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
10009 | if (list_empty(&space_info->block_groups[index])) |
10010 | first = true; | |
10011 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
10012 | up_write(&space_info->groups_sem); | |
10013 | ||
10014 | if (first) { | |
75cb379d JM |
10015 | struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
10016 | if (!rkobj) { | |
10017 | btrfs_warn(cache->fs_info, | |
10018 | "couldn't alloc memory for raid level kobject"); | |
10019 | return; | |
6ab0a202 | 10020 | } |
75cb379d JM |
10021 | rkobj->flags = cache->flags; |
10022 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
10023 | ||
10024 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
10025 | list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs); | |
10026 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
c1895442 | 10027 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 10028 | } |
b742bb82 YZ |
10029 | } |
10030 | ||
920e4a58 | 10031 | static struct btrfs_block_group_cache * |
2ff7e61e JM |
10032 | btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info, |
10033 | u64 start, u64 size) | |
920e4a58 MX |
10034 | { |
10035 | struct btrfs_block_group_cache *cache; | |
10036 | ||
10037 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
10038 | if (!cache) | |
10039 | return NULL; | |
10040 | ||
10041 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
10042 | GFP_NOFS); | |
10043 | if (!cache->free_space_ctl) { | |
10044 | kfree(cache); | |
10045 | return NULL; | |
10046 | } | |
10047 | ||
10048 | cache->key.objectid = start; | |
10049 | cache->key.offset = size; | |
10050 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
10051 | ||
0b246afa | 10052 | cache->fs_info = fs_info; |
e4ff5fb5 | 10053 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); |
1e144fb8 OS |
10054 | set_free_space_tree_thresholds(cache); |
10055 | ||
920e4a58 MX |
10056 | atomic_set(&cache->count, 1); |
10057 | spin_lock_init(&cache->lock); | |
e570fd27 | 10058 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
10059 | INIT_LIST_HEAD(&cache->list); |
10060 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 10061 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 10062 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 10063 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 10064 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 10065 | btrfs_init_free_space_ctl(cache); |
04216820 | 10066 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 10067 | mutex_init(&cache->free_space_lock); |
0966a7b1 | 10068 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); |
920e4a58 MX |
10069 | |
10070 | return cache; | |
10071 | } | |
10072 | ||
5b4aacef | 10073 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
9078a3e1 CM |
10074 | { |
10075 | struct btrfs_path *path; | |
10076 | int ret; | |
9078a3e1 | 10077 | struct btrfs_block_group_cache *cache; |
6324fbf3 | 10078 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
10079 | struct btrfs_key key; |
10080 | struct btrfs_key found_key; | |
5f39d397 | 10081 | struct extent_buffer *leaf; |
0af3d00b JB |
10082 | int need_clear = 0; |
10083 | u64 cache_gen; | |
49303381 LB |
10084 | u64 feature; |
10085 | int mixed; | |
10086 | ||
10087 | feature = btrfs_super_incompat_flags(info->super_copy); | |
10088 | mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS); | |
96b5179d | 10089 | |
9078a3e1 | 10090 | key.objectid = 0; |
0b86a832 | 10091 | key.offset = 0; |
962a298f | 10092 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
10093 | path = btrfs_alloc_path(); |
10094 | if (!path) | |
10095 | return -ENOMEM; | |
e4058b54 | 10096 | path->reada = READA_FORWARD; |
9078a3e1 | 10097 | |
0b246afa JM |
10098 | cache_gen = btrfs_super_cache_generation(info->super_copy); |
10099 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
10100 | btrfs_super_generation(info->super_copy) != cache_gen) | |
0af3d00b | 10101 | need_clear = 1; |
0b246afa | 10102 | if (btrfs_test_opt(info, CLEAR_CACHE)) |
88c2ba3b | 10103 | need_clear = 1; |
0af3d00b | 10104 | |
d397712b | 10105 | while (1) { |
6bccf3ab | 10106 | ret = find_first_block_group(info, path, &key); |
b742bb82 YZ |
10107 | if (ret > 0) |
10108 | break; | |
0b86a832 CM |
10109 | if (ret != 0) |
10110 | goto error; | |
920e4a58 | 10111 | |
5f39d397 CM |
10112 | leaf = path->nodes[0]; |
10113 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 | 10114 | |
2ff7e61e | 10115 | cache = btrfs_create_block_group_cache(info, found_key.objectid, |
920e4a58 | 10116 | found_key.offset); |
9078a3e1 | 10117 | if (!cache) { |
0b86a832 | 10118 | ret = -ENOMEM; |
f0486c68 | 10119 | goto error; |
9078a3e1 | 10120 | } |
96303081 | 10121 | |
cf7c1ef6 LB |
10122 | if (need_clear) { |
10123 | /* | |
10124 | * When we mount with old space cache, we need to | |
10125 | * set BTRFS_DC_CLEAR and set dirty flag. | |
10126 | * | |
10127 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
10128 | * truncate the old free space cache inode and | |
10129 | * setup a new one. | |
10130 | * b) Setting 'dirty flag' makes sure that we flush | |
10131 | * the new space cache info onto disk. | |
10132 | */ | |
0b246afa | 10133 | if (btrfs_test_opt(info, SPACE_CACHE)) |
ce93ec54 | 10134 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 10135 | } |
0af3d00b | 10136 | |
5f39d397 CM |
10137 | read_extent_buffer(leaf, &cache->item, |
10138 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
10139 | sizeof(cache->item)); | |
920e4a58 | 10140 | cache->flags = btrfs_block_group_flags(&cache->item); |
49303381 LB |
10141 | if (!mixed && |
10142 | ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && | |
10143 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
10144 | btrfs_err(info, | |
10145 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
10146 | cache->key.objectid); | |
10147 | ret = -EINVAL; | |
10148 | goto error; | |
10149 | } | |
0b86a832 | 10150 | |
9078a3e1 | 10151 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 10152 | btrfs_release_path(path); |
34d52cb6 | 10153 | |
3c14874a JB |
10154 | /* |
10155 | * We need to exclude the super stripes now so that the space | |
10156 | * info has super bytes accounted for, otherwise we'll think | |
10157 | * we have more space than we actually do. | |
10158 | */ | |
2ff7e61e | 10159 | ret = exclude_super_stripes(info, cache); |
835d974f JB |
10160 | if (ret) { |
10161 | /* | |
10162 | * We may have excluded something, so call this just in | |
10163 | * case. | |
10164 | */ | |
2ff7e61e | 10165 | free_excluded_extents(info, cache); |
920e4a58 | 10166 | btrfs_put_block_group(cache); |
835d974f JB |
10167 | goto error; |
10168 | } | |
3c14874a | 10169 | |
817d52f8 JB |
10170 | /* |
10171 | * check for two cases, either we are full, and therefore | |
10172 | * don't need to bother with the caching work since we won't | |
10173 | * find any space, or we are empty, and we can just add all | |
10174 | * the space in and be done with it. This saves us _alot_ of | |
10175 | * time, particularly in the full case. | |
10176 | */ | |
10177 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 10178 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10179 | cache->cached = BTRFS_CACHE_FINISHED; |
2ff7e61e | 10180 | free_excluded_extents(info, cache); |
817d52f8 | 10181 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 10182 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10183 | cache->cached = BTRFS_CACHE_FINISHED; |
0b246afa | 10184 | add_new_free_space(cache, info, |
817d52f8 JB |
10185 | found_key.objectid, |
10186 | found_key.objectid + | |
10187 | found_key.offset); | |
2ff7e61e | 10188 | free_excluded_extents(info, cache); |
817d52f8 | 10189 | } |
96b5179d | 10190 | |
0b246afa | 10191 | ret = btrfs_add_block_group_cache(info, cache); |
8c579fe7 JB |
10192 | if (ret) { |
10193 | btrfs_remove_free_space_cache(cache); | |
10194 | btrfs_put_block_group(cache); | |
10195 | goto error; | |
10196 | } | |
10197 | ||
0b246afa | 10198 | trace_btrfs_add_block_group(info, cache, 0); |
d2006e6d NB |
10199 | update_space_info(info, cache->flags, found_key.offset, |
10200 | btrfs_block_group_used(&cache->item), | |
10201 | cache->bytes_super, &space_info); | |
8c579fe7 | 10202 | |
6324fbf3 | 10203 | cache->space_info = space_info; |
1b2da372 | 10204 | |
c434d21c | 10205 | link_block_group(cache); |
0f9dd46c | 10206 | |
0b246afa | 10207 | set_avail_alloc_bits(info, cache->flags); |
2ff7e61e | 10208 | if (btrfs_chunk_readonly(info, cache->key.objectid)) { |
868f401a | 10209 | inc_block_group_ro(cache, 1); |
47ab2a6c JB |
10210 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
10211 | spin_lock(&info->unused_bgs_lock); | |
10212 | /* Should always be true but just in case. */ | |
10213 | if (list_empty(&cache->bg_list)) { | |
10214 | btrfs_get_block_group(cache); | |
10215 | list_add_tail(&cache->bg_list, | |
10216 | &info->unused_bgs); | |
10217 | } | |
10218 | spin_unlock(&info->unused_bgs_lock); | |
10219 | } | |
9078a3e1 | 10220 | } |
b742bb82 | 10221 | |
0b246afa | 10222 | list_for_each_entry_rcu(space_info, &info->space_info, list) { |
2ff7e61e | 10223 | if (!(get_alloc_profile(info, space_info->flags) & |
b742bb82 YZ |
10224 | (BTRFS_BLOCK_GROUP_RAID10 | |
10225 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
10226 | BTRFS_BLOCK_GROUP_RAID5 | |
10227 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
10228 | BTRFS_BLOCK_GROUP_DUP))) |
10229 | continue; | |
10230 | /* | |
10231 | * avoid allocating from un-mirrored block group if there are | |
10232 | * mirrored block groups. | |
10233 | */ | |
1095cc0d | 10234 | list_for_each_entry(cache, |
10235 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
10236 | list) | |
868f401a | 10237 | inc_block_group_ro(cache, 1); |
1095cc0d | 10238 | list_for_each_entry(cache, |
10239 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
10240 | list) | |
868f401a | 10241 | inc_block_group_ro(cache, 1); |
9078a3e1 | 10242 | } |
f0486c68 | 10243 | |
75cb379d | 10244 | btrfs_add_raid_kobjects(info); |
f0486c68 | 10245 | init_global_block_rsv(info); |
0b86a832 CM |
10246 | ret = 0; |
10247 | error: | |
9078a3e1 | 10248 | btrfs_free_path(path); |
0b86a832 | 10249 | return ret; |
9078a3e1 | 10250 | } |
6324fbf3 | 10251 | |
6c686b35 | 10252 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) |
ea658bad | 10253 | { |
6c686b35 | 10254 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ea658bad | 10255 | struct btrfs_block_group_cache *block_group, *tmp; |
0b246afa | 10256 | struct btrfs_root *extent_root = fs_info->extent_root; |
ea658bad JB |
10257 | struct btrfs_block_group_item item; |
10258 | struct btrfs_key key; | |
10259 | int ret = 0; | |
d9a0540a | 10260 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
ea658bad | 10261 | |
d9a0540a | 10262 | trans->can_flush_pending_bgs = false; |
47ab2a6c | 10263 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { |
ea658bad | 10264 | if (ret) |
c92f6be3 | 10265 | goto next; |
ea658bad JB |
10266 | |
10267 | spin_lock(&block_group->lock); | |
10268 | memcpy(&item, &block_group->item, sizeof(item)); | |
10269 | memcpy(&key, &block_group->key, sizeof(key)); | |
10270 | spin_unlock(&block_group->lock); | |
10271 | ||
10272 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
10273 | sizeof(item)); | |
10274 | if (ret) | |
66642832 | 10275 | btrfs_abort_transaction(trans, ret); |
0b246afa JM |
10276 | ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid, |
10277 | key.offset); | |
6df9a95e | 10278 | if (ret) |
66642832 | 10279 | btrfs_abort_transaction(trans, ret); |
0b246afa | 10280 | add_block_group_free_space(trans, fs_info, block_group); |
1e144fb8 | 10281 | /* already aborted the transaction if it failed. */ |
c92f6be3 FM |
10282 | next: |
10283 | list_del_init(&block_group->bg_list); | |
ea658bad | 10284 | } |
d9a0540a | 10285 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
ea658bad JB |
10286 | } |
10287 | ||
6324fbf3 | 10288 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 10289 | struct btrfs_fs_info *fs_info, u64 bytes_used, |
0174484d | 10290 | u64 type, u64 chunk_offset, u64 size) |
6324fbf3 | 10291 | { |
6324fbf3 | 10292 | struct btrfs_block_group_cache *cache; |
0b246afa | 10293 | int ret; |
6324fbf3 | 10294 | |
0b246afa | 10295 | btrfs_set_log_full_commit(fs_info, trans); |
e02119d5 | 10296 | |
2ff7e61e | 10297 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size); |
0f9dd46c JB |
10298 | if (!cache) |
10299 | return -ENOMEM; | |
34d52cb6 | 10300 | |
6324fbf3 | 10301 | btrfs_set_block_group_used(&cache->item, bytes_used); |
0174484d NB |
10302 | btrfs_set_block_group_chunk_objectid(&cache->item, |
10303 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
6324fbf3 CM |
10304 | btrfs_set_block_group_flags(&cache->item, type); |
10305 | ||
920e4a58 | 10306 | cache->flags = type; |
11833d66 | 10307 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10308 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 10309 | cache->needs_free_space = 1; |
2ff7e61e | 10310 | ret = exclude_super_stripes(fs_info, cache); |
835d974f JB |
10311 | if (ret) { |
10312 | /* | |
10313 | * We may have excluded something, so call this just in | |
10314 | * case. | |
10315 | */ | |
2ff7e61e | 10316 | free_excluded_extents(fs_info, cache); |
920e4a58 | 10317 | btrfs_put_block_group(cache); |
835d974f JB |
10318 | return ret; |
10319 | } | |
96303081 | 10320 | |
0b246afa | 10321 | add_new_free_space(cache, fs_info, chunk_offset, chunk_offset + size); |
817d52f8 | 10322 | |
2ff7e61e | 10323 | free_excluded_extents(fs_info, cache); |
11833d66 | 10324 | |
d0bd4560 | 10325 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 10326 | if (btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
10327 | u64 new_bytes_used = size - bytes_used; |
10328 | ||
10329 | bytes_used += new_bytes_used >> 1; | |
2ff7e61e | 10330 | fragment_free_space(cache); |
d0bd4560 JB |
10331 | } |
10332 | #endif | |
2e6e5183 | 10333 | /* |
2be12ef7 NB |
10334 | * Ensure the corresponding space_info object is created and |
10335 | * assigned to our block group. We want our bg to be added to the rbtree | |
10336 | * with its ->space_info set. | |
2e6e5183 | 10337 | */ |
2be12ef7 | 10338 | cache->space_info = __find_space_info(fs_info, cache->flags); |
dc2d3005 | 10339 | ASSERT(cache->space_info); |
2e6e5183 | 10340 | |
0b246afa | 10341 | ret = btrfs_add_block_group_cache(fs_info, cache); |
8c579fe7 JB |
10342 | if (ret) { |
10343 | btrfs_remove_free_space_cache(cache); | |
10344 | btrfs_put_block_group(cache); | |
10345 | return ret; | |
10346 | } | |
10347 | ||
2e6e5183 FM |
10348 | /* |
10349 | * Now that our block group has its ->space_info set and is inserted in | |
10350 | * the rbtree, update the space info's counters. | |
10351 | */ | |
0b246afa | 10352 | trace_btrfs_add_block_group(fs_info, cache, 1); |
d2006e6d | 10353 | update_space_info(fs_info, cache->flags, size, bytes_used, |
e40edf2d | 10354 | cache->bytes_super, &cache->space_info); |
0b246afa | 10355 | update_global_block_rsv(fs_info); |
1b2da372 | 10356 | |
c434d21c | 10357 | link_block_group(cache); |
6324fbf3 | 10358 | |
47ab2a6c | 10359 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
6324fbf3 | 10360 | |
0b246afa | 10361 | set_avail_alloc_bits(fs_info, type); |
6324fbf3 CM |
10362 | return 0; |
10363 | } | |
1a40e23b | 10364 | |
10ea00f5 ID |
10365 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
10366 | { | |
899c81ea ID |
10367 | u64 extra_flags = chunk_to_extended(flags) & |
10368 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 10369 | |
de98ced9 | 10370 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10371 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
10372 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10373 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10374 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10375 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10376 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 10377 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10378 | } |
10379 | ||
1a40e23b | 10380 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
6bccf3ab | 10381 | struct btrfs_fs_info *fs_info, u64 group_start, |
04216820 | 10382 | struct extent_map *em) |
1a40e23b | 10383 | { |
6bccf3ab | 10384 | struct btrfs_root *root = fs_info->extent_root; |
1a40e23b ZY |
10385 | struct btrfs_path *path; |
10386 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 10387 | struct btrfs_free_cluster *cluster; |
0b246afa | 10388 | struct btrfs_root *tree_root = fs_info->tree_root; |
1a40e23b | 10389 | struct btrfs_key key; |
0af3d00b | 10390 | struct inode *inode; |
c1895442 | 10391 | struct kobject *kobj = NULL; |
1a40e23b | 10392 | int ret; |
10ea00f5 | 10393 | int index; |
89a55897 | 10394 | int factor; |
4f69cb98 | 10395 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 10396 | bool remove_em; |
1a40e23b | 10397 | |
6bccf3ab | 10398 | block_group = btrfs_lookup_block_group(fs_info, group_start); |
1a40e23b | 10399 | BUG_ON(!block_group); |
c146afad | 10400 | BUG_ON(!block_group->ro); |
1a40e23b | 10401 | |
9f7c43c9 | 10402 | /* |
10403 | * Free the reserved super bytes from this block group before | |
10404 | * remove it. | |
10405 | */ | |
2ff7e61e | 10406 | free_excluded_extents(fs_info, block_group); |
fd708b81 JB |
10407 | btrfs_free_ref_tree_range(fs_info, block_group->key.objectid, |
10408 | block_group->key.offset); | |
9f7c43c9 | 10409 | |
1a40e23b | 10410 | memcpy(&key, &block_group->key, sizeof(key)); |
3e72ee88 | 10411 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
89a55897 JB |
10412 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
10413 | BTRFS_BLOCK_GROUP_RAID1 | | |
10414 | BTRFS_BLOCK_GROUP_RAID10)) | |
10415 | factor = 2; | |
10416 | else | |
10417 | factor = 1; | |
1a40e23b | 10418 | |
44fb5511 | 10419 | /* make sure this block group isn't part of an allocation cluster */ |
0b246afa | 10420 | cluster = &fs_info->data_alloc_cluster; |
44fb5511 CM |
10421 | spin_lock(&cluster->refill_lock); |
10422 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10423 | spin_unlock(&cluster->refill_lock); | |
10424 | ||
10425 | /* | |
10426 | * make sure this block group isn't part of a metadata | |
10427 | * allocation cluster | |
10428 | */ | |
0b246afa | 10429 | cluster = &fs_info->meta_alloc_cluster; |
44fb5511 CM |
10430 | spin_lock(&cluster->refill_lock); |
10431 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10432 | spin_unlock(&cluster->refill_lock); | |
10433 | ||
1a40e23b | 10434 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10435 | if (!path) { |
10436 | ret = -ENOMEM; | |
10437 | goto out; | |
10438 | } | |
1a40e23b | 10439 | |
1bbc621e CM |
10440 | /* |
10441 | * get the inode first so any iput calls done for the io_list | |
10442 | * aren't the final iput (no unlinks allowed now) | |
10443 | */ | |
77ab86bf | 10444 | inode = lookup_free_space_inode(fs_info, block_group, path); |
1bbc621e CM |
10445 | |
10446 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10447 | /* | |
10448 | * make sure our free spache cache IO is done before remove the | |
10449 | * free space inode | |
10450 | */ | |
10451 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10452 | if (!list_empty(&block_group->io_list)) { | |
10453 | list_del_init(&block_group->io_list); | |
10454 | ||
10455 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10456 | ||
10457 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
afdb5718 | 10458 | btrfs_wait_cache_io(trans, block_group, path); |
1bbc621e CM |
10459 | btrfs_put_block_group(block_group); |
10460 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10461 | } | |
10462 | ||
10463 | if (!list_empty(&block_group->dirty_list)) { | |
10464 | list_del_init(&block_group->dirty_list); | |
10465 | btrfs_put_block_group(block_group); | |
10466 | } | |
10467 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10468 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10469 | ||
0af3d00b | 10470 | if (!IS_ERR(inode)) { |
73f2e545 | 10471 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
79787eaa JM |
10472 | if (ret) { |
10473 | btrfs_add_delayed_iput(inode); | |
10474 | goto out; | |
10475 | } | |
0af3d00b JB |
10476 | clear_nlink(inode); |
10477 | /* One for the block groups ref */ | |
10478 | spin_lock(&block_group->lock); | |
10479 | if (block_group->iref) { | |
10480 | block_group->iref = 0; | |
10481 | block_group->inode = NULL; | |
10482 | spin_unlock(&block_group->lock); | |
10483 | iput(inode); | |
10484 | } else { | |
10485 | spin_unlock(&block_group->lock); | |
10486 | } | |
10487 | /* One for our lookup ref */ | |
455757c3 | 10488 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10489 | } |
10490 | ||
10491 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10492 | key.offset = block_group->key.objectid; | |
10493 | key.type = 0; | |
10494 | ||
10495 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10496 | if (ret < 0) | |
10497 | goto out; | |
10498 | if (ret > 0) | |
b3b4aa74 | 10499 | btrfs_release_path(path); |
0af3d00b JB |
10500 | if (ret == 0) { |
10501 | ret = btrfs_del_item(trans, tree_root, path); | |
10502 | if (ret) | |
10503 | goto out; | |
b3b4aa74 | 10504 | btrfs_release_path(path); |
0af3d00b JB |
10505 | } |
10506 | ||
0b246afa | 10507 | spin_lock(&fs_info->block_group_cache_lock); |
1a40e23b | 10508 | rb_erase(&block_group->cache_node, |
0b246afa | 10509 | &fs_info->block_group_cache_tree); |
292cbd51 | 10510 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd | 10511 | |
0b246afa JM |
10512 | if (fs_info->first_logical_byte == block_group->key.objectid) |
10513 | fs_info->first_logical_byte = (u64)-1; | |
10514 | spin_unlock(&fs_info->block_group_cache_lock); | |
817d52f8 | 10515 | |
80eb234a | 10516 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10517 | /* |
10518 | * we must use list_del_init so people can check to see if they | |
10519 | * are still on the list after taking the semaphore | |
10520 | */ | |
10521 | list_del_init(&block_group->list); | |
6ab0a202 | 10522 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10523 | kobj = block_group->space_info->block_group_kobjs[index]; |
10524 | block_group->space_info->block_group_kobjs[index] = NULL; | |
0b246afa | 10525 | clear_avail_alloc_bits(fs_info, block_group->flags); |
6ab0a202 | 10526 | } |
80eb234a | 10527 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
10528 | if (kobj) { |
10529 | kobject_del(kobj); | |
10530 | kobject_put(kobj); | |
10531 | } | |
1a40e23b | 10532 | |
4f69cb98 FM |
10533 | if (block_group->has_caching_ctl) |
10534 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10535 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10536 | wait_block_group_cache_done(block_group); |
4f69cb98 | 10537 | if (block_group->has_caching_ctl) { |
0b246afa | 10538 | down_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10539 | if (!caching_ctl) { |
10540 | struct btrfs_caching_control *ctl; | |
10541 | ||
10542 | list_for_each_entry(ctl, | |
0b246afa | 10543 | &fs_info->caching_block_groups, list) |
4f69cb98 FM |
10544 | if (ctl->block_group == block_group) { |
10545 | caching_ctl = ctl; | |
1e4f4714 | 10546 | refcount_inc(&caching_ctl->count); |
4f69cb98 FM |
10547 | break; |
10548 | } | |
10549 | } | |
10550 | if (caching_ctl) | |
10551 | list_del_init(&caching_ctl->list); | |
0b246afa | 10552 | up_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10553 | if (caching_ctl) { |
10554 | /* Once for the caching bgs list and once for us. */ | |
10555 | put_caching_control(caching_ctl); | |
10556 | put_caching_control(caching_ctl); | |
10557 | } | |
10558 | } | |
817d52f8 | 10559 | |
ce93ec54 JB |
10560 | spin_lock(&trans->transaction->dirty_bgs_lock); |
10561 | if (!list_empty(&block_group->dirty_list)) { | |
1bbc621e CM |
10562 | WARN_ON(1); |
10563 | } | |
10564 | if (!list_empty(&block_group->io_list)) { | |
10565 | WARN_ON(1); | |
ce93ec54 JB |
10566 | } |
10567 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
817d52f8 JB |
10568 | btrfs_remove_free_space_cache(block_group); |
10569 | ||
c146afad | 10570 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10571 | list_del_init(&block_group->ro_list); |
18d018ad | 10572 | |
0b246afa | 10573 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
18d018ad ZL |
10574 | WARN_ON(block_group->space_info->total_bytes |
10575 | < block_group->key.offset); | |
10576 | WARN_ON(block_group->space_info->bytes_readonly | |
10577 | < block_group->key.offset); | |
10578 | WARN_ON(block_group->space_info->disk_total | |
10579 | < block_group->key.offset * factor); | |
10580 | } | |
c146afad YZ |
10581 | block_group->space_info->total_bytes -= block_group->key.offset; |
10582 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10583 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10584 | |
c146afad | 10585 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10586 | |
0af3d00b JB |
10587 | memcpy(&key, &block_group->key, sizeof(key)); |
10588 | ||
34441361 | 10589 | mutex_lock(&fs_info->chunk_mutex); |
495e64f4 FM |
10590 | if (!list_empty(&em->list)) { |
10591 | /* We're in the transaction->pending_chunks list. */ | |
10592 | free_extent_map(em); | |
10593 | } | |
04216820 FM |
10594 | spin_lock(&block_group->lock); |
10595 | block_group->removed = 1; | |
10596 | /* | |
10597 | * At this point trimming can't start on this block group, because we | |
10598 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10599 | * so no one can't find it anymore and even if someone already got this | |
10600 | * block group before we removed it from the rbtree, they have already | |
10601 | * incremented block_group->trimming - if they didn't, they won't find | |
10602 | * any free space entries because we already removed them all when we | |
10603 | * called btrfs_remove_free_space_cache(). | |
10604 | * | |
10605 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10606 | * to prevent the same logical address range and physical device space | |
10607 | * ranges from being reused for a new block group. This is because our | |
10608 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10609 | * completely transactionless, so while it is trimming a range the | |
10610 | * currently running transaction might finish and a new one start, | |
10611 | * allowing for new block groups to be created that can reuse the same | |
10612 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10613 | * |
10614 | * There may also be an implicit trim operation if the file system | |
10615 | * is mounted with -odiscard. The same protections must remain | |
10616 | * in place until the extents have been discarded completely when | |
10617 | * the transaction commit has completed. | |
04216820 FM |
10618 | */ |
10619 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
10620 | /* | |
10621 | * Make sure a trimmer task always sees the em in the pinned_chunks list | |
10622 | * if it sees block_group->removed == 1 (needs to lock block_group->lock | |
10623 | * before checking block_group->removed). | |
10624 | */ | |
10625 | if (!remove_em) { | |
10626 | /* | |
10627 | * Our em might be in trans->transaction->pending_chunks which | |
10628 | * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks), | |
10629 | * and so is the fs_info->pinned_chunks list. | |
10630 | * | |
10631 | * So at this point we must be holding the chunk_mutex to avoid | |
10632 | * any races with chunk allocation (more specifically at | |
10633 | * volumes.c:contains_pending_extent()), to ensure it always | |
10634 | * sees the em, either in the pending_chunks list or in the | |
10635 | * pinned_chunks list. | |
10636 | */ | |
0b246afa | 10637 | list_move_tail(&em->list, &fs_info->pinned_chunks); |
04216820 FM |
10638 | } |
10639 | spin_unlock(&block_group->lock); | |
04216820 FM |
10640 | |
10641 | if (remove_em) { | |
10642 | struct extent_map_tree *em_tree; | |
10643 | ||
0b246afa | 10644 | em_tree = &fs_info->mapping_tree.map_tree; |
04216820 | 10645 | write_lock(&em_tree->lock); |
8dbcd10f FM |
10646 | /* |
10647 | * The em might be in the pending_chunks list, so make sure the | |
10648 | * chunk mutex is locked, since remove_extent_mapping() will | |
10649 | * delete us from that list. | |
10650 | */ | |
04216820 FM |
10651 | remove_extent_mapping(em_tree, em); |
10652 | write_unlock(&em_tree->lock); | |
10653 | /* once for the tree */ | |
10654 | free_extent_map(em); | |
10655 | } | |
10656 | ||
34441361 | 10657 | mutex_unlock(&fs_info->chunk_mutex); |
8dbcd10f | 10658 | |
0b246afa | 10659 | ret = remove_block_group_free_space(trans, fs_info, block_group); |
1e144fb8 OS |
10660 | if (ret) |
10661 | goto out; | |
10662 | ||
fa9c0d79 CM |
10663 | btrfs_put_block_group(block_group); |
10664 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10665 | |
10666 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10667 | if (ret > 0) | |
10668 | ret = -EIO; | |
10669 | if (ret < 0) | |
10670 | goto out; | |
10671 | ||
10672 | ret = btrfs_del_item(trans, root, path); | |
10673 | out: | |
10674 | btrfs_free_path(path); | |
10675 | return ret; | |
10676 | } | |
acce952b | 10677 | |
8eab77ff | 10678 | struct btrfs_trans_handle * |
7fd01182 FM |
10679 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10680 | const u64 chunk_offset) | |
8eab77ff | 10681 | { |
7fd01182 FM |
10682 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; |
10683 | struct extent_map *em; | |
10684 | struct map_lookup *map; | |
10685 | unsigned int num_items; | |
10686 | ||
10687 | read_lock(&em_tree->lock); | |
10688 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10689 | read_unlock(&em_tree->lock); | |
10690 | ASSERT(em && em->start == chunk_offset); | |
10691 | ||
8eab77ff | 10692 | /* |
7fd01182 FM |
10693 | * We need to reserve 3 + N units from the metadata space info in order |
10694 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10695 | * btrfs_remove_block_group()), which are used for: | |
10696 | * | |
8eab77ff FM |
10697 | * 1 unit for adding the free space inode's orphan (located in the tree |
10698 | * of tree roots). | |
7fd01182 FM |
10699 | * 1 unit for deleting the block group item (located in the extent |
10700 | * tree). | |
10701 | * 1 unit for deleting the free space item (located in tree of tree | |
10702 | * roots). | |
10703 | * N units for deleting N device extent items corresponding to each | |
10704 | * stripe (located in the device tree). | |
10705 | * | |
10706 | * In order to remove a block group we also need to reserve units in the | |
10707 | * system space info in order to update the chunk tree (update one or | |
10708 | * more device items and remove one chunk item), but this is done at | |
10709 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10710 | */ |
95617d69 | 10711 | map = em->map_lookup; |
7fd01182 FM |
10712 | num_items = 3 + map->num_stripes; |
10713 | free_extent_map(em); | |
10714 | ||
8eab77ff | 10715 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10716 | num_items, 1); |
8eab77ff FM |
10717 | } |
10718 | ||
47ab2a6c JB |
10719 | /* |
10720 | * Process the unused_bgs list and remove any that don't have any allocated | |
10721 | * space inside of them. | |
10722 | */ | |
10723 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10724 | { | |
10725 | struct btrfs_block_group_cache *block_group; | |
10726 | struct btrfs_space_info *space_info; | |
47ab2a6c JB |
10727 | struct btrfs_trans_handle *trans; |
10728 | int ret = 0; | |
10729 | ||
afcdd129 | 10730 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) |
47ab2a6c JB |
10731 | return; |
10732 | ||
10733 | spin_lock(&fs_info->unused_bgs_lock); | |
10734 | while (!list_empty(&fs_info->unused_bgs)) { | |
10735 | u64 start, end; | |
e33e17ee | 10736 | int trimming; |
47ab2a6c JB |
10737 | |
10738 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10739 | struct btrfs_block_group_cache, | |
10740 | bg_list); | |
47ab2a6c | 10741 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10742 | |
10743 | space_info = block_group->space_info; | |
10744 | ||
47ab2a6c JB |
10745 | if (ret || btrfs_mixed_space_info(space_info)) { |
10746 | btrfs_put_block_group(block_group); | |
10747 | continue; | |
10748 | } | |
10749 | spin_unlock(&fs_info->unused_bgs_lock); | |
10750 | ||
d5f2e33b | 10751 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10752 | |
47ab2a6c JB |
10753 | /* Don't want to race with allocators so take the groups_sem */ |
10754 | down_write(&space_info->groups_sem); | |
10755 | spin_lock(&block_group->lock); | |
10756 | if (block_group->reserved || | |
10757 | btrfs_block_group_used(&block_group->item) || | |
19c4d2f9 | 10758 | block_group->ro || |
aefbe9a6 | 10759 | list_is_singular(&block_group->list)) { |
47ab2a6c JB |
10760 | /* |
10761 | * We want to bail if we made new allocations or have | |
10762 | * outstanding allocations in this block group. We do | |
10763 | * the ro check in case balance is currently acting on | |
10764 | * this block group. | |
10765 | */ | |
10766 | spin_unlock(&block_group->lock); | |
10767 | up_write(&space_info->groups_sem); | |
10768 | goto next; | |
10769 | } | |
10770 | spin_unlock(&block_group->lock); | |
10771 | ||
10772 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 10773 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
10774 | up_write(&space_info->groups_sem); |
10775 | if (ret < 0) { | |
10776 | ret = 0; | |
10777 | goto next; | |
10778 | } | |
10779 | ||
10780 | /* | |
10781 | * Want to do this before we do anything else so we can recover | |
10782 | * properly if we fail to join the transaction. | |
10783 | */ | |
7fd01182 FM |
10784 | trans = btrfs_start_trans_remove_block_group(fs_info, |
10785 | block_group->key.objectid); | |
47ab2a6c | 10786 | if (IS_ERR(trans)) { |
2ff7e61e | 10787 | btrfs_dec_block_group_ro(block_group); |
47ab2a6c JB |
10788 | ret = PTR_ERR(trans); |
10789 | goto next; | |
10790 | } | |
10791 | ||
10792 | /* | |
10793 | * We could have pending pinned extents for this block group, | |
10794 | * just delete them, we don't care about them anymore. | |
10795 | */ | |
10796 | start = block_group->key.objectid; | |
10797 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
10798 | /* |
10799 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
10800 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
10801 | * another task might be running finish_extent_commit() for the | |
10802 | * previous transaction N - 1, and have seen a range belonging | |
10803 | * to the block group in freed_extents[] before we were able to | |
10804 | * clear the whole block group range from freed_extents[]. This | |
10805 | * means that task can lookup for the block group after we | |
10806 | * unpinned it from freed_extents[] and removed it, leading to | |
10807 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
10808 | */ | |
10809 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 10810 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
91166212 | 10811 | EXTENT_DIRTY); |
758eb51e | 10812 | if (ret) { |
d4b450cd | 10813 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10814 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10815 | goto end_trans; |
10816 | } | |
10817 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
91166212 | 10818 | EXTENT_DIRTY); |
758eb51e | 10819 | if (ret) { |
d4b450cd | 10820 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10821 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10822 | goto end_trans; |
10823 | } | |
d4b450cd | 10824 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
10825 | |
10826 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
10827 | spin_lock(&space_info->lock); |
10828 | spin_lock(&block_group->lock); | |
10829 | ||
10830 | space_info->bytes_pinned -= block_group->pinned; | |
10831 | space_info->bytes_readonly += block_group->pinned; | |
10832 | percpu_counter_add(&space_info->total_bytes_pinned, | |
10833 | -block_group->pinned); | |
47ab2a6c JB |
10834 | block_group->pinned = 0; |
10835 | ||
c30666d4 ZL |
10836 | spin_unlock(&block_group->lock); |
10837 | spin_unlock(&space_info->lock); | |
10838 | ||
e33e17ee | 10839 | /* DISCARD can flip during remount */ |
0b246afa | 10840 | trimming = btrfs_test_opt(fs_info, DISCARD); |
e33e17ee JM |
10841 | |
10842 | /* Implicit trim during transaction commit. */ | |
10843 | if (trimming) | |
10844 | btrfs_get_block_group_trimming(block_group); | |
10845 | ||
47ab2a6c JB |
10846 | /* |
10847 | * Btrfs_remove_chunk will abort the transaction if things go | |
10848 | * horribly wrong. | |
10849 | */ | |
5b4aacef | 10850 | ret = btrfs_remove_chunk(trans, fs_info, |
47ab2a6c | 10851 | block_group->key.objectid); |
e33e17ee JM |
10852 | |
10853 | if (ret) { | |
10854 | if (trimming) | |
10855 | btrfs_put_block_group_trimming(block_group); | |
10856 | goto end_trans; | |
10857 | } | |
10858 | ||
10859 | /* | |
10860 | * If we're not mounted with -odiscard, we can just forget | |
10861 | * about this block group. Otherwise we'll need to wait | |
10862 | * until transaction commit to do the actual discard. | |
10863 | */ | |
10864 | if (trimming) { | |
348a0013 FM |
10865 | spin_lock(&fs_info->unused_bgs_lock); |
10866 | /* | |
10867 | * A concurrent scrub might have added us to the list | |
10868 | * fs_info->unused_bgs, so use a list_move operation | |
10869 | * to add the block group to the deleted_bgs list. | |
10870 | */ | |
e33e17ee JM |
10871 | list_move(&block_group->bg_list, |
10872 | &trans->transaction->deleted_bgs); | |
348a0013 | 10873 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
10874 | btrfs_get_block_group(block_group); |
10875 | } | |
758eb51e | 10876 | end_trans: |
3a45bb20 | 10877 | btrfs_end_transaction(trans); |
47ab2a6c | 10878 | next: |
d5f2e33b | 10879 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
10880 | btrfs_put_block_group(block_group); |
10881 | spin_lock(&fs_info->unused_bgs_lock); | |
10882 | } | |
10883 | spin_unlock(&fs_info->unused_bgs_lock); | |
10884 | } | |
10885 | ||
c59021f8 | 10886 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
10887 | { | |
10888 | struct btrfs_space_info *space_info; | |
1aba86d6 | 10889 | struct btrfs_super_block *disk_super; |
10890 | u64 features; | |
10891 | u64 flags; | |
10892 | int mixed = 0; | |
c59021f8 | 10893 | int ret; |
10894 | ||
6c41761f | 10895 | disk_super = fs_info->super_copy; |
1aba86d6 | 10896 | if (!btrfs_super_root(disk_super)) |
0dc924c5 | 10897 | return -EINVAL; |
c59021f8 | 10898 | |
1aba86d6 | 10899 | features = btrfs_super_incompat_flags(disk_super); |
10900 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
10901 | mixed = 1; | |
c59021f8 | 10902 | |
1aba86d6 | 10903 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
2be12ef7 | 10904 | ret = create_space_info(fs_info, flags, &space_info); |
c59021f8 | 10905 | if (ret) |
1aba86d6 | 10906 | goto out; |
c59021f8 | 10907 | |
1aba86d6 | 10908 | if (mixed) { |
10909 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
2be12ef7 | 10910 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 10911 | } else { |
10912 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
2be12ef7 | 10913 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 10914 | if (ret) |
10915 | goto out; | |
10916 | ||
10917 | flags = BTRFS_BLOCK_GROUP_DATA; | |
2be12ef7 | 10918 | ret = create_space_info(fs_info, flags, &space_info); |
1aba86d6 | 10919 | } |
10920 | out: | |
c59021f8 | 10921 | return ret; |
10922 | } | |
10923 | ||
2ff7e61e JM |
10924 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
10925 | u64 start, u64 end) | |
acce952b | 10926 | { |
2ff7e61e | 10927 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 10928 | } |
10929 | ||
499f377f JM |
10930 | /* |
10931 | * It used to be that old block groups would be left around forever. | |
10932 | * Iterating over them would be enough to trim unused space. Since we | |
10933 | * now automatically remove them, we also need to iterate over unallocated | |
10934 | * space. | |
10935 | * | |
10936 | * We don't want a transaction for this since the discard may take a | |
10937 | * substantial amount of time. We don't require that a transaction be | |
10938 | * running, but we do need to take a running transaction into account | |
10939 | * to ensure that we're not discarding chunks that were released in | |
10940 | * the current transaction. | |
10941 | * | |
10942 | * Holding the chunks lock will prevent other threads from allocating | |
10943 | * or releasing chunks, but it won't prevent a running transaction | |
10944 | * from committing and releasing the memory that the pending chunks | |
10945 | * list head uses. For that, we need to take a reference to the | |
10946 | * transaction. | |
10947 | */ | |
10948 | static int btrfs_trim_free_extents(struct btrfs_device *device, | |
10949 | u64 minlen, u64 *trimmed) | |
10950 | { | |
10951 | u64 start = 0, len = 0; | |
10952 | int ret; | |
10953 | ||
10954 | *trimmed = 0; | |
10955 | ||
10956 | /* Not writeable = nothing to do. */ | |
ebbede42 | 10957 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
10958 | return 0; |
10959 | ||
10960 | /* No free space = nothing to do. */ | |
10961 | if (device->total_bytes <= device->bytes_used) | |
10962 | return 0; | |
10963 | ||
10964 | ret = 0; | |
10965 | ||
10966 | while (1) { | |
fb456252 | 10967 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
10968 | struct btrfs_transaction *trans; |
10969 | u64 bytes; | |
10970 | ||
10971 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
10972 | if (ret) | |
10973 | return ret; | |
10974 | ||
10975 | down_read(&fs_info->commit_root_sem); | |
10976 | ||
10977 | spin_lock(&fs_info->trans_lock); | |
10978 | trans = fs_info->running_transaction; | |
10979 | if (trans) | |
9b64f57d | 10980 | refcount_inc(&trans->use_count); |
499f377f JM |
10981 | spin_unlock(&fs_info->trans_lock); |
10982 | ||
10983 | ret = find_free_dev_extent_start(trans, device, minlen, start, | |
10984 | &start, &len); | |
10985 | if (trans) | |
10986 | btrfs_put_transaction(trans); | |
10987 | ||
10988 | if (ret) { | |
10989 | up_read(&fs_info->commit_root_sem); | |
10990 | mutex_unlock(&fs_info->chunk_mutex); | |
10991 | if (ret == -ENOSPC) | |
10992 | ret = 0; | |
10993 | break; | |
10994 | } | |
10995 | ||
10996 | ret = btrfs_issue_discard(device->bdev, start, len, &bytes); | |
10997 | up_read(&fs_info->commit_root_sem); | |
10998 | mutex_unlock(&fs_info->chunk_mutex); | |
10999 | ||
11000 | if (ret) | |
11001 | break; | |
11002 | ||
11003 | start += len; | |
11004 | *trimmed += bytes; | |
11005 | ||
11006 | if (fatal_signal_pending(current)) { | |
11007 | ret = -ERESTARTSYS; | |
11008 | break; | |
11009 | } | |
11010 | ||
11011 | cond_resched(); | |
11012 | } | |
11013 | ||
11014 | return ret; | |
11015 | } | |
11016 | ||
2ff7e61e | 11017 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 11018 | { |
f7039b1d | 11019 | struct btrfs_block_group_cache *cache = NULL; |
499f377f JM |
11020 | struct btrfs_device *device; |
11021 | struct list_head *devices; | |
f7039b1d LD |
11022 | u64 group_trimmed; |
11023 | u64 start; | |
11024 | u64 end; | |
11025 | u64 trimmed = 0; | |
2cac13e4 | 11026 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
11027 | int ret = 0; |
11028 | ||
2cac13e4 LB |
11029 | /* |
11030 | * try to trim all FS space, our block group may start from non-zero. | |
11031 | */ | |
11032 | if (range->len == total_bytes) | |
11033 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
11034 | else | |
11035 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
11036 | |
11037 | while (cache) { | |
11038 | if (cache->key.objectid >= (range->start + range->len)) { | |
11039 | btrfs_put_block_group(cache); | |
11040 | break; | |
11041 | } | |
11042 | ||
11043 | start = max(range->start, cache->key.objectid); | |
11044 | end = min(range->start + range->len, | |
11045 | cache->key.objectid + cache->key.offset); | |
11046 | ||
11047 | if (end - start >= range->minlen) { | |
11048 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 11049 | ret = cache_block_group(cache, 0); |
1be41b78 JB |
11050 | if (ret) { |
11051 | btrfs_put_block_group(cache); | |
11052 | break; | |
11053 | } | |
11054 | ret = wait_block_group_cache_done(cache); | |
11055 | if (ret) { | |
11056 | btrfs_put_block_group(cache); | |
11057 | break; | |
11058 | } | |
f7039b1d LD |
11059 | } |
11060 | ret = btrfs_trim_block_group(cache, | |
11061 | &group_trimmed, | |
11062 | start, | |
11063 | end, | |
11064 | range->minlen); | |
11065 | ||
11066 | trimmed += group_trimmed; | |
11067 | if (ret) { | |
11068 | btrfs_put_block_group(cache); | |
11069 | break; | |
11070 | } | |
11071 | } | |
11072 | ||
2ff7e61e | 11073 | cache = next_block_group(fs_info, cache); |
f7039b1d LD |
11074 | } |
11075 | ||
0b246afa JM |
11076 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
11077 | devices = &fs_info->fs_devices->alloc_list; | |
499f377f JM |
11078 | list_for_each_entry(device, devices, dev_alloc_list) { |
11079 | ret = btrfs_trim_free_extents(device, range->minlen, | |
11080 | &group_trimmed); | |
11081 | if (ret) | |
11082 | break; | |
11083 | ||
11084 | trimmed += group_trimmed; | |
11085 | } | |
0b246afa | 11086 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 11087 | |
f7039b1d LD |
11088 | range->len = trimmed; |
11089 | return ret; | |
11090 | } | |
8257b2dc MX |
11091 | |
11092 | /* | |
ea14b57f | 11093 | * btrfs_{start,end}_write_no_snapshotting() are similar to |
9ea24bbe FM |
11094 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing |
11095 | * data into the page cache through nocow before the subvolume is snapshoted, | |
11096 | * but flush the data into disk after the snapshot creation, or to prevent | |
ea14b57f | 11097 | * operations while snapshotting is ongoing and that cause the snapshot to be |
9ea24bbe | 11098 | * inconsistent (writes followed by expanding truncates for example). |
8257b2dc | 11099 | */ |
ea14b57f | 11100 | void btrfs_end_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc MX |
11101 | { |
11102 | percpu_counter_dec(&root->subv_writers->counter); | |
11103 | /* | |
a83342aa | 11104 | * Make sure counter is updated before we wake up waiters. |
8257b2dc MX |
11105 | */ |
11106 | smp_mb(); | |
11107 | if (waitqueue_active(&root->subv_writers->wait)) | |
11108 | wake_up(&root->subv_writers->wait); | |
11109 | } | |
11110 | ||
ea14b57f | 11111 | int btrfs_start_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc | 11112 | { |
ea14b57f | 11113 | if (atomic_read(&root->will_be_snapshotted)) |
8257b2dc MX |
11114 | return 0; |
11115 | ||
11116 | percpu_counter_inc(&root->subv_writers->counter); | |
11117 | /* | |
11118 | * Make sure counter is updated before we check for snapshot creation. | |
11119 | */ | |
11120 | smp_mb(); | |
ea14b57f DS |
11121 | if (atomic_read(&root->will_be_snapshotted)) { |
11122 | btrfs_end_write_no_snapshotting(root); | |
8257b2dc MX |
11123 | return 0; |
11124 | } | |
11125 | return 1; | |
11126 | } | |
0bc19f90 | 11127 | |
0bc19f90 ZL |
11128 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) |
11129 | { | |
11130 | while (true) { | |
11131 | int ret; | |
11132 | ||
ea14b57f | 11133 | ret = btrfs_start_write_no_snapshotting(root); |
0bc19f90 ZL |
11134 | if (ret) |
11135 | break; | |
4625956a PZ |
11136 | wait_var_event(&root->will_be_snapshotted, |
11137 | !atomic_read(&root->will_be_snapshotted)); | |
0bc19f90 ZL |
11138 | } |
11139 | } |