]>
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, |
e72cb923 NB |
55 | struct btrfs_delayed_ref_node *node, u64 parent, |
56 | u64 root_objectid, u64 owner_objectid, | |
57 | u64 owner_offset, int refs_to_drop, | |
58 | struct btrfs_delayed_extent_op *extra_op); | |
5d4f98a2 YZ |
59 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
60 | struct extent_buffer *leaf, | |
61 | struct btrfs_extent_item *ei); | |
62 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
63 | u64 parent, u64 root_objectid, |
64 | u64 flags, u64 owner, u64 offset, | |
65 | struct btrfs_key *ins, int ref_mod); | |
66 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, | |
4e6bd4e0 | 67 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 68 | struct btrfs_delayed_extent_op *extent_op); |
01458828 | 69 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, |
698d0082 | 70 | int force); |
11833d66 YZ |
71 | static int find_next_key(struct btrfs_path *path, int level, |
72 | struct btrfs_key *key); | |
ab8d0fc4 JM |
73 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
74 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 75 | int dump_block_groups); |
5d80366e JB |
76 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
77 | u64 num_bytes); | |
957780eb JB |
78 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, |
79 | struct btrfs_space_info *space_info, | |
80 | u64 num_bytes); | |
81 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
82 | struct btrfs_space_info *space_info, | |
83 | u64 num_bytes); | |
6a63209f | 84 | |
817d52f8 JB |
85 | static noinline int |
86 | block_group_cache_done(struct btrfs_block_group_cache *cache) | |
87 | { | |
88 | smp_mb(); | |
36cce922 JB |
89 | return cache->cached == BTRFS_CACHE_FINISHED || |
90 | cache->cached == BTRFS_CACHE_ERROR; | |
817d52f8 JB |
91 | } |
92 | ||
0f9dd46c JB |
93 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
94 | { | |
95 | return (cache->flags & bits) == bits; | |
96 | } | |
97 | ||
758f2dfc | 98 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
11dfe35a JB |
99 | { |
100 | atomic_inc(&cache->count); | |
101 | } | |
102 | ||
103 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) | |
104 | { | |
f0486c68 YZ |
105 | if (atomic_dec_and_test(&cache->count)) { |
106 | WARN_ON(cache->pinned > 0); | |
107 | WARN_ON(cache->reserved > 0); | |
0966a7b1 QW |
108 | |
109 | /* | |
110 | * If not empty, someone is still holding mutex of | |
111 | * full_stripe_lock, which can only be released by caller. | |
112 | * And it will definitely cause use-after-free when caller | |
113 | * tries to release full stripe lock. | |
114 | * | |
115 | * No better way to resolve, but only to warn. | |
116 | */ | |
117 | WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); | |
34d52cb6 | 118 | kfree(cache->free_space_ctl); |
11dfe35a | 119 | kfree(cache); |
f0486c68 | 120 | } |
11dfe35a JB |
121 | } |
122 | ||
0f9dd46c JB |
123 | /* |
124 | * this adds the block group to the fs_info rb tree for the block group | |
125 | * cache | |
126 | */ | |
b2950863 | 127 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
0f9dd46c JB |
128 | struct btrfs_block_group_cache *block_group) |
129 | { | |
130 | struct rb_node **p; | |
131 | struct rb_node *parent = NULL; | |
132 | struct btrfs_block_group_cache *cache; | |
133 | ||
134 | spin_lock(&info->block_group_cache_lock); | |
135 | p = &info->block_group_cache_tree.rb_node; | |
136 | ||
137 | while (*p) { | |
138 | parent = *p; | |
139 | cache = rb_entry(parent, struct btrfs_block_group_cache, | |
140 | cache_node); | |
141 | if (block_group->key.objectid < cache->key.objectid) { | |
142 | p = &(*p)->rb_left; | |
143 | } else if (block_group->key.objectid > cache->key.objectid) { | |
144 | p = &(*p)->rb_right; | |
145 | } else { | |
146 | spin_unlock(&info->block_group_cache_lock); | |
147 | return -EEXIST; | |
148 | } | |
149 | } | |
150 | ||
151 | rb_link_node(&block_group->cache_node, parent, p); | |
152 | rb_insert_color(&block_group->cache_node, | |
153 | &info->block_group_cache_tree); | |
a1897fdd LB |
154 | |
155 | if (info->first_logical_byte > block_group->key.objectid) | |
156 | info->first_logical_byte = block_group->key.objectid; | |
157 | ||
0f9dd46c JB |
158 | spin_unlock(&info->block_group_cache_lock); |
159 | ||
160 | return 0; | |
161 | } | |
162 | ||
163 | /* | |
164 | * This will return the block group at or after bytenr if contains is 0, else | |
165 | * it will return the block group that contains the bytenr | |
166 | */ | |
167 | static struct btrfs_block_group_cache * | |
168 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, | |
169 | int contains) | |
170 | { | |
171 | struct btrfs_block_group_cache *cache, *ret = NULL; | |
172 | struct rb_node *n; | |
173 | u64 end, start; | |
174 | ||
175 | spin_lock(&info->block_group_cache_lock); | |
176 | n = info->block_group_cache_tree.rb_node; | |
177 | ||
178 | while (n) { | |
179 | cache = rb_entry(n, struct btrfs_block_group_cache, | |
180 | cache_node); | |
181 | end = cache->key.objectid + cache->key.offset - 1; | |
182 | start = cache->key.objectid; | |
183 | ||
184 | if (bytenr < start) { | |
185 | if (!contains && (!ret || start < ret->key.objectid)) | |
186 | ret = cache; | |
187 | n = n->rb_left; | |
188 | } else if (bytenr > start) { | |
189 | if (contains && bytenr <= end) { | |
190 | ret = cache; | |
191 | break; | |
192 | } | |
193 | n = n->rb_right; | |
194 | } else { | |
195 | ret = cache; | |
196 | break; | |
197 | } | |
198 | } | |
a1897fdd | 199 | if (ret) { |
11dfe35a | 200 | btrfs_get_block_group(ret); |
a1897fdd LB |
201 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
202 | info->first_logical_byte = ret->key.objectid; | |
203 | } | |
0f9dd46c JB |
204 | spin_unlock(&info->block_group_cache_lock); |
205 | ||
206 | return ret; | |
207 | } | |
208 | ||
2ff7e61e | 209 | static int add_excluded_extent(struct btrfs_fs_info *fs_info, |
11833d66 | 210 | u64 start, u64 num_bytes) |
817d52f8 | 211 | { |
11833d66 | 212 | u64 end = start + num_bytes - 1; |
0b246afa | 213 | set_extent_bits(&fs_info->freed_extents[0], |
ceeb0ae7 | 214 | start, end, EXTENT_UPTODATE); |
0b246afa | 215 | set_extent_bits(&fs_info->freed_extents[1], |
ceeb0ae7 | 216 | start, end, EXTENT_UPTODATE); |
11833d66 YZ |
217 | return 0; |
218 | } | |
817d52f8 | 219 | |
9e715da8 | 220 | static void free_excluded_extents(struct btrfs_block_group_cache *cache) |
11833d66 | 221 | { |
9e715da8 | 222 | struct btrfs_fs_info *fs_info = cache->fs_info; |
11833d66 | 223 | u64 start, end; |
817d52f8 | 224 | |
11833d66 YZ |
225 | start = cache->key.objectid; |
226 | end = start + cache->key.offset - 1; | |
227 | ||
0b246afa | 228 | clear_extent_bits(&fs_info->freed_extents[0], |
91166212 | 229 | start, end, EXTENT_UPTODATE); |
0b246afa | 230 | clear_extent_bits(&fs_info->freed_extents[1], |
91166212 | 231 | start, end, EXTENT_UPTODATE); |
817d52f8 JB |
232 | } |
233 | ||
3c4da657 | 234 | static int exclude_super_stripes(struct btrfs_block_group_cache *cache) |
817d52f8 | 235 | { |
3c4da657 | 236 | struct btrfs_fs_info *fs_info = cache->fs_info; |
817d52f8 JB |
237 | u64 bytenr; |
238 | u64 *logical; | |
239 | int stripe_len; | |
240 | int i, nr, ret; | |
241 | ||
06b2331f YZ |
242 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
243 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; | |
244 | cache->bytes_super += stripe_len; | |
2ff7e61e | 245 | ret = add_excluded_extent(fs_info, cache->key.objectid, |
06b2331f | 246 | stripe_len); |
835d974f JB |
247 | if (ret) |
248 | return ret; | |
06b2331f YZ |
249 | } |
250 | ||
817d52f8 JB |
251 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
252 | bytenr = btrfs_sb_offset(i); | |
0b246afa | 253 | ret = btrfs_rmap_block(fs_info, cache->key.objectid, |
63a9c7b9 | 254 | bytenr, &logical, &nr, &stripe_len); |
835d974f JB |
255 | if (ret) |
256 | return ret; | |
11833d66 | 257 | |
817d52f8 | 258 | while (nr--) { |
51bf5f0b JB |
259 | u64 start, len; |
260 | ||
261 | if (logical[nr] > cache->key.objectid + | |
262 | cache->key.offset) | |
263 | continue; | |
264 | ||
265 | if (logical[nr] + stripe_len <= cache->key.objectid) | |
266 | continue; | |
267 | ||
268 | start = logical[nr]; | |
269 | if (start < cache->key.objectid) { | |
270 | start = cache->key.objectid; | |
271 | len = (logical[nr] + stripe_len) - start; | |
272 | } else { | |
273 | len = min_t(u64, stripe_len, | |
274 | cache->key.objectid + | |
275 | cache->key.offset - start); | |
276 | } | |
277 | ||
278 | cache->bytes_super += len; | |
2ff7e61e | 279 | ret = add_excluded_extent(fs_info, start, len); |
835d974f JB |
280 | if (ret) { |
281 | kfree(logical); | |
282 | return ret; | |
283 | } | |
817d52f8 | 284 | } |
11833d66 | 285 | |
817d52f8 JB |
286 | kfree(logical); |
287 | } | |
817d52f8 JB |
288 | return 0; |
289 | } | |
290 | ||
11833d66 YZ |
291 | static struct btrfs_caching_control * |
292 | get_caching_control(struct btrfs_block_group_cache *cache) | |
293 | { | |
294 | struct btrfs_caching_control *ctl; | |
295 | ||
296 | spin_lock(&cache->lock); | |
dde5abee JB |
297 | if (!cache->caching_ctl) { |
298 | spin_unlock(&cache->lock); | |
11833d66 YZ |
299 | return NULL; |
300 | } | |
301 | ||
302 | ctl = cache->caching_ctl; | |
1e4f4714 | 303 | refcount_inc(&ctl->count); |
11833d66 YZ |
304 | spin_unlock(&cache->lock); |
305 | return ctl; | |
306 | } | |
307 | ||
308 | static void put_caching_control(struct btrfs_caching_control *ctl) | |
309 | { | |
1e4f4714 | 310 | if (refcount_dec_and_test(&ctl->count)) |
11833d66 YZ |
311 | kfree(ctl); |
312 | } | |
313 | ||
d0bd4560 | 314 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 315 | static void fragment_free_space(struct btrfs_block_group_cache *block_group) |
d0bd4560 | 316 | { |
2ff7e61e | 317 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
d0bd4560 JB |
318 | u64 start = block_group->key.objectid; |
319 | u64 len = block_group->key.offset; | |
320 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? | |
0b246afa | 321 | fs_info->nodesize : fs_info->sectorsize; |
d0bd4560 JB |
322 | u64 step = chunk << 1; |
323 | ||
324 | while (len > chunk) { | |
325 | btrfs_remove_free_space(block_group, start, chunk); | |
326 | start += step; | |
327 | if (len < step) | |
328 | len = 0; | |
329 | else | |
330 | len -= step; | |
331 | } | |
332 | } | |
333 | #endif | |
334 | ||
0f9dd46c JB |
335 | /* |
336 | * this is only called by cache_block_group, since we could have freed extents | |
337 | * we need to check the pinned_extents for any extents that can't be used yet | |
338 | * since their free space will be released as soon as the transaction commits. | |
339 | */ | |
a5ed9182 | 340 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
4457c1c7 | 341 | u64 start, u64 end) |
0f9dd46c | 342 | { |
4457c1c7 | 343 | struct btrfs_fs_info *info = block_group->fs_info; |
817d52f8 | 344 | u64 extent_start, extent_end, size, total_added = 0; |
0f9dd46c JB |
345 | int ret; |
346 | ||
347 | while (start < end) { | |
11833d66 | 348 | ret = find_first_extent_bit(info->pinned_extents, start, |
0f9dd46c | 349 | &extent_start, &extent_end, |
e6138876 JB |
350 | EXTENT_DIRTY | EXTENT_UPTODATE, |
351 | NULL); | |
0f9dd46c JB |
352 | if (ret) |
353 | break; | |
354 | ||
06b2331f | 355 | if (extent_start <= start) { |
0f9dd46c JB |
356 | start = extent_end + 1; |
357 | } else if (extent_start > start && extent_start < end) { | |
358 | size = extent_start - start; | |
817d52f8 | 359 | total_added += size; |
ea6a478e JB |
360 | ret = btrfs_add_free_space(block_group, start, |
361 | size); | |
79787eaa | 362 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
363 | start = extent_end + 1; |
364 | } else { | |
365 | break; | |
366 | } | |
367 | } | |
368 | ||
369 | if (start < end) { | |
370 | size = end - start; | |
817d52f8 | 371 | total_added += size; |
ea6a478e | 372 | ret = btrfs_add_free_space(block_group, start, size); |
79787eaa | 373 | BUG_ON(ret); /* -ENOMEM or logic error */ |
0f9dd46c JB |
374 | } |
375 | ||
817d52f8 | 376 | return total_added; |
0f9dd46c JB |
377 | } |
378 | ||
73fa48b6 | 379 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) |
e37c9e69 | 380 | { |
0b246afa JM |
381 | struct btrfs_block_group_cache *block_group = caching_ctl->block_group; |
382 | struct btrfs_fs_info *fs_info = block_group->fs_info; | |
383 | struct btrfs_root *extent_root = fs_info->extent_root; | |
e37c9e69 | 384 | struct btrfs_path *path; |
5f39d397 | 385 | struct extent_buffer *leaf; |
11833d66 | 386 | struct btrfs_key key; |
817d52f8 | 387 | u64 total_found = 0; |
11833d66 YZ |
388 | u64 last = 0; |
389 | u32 nritems; | |
73fa48b6 | 390 | int ret; |
d0bd4560 | 391 | bool wakeup = true; |
f510cfec | 392 | |
e37c9e69 CM |
393 | path = btrfs_alloc_path(); |
394 | if (!path) | |
73fa48b6 | 395 | return -ENOMEM; |
7d7d6068 | 396 | |
817d52f8 | 397 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
11833d66 | 398 | |
d0bd4560 JB |
399 | #ifdef CONFIG_BTRFS_DEBUG |
400 | /* | |
401 | * If we're fragmenting we don't want to make anybody think we can | |
402 | * allocate from this block group until we've had a chance to fragment | |
403 | * the free space. | |
404 | */ | |
2ff7e61e | 405 | if (btrfs_should_fragment_free_space(block_group)) |
d0bd4560 JB |
406 | wakeup = false; |
407 | #endif | |
5cd57b2c | 408 | /* |
817d52f8 JB |
409 | * We don't want to deadlock with somebody trying to allocate a new |
410 | * extent for the extent root while also trying to search the extent | |
411 | * root to add free space. So we skip locking and search the commit | |
412 | * root, since its read-only | |
5cd57b2c CM |
413 | */ |
414 | path->skip_locking = 1; | |
817d52f8 | 415 | path->search_commit_root = 1; |
e4058b54 | 416 | path->reada = READA_FORWARD; |
817d52f8 | 417 | |
e4404d6e | 418 | key.objectid = last; |
e37c9e69 | 419 | key.offset = 0; |
11833d66 | 420 | key.type = BTRFS_EXTENT_ITEM_KEY; |
013f1b12 | 421 | |
52ee28d2 | 422 | next: |
11833d66 | 423 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
e37c9e69 | 424 | if (ret < 0) |
73fa48b6 | 425 | goto out; |
a512bbf8 | 426 | |
11833d66 YZ |
427 | leaf = path->nodes[0]; |
428 | nritems = btrfs_header_nritems(leaf); | |
429 | ||
d397712b | 430 | while (1) { |
7841cb28 | 431 | if (btrfs_fs_closing(fs_info) > 1) { |
f25784b3 | 432 | last = (u64)-1; |
817d52f8 | 433 | break; |
f25784b3 | 434 | } |
817d52f8 | 435 | |
11833d66 YZ |
436 | if (path->slots[0] < nritems) { |
437 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
438 | } else { | |
439 | ret = find_next_key(path, 0, &key); | |
440 | if (ret) | |
e37c9e69 | 441 | break; |
817d52f8 | 442 | |
c9ea7b24 | 443 | if (need_resched() || |
9e351cc8 | 444 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
d0bd4560 JB |
445 | if (wakeup) |
446 | caching_ctl->progress = last; | |
ff5714cc | 447 | btrfs_release_path(path); |
9e351cc8 | 448 | up_read(&fs_info->commit_root_sem); |
589d8ade | 449 | mutex_unlock(&caching_ctl->mutex); |
11833d66 | 450 | cond_resched(); |
73fa48b6 OS |
451 | mutex_lock(&caching_ctl->mutex); |
452 | down_read(&fs_info->commit_root_sem); | |
453 | goto next; | |
589d8ade | 454 | } |
0a3896d0 JB |
455 | |
456 | ret = btrfs_next_leaf(extent_root, path); | |
457 | if (ret < 0) | |
73fa48b6 | 458 | goto out; |
0a3896d0 JB |
459 | if (ret) |
460 | break; | |
589d8ade JB |
461 | leaf = path->nodes[0]; |
462 | nritems = btrfs_header_nritems(leaf); | |
463 | continue; | |
11833d66 | 464 | } |
817d52f8 | 465 | |
52ee28d2 LB |
466 | if (key.objectid < last) { |
467 | key.objectid = last; | |
468 | key.offset = 0; | |
469 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
470 | ||
d0bd4560 JB |
471 | if (wakeup) |
472 | caching_ctl->progress = last; | |
52ee28d2 LB |
473 | btrfs_release_path(path); |
474 | goto next; | |
475 | } | |
476 | ||
11833d66 YZ |
477 | if (key.objectid < block_group->key.objectid) { |
478 | path->slots[0]++; | |
817d52f8 | 479 | continue; |
e37c9e69 | 480 | } |
0f9dd46c | 481 | |
e37c9e69 | 482 | if (key.objectid >= block_group->key.objectid + |
0f9dd46c | 483 | block_group->key.offset) |
e37c9e69 | 484 | break; |
7d7d6068 | 485 | |
3173a18f JB |
486 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
487 | key.type == BTRFS_METADATA_ITEM_KEY) { | |
4457c1c7 | 488 | total_found += add_new_free_space(block_group, last, |
817d52f8 | 489 | key.objectid); |
3173a18f JB |
490 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
491 | last = key.objectid + | |
da17066c | 492 | fs_info->nodesize; |
3173a18f JB |
493 | else |
494 | last = key.objectid + key.offset; | |
817d52f8 | 495 | |
73fa48b6 | 496 | if (total_found > CACHING_CTL_WAKE_UP) { |
11833d66 | 497 | total_found = 0; |
d0bd4560 JB |
498 | if (wakeup) |
499 | wake_up(&caching_ctl->wait); | |
11833d66 | 500 | } |
817d52f8 | 501 | } |
e37c9e69 CM |
502 | path->slots[0]++; |
503 | } | |
817d52f8 | 504 | ret = 0; |
e37c9e69 | 505 | |
4457c1c7 | 506 | total_found += add_new_free_space(block_group, last, |
817d52f8 JB |
507 | block_group->key.objectid + |
508 | block_group->key.offset); | |
11833d66 | 509 | caching_ctl->progress = (u64)-1; |
817d52f8 | 510 | |
73fa48b6 OS |
511 | out: |
512 | btrfs_free_path(path); | |
513 | return ret; | |
514 | } | |
515 | ||
516 | static noinline void caching_thread(struct btrfs_work *work) | |
517 | { | |
518 | struct btrfs_block_group_cache *block_group; | |
519 | struct btrfs_fs_info *fs_info; | |
520 | struct btrfs_caching_control *caching_ctl; | |
521 | int ret; | |
522 | ||
523 | caching_ctl = container_of(work, struct btrfs_caching_control, work); | |
524 | block_group = caching_ctl->block_group; | |
525 | fs_info = block_group->fs_info; | |
526 | ||
527 | mutex_lock(&caching_ctl->mutex); | |
528 | down_read(&fs_info->commit_root_sem); | |
529 | ||
1e144fb8 OS |
530 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
531 | ret = load_free_space_tree(caching_ctl); | |
532 | else | |
533 | ret = load_extent_tree_free(caching_ctl); | |
73fa48b6 | 534 | |
817d52f8 | 535 | spin_lock(&block_group->lock); |
11833d66 | 536 | block_group->caching_ctl = NULL; |
73fa48b6 | 537 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; |
817d52f8 | 538 | spin_unlock(&block_group->lock); |
0f9dd46c | 539 | |
d0bd4560 | 540 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 541 | if (btrfs_should_fragment_free_space(block_group)) { |
d0bd4560 JB |
542 | u64 bytes_used; |
543 | ||
544 | spin_lock(&block_group->space_info->lock); | |
545 | spin_lock(&block_group->lock); | |
546 | bytes_used = block_group->key.offset - | |
547 | btrfs_block_group_used(&block_group->item); | |
548 | block_group->space_info->bytes_used += bytes_used >> 1; | |
549 | spin_unlock(&block_group->lock); | |
550 | spin_unlock(&block_group->space_info->lock); | |
2ff7e61e | 551 | fragment_free_space(block_group); |
d0bd4560 JB |
552 | } |
553 | #endif | |
554 | ||
555 | caching_ctl->progress = (u64)-1; | |
11833d66 | 556 | |
9e351cc8 | 557 | up_read(&fs_info->commit_root_sem); |
9e715da8 | 558 | free_excluded_extents(block_group); |
11833d66 | 559 | mutex_unlock(&caching_ctl->mutex); |
73fa48b6 | 560 | |
11833d66 YZ |
561 | wake_up(&caching_ctl->wait); |
562 | ||
563 | put_caching_control(caching_ctl); | |
11dfe35a | 564 | btrfs_put_block_group(block_group); |
817d52f8 JB |
565 | } |
566 | ||
9d66e233 | 567 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
9d66e233 | 568 | int load_cache_only) |
817d52f8 | 569 | { |
291c7d2f | 570 | DEFINE_WAIT(wait); |
11833d66 YZ |
571 | struct btrfs_fs_info *fs_info = cache->fs_info; |
572 | struct btrfs_caching_control *caching_ctl; | |
817d52f8 JB |
573 | int ret = 0; |
574 | ||
291c7d2f | 575 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
79787eaa JM |
576 | if (!caching_ctl) |
577 | return -ENOMEM; | |
291c7d2f JB |
578 | |
579 | INIT_LIST_HEAD(&caching_ctl->list); | |
580 | mutex_init(&caching_ctl->mutex); | |
581 | init_waitqueue_head(&caching_ctl->wait); | |
582 | caching_ctl->block_group = cache; | |
583 | caching_ctl->progress = cache->key.objectid; | |
1e4f4714 | 584 | refcount_set(&caching_ctl->count, 1); |
9e0af237 LB |
585 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, |
586 | caching_thread, NULL, NULL); | |
291c7d2f JB |
587 | |
588 | spin_lock(&cache->lock); | |
589 | /* | |
590 | * This should be a rare occasion, but this could happen I think in the | |
591 | * case where one thread starts to load the space cache info, and then | |
592 | * some other thread starts a transaction commit which tries to do an | |
593 | * allocation while the other thread is still loading the space cache | |
594 | * info. The previous loop should have kept us from choosing this block | |
595 | * group, but if we've moved to the state where we will wait on caching | |
596 | * block groups we need to first check if we're doing a fast load here, | |
597 | * so we can wait for it to finish, otherwise we could end up allocating | |
598 | * from a block group who's cache gets evicted for one reason or | |
599 | * another. | |
600 | */ | |
601 | while (cache->cached == BTRFS_CACHE_FAST) { | |
602 | struct btrfs_caching_control *ctl; | |
603 | ||
604 | ctl = cache->caching_ctl; | |
1e4f4714 | 605 | refcount_inc(&ctl->count); |
291c7d2f JB |
606 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); |
607 | spin_unlock(&cache->lock); | |
608 | ||
609 | schedule(); | |
610 | ||
611 | finish_wait(&ctl->wait, &wait); | |
612 | put_caching_control(ctl); | |
613 | spin_lock(&cache->lock); | |
614 | } | |
615 | ||
616 | if (cache->cached != BTRFS_CACHE_NO) { | |
617 | spin_unlock(&cache->lock); | |
618 | kfree(caching_ctl); | |
11833d66 | 619 | return 0; |
291c7d2f JB |
620 | } |
621 | WARN_ON(cache->caching_ctl); | |
622 | cache->caching_ctl = caching_ctl; | |
623 | cache->cached = BTRFS_CACHE_FAST; | |
624 | spin_unlock(&cache->lock); | |
11833d66 | 625 | |
d8953d69 | 626 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cb83b7b8 | 627 | mutex_lock(&caching_ctl->mutex); |
9d66e233 JB |
628 | ret = load_free_space_cache(fs_info, cache); |
629 | ||
630 | spin_lock(&cache->lock); | |
631 | if (ret == 1) { | |
291c7d2f | 632 | cache->caching_ctl = NULL; |
9d66e233 JB |
633 | cache->cached = BTRFS_CACHE_FINISHED; |
634 | cache->last_byte_to_unpin = (u64)-1; | |
cb83b7b8 | 635 | caching_ctl->progress = (u64)-1; |
9d66e233 | 636 | } else { |
291c7d2f JB |
637 | if (load_cache_only) { |
638 | cache->caching_ctl = NULL; | |
639 | cache->cached = BTRFS_CACHE_NO; | |
640 | } else { | |
641 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 642 | cache->has_caching_ctl = 1; |
291c7d2f | 643 | } |
9d66e233 JB |
644 | } |
645 | spin_unlock(&cache->lock); | |
d0bd4560 JB |
646 | #ifdef CONFIG_BTRFS_DEBUG |
647 | if (ret == 1 && | |
2ff7e61e | 648 | btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
649 | u64 bytes_used; |
650 | ||
651 | spin_lock(&cache->space_info->lock); | |
652 | spin_lock(&cache->lock); | |
653 | bytes_used = cache->key.offset - | |
654 | btrfs_block_group_used(&cache->item); | |
655 | cache->space_info->bytes_used += bytes_used >> 1; | |
656 | spin_unlock(&cache->lock); | |
657 | spin_unlock(&cache->space_info->lock); | |
2ff7e61e | 658 | fragment_free_space(cache); |
d0bd4560 JB |
659 | } |
660 | #endif | |
cb83b7b8 JB |
661 | mutex_unlock(&caching_ctl->mutex); |
662 | ||
291c7d2f | 663 | wake_up(&caching_ctl->wait); |
3c14874a | 664 | if (ret == 1) { |
291c7d2f | 665 | put_caching_control(caching_ctl); |
9e715da8 | 666 | free_excluded_extents(cache); |
9d66e233 | 667 | return 0; |
3c14874a | 668 | } |
291c7d2f JB |
669 | } else { |
670 | /* | |
1e144fb8 OS |
671 | * We're either using the free space tree or no caching at all. |
672 | * Set cached to the appropriate value and wakeup any waiters. | |
291c7d2f JB |
673 | */ |
674 | spin_lock(&cache->lock); | |
675 | if (load_cache_only) { | |
676 | cache->caching_ctl = NULL; | |
677 | cache->cached = BTRFS_CACHE_NO; | |
678 | } else { | |
679 | cache->cached = BTRFS_CACHE_STARTED; | |
4f69cb98 | 680 | cache->has_caching_ctl = 1; |
291c7d2f JB |
681 | } |
682 | spin_unlock(&cache->lock); | |
683 | wake_up(&caching_ctl->wait); | |
9d66e233 JB |
684 | } |
685 | ||
291c7d2f JB |
686 | if (load_cache_only) { |
687 | put_caching_control(caching_ctl); | |
11833d66 | 688 | return 0; |
817d52f8 | 689 | } |
817d52f8 | 690 | |
9e351cc8 | 691 | down_write(&fs_info->commit_root_sem); |
1e4f4714 | 692 | refcount_inc(&caching_ctl->count); |
11833d66 | 693 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
9e351cc8 | 694 | up_write(&fs_info->commit_root_sem); |
11833d66 | 695 | |
11dfe35a | 696 | btrfs_get_block_group(cache); |
11833d66 | 697 | |
e66f0bb1 | 698 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
817d52f8 | 699 | |
ef8bbdfe | 700 | return ret; |
e37c9e69 CM |
701 | } |
702 | ||
0f9dd46c JB |
703 | /* |
704 | * return the block group that starts at or after bytenr | |
705 | */ | |
d397712b CM |
706 | static struct btrfs_block_group_cache * |
707 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) | |
0ef3e66b | 708 | { |
e2c89907 | 709 | return block_group_cache_tree_search(info, bytenr, 0); |
0ef3e66b CM |
710 | } |
711 | ||
0f9dd46c | 712 | /* |
9f55684c | 713 | * return the block group that contains the given bytenr |
0f9dd46c | 714 | */ |
d397712b CM |
715 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
716 | struct btrfs_fs_info *info, | |
717 | u64 bytenr) | |
be744175 | 718 | { |
e2c89907 | 719 | return block_group_cache_tree_search(info, bytenr, 1); |
be744175 | 720 | } |
0b86a832 | 721 | |
0f9dd46c JB |
722 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
723 | u64 flags) | |
6324fbf3 | 724 | { |
0f9dd46c | 725 | struct list_head *head = &info->space_info; |
0f9dd46c | 726 | struct btrfs_space_info *found; |
4184ea7f | 727 | |
52ba6929 | 728 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
b742bb82 | 729 | |
4184ea7f CM |
730 | rcu_read_lock(); |
731 | list_for_each_entry_rcu(found, head, list) { | |
67377734 | 732 | if (found->flags & flags) { |
4184ea7f | 733 | rcu_read_unlock(); |
0f9dd46c | 734 | return found; |
4184ea7f | 735 | } |
0f9dd46c | 736 | } |
4184ea7f | 737 | rcu_read_unlock(); |
0f9dd46c | 738 | return NULL; |
6324fbf3 CM |
739 | } |
740 | ||
0d9f824d | 741 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes, |
29d2b84c | 742 | bool metadata, u64 root_objectid) |
0d9f824d OS |
743 | { |
744 | struct btrfs_space_info *space_info; | |
745 | u64 flags; | |
746 | ||
29d2b84c | 747 | if (metadata) { |
0d9f824d OS |
748 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) |
749 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
750 | else | |
751 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
752 | } else { | |
753 | flags = BTRFS_BLOCK_GROUP_DATA; | |
754 | } | |
755 | ||
756 | space_info = __find_space_info(fs_info, flags); | |
55e8196a | 757 | ASSERT(space_info); |
dec59fa3 EL |
758 | percpu_counter_add_batch(&space_info->total_bytes_pinned, num_bytes, |
759 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
0d9f824d OS |
760 | } |
761 | ||
4184ea7f CM |
762 | /* |
763 | * after adding space to the filesystem, we need to clear the full flags | |
764 | * on all the space infos. | |
765 | */ | |
766 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) | |
767 | { | |
768 | struct list_head *head = &info->space_info; | |
769 | struct btrfs_space_info *found; | |
770 | ||
771 | rcu_read_lock(); | |
772 | list_for_each_entry_rcu(found, head, list) | |
773 | found->full = 0; | |
774 | rcu_read_unlock(); | |
775 | } | |
776 | ||
1a4ed8fd | 777 | /* simple helper to search for an existing data extent at a given offset */ |
2ff7e61e | 778 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
e02119d5 CM |
779 | { |
780 | int ret; | |
781 | struct btrfs_key key; | |
31840ae1 | 782 | struct btrfs_path *path; |
e02119d5 | 783 | |
31840ae1 | 784 | path = btrfs_alloc_path(); |
d8926bb3 MF |
785 | if (!path) |
786 | return -ENOMEM; | |
787 | ||
e02119d5 CM |
788 | key.objectid = start; |
789 | key.offset = len; | |
3173a18f | 790 | key.type = BTRFS_EXTENT_ITEM_KEY; |
0b246afa | 791 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
31840ae1 | 792 | btrfs_free_path(path); |
7bb86316 CM |
793 | return ret; |
794 | } | |
795 | ||
a22285a6 | 796 | /* |
3173a18f | 797 | * helper function to lookup reference count and flags of a tree block. |
a22285a6 YZ |
798 | * |
799 | * the head node for delayed ref is used to store the sum of all the | |
800 | * reference count modifications queued up in the rbtree. the head | |
801 | * node may also store the extent flags to set. This way you can check | |
802 | * to see what the reference count and extent flags would be if all of | |
803 | * the delayed refs are not processed. | |
804 | */ | |
805 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, | |
2ff7e61e | 806 | struct btrfs_fs_info *fs_info, u64 bytenr, |
3173a18f | 807 | u64 offset, int metadata, u64 *refs, u64 *flags) |
a22285a6 YZ |
808 | { |
809 | struct btrfs_delayed_ref_head *head; | |
810 | struct btrfs_delayed_ref_root *delayed_refs; | |
811 | struct btrfs_path *path; | |
812 | struct btrfs_extent_item *ei; | |
813 | struct extent_buffer *leaf; | |
814 | struct btrfs_key key; | |
815 | u32 item_size; | |
816 | u64 num_refs; | |
817 | u64 extent_flags; | |
818 | int ret; | |
819 | ||
3173a18f JB |
820 | /* |
821 | * If we don't have skinny metadata, don't bother doing anything | |
822 | * different | |
823 | */ | |
0b246afa JM |
824 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
825 | offset = fs_info->nodesize; | |
3173a18f JB |
826 | metadata = 0; |
827 | } | |
828 | ||
a22285a6 YZ |
829 | path = btrfs_alloc_path(); |
830 | if (!path) | |
831 | return -ENOMEM; | |
832 | ||
a22285a6 YZ |
833 | if (!trans) { |
834 | path->skip_locking = 1; | |
835 | path->search_commit_root = 1; | |
836 | } | |
639eefc8 FDBM |
837 | |
838 | search_again: | |
839 | key.objectid = bytenr; | |
840 | key.offset = offset; | |
841 | if (metadata) | |
842 | key.type = BTRFS_METADATA_ITEM_KEY; | |
843 | else | |
844 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
845 | ||
0b246afa | 846 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
a22285a6 YZ |
847 | if (ret < 0) |
848 | goto out_free; | |
849 | ||
3173a18f | 850 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
74be9510 FDBM |
851 | if (path->slots[0]) { |
852 | path->slots[0]--; | |
853 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
854 | path->slots[0]); | |
855 | if (key.objectid == bytenr && | |
856 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
0b246afa | 857 | key.offset == fs_info->nodesize) |
74be9510 FDBM |
858 | ret = 0; |
859 | } | |
3173a18f JB |
860 | } |
861 | ||
a22285a6 YZ |
862 | if (ret == 0) { |
863 | leaf = path->nodes[0]; | |
864 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
865 | if (item_size >= sizeof(*ei)) { | |
866 | ei = btrfs_item_ptr(leaf, path->slots[0], | |
867 | struct btrfs_extent_item); | |
868 | num_refs = btrfs_extent_refs(leaf, ei); | |
869 | extent_flags = btrfs_extent_flags(leaf, ei); | |
870 | } else { | |
ba3c2b19 NB |
871 | ret = -EINVAL; |
872 | btrfs_print_v0_err(fs_info); | |
873 | if (trans) | |
874 | btrfs_abort_transaction(trans, ret); | |
875 | else | |
876 | btrfs_handle_fs_error(fs_info, ret, NULL); | |
877 | ||
878 | goto out_free; | |
a22285a6 | 879 | } |
ba3c2b19 | 880 | |
a22285a6 YZ |
881 | BUG_ON(num_refs == 0); |
882 | } else { | |
883 | num_refs = 0; | |
884 | extent_flags = 0; | |
885 | ret = 0; | |
886 | } | |
887 | ||
888 | if (!trans) | |
889 | goto out; | |
890 | ||
891 | delayed_refs = &trans->transaction->delayed_refs; | |
892 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 893 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
a22285a6 YZ |
894 | if (head) { |
895 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 896 | refcount_inc(&head->refs); |
a22285a6 YZ |
897 | spin_unlock(&delayed_refs->lock); |
898 | ||
b3b4aa74 | 899 | btrfs_release_path(path); |
a22285a6 | 900 | |
8cc33e5c DS |
901 | /* |
902 | * Mutex was contended, block until it's released and try | |
903 | * again | |
904 | */ | |
a22285a6 YZ |
905 | mutex_lock(&head->mutex); |
906 | mutex_unlock(&head->mutex); | |
d278850e | 907 | btrfs_put_delayed_ref_head(head); |
639eefc8 | 908 | goto search_again; |
a22285a6 | 909 | } |
d7df2c79 | 910 | spin_lock(&head->lock); |
a22285a6 YZ |
911 | if (head->extent_op && head->extent_op->update_flags) |
912 | extent_flags |= head->extent_op->flags_to_set; | |
913 | else | |
914 | BUG_ON(num_refs == 0); | |
915 | ||
d278850e | 916 | num_refs += head->ref_mod; |
d7df2c79 | 917 | spin_unlock(&head->lock); |
a22285a6 YZ |
918 | mutex_unlock(&head->mutex); |
919 | } | |
920 | spin_unlock(&delayed_refs->lock); | |
921 | out: | |
922 | WARN_ON(num_refs == 0); | |
923 | if (refs) | |
924 | *refs = num_refs; | |
925 | if (flags) | |
926 | *flags = extent_flags; | |
927 | out_free: | |
928 | btrfs_free_path(path); | |
929 | return ret; | |
930 | } | |
931 | ||
d8d5f3e1 CM |
932 | /* |
933 | * Back reference rules. Back refs have three main goals: | |
934 | * | |
935 | * 1) differentiate between all holders of references to an extent so that | |
936 | * when a reference is dropped we can make sure it was a valid reference | |
937 | * before freeing the extent. | |
938 | * | |
939 | * 2) Provide enough information to quickly find the holders of an extent | |
940 | * if we notice a given block is corrupted or bad. | |
941 | * | |
942 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool | |
943 | * maintenance. This is actually the same as #2, but with a slightly | |
944 | * different use case. | |
945 | * | |
5d4f98a2 YZ |
946 | * There are two kinds of back refs. The implicit back refs is optimized |
947 | * for pointers in non-shared tree blocks. For a given pointer in a block, | |
948 | * back refs of this kind provide information about the block's owner tree | |
949 | * and the pointer's key. These information allow us to find the block by | |
950 | * b-tree searching. The full back refs is for pointers in tree blocks not | |
951 | * referenced by their owner trees. The location of tree block is recorded | |
952 | * in the back refs. Actually the full back refs is generic, and can be | |
953 | * used in all cases the implicit back refs is used. The major shortcoming | |
954 | * of the full back refs is its overhead. Every time a tree block gets | |
955 | * COWed, we have to update back refs entry for all pointers in it. | |
956 | * | |
957 | * For a newly allocated tree block, we use implicit back refs for | |
958 | * pointers in it. This means most tree related operations only involve | |
959 | * implicit back refs. For a tree block created in old transaction, the | |
960 | * only way to drop a reference to it is COW it. So we can detect the | |
961 | * event that tree block loses its owner tree's reference and do the | |
962 | * back refs conversion. | |
963 | * | |
01327610 | 964 | * When a tree block is COWed through a tree, there are four cases: |
5d4f98a2 YZ |
965 | * |
966 | * The reference count of the block is one and the tree is the block's | |
967 | * owner tree. Nothing to do in this case. | |
968 | * | |
969 | * The reference count of the block is one and the tree is not the | |
970 | * block's owner tree. In this case, full back refs is used for pointers | |
971 | * in the block. Remove these full back refs, add implicit back refs for | |
972 | * every pointers in the new block. | |
973 | * | |
974 | * The reference count of the block is greater than one and the tree is | |
975 | * the block's owner tree. In this case, implicit back refs is used for | |
976 | * pointers in the block. Add full back refs for every pointers in the | |
977 | * block, increase lower level extents' reference counts. The original | |
978 | * implicit back refs are entailed to the new block. | |
979 | * | |
980 | * The reference count of the block is greater than one and the tree is | |
981 | * not the block's owner tree. Add implicit back refs for every pointer in | |
982 | * the new block, increase lower level extents' reference count. | |
983 | * | |
984 | * Back Reference Key composing: | |
985 | * | |
986 | * The key objectid corresponds to the first byte in the extent, | |
987 | * The key type is used to differentiate between types of back refs. | |
988 | * There are different meanings of the key offset for different types | |
989 | * of back refs. | |
990 | * | |
d8d5f3e1 CM |
991 | * File extents can be referenced by: |
992 | * | |
993 | * - multiple snapshots, subvolumes, or different generations in one subvol | |
31840ae1 | 994 | * - different files inside a single subvolume |
d8d5f3e1 CM |
995 | * - different offsets inside a file (bookend extents in file.c) |
996 | * | |
5d4f98a2 | 997 | * The extent ref structure for the implicit back refs has fields for: |
d8d5f3e1 CM |
998 | * |
999 | * - Objectid of the subvolume root | |
d8d5f3e1 | 1000 | * - objectid of the file holding the reference |
5d4f98a2 YZ |
1001 | * - original offset in the file |
1002 | * - how many bookend extents | |
d8d5f3e1 | 1003 | * |
5d4f98a2 YZ |
1004 | * The key offset for the implicit back refs is hash of the first |
1005 | * three fields. | |
d8d5f3e1 | 1006 | * |
5d4f98a2 | 1007 | * The extent ref structure for the full back refs has field for: |
d8d5f3e1 | 1008 | * |
5d4f98a2 | 1009 | * - number of pointers in the tree leaf |
d8d5f3e1 | 1010 | * |
5d4f98a2 YZ |
1011 | * The key offset for the implicit back refs is the first byte of |
1012 | * the tree leaf | |
d8d5f3e1 | 1013 | * |
5d4f98a2 YZ |
1014 | * When a file extent is allocated, The implicit back refs is used. |
1015 | * the fields are filled in: | |
d8d5f3e1 | 1016 | * |
5d4f98a2 | 1017 | * (root_key.objectid, inode objectid, offset in file, 1) |
d8d5f3e1 | 1018 | * |
5d4f98a2 YZ |
1019 | * When a file extent is removed file truncation, we find the |
1020 | * corresponding implicit back refs and check the following fields: | |
d8d5f3e1 | 1021 | * |
5d4f98a2 | 1022 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
d8d5f3e1 | 1023 | * |
5d4f98a2 | 1024 | * Btree extents can be referenced by: |
d8d5f3e1 | 1025 | * |
5d4f98a2 | 1026 | * - Different subvolumes |
d8d5f3e1 | 1027 | * |
5d4f98a2 YZ |
1028 | * Both the implicit back refs and the full back refs for tree blocks |
1029 | * only consist of key. The key offset for the implicit back refs is | |
1030 | * objectid of block's owner tree. The key offset for the full back refs | |
1031 | * is the first byte of parent block. | |
d8d5f3e1 | 1032 | * |
5d4f98a2 YZ |
1033 | * When implicit back refs is used, information about the lowest key and |
1034 | * level of the tree block are required. These information are stored in | |
1035 | * tree block info structure. | |
d8d5f3e1 | 1036 | */ |
31840ae1 | 1037 | |
167ce953 LB |
1038 | /* |
1039 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, | |
1040 | * is_data == BTRFS_REF_TYPE_DATA, data type is requried, | |
1041 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. | |
1042 | */ | |
1043 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, | |
1044 | struct btrfs_extent_inline_ref *iref, | |
1045 | enum btrfs_inline_ref_type is_data) | |
1046 | { | |
1047 | int type = btrfs_extent_inline_ref_type(eb, iref); | |
64ecdb64 | 1048 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
167ce953 LB |
1049 | |
1050 | if (type == BTRFS_TREE_BLOCK_REF_KEY || | |
1051 | type == BTRFS_SHARED_BLOCK_REF_KEY || | |
1052 | type == BTRFS_SHARED_DATA_REF_KEY || | |
1053 | type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1054 | if (is_data == BTRFS_REF_TYPE_BLOCK) { | |
64ecdb64 | 1055 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
167ce953 | 1056 | return type; |
64ecdb64 LB |
1057 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
1058 | ASSERT(eb->fs_info); | |
1059 | /* | |
1060 | * Every shared one has parent tree | |
1061 | * block, which must be aligned to | |
1062 | * nodesize. | |
1063 | */ | |
1064 | if (offset && | |
1065 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1066 | return type; | |
1067 | } | |
167ce953 | 1068 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
64ecdb64 | 1069 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
167ce953 | 1070 | return type; |
64ecdb64 LB |
1071 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
1072 | ASSERT(eb->fs_info); | |
1073 | /* | |
1074 | * Every shared one has parent tree | |
1075 | * block, which must be aligned to | |
1076 | * nodesize. | |
1077 | */ | |
1078 | if (offset && | |
1079 | IS_ALIGNED(offset, eb->fs_info->nodesize)) | |
1080 | return type; | |
1081 | } | |
167ce953 LB |
1082 | } else { |
1083 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); | |
1084 | return type; | |
1085 | } | |
1086 | } | |
1087 | ||
1088 | btrfs_print_leaf((struct extent_buffer *)eb); | |
1089 | btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d", | |
1090 | eb->start, type); | |
1091 | WARN_ON(1); | |
1092 | ||
1093 | return BTRFS_REF_TYPE_INVALID; | |
1094 | } | |
1095 | ||
5d4f98a2 YZ |
1096 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
1097 | { | |
1098 | u32 high_crc = ~(u32)0; | |
1099 | u32 low_crc = ~(u32)0; | |
1100 | __le64 lenum; | |
1101 | ||
1102 | lenum = cpu_to_le64(root_objectid); | |
9678c543 | 1103 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1104 | lenum = cpu_to_le64(owner); |
9678c543 | 1105 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 | 1106 | lenum = cpu_to_le64(offset); |
9678c543 | 1107 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
5d4f98a2 YZ |
1108 | |
1109 | return ((u64)high_crc << 31) ^ (u64)low_crc; | |
1110 | } | |
1111 | ||
1112 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, | |
1113 | struct btrfs_extent_data_ref *ref) | |
1114 | { | |
1115 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), | |
1116 | btrfs_extent_data_ref_objectid(leaf, ref), | |
1117 | btrfs_extent_data_ref_offset(leaf, ref)); | |
1118 | } | |
1119 | ||
1120 | static int match_extent_data_ref(struct extent_buffer *leaf, | |
1121 | struct btrfs_extent_data_ref *ref, | |
1122 | u64 root_objectid, u64 owner, u64 offset) | |
1123 | { | |
1124 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || | |
1125 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || | |
1126 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
1127 | return 0; | |
1128 | return 1; | |
1129 | } | |
1130 | ||
1131 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1132 | struct btrfs_path *path, |
1133 | u64 bytenr, u64 parent, | |
1134 | u64 root_objectid, | |
1135 | u64 owner, u64 offset) | |
1136 | { | |
bd1d53ef | 1137 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 YZ |
1138 | struct btrfs_key key; |
1139 | struct btrfs_extent_data_ref *ref; | |
31840ae1 | 1140 | struct extent_buffer *leaf; |
5d4f98a2 | 1141 | u32 nritems; |
74493f7a | 1142 | int ret; |
5d4f98a2 YZ |
1143 | int recow; |
1144 | int err = -ENOENT; | |
74493f7a | 1145 | |
31840ae1 | 1146 | key.objectid = bytenr; |
5d4f98a2 YZ |
1147 | if (parent) { |
1148 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1149 | key.offset = parent; | |
1150 | } else { | |
1151 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1152 | key.offset = hash_extent_data_ref(root_objectid, | |
1153 | owner, offset); | |
1154 | } | |
1155 | again: | |
1156 | recow = 0; | |
1157 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1158 | if (ret < 0) { | |
1159 | err = ret; | |
1160 | goto fail; | |
1161 | } | |
31840ae1 | 1162 | |
5d4f98a2 YZ |
1163 | if (parent) { |
1164 | if (!ret) | |
1165 | return 0; | |
5d4f98a2 | 1166 | goto fail; |
31840ae1 ZY |
1167 | } |
1168 | ||
1169 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1170 | nritems = btrfs_header_nritems(leaf); |
1171 | while (1) { | |
1172 | if (path->slots[0] >= nritems) { | |
1173 | ret = btrfs_next_leaf(root, path); | |
1174 | if (ret < 0) | |
1175 | err = ret; | |
1176 | if (ret) | |
1177 | goto fail; | |
1178 | ||
1179 | leaf = path->nodes[0]; | |
1180 | nritems = btrfs_header_nritems(leaf); | |
1181 | recow = 1; | |
1182 | } | |
1183 | ||
1184 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1185 | if (key.objectid != bytenr || | |
1186 | key.type != BTRFS_EXTENT_DATA_REF_KEY) | |
1187 | goto fail; | |
1188 | ||
1189 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1190 | struct btrfs_extent_data_ref); | |
1191 | ||
1192 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1193 | owner, offset)) { | |
1194 | if (recow) { | |
b3b4aa74 | 1195 | btrfs_release_path(path); |
5d4f98a2 YZ |
1196 | goto again; |
1197 | } | |
1198 | err = 0; | |
1199 | break; | |
1200 | } | |
1201 | path->slots[0]++; | |
31840ae1 | 1202 | } |
5d4f98a2 YZ |
1203 | fail: |
1204 | return err; | |
31840ae1 ZY |
1205 | } |
1206 | ||
5d4f98a2 | 1207 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1208 | struct btrfs_path *path, |
1209 | u64 bytenr, u64 parent, | |
1210 | u64 root_objectid, u64 owner, | |
1211 | u64 offset, int refs_to_add) | |
31840ae1 | 1212 | { |
62b895af | 1213 | struct btrfs_root *root = trans->fs_info->extent_root; |
31840ae1 ZY |
1214 | struct btrfs_key key; |
1215 | struct extent_buffer *leaf; | |
5d4f98a2 | 1216 | u32 size; |
31840ae1 ZY |
1217 | u32 num_refs; |
1218 | int ret; | |
74493f7a | 1219 | |
74493f7a | 1220 | key.objectid = bytenr; |
5d4f98a2 YZ |
1221 | if (parent) { |
1222 | key.type = BTRFS_SHARED_DATA_REF_KEY; | |
1223 | key.offset = parent; | |
1224 | size = sizeof(struct btrfs_shared_data_ref); | |
1225 | } else { | |
1226 | key.type = BTRFS_EXTENT_DATA_REF_KEY; | |
1227 | key.offset = hash_extent_data_ref(root_objectid, | |
1228 | owner, offset); | |
1229 | size = sizeof(struct btrfs_extent_data_ref); | |
1230 | } | |
74493f7a | 1231 | |
5d4f98a2 YZ |
1232 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
1233 | if (ret && ret != -EEXIST) | |
1234 | goto fail; | |
1235 | ||
1236 | leaf = path->nodes[0]; | |
1237 | if (parent) { | |
1238 | struct btrfs_shared_data_ref *ref; | |
31840ae1 | 1239 | ref = btrfs_item_ptr(leaf, path->slots[0], |
5d4f98a2 YZ |
1240 | struct btrfs_shared_data_ref); |
1241 | if (ret == 0) { | |
1242 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); | |
1243 | } else { | |
1244 | num_refs = btrfs_shared_data_ref_count(leaf, ref); | |
1245 | num_refs += refs_to_add; | |
1246 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1247 | } |
5d4f98a2 YZ |
1248 | } else { |
1249 | struct btrfs_extent_data_ref *ref; | |
1250 | while (ret == -EEXIST) { | |
1251 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1252 | struct btrfs_extent_data_ref); | |
1253 | if (match_extent_data_ref(leaf, ref, root_objectid, | |
1254 | owner, offset)) | |
1255 | break; | |
b3b4aa74 | 1256 | btrfs_release_path(path); |
5d4f98a2 YZ |
1257 | key.offset++; |
1258 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1259 | size); | |
1260 | if (ret && ret != -EEXIST) | |
1261 | goto fail; | |
31840ae1 | 1262 | |
5d4f98a2 YZ |
1263 | leaf = path->nodes[0]; |
1264 | } | |
1265 | ref = btrfs_item_ptr(leaf, path->slots[0], | |
1266 | struct btrfs_extent_data_ref); | |
1267 | if (ret == 0) { | |
1268 | btrfs_set_extent_data_ref_root(leaf, ref, | |
1269 | root_objectid); | |
1270 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
1271 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
1272 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); | |
1273 | } else { | |
1274 | num_refs = btrfs_extent_data_ref_count(leaf, ref); | |
1275 | num_refs += refs_to_add; | |
1276 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); | |
31840ae1 | 1277 | } |
31840ae1 | 1278 | } |
5d4f98a2 YZ |
1279 | btrfs_mark_buffer_dirty(leaf); |
1280 | ret = 0; | |
1281 | fail: | |
b3b4aa74 | 1282 | btrfs_release_path(path); |
7bb86316 | 1283 | return ret; |
74493f7a CM |
1284 | } |
1285 | ||
5d4f98a2 | 1286 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 | 1287 | struct btrfs_path *path, |
fcebe456 | 1288 | int refs_to_drop, int *last_ref) |
31840ae1 | 1289 | { |
5d4f98a2 YZ |
1290 | struct btrfs_key key; |
1291 | struct btrfs_extent_data_ref *ref1 = NULL; | |
1292 | struct btrfs_shared_data_ref *ref2 = NULL; | |
31840ae1 | 1293 | struct extent_buffer *leaf; |
5d4f98a2 | 1294 | u32 num_refs = 0; |
31840ae1 ZY |
1295 | int ret = 0; |
1296 | ||
1297 | leaf = path->nodes[0]; | |
5d4f98a2 YZ |
1298 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
1299 | ||
1300 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1301 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1302 | struct btrfs_extent_data_ref); | |
1303 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1304 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1305 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1306 | struct btrfs_shared_data_ref); | |
1307 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
6d8ff4e4 | 1308 | } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { |
ba3c2b19 NB |
1309 | btrfs_print_v0_err(trans->fs_info); |
1310 | btrfs_abort_transaction(trans, -EINVAL); | |
1311 | return -EINVAL; | |
5d4f98a2 YZ |
1312 | } else { |
1313 | BUG(); | |
1314 | } | |
1315 | ||
56bec294 CM |
1316 | BUG_ON(num_refs < refs_to_drop); |
1317 | num_refs -= refs_to_drop; | |
5d4f98a2 | 1318 | |
31840ae1 | 1319 | if (num_refs == 0) { |
e9f6290d | 1320 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
fcebe456 | 1321 | *last_ref = 1; |
31840ae1 | 1322 | } else { |
5d4f98a2 YZ |
1323 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
1324 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); | |
1325 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) | |
1326 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); | |
31840ae1 ZY |
1327 | btrfs_mark_buffer_dirty(leaf); |
1328 | } | |
31840ae1 ZY |
1329 | return ret; |
1330 | } | |
1331 | ||
9ed0dea0 | 1332 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
5d4f98a2 | 1333 | struct btrfs_extent_inline_ref *iref) |
15916de8 | 1334 | { |
5d4f98a2 YZ |
1335 | struct btrfs_key key; |
1336 | struct extent_buffer *leaf; | |
1337 | struct btrfs_extent_data_ref *ref1; | |
1338 | struct btrfs_shared_data_ref *ref2; | |
1339 | u32 num_refs = 0; | |
3de28d57 | 1340 | int type; |
5d4f98a2 YZ |
1341 | |
1342 | leaf = path->nodes[0]; | |
1343 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
ba3c2b19 NB |
1344 | |
1345 | BUG_ON(key.type == BTRFS_EXTENT_REF_V0_KEY); | |
5d4f98a2 | 1346 | if (iref) { |
3de28d57 LB |
1347 | /* |
1348 | * If type is invalid, we should have bailed out earlier than | |
1349 | * this call. | |
1350 | */ | |
1351 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
1352 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
1353 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
5d4f98a2 YZ |
1354 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
1355 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1356 | } else { | |
1357 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); | |
1358 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
1359 | } | |
1360 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1361 | ref1 = btrfs_item_ptr(leaf, path->slots[0], | |
1362 | struct btrfs_extent_data_ref); | |
1363 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); | |
1364 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { | |
1365 | ref2 = btrfs_item_ptr(leaf, path->slots[0], | |
1366 | struct btrfs_shared_data_ref); | |
1367 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); | |
5d4f98a2 YZ |
1368 | } else { |
1369 | WARN_ON(1); | |
1370 | } | |
1371 | return num_refs; | |
1372 | } | |
15916de8 | 1373 | |
5d4f98a2 | 1374 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1375 | struct btrfs_path *path, |
1376 | u64 bytenr, u64 parent, | |
1377 | u64 root_objectid) | |
1f3c79a2 | 1378 | { |
b8582eea | 1379 | struct btrfs_root *root = trans->fs_info->extent_root; |
5d4f98a2 | 1380 | struct btrfs_key key; |
1f3c79a2 | 1381 | int ret; |
1f3c79a2 | 1382 | |
5d4f98a2 YZ |
1383 | key.objectid = bytenr; |
1384 | if (parent) { | |
1385 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1386 | key.offset = parent; | |
1387 | } else { | |
1388 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1389 | key.offset = root_objectid; | |
1f3c79a2 LH |
1390 | } |
1391 | ||
5d4f98a2 YZ |
1392 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
1393 | if (ret > 0) | |
1394 | ret = -ENOENT; | |
5d4f98a2 | 1395 | return ret; |
1f3c79a2 LH |
1396 | } |
1397 | ||
5d4f98a2 | 1398 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1399 | struct btrfs_path *path, |
1400 | u64 bytenr, u64 parent, | |
1401 | u64 root_objectid) | |
31840ae1 | 1402 | { |
5d4f98a2 | 1403 | struct btrfs_key key; |
31840ae1 | 1404 | int ret; |
31840ae1 | 1405 | |
5d4f98a2 YZ |
1406 | key.objectid = bytenr; |
1407 | if (parent) { | |
1408 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1409 | key.offset = parent; | |
1410 | } else { | |
1411 | key.type = BTRFS_TREE_BLOCK_REF_KEY; | |
1412 | key.offset = root_objectid; | |
1413 | } | |
1414 | ||
10728404 | 1415 | ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root, |
87bde3cd | 1416 | path, &key, 0); |
b3b4aa74 | 1417 | btrfs_release_path(path); |
31840ae1 ZY |
1418 | return ret; |
1419 | } | |
1420 | ||
5d4f98a2 | 1421 | static inline int extent_ref_type(u64 parent, u64 owner) |
31840ae1 | 1422 | { |
5d4f98a2 YZ |
1423 | int type; |
1424 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1425 | if (parent > 0) | |
1426 | type = BTRFS_SHARED_BLOCK_REF_KEY; | |
1427 | else | |
1428 | type = BTRFS_TREE_BLOCK_REF_KEY; | |
1429 | } else { | |
1430 | if (parent > 0) | |
1431 | type = BTRFS_SHARED_DATA_REF_KEY; | |
1432 | else | |
1433 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
1434 | } | |
1435 | return type; | |
31840ae1 | 1436 | } |
56bec294 | 1437 | |
2c47e605 YZ |
1438 | static int find_next_key(struct btrfs_path *path, int level, |
1439 | struct btrfs_key *key) | |
56bec294 | 1440 | |
02217ed2 | 1441 | { |
2c47e605 | 1442 | for (; level < BTRFS_MAX_LEVEL; level++) { |
5d4f98a2 YZ |
1443 | if (!path->nodes[level]) |
1444 | break; | |
5d4f98a2 YZ |
1445 | if (path->slots[level] + 1 >= |
1446 | btrfs_header_nritems(path->nodes[level])) | |
1447 | continue; | |
1448 | if (level == 0) | |
1449 | btrfs_item_key_to_cpu(path->nodes[level], key, | |
1450 | path->slots[level] + 1); | |
1451 | else | |
1452 | btrfs_node_key_to_cpu(path->nodes[level], key, | |
1453 | path->slots[level] + 1); | |
1454 | return 0; | |
1455 | } | |
1456 | return 1; | |
1457 | } | |
037e6390 | 1458 | |
5d4f98a2 YZ |
1459 | /* |
1460 | * look for inline back ref. if back ref is found, *ref_ret is set | |
1461 | * to the address of inline back ref, and 0 is returned. | |
1462 | * | |
1463 | * if back ref isn't found, *ref_ret is set to the address where it | |
1464 | * should be inserted, and -ENOENT is returned. | |
1465 | * | |
1466 | * if insert is true and there are too many inline back refs, the path | |
1467 | * points to the extent item, and -EAGAIN is returned. | |
1468 | * | |
1469 | * NOTE: inline back refs are ordered in the same way that back ref | |
1470 | * items in the tree are ordered. | |
1471 | */ | |
1472 | static noinline_for_stack | |
1473 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1474 | struct btrfs_path *path, |
1475 | struct btrfs_extent_inline_ref **ref_ret, | |
1476 | u64 bytenr, u64 num_bytes, | |
1477 | u64 parent, u64 root_objectid, | |
1478 | u64 owner, u64 offset, int insert) | |
1479 | { | |
867cc1fb | 1480 | struct btrfs_fs_info *fs_info = trans->fs_info; |
87bde3cd | 1481 | struct btrfs_root *root = fs_info->extent_root; |
5d4f98a2 YZ |
1482 | struct btrfs_key key; |
1483 | struct extent_buffer *leaf; | |
1484 | struct btrfs_extent_item *ei; | |
1485 | struct btrfs_extent_inline_ref *iref; | |
1486 | u64 flags; | |
1487 | u64 item_size; | |
1488 | unsigned long ptr; | |
1489 | unsigned long end; | |
1490 | int extra_size; | |
1491 | int type; | |
1492 | int want; | |
1493 | int ret; | |
1494 | int err = 0; | |
0b246afa | 1495 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3de28d57 | 1496 | int needed; |
26b8003f | 1497 | |
db94535d | 1498 | key.objectid = bytenr; |
31840ae1 | 1499 | key.type = BTRFS_EXTENT_ITEM_KEY; |
56bec294 | 1500 | key.offset = num_bytes; |
31840ae1 | 1501 | |
5d4f98a2 YZ |
1502 | want = extent_ref_type(parent, owner); |
1503 | if (insert) { | |
1504 | extra_size = btrfs_extent_inline_ref_size(want); | |
85d4198e | 1505 | path->keep_locks = 1; |
5d4f98a2 YZ |
1506 | } else |
1507 | extra_size = -1; | |
3173a18f JB |
1508 | |
1509 | /* | |
16d1c062 NB |
1510 | * Owner is our level, so we can just add one to get the level for the |
1511 | * block we are interested in. | |
3173a18f JB |
1512 | */ |
1513 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1514 | key.type = BTRFS_METADATA_ITEM_KEY; | |
1515 | key.offset = owner; | |
1516 | } | |
1517 | ||
1518 | again: | |
5d4f98a2 | 1519 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
b9473439 | 1520 | if (ret < 0) { |
5d4f98a2 YZ |
1521 | err = ret; |
1522 | goto out; | |
1523 | } | |
3173a18f JB |
1524 | |
1525 | /* | |
1526 | * We may be a newly converted file system which still has the old fat | |
1527 | * extent entries for metadata, so try and see if we have one of those. | |
1528 | */ | |
1529 | if (ret > 0 && skinny_metadata) { | |
1530 | skinny_metadata = false; | |
1531 | if (path->slots[0]) { | |
1532 | path->slots[0]--; | |
1533 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
1534 | path->slots[0]); | |
1535 | if (key.objectid == bytenr && | |
1536 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
1537 | key.offset == num_bytes) | |
1538 | ret = 0; | |
1539 | } | |
1540 | if (ret) { | |
9ce49a0b | 1541 | key.objectid = bytenr; |
3173a18f JB |
1542 | key.type = BTRFS_EXTENT_ITEM_KEY; |
1543 | key.offset = num_bytes; | |
1544 | btrfs_release_path(path); | |
1545 | goto again; | |
1546 | } | |
1547 | } | |
1548 | ||
79787eaa JM |
1549 | if (ret && !insert) { |
1550 | err = -ENOENT; | |
1551 | goto out; | |
fae7f21c | 1552 | } else if (WARN_ON(ret)) { |
492104c8 | 1553 | err = -EIO; |
492104c8 | 1554 | goto out; |
79787eaa | 1555 | } |
5d4f98a2 YZ |
1556 | |
1557 | leaf = path->nodes[0]; | |
1558 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
6d8ff4e4 | 1559 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
1560 | err = -EINVAL; |
1561 | btrfs_print_v0_err(fs_info); | |
1562 | btrfs_abort_transaction(trans, err); | |
1563 | goto out; | |
1564 | } | |
5d4f98a2 | 1565 | |
5d4f98a2 YZ |
1566 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1567 | flags = btrfs_extent_flags(leaf, ei); | |
1568 | ||
1569 | ptr = (unsigned long)(ei + 1); | |
1570 | end = (unsigned long)ei + item_size; | |
1571 | ||
3173a18f | 1572 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
5d4f98a2 YZ |
1573 | ptr += sizeof(struct btrfs_tree_block_info); |
1574 | BUG_ON(ptr > end); | |
5d4f98a2 YZ |
1575 | } |
1576 | ||
3de28d57 LB |
1577 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
1578 | needed = BTRFS_REF_TYPE_DATA; | |
1579 | else | |
1580 | needed = BTRFS_REF_TYPE_BLOCK; | |
1581 | ||
5d4f98a2 YZ |
1582 | err = -ENOENT; |
1583 | while (1) { | |
1584 | if (ptr >= end) { | |
1585 | WARN_ON(ptr > end); | |
1586 | break; | |
1587 | } | |
1588 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
3de28d57 LB |
1589 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
1590 | if (type == BTRFS_REF_TYPE_INVALID) { | |
af431dcb | 1591 | err = -EUCLEAN; |
3de28d57 LB |
1592 | goto out; |
1593 | } | |
1594 | ||
5d4f98a2 YZ |
1595 | if (want < type) |
1596 | break; | |
1597 | if (want > type) { | |
1598 | ptr += btrfs_extent_inline_ref_size(type); | |
1599 | continue; | |
1600 | } | |
1601 | ||
1602 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1603 | struct btrfs_extent_data_ref *dref; | |
1604 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1605 | if (match_extent_data_ref(leaf, dref, root_objectid, | |
1606 | owner, offset)) { | |
1607 | err = 0; | |
1608 | break; | |
1609 | } | |
1610 | if (hash_extent_data_ref_item(leaf, dref) < | |
1611 | hash_extent_data_ref(root_objectid, owner, offset)) | |
1612 | break; | |
1613 | } else { | |
1614 | u64 ref_offset; | |
1615 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); | |
1616 | if (parent > 0) { | |
1617 | if (parent == ref_offset) { | |
1618 | err = 0; | |
1619 | break; | |
1620 | } | |
1621 | if (ref_offset < parent) | |
1622 | break; | |
1623 | } else { | |
1624 | if (root_objectid == ref_offset) { | |
1625 | err = 0; | |
1626 | break; | |
1627 | } | |
1628 | if (ref_offset < root_objectid) | |
1629 | break; | |
1630 | } | |
1631 | } | |
1632 | ptr += btrfs_extent_inline_ref_size(type); | |
1633 | } | |
1634 | if (err == -ENOENT && insert) { | |
1635 | if (item_size + extra_size >= | |
1636 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { | |
1637 | err = -EAGAIN; | |
1638 | goto out; | |
1639 | } | |
1640 | /* | |
1641 | * To add new inline back ref, we have to make sure | |
1642 | * there is no corresponding back ref item. | |
1643 | * For simplicity, we just do not add new inline back | |
1644 | * ref if there is any kind of item for this block | |
1645 | */ | |
2c47e605 YZ |
1646 | if (find_next_key(path, 0, &key) == 0 && |
1647 | key.objectid == bytenr && | |
85d4198e | 1648 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
5d4f98a2 YZ |
1649 | err = -EAGAIN; |
1650 | goto out; | |
1651 | } | |
1652 | } | |
1653 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; | |
1654 | out: | |
85d4198e | 1655 | if (insert) { |
5d4f98a2 YZ |
1656 | path->keep_locks = 0; |
1657 | btrfs_unlock_up_safe(path, 1); | |
1658 | } | |
1659 | return err; | |
1660 | } | |
1661 | ||
1662 | /* | |
1663 | * helper to add new inline back ref | |
1664 | */ | |
1665 | static noinline_for_stack | |
87bde3cd | 1666 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
143bede5 JM |
1667 | struct btrfs_path *path, |
1668 | struct btrfs_extent_inline_ref *iref, | |
1669 | u64 parent, u64 root_objectid, | |
1670 | u64 owner, u64 offset, int refs_to_add, | |
1671 | struct btrfs_delayed_extent_op *extent_op) | |
5d4f98a2 YZ |
1672 | { |
1673 | struct extent_buffer *leaf; | |
1674 | struct btrfs_extent_item *ei; | |
1675 | unsigned long ptr; | |
1676 | unsigned long end; | |
1677 | unsigned long item_offset; | |
1678 | u64 refs; | |
1679 | int size; | |
1680 | int type; | |
5d4f98a2 YZ |
1681 | |
1682 | leaf = path->nodes[0]; | |
1683 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1684 | item_offset = (unsigned long)iref - (unsigned long)ei; | |
1685 | ||
1686 | type = extent_ref_type(parent, owner); | |
1687 | size = btrfs_extent_inline_ref_size(type); | |
1688 | ||
87bde3cd | 1689 | btrfs_extend_item(fs_info, path, size); |
5d4f98a2 YZ |
1690 | |
1691 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); | |
1692 | refs = btrfs_extent_refs(leaf, ei); | |
1693 | refs += refs_to_add; | |
1694 | btrfs_set_extent_refs(leaf, ei, refs); | |
1695 | if (extent_op) | |
1696 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1697 | ||
1698 | ptr = (unsigned long)ei + item_offset; | |
1699 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); | |
1700 | if (ptr < end - size) | |
1701 | memmove_extent_buffer(leaf, ptr + size, ptr, | |
1702 | end - size - ptr); | |
1703 | ||
1704 | iref = (struct btrfs_extent_inline_ref *)ptr; | |
1705 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
1706 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1707 | struct btrfs_extent_data_ref *dref; | |
1708 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1709 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); | |
1710 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); | |
1711 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); | |
1712 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); | |
1713 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1714 | struct btrfs_shared_data_ref *sref; | |
1715 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1716 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); | |
1717 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1718 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { | |
1719 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
1720 | } else { | |
1721 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); | |
1722 | } | |
1723 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1724 | } |
1725 | ||
1726 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1727 | struct btrfs_path *path, |
1728 | struct btrfs_extent_inline_ref **ref_ret, | |
1729 | u64 bytenr, u64 num_bytes, u64 parent, | |
1730 | u64 root_objectid, u64 owner, u64 offset) | |
1731 | { | |
1732 | int ret; | |
1733 | ||
867cc1fb NB |
1734 | ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr, |
1735 | num_bytes, parent, root_objectid, | |
1736 | owner, offset, 0); | |
5d4f98a2 | 1737 | if (ret != -ENOENT) |
54aa1f4d | 1738 | return ret; |
5d4f98a2 | 1739 | |
b3b4aa74 | 1740 | btrfs_release_path(path); |
5d4f98a2 YZ |
1741 | *ref_ret = NULL; |
1742 | ||
1743 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
b8582eea NB |
1744 | ret = lookup_tree_block_ref(trans, path, bytenr, parent, |
1745 | root_objectid); | |
5d4f98a2 | 1746 | } else { |
bd1d53ef NB |
1747 | ret = lookup_extent_data_ref(trans, path, bytenr, parent, |
1748 | root_objectid, owner, offset); | |
b9473439 | 1749 | } |
5d4f98a2 YZ |
1750 | return ret; |
1751 | } | |
31840ae1 | 1752 | |
5d4f98a2 YZ |
1753 | /* |
1754 | * helper to update/remove inline back ref | |
1755 | */ | |
1756 | static noinline_for_stack | |
61a18f1c | 1757 | void update_inline_extent_backref(struct btrfs_path *path, |
143bede5 JM |
1758 | struct btrfs_extent_inline_ref *iref, |
1759 | int refs_to_mod, | |
fcebe456 JB |
1760 | struct btrfs_delayed_extent_op *extent_op, |
1761 | int *last_ref) | |
5d4f98a2 | 1762 | { |
61a18f1c NB |
1763 | struct extent_buffer *leaf = path->nodes[0]; |
1764 | struct btrfs_fs_info *fs_info = leaf->fs_info; | |
5d4f98a2 YZ |
1765 | struct btrfs_extent_item *ei; |
1766 | struct btrfs_extent_data_ref *dref = NULL; | |
1767 | struct btrfs_shared_data_ref *sref = NULL; | |
1768 | unsigned long ptr; | |
1769 | unsigned long end; | |
1770 | u32 item_size; | |
1771 | int size; | |
1772 | int type; | |
5d4f98a2 YZ |
1773 | u64 refs; |
1774 | ||
5d4f98a2 YZ |
1775 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
1776 | refs = btrfs_extent_refs(leaf, ei); | |
1777 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); | |
1778 | refs += refs_to_mod; | |
1779 | btrfs_set_extent_refs(leaf, ei, refs); | |
1780 | if (extent_op) | |
1781 | __run_delayed_extent_op(extent_op, leaf, ei); | |
1782 | ||
3de28d57 LB |
1783 | /* |
1784 | * If type is invalid, we should have bailed out after | |
1785 | * lookup_inline_extent_backref(). | |
1786 | */ | |
1787 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); | |
1788 | ASSERT(type != BTRFS_REF_TYPE_INVALID); | |
5d4f98a2 YZ |
1789 | |
1790 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { | |
1791 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
1792 | refs = btrfs_extent_data_ref_count(leaf, dref); | |
1793 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { | |
1794 | sref = (struct btrfs_shared_data_ref *)(iref + 1); | |
1795 | refs = btrfs_shared_data_ref_count(leaf, sref); | |
1796 | } else { | |
1797 | refs = 1; | |
1798 | BUG_ON(refs_to_mod != -1); | |
56bec294 | 1799 | } |
31840ae1 | 1800 | |
5d4f98a2 YZ |
1801 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
1802 | refs += refs_to_mod; | |
1803 | ||
1804 | if (refs > 0) { | |
1805 | if (type == BTRFS_EXTENT_DATA_REF_KEY) | |
1806 | btrfs_set_extent_data_ref_count(leaf, dref, refs); | |
1807 | else | |
1808 | btrfs_set_shared_data_ref_count(leaf, sref, refs); | |
1809 | } else { | |
fcebe456 | 1810 | *last_ref = 1; |
5d4f98a2 YZ |
1811 | size = btrfs_extent_inline_ref_size(type); |
1812 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
1813 | ptr = (unsigned long)iref; | |
1814 | end = (unsigned long)ei + item_size; | |
1815 | if (ptr + size < end) | |
1816 | memmove_extent_buffer(leaf, ptr, ptr + size, | |
1817 | end - ptr - size); | |
1818 | item_size -= size; | |
87bde3cd | 1819 | btrfs_truncate_item(fs_info, path, item_size, 1); |
5d4f98a2 YZ |
1820 | } |
1821 | btrfs_mark_buffer_dirty(leaf); | |
5d4f98a2 YZ |
1822 | } |
1823 | ||
1824 | static noinline_for_stack | |
1825 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, | |
5d4f98a2 YZ |
1826 | struct btrfs_path *path, |
1827 | u64 bytenr, u64 num_bytes, u64 parent, | |
1828 | u64 root_objectid, u64 owner, | |
1829 | u64 offset, int refs_to_add, | |
1830 | struct btrfs_delayed_extent_op *extent_op) | |
1831 | { | |
1832 | struct btrfs_extent_inline_ref *iref; | |
1833 | int ret; | |
1834 | ||
867cc1fb NB |
1835 | ret = lookup_inline_extent_backref(trans, path, &iref, bytenr, |
1836 | num_bytes, parent, root_objectid, | |
1837 | owner, offset, 1); | |
5d4f98a2 YZ |
1838 | if (ret == 0) { |
1839 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); | |
61a18f1c NB |
1840 | update_inline_extent_backref(path, iref, refs_to_add, |
1841 | extent_op, NULL); | |
5d4f98a2 | 1842 | } else if (ret == -ENOENT) { |
a639cdeb | 1843 | setup_inline_extent_backref(trans->fs_info, path, iref, parent, |
143bede5 JM |
1844 | root_objectid, owner, offset, |
1845 | refs_to_add, extent_op); | |
1846 | ret = 0; | |
771ed689 | 1847 | } |
5d4f98a2 YZ |
1848 | return ret; |
1849 | } | |
31840ae1 | 1850 | |
5d4f98a2 | 1851 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1852 | struct btrfs_path *path, |
1853 | u64 bytenr, u64 parent, u64 root_objectid, | |
1854 | u64 owner, u64 offset, int refs_to_add) | |
1855 | { | |
1856 | int ret; | |
1857 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { | |
1858 | BUG_ON(refs_to_add != 1); | |
10728404 NB |
1859 | ret = insert_tree_block_ref(trans, path, bytenr, parent, |
1860 | root_objectid); | |
5d4f98a2 | 1861 | } else { |
62b895af NB |
1862 | ret = insert_extent_data_ref(trans, path, bytenr, parent, |
1863 | root_objectid, owner, offset, | |
1864 | refs_to_add); | |
5d4f98a2 YZ |
1865 | } |
1866 | return ret; | |
1867 | } | |
56bec294 | 1868 | |
5d4f98a2 | 1869 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
1870 | struct btrfs_path *path, |
1871 | struct btrfs_extent_inline_ref *iref, | |
fcebe456 | 1872 | int refs_to_drop, int is_data, int *last_ref) |
5d4f98a2 | 1873 | { |
143bede5 | 1874 | int ret = 0; |
b9473439 | 1875 | |
5d4f98a2 YZ |
1876 | BUG_ON(!is_data && refs_to_drop != 1); |
1877 | if (iref) { | |
61a18f1c NB |
1878 | update_inline_extent_backref(path, iref, -refs_to_drop, NULL, |
1879 | last_ref); | |
5d4f98a2 | 1880 | } else if (is_data) { |
e9f6290d | 1881 | ret = remove_extent_data_ref(trans, path, refs_to_drop, |
fcebe456 | 1882 | last_ref); |
5d4f98a2 | 1883 | } else { |
fcebe456 | 1884 | *last_ref = 1; |
87cc7a8a | 1885 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
5d4f98a2 YZ |
1886 | } |
1887 | return ret; | |
1888 | } | |
1889 | ||
86557861 | 1890 | #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len)) |
d04c6b88 JM |
1891 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
1892 | u64 *discarded_bytes) | |
5d4f98a2 | 1893 | { |
86557861 JM |
1894 | int j, ret = 0; |
1895 | u64 bytes_left, end; | |
4d89d377 | 1896 | u64 aligned_start = ALIGN(start, 1 << 9); |
d04c6b88 | 1897 | |
4d89d377 JM |
1898 | if (WARN_ON(start != aligned_start)) { |
1899 | len -= aligned_start - start; | |
1900 | len = round_down(len, 1 << 9); | |
1901 | start = aligned_start; | |
1902 | } | |
d04c6b88 | 1903 | |
4d89d377 | 1904 | *discarded_bytes = 0; |
86557861 JM |
1905 | |
1906 | if (!len) | |
1907 | return 0; | |
1908 | ||
1909 | end = start + len; | |
1910 | bytes_left = len; | |
1911 | ||
1912 | /* Skip any superblocks on this device. */ | |
1913 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { | |
1914 | u64 sb_start = btrfs_sb_offset(j); | |
1915 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; | |
1916 | u64 size = sb_start - start; | |
1917 | ||
1918 | if (!in_range(sb_start, start, bytes_left) && | |
1919 | !in_range(sb_end, start, bytes_left) && | |
1920 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) | |
1921 | continue; | |
1922 | ||
1923 | /* | |
1924 | * Superblock spans beginning of range. Adjust start and | |
1925 | * try again. | |
1926 | */ | |
1927 | if (sb_start <= start) { | |
1928 | start += sb_end - start; | |
1929 | if (start > end) { | |
1930 | bytes_left = 0; | |
1931 | break; | |
1932 | } | |
1933 | bytes_left = end - start; | |
1934 | continue; | |
1935 | } | |
1936 | ||
1937 | if (size) { | |
1938 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, | |
1939 | GFP_NOFS, 0); | |
1940 | if (!ret) | |
1941 | *discarded_bytes += size; | |
1942 | else if (ret != -EOPNOTSUPP) | |
1943 | return ret; | |
1944 | } | |
1945 | ||
1946 | start = sb_end; | |
1947 | if (start > end) { | |
1948 | bytes_left = 0; | |
1949 | break; | |
1950 | } | |
1951 | bytes_left = end - start; | |
1952 | } | |
1953 | ||
1954 | if (bytes_left) { | |
1955 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, | |
4d89d377 JM |
1956 | GFP_NOFS, 0); |
1957 | if (!ret) | |
86557861 | 1958 | *discarded_bytes += bytes_left; |
4d89d377 | 1959 | } |
d04c6b88 | 1960 | return ret; |
5d4f98a2 | 1961 | } |
5d4f98a2 | 1962 | |
2ff7e61e | 1963 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
1edb647b | 1964 | u64 num_bytes, u64 *actual_bytes) |
5d4f98a2 | 1965 | { |
5d4f98a2 | 1966 | int ret; |
5378e607 | 1967 | u64 discarded_bytes = 0; |
a1d3c478 | 1968 | struct btrfs_bio *bbio = NULL; |
5d4f98a2 | 1969 | |
e244a0ae | 1970 | |
2999241d FM |
1971 | /* |
1972 | * Avoid races with device replace and make sure our bbio has devices | |
1973 | * associated to its stripes that don't go away while we are discarding. | |
1974 | */ | |
0b246afa | 1975 | btrfs_bio_counter_inc_blocked(fs_info); |
5d4f98a2 | 1976 | /* Tell the block device(s) that the sectors can be discarded */ |
0b246afa JM |
1977 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes, |
1978 | &bbio, 0); | |
79787eaa | 1979 | /* Error condition is -ENOMEM */ |
5d4f98a2 | 1980 | if (!ret) { |
a1d3c478 | 1981 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
5d4f98a2 YZ |
1982 | int i; |
1983 | ||
5d4f98a2 | 1984 | |
a1d3c478 | 1985 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
d04c6b88 | 1986 | u64 bytes; |
38b5f68e AJ |
1987 | struct request_queue *req_q; |
1988 | ||
627e0873 FM |
1989 | if (!stripe->dev->bdev) { |
1990 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); | |
1991 | continue; | |
1992 | } | |
38b5f68e AJ |
1993 | req_q = bdev_get_queue(stripe->dev->bdev); |
1994 | if (!blk_queue_discard(req_q)) | |
d5e2003c JB |
1995 | continue; |
1996 | ||
5378e607 LD |
1997 | ret = btrfs_issue_discard(stripe->dev->bdev, |
1998 | stripe->physical, | |
d04c6b88 JM |
1999 | stripe->length, |
2000 | &bytes); | |
5378e607 | 2001 | if (!ret) |
d04c6b88 | 2002 | discarded_bytes += bytes; |
5378e607 | 2003 | else if (ret != -EOPNOTSUPP) |
79787eaa | 2004 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
d5e2003c JB |
2005 | |
2006 | /* | |
2007 | * Just in case we get back EOPNOTSUPP for some reason, | |
2008 | * just ignore the return value so we don't screw up | |
2009 | * people calling discard_extent. | |
2010 | */ | |
2011 | ret = 0; | |
5d4f98a2 | 2012 | } |
6e9606d2 | 2013 | btrfs_put_bbio(bbio); |
5d4f98a2 | 2014 | } |
0b246afa | 2015 | btrfs_bio_counter_dec(fs_info); |
5378e607 LD |
2016 | |
2017 | if (actual_bytes) | |
2018 | *actual_bytes = discarded_bytes; | |
2019 | ||
5d4f98a2 | 2020 | |
53b381b3 DW |
2021 | if (ret == -EOPNOTSUPP) |
2022 | ret = 0; | |
5d4f98a2 | 2023 | return ret; |
5d4f98a2 YZ |
2024 | } |
2025 | ||
79787eaa | 2026 | /* Can return -ENOMEM */ |
5d4f98a2 | 2027 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
84f7d8e6 | 2028 | struct btrfs_root *root, |
5d4f98a2 | 2029 | u64 bytenr, u64 num_bytes, u64 parent, |
b06c4bf5 | 2030 | u64 root_objectid, u64 owner, u64 offset) |
5d4f98a2 | 2031 | { |
84f7d8e6 | 2032 | struct btrfs_fs_info *fs_info = root->fs_info; |
d7eae340 | 2033 | int old_ref_mod, new_ref_mod; |
5d4f98a2 | 2034 | int ret; |
66d7e7f0 | 2035 | |
5d4f98a2 YZ |
2036 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
2037 | root_objectid == BTRFS_TREE_LOG_OBJECTID); | |
2038 | ||
fd708b81 JB |
2039 | btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid, |
2040 | owner, offset, BTRFS_ADD_DELAYED_REF); | |
2041 | ||
5d4f98a2 | 2042 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
44e1c47d | 2043 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, |
7be07912 OS |
2044 | num_bytes, parent, |
2045 | root_objectid, (int)owner, | |
2046 | BTRFS_ADD_DELAYED_REF, NULL, | |
d7eae340 | 2047 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 2048 | } else { |
88a979c6 | 2049 | ret = btrfs_add_delayed_data_ref(trans, bytenr, |
7be07912 OS |
2050 | num_bytes, parent, |
2051 | root_objectid, owner, offset, | |
d7eae340 OS |
2052 | 0, BTRFS_ADD_DELAYED_REF, |
2053 | &old_ref_mod, &new_ref_mod); | |
5d4f98a2 | 2054 | } |
d7eae340 | 2055 | |
29d2b84c NB |
2056 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) { |
2057 | bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID; | |
2058 | ||
2059 | add_pinned_bytes(fs_info, -num_bytes, metadata, root_objectid); | |
2060 | } | |
d7eae340 | 2061 | |
5d4f98a2 YZ |
2062 | return ret; |
2063 | } | |
2064 | ||
bd3c685e NB |
2065 | /* |
2066 | * __btrfs_inc_extent_ref - insert backreference for a given extent | |
2067 | * | |
2068 | * @trans: Handle of transaction | |
2069 | * | |
2070 | * @node: The delayed ref node used to get the bytenr/length for | |
2071 | * extent whose references are incremented. | |
2072 | * | |
2073 | * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/ | |
2074 | * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical | |
2075 | * bytenr of the parent block. Since new extents are always | |
2076 | * created with indirect references, this will only be the case | |
2077 | * when relocating a shared extent. In that case, root_objectid | |
2078 | * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must | |
2079 | * be 0 | |
2080 | * | |
2081 | * @root_objectid: The id of the root where this modification has originated, | |
2082 | * this can be either one of the well-known metadata trees or | |
2083 | * the subvolume id which references this extent. | |
2084 | * | |
2085 | * @owner: For data extents it is the inode number of the owning file. | |
2086 | * For metadata extents this parameter holds the level in the | |
2087 | * tree of the extent. | |
2088 | * | |
2089 | * @offset: For metadata extents the offset is ignored and is currently | |
2090 | * always passed as 0. For data extents it is the fileoffset | |
2091 | * this extent belongs to. | |
2092 | * | |
2093 | * @refs_to_add Number of references to add | |
2094 | * | |
2095 | * @extent_op Pointer to a structure, holding information necessary when | |
2096 | * updating a tree block's flags | |
2097 | * | |
2098 | */ | |
5d4f98a2 | 2099 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
c682f9b3 | 2100 | struct btrfs_delayed_ref_node *node, |
5d4f98a2 YZ |
2101 | u64 parent, u64 root_objectid, |
2102 | u64 owner, u64 offset, int refs_to_add, | |
2103 | struct btrfs_delayed_extent_op *extent_op) | |
2104 | { | |
2105 | struct btrfs_path *path; | |
2106 | struct extent_buffer *leaf; | |
2107 | struct btrfs_extent_item *item; | |
fcebe456 | 2108 | struct btrfs_key key; |
c682f9b3 QW |
2109 | u64 bytenr = node->bytenr; |
2110 | u64 num_bytes = node->num_bytes; | |
5d4f98a2 YZ |
2111 | u64 refs; |
2112 | int ret; | |
5d4f98a2 YZ |
2113 | |
2114 | path = btrfs_alloc_path(); | |
2115 | if (!path) | |
2116 | return -ENOMEM; | |
2117 | ||
e4058b54 | 2118 | path->reada = READA_FORWARD; |
5d4f98a2 YZ |
2119 | path->leave_spinning = 1; |
2120 | /* this will setup the path even if it fails to insert the back ref */ | |
a639cdeb NB |
2121 | ret = insert_inline_extent_backref(trans, path, bytenr, num_bytes, |
2122 | parent, root_objectid, owner, | |
2123 | offset, refs_to_add, extent_op); | |
0ed4792a | 2124 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
5d4f98a2 | 2125 | goto out; |
fcebe456 JB |
2126 | |
2127 | /* | |
2128 | * Ok we had -EAGAIN which means we didn't have space to insert and | |
2129 | * inline extent ref, so just update the reference count and add a | |
2130 | * normal backref. | |
2131 | */ | |
5d4f98a2 | 2132 | leaf = path->nodes[0]; |
fcebe456 | 2133 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
5d4f98a2 YZ |
2134 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2135 | refs = btrfs_extent_refs(leaf, item); | |
2136 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); | |
2137 | if (extent_op) | |
2138 | __run_delayed_extent_op(extent_op, leaf, item); | |
56bec294 | 2139 | |
5d4f98a2 | 2140 | btrfs_mark_buffer_dirty(leaf); |
b3b4aa74 | 2141 | btrfs_release_path(path); |
56bec294 | 2142 | |
e4058b54 | 2143 | path->reada = READA_FORWARD; |
b9473439 | 2144 | path->leave_spinning = 1; |
56bec294 | 2145 | /* now insert the actual backref */ |
37593410 NB |
2146 | ret = insert_extent_backref(trans, path, bytenr, parent, root_objectid, |
2147 | owner, offset, refs_to_add); | |
79787eaa | 2148 | if (ret) |
66642832 | 2149 | btrfs_abort_transaction(trans, ret); |
5d4f98a2 | 2150 | out: |
56bec294 | 2151 | btrfs_free_path(path); |
30d133fc | 2152 | return ret; |
56bec294 CM |
2153 | } |
2154 | ||
5d4f98a2 | 2155 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2156 | struct btrfs_delayed_ref_node *node, |
2157 | struct btrfs_delayed_extent_op *extent_op, | |
2158 | int insert_reserved) | |
56bec294 | 2159 | { |
5d4f98a2 YZ |
2160 | int ret = 0; |
2161 | struct btrfs_delayed_data_ref *ref; | |
2162 | struct btrfs_key ins; | |
2163 | u64 parent = 0; | |
2164 | u64 ref_root = 0; | |
2165 | u64 flags = 0; | |
2166 | ||
2167 | ins.objectid = node->bytenr; | |
2168 | ins.offset = node->num_bytes; | |
2169 | ins.type = BTRFS_EXTENT_ITEM_KEY; | |
2170 | ||
2171 | ref = btrfs_delayed_node_to_data_ref(node); | |
2bf98ef3 | 2172 | trace_run_delayed_data_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 2173 | |
5d4f98a2 YZ |
2174 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
2175 | parent = ref->parent; | |
fcebe456 | 2176 | ref_root = ref->root; |
5d4f98a2 YZ |
2177 | |
2178 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { | |
3173a18f | 2179 | if (extent_op) |
5d4f98a2 | 2180 | flags |= extent_op->flags_to_set; |
ef89b824 NB |
2181 | ret = alloc_reserved_file_extent(trans, parent, ref_root, |
2182 | flags, ref->objectid, | |
2183 | ref->offset, &ins, | |
2184 | node->ref_mod); | |
5d4f98a2 | 2185 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
2186 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
2187 | ref->objectid, ref->offset, | |
2188 | node->ref_mod, extent_op); | |
5d4f98a2 | 2189 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 2190 | ret = __btrfs_free_extent(trans, node, parent, |
5d4f98a2 YZ |
2191 | ref_root, ref->objectid, |
2192 | ref->offset, node->ref_mod, | |
c682f9b3 | 2193 | extent_op); |
5d4f98a2 YZ |
2194 | } else { |
2195 | BUG(); | |
2196 | } | |
2197 | return ret; | |
2198 | } | |
2199 | ||
2200 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, | |
2201 | struct extent_buffer *leaf, | |
2202 | struct btrfs_extent_item *ei) | |
2203 | { | |
2204 | u64 flags = btrfs_extent_flags(leaf, ei); | |
2205 | if (extent_op->update_flags) { | |
2206 | flags |= extent_op->flags_to_set; | |
2207 | btrfs_set_extent_flags(leaf, ei, flags); | |
2208 | } | |
2209 | ||
2210 | if (extent_op->update_key) { | |
2211 | struct btrfs_tree_block_info *bi; | |
2212 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); | |
2213 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
2214 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); | |
2215 | } | |
2216 | } | |
2217 | ||
2218 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, | |
d278850e | 2219 | struct btrfs_delayed_ref_head *head, |
5d4f98a2 YZ |
2220 | struct btrfs_delayed_extent_op *extent_op) |
2221 | { | |
20b9a2d6 | 2222 | struct btrfs_fs_info *fs_info = trans->fs_info; |
5d4f98a2 YZ |
2223 | struct btrfs_key key; |
2224 | struct btrfs_path *path; | |
2225 | struct btrfs_extent_item *ei; | |
2226 | struct extent_buffer *leaf; | |
2227 | u32 item_size; | |
56bec294 | 2228 | int ret; |
5d4f98a2 | 2229 | int err = 0; |
b1c79e09 | 2230 | int metadata = !extent_op->is_data; |
5d4f98a2 | 2231 | |
79787eaa JM |
2232 | if (trans->aborted) |
2233 | return 0; | |
2234 | ||
0b246afa | 2235 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
3173a18f JB |
2236 | metadata = 0; |
2237 | ||
5d4f98a2 YZ |
2238 | path = btrfs_alloc_path(); |
2239 | if (!path) | |
2240 | return -ENOMEM; | |
2241 | ||
d278850e | 2242 | key.objectid = head->bytenr; |
5d4f98a2 | 2243 | |
3173a18f | 2244 | if (metadata) { |
3173a18f | 2245 | key.type = BTRFS_METADATA_ITEM_KEY; |
b1c79e09 | 2246 | key.offset = extent_op->level; |
3173a18f JB |
2247 | } else { |
2248 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
d278850e | 2249 | key.offset = head->num_bytes; |
3173a18f JB |
2250 | } |
2251 | ||
2252 | again: | |
e4058b54 | 2253 | path->reada = READA_FORWARD; |
5d4f98a2 | 2254 | path->leave_spinning = 1; |
0b246afa | 2255 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
5d4f98a2 YZ |
2256 | if (ret < 0) { |
2257 | err = ret; | |
2258 | goto out; | |
2259 | } | |
2260 | if (ret > 0) { | |
3173a18f | 2261 | if (metadata) { |
55994887 FDBM |
2262 | if (path->slots[0] > 0) { |
2263 | path->slots[0]--; | |
2264 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
2265 | path->slots[0]); | |
d278850e | 2266 | if (key.objectid == head->bytenr && |
55994887 | 2267 | key.type == BTRFS_EXTENT_ITEM_KEY && |
d278850e | 2268 | key.offset == head->num_bytes) |
55994887 FDBM |
2269 | ret = 0; |
2270 | } | |
2271 | if (ret > 0) { | |
2272 | btrfs_release_path(path); | |
2273 | metadata = 0; | |
3173a18f | 2274 | |
d278850e JB |
2275 | key.objectid = head->bytenr; |
2276 | key.offset = head->num_bytes; | |
55994887 FDBM |
2277 | key.type = BTRFS_EXTENT_ITEM_KEY; |
2278 | goto again; | |
2279 | } | |
2280 | } else { | |
2281 | err = -EIO; | |
2282 | goto out; | |
3173a18f | 2283 | } |
5d4f98a2 YZ |
2284 | } |
2285 | ||
2286 | leaf = path->nodes[0]; | |
2287 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
ba3c2b19 | 2288 | |
6d8ff4e4 | 2289 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
2290 | err = -EINVAL; |
2291 | btrfs_print_v0_err(fs_info); | |
2292 | btrfs_abort_transaction(trans, err); | |
2293 | goto out; | |
2294 | } | |
2295 | ||
5d4f98a2 YZ |
2296 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
2297 | __run_delayed_extent_op(extent_op, leaf, ei); | |
56bec294 | 2298 | |
5d4f98a2 YZ |
2299 | btrfs_mark_buffer_dirty(leaf); |
2300 | out: | |
2301 | btrfs_free_path(path); | |
2302 | return err; | |
56bec294 CM |
2303 | } |
2304 | ||
5d4f98a2 | 2305 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2306 | struct btrfs_delayed_ref_node *node, |
2307 | struct btrfs_delayed_extent_op *extent_op, | |
2308 | int insert_reserved) | |
56bec294 CM |
2309 | { |
2310 | int ret = 0; | |
5d4f98a2 | 2311 | struct btrfs_delayed_tree_ref *ref; |
5d4f98a2 YZ |
2312 | u64 parent = 0; |
2313 | u64 ref_root = 0; | |
56bec294 | 2314 | |
5d4f98a2 | 2315 | ref = btrfs_delayed_node_to_tree_ref(node); |
f97806f2 | 2316 | trace_run_delayed_tree_ref(trans->fs_info, node, ref, node->action); |
599c75ec | 2317 | |
5d4f98a2 YZ |
2318 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
2319 | parent = ref->parent; | |
fcebe456 | 2320 | ref_root = ref->root; |
5d4f98a2 | 2321 | |
02794222 | 2322 | if (node->ref_mod != 1) { |
f97806f2 | 2323 | btrfs_err(trans->fs_info, |
02794222 LB |
2324 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
2325 | node->bytenr, node->ref_mod, node->action, ref_root, | |
2326 | parent); | |
2327 | return -EIO; | |
2328 | } | |
5d4f98a2 | 2329 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
3173a18f | 2330 | BUG_ON(!extent_op || !extent_op->update_flags); |
21ebfbe7 | 2331 | ret = alloc_reserved_tree_block(trans, node, extent_op); |
5d4f98a2 | 2332 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
2590d0f1 NB |
2333 | ret = __btrfs_inc_extent_ref(trans, node, parent, ref_root, |
2334 | ref->level, 0, 1, extent_op); | |
5d4f98a2 | 2335 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
e72cb923 | 2336 | ret = __btrfs_free_extent(trans, node, parent, ref_root, |
c682f9b3 | 2337 | ref->level, 0, 1, extent_op); |
5d4f98a2 YZ |
2338 | } else { |
2339 | BUG(); | |
2340 | } | |
56bec294 CM |
2341 | return ret; |
2342 | } | |
2343 | ||
2344 | /* helper function to actually process a single delayed ref entry */ | |
5d4f98a2 | 2345 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
2346 | struct btrfs_delayed_ref_node *node, |
2347 | struct btrfs_delayed_extent_op *extent_op, | |
2348 | int insert_reserved) | |
56bec294 | 2349 | { |
79787eaa JM |
2350 | int ret = 0; |
2351 | ||
857cc2fc JB |
2352 | if (trans->aborted) { |
2353 | if (insert_reserved) | |
5fac7f9e | 2354 | btrfs_pin_extent(trans->fs_info, node->bytenr, |
857cc2fc | 2355 | node->num_bytes, 1); |
79787eaa | 2356 | return 0; |
857cc2fc | 2357 | } |
79787eaa | 2358 | |
5d4f98a2 YZ |
2359 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
2360 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) | |
f97806f2 | 2361 | ret = run_delayed_tree_ref(trans, node, extent_op, |
5d4f98a2 YZ |
2362 | insert_reserved); |
2363 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || | |
2364 | node->type == BTRFS_SHARED_DATA_REF_KEY) | |
2bf98ef3 | 2365 | ret = run_delayed_data_ref(trans, node, extent_op, |
5d4f98a2 YZ |
2366 | insert_reserved); |
2367 | else | |
2368 | BUG(); | |
2369 | return ret; | |
56bec294 CM |
2370 | } |
2371 | ||
c6fc2454 | 2372 | static inline struct btrfs_delayed_ref_node * |
56bec294 CM |
2373 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
2374 | { | |
cffc3374 FM |
2375 | struct btrfs_delayed_ref_node *ref; |
2376 | ||
0e0adbcf | 2377 | if (RB_EMPTY_ROOT(&head->ref_tree)) |
c6fc2454 | 2378 | return NULL; |
d7df2c79 | 2379 | |
cffc3374 FM |
2380 | /* |
2381 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. | |
2382 | * This is to prevent a ref count from going down to zero, which deletes | |
2383 | * the extent item from the extent tree, when there still are references | |
2384 | * to add, which would fail because they would not find the extent item. | |
2385 | */ | |
1d57ee94 WX |
2386 | if (!list_empty(&head->ref_add_list)) |
2387 | return list_first_entry(&head->ref_add_list, | |
2388 | struct btrfs_delayed_ref_node, add_list); | |
2389 | ||
0e0adbcf JB |
2390 | ref = rb_entry(rb_first(&head->ref_tree), |
2391 | struct btrfs_delayed_ref_node, ref_node); | |
1d57ee94 WX |
2392 | ASSERT(list_empty(&ref->add_list)); |
2393 | return ref; | |
56bec294 CM |
2394 | } |
2395 | ||
2eadaa22 JB |
2396 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
2397 | struct btrfs_delayed_ref_head *head) | |
2398 | { | |
2399 | spin_lock(&delayed_refs->lock); | |
2400 | head->processing = 0; | |
2401 | delayed_refs->num_heads_ready++; | |
2402 | spin_unlock(&delayed_refs->lock); | |
2403 | btrfs_delayed_ref_unlock(head); | |
2404 | } | |
2405 | ||
b00e6250 | 2406 | static int cleanup_extent_op(struct btrfs_trans_handle *trans, |
b00e6250 JB |
2407 | struct btrfs_delayed_ref_head *head) |
2408 | { | |
2409 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; | |
2410 | int ret; | |
2411 | ||
2412 | if (!extent_op) | |
2413 | return 0; | |
2414 | head->extent_op = NULL; | |
2415 | if (head->must_insert_reserved) { | |
2416 | btrfs_free_delayed_extent_op(extent_op); | |
2417 | return 0; | |
2418 | } | |
2419 | spin_unlock(&head->lock); | |
20b9a2d6 | 2420 | ret = run_delayed_extent_op(trans, head, extent_op); |
b00e6250 JB |
2421 | btrfs_free_delayed_extent_op(extent_op); |
2422 | return ret ? ret : 1; | |
2423 | } | |
2424 | ||
194ab0bc | 2425 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
194ab0bc JB |
2426 | struct btrfs_delayed_ref_head *head) |
2427 | { | |
f9871edd NB |
2428 | |
2429 | struct btrfs_fs_info *fs_info = trans->fs_info; | |
194ab0bc JB |
2430 | struct btrfs_delayed_ref_root *delayed_refs; |
2431 | int ret; | |
2432 | ||
2433 | delayed_refs = &trans->transaction->delayed_refs; | |
2434 | ||
c4d56d4a | 2435 | ret = cleanup_extent_op(trans, head); |
194ab0bc JB |
2436 | if (ret < 0) { |
2437 | unselect_delayed_ref_head(delayed_refs, head); | |
2438 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); | |
2439 | return ret; | |
2440 | } else if (ret) { | |
2441 | return ret; | |
2442 | } | |
2443 | ||
2444 | /* | |
2445 | * Need to drop our head ref lock and re-acquire the delayed ref lock | |
2446 | * and then re-check to make sure nobody got added. | |
2447 | */ | |
2448 | spin_unlock(&head->lock); | |
2449 | spin_lock(&delayed_refs->lock); | |
2450 | spin_lock(&head->lock); | |
0e0adbcf | 2451 | if (!RB_EMPTY_ROOT(&head->ref_tree) || head->extent_op) { |
194ab0bc JB |
2452 | spin_unlock(&head->lock); |
2453 | spin_unlock(&delayed_refs->lock); | |
2454 | return 1; | |
2455 | } | |
194ab0bc JB |
2456 | delayed_refs->num_heads--; |
2457 | rb_erase(&head->href_node, &delayed_refs->href_root); | |
d278850e | 2458 | RB_CLEAR_NODE(&head->href_node); |
c1103f7a | 2459 | spin_unlock(&head->lock); |
1e7a1421 | 2460 | spin_unlock(&delayed_refs->lock); |
c1103f7a JB |
2461 | atomic_dec(&delayed_refs->num_entries); |
2462 | ||
d278850e | 2463 | trace_run_delayed_ref_head(fs_info, head, 0); |
c1103f7a JB |
2464 | |
2465 | if (head->total_ref_mod < 0) { | |
5e388e95 NB |
2466 | struct btrfs_space_info *space_info; |
2467 | u64 flags; | |
c1103f7a | 2468 | |
5e388e95 NB |
2469 | if (head->is_data) |
2470 | flags = BTRFS_BLOCK_GROUP_DATA; | |
2471 | else if (head->is_system) | |
2472 | flags = BTRFS_BLOCK_GROUP_SYSTEM; | |
2473 | else | |
2474 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
2475 | space_info = __find_space_info(fs_info, flags); | |
2476 | ASSERT(space_info); | |
dec59fa3 EL |
2477 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
2478 | -head->num_bytes, | |
2479 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
c1103f7a JB |
2480 | |
2481 | if (head->is_data) { | |
2482 | spin_lock(&delayed_refs->lock); | |
d278850e | 2483 | delayed_refs->pending_csums -= head->num_bytes; |
c1103f7a JB |
2484 | spin_unlock(&delayed_refs->lock); |
2485 | } | |
2486 | } | |
2487 | ||
2488 | if (head->must_insert_reserved) { | |
d278850e JB |
2489 | btrfs_pin_extent(fs_info, head->bytenr, |
2490 | head->num_bytes, 1); | |
c1103f7a | 2491 | if (head->is_data) { |
d278850e JB |
2492 | ret = btrfs_del_csums(trans, fs_info, head->bytenr, |
2493 | head->num_bytes); | |
c1103f7a JB |
2494 | } |
2495 | } | |
2496 | ||
2497 | /* Also free its reserved qgroup space */ | |
2498 | btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root, | |
2499 | head->qgroup_reserved); | |
2500 | btrfs_delayed_ref_unlock(head); | |
d278850e | 2501 | btrfs_put_delayed_ref_head(head); |
194ab0bc JB |
2502 | return 0; |
2503 | } | |
2504 | ||
79787eaa JM |
2505 | /* |
2506 | * Returns 0 on success or if called with an already aborted transaction. | |
2507 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. | |
2508 | */ | |
d7df2c79 | 2509 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
d7df2c79 | 2510 | unsigned long nr) |
56bec294 | 2511 | { |
0a1e458a | 2512 | struct btrfs_fs_info *fs_info = trans->fs_info; |
56bec294 CM |
2513 | struct btrfs_delayed_ref_root *delayed_refs; |
2514 | struct btrfs_delayed_ref_node *ref; | |
2515 | struct btrfs_delayed_ref_head *locked_ref = NULL; | |
5d4f98a2 | 2516 | struct btrfs_delayed_extent_op *extent_op; |
0a2b2a84 | 2517 | ktime_t start = ktime_get(); |
56bec294 | 2518 | int ret; |
d7df2c79 | 2519 | unsigned long count = 0; |
0a2b2a84 | 2520 | unsigned long actual_count = 0; |
56bec294 | 2521 | int must_insert_reserved = 0; |
56bec294 CM |
2522 | |
2523 | delayed_refs = &trans->transaction->delayed_refs; | |
56bec294 CM |
2524 | while (1) { |
2525 | if (!locked_ref) { | |
d7df2c79 | 2526 | if (count >= nr) |
56bec294 | 2527 | break; |
56bec294 | 2528 | |
d7df2c79 JB |
2529 | spin_lock(&delayed_refs->lock); |
2530 | locked_ref = btrfs_select_ref_head(trans); | |
2531 | if (!locked_ref) { | |
2532 | spin_unlock(&delayed_refs->lock); | |
2533 | break; | |
2534 | } | |
c3e69d58 CM |
2535 | |
2536 | /* grab the lock that says we are going to process | |
2537 | * all the refs for this head */ | |
2538 | ret = btrfs_delayed_ref_lock(trans, locked_ref); | |
d7df2c79 | 2539 | spin_unlock(&delayed_refs->lock); |
c3e69d58 CM |
2540 | /* |
2541 | * we may have dropped the spin lock to get the head | |
2542 | * mutex lock, and that might have given someone else | |
2543 | * time to free the head. If that's true, it has been | |
2544 | * removed from our list and we can move on. | |
2545 | */ | |
2546 | if (ret == -EAGAIN) { | |
2547 | locked_ref = NULL; | |
2548 | count++; | |
2549 | continue; | |
56bec294 CM |
2550 | } |
2551 | } | |
a28ec197 | 2552 | |
2c3cf7d5 FM |
2553 | /* |
2554 | * We need to try and merge add/drops of the same ref since we | |
2555 | * can run into issues with relocate dropping the implicit ref | |
2556 | * and then it being added back again before the drop can | |
2557 | * finish. If we merged anything we need to re-loop so we can | |
2558 | * get a good ref. | |
2559 | * Or we can get node references of the same type that weren't | |
2560 | * merged when created due to bumps in the tree mod seq, and | |
2561 | * we need to merge them to prevent adding an inline extent | |
2562 | * backref before dropping it (triggering a BUG_ON at | |
2563 | * insert_inline_extent_backref()). | |
2564 | */ | |
d7df2c79 | 2565 | spin_lock(&locked_ref->lock); |
be97f133 | 2566 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); |
ae1e206b | 2567 | |
d1270cd9 AJ |
2568 | ref = select_delayed_ref(locked_ref); |
2569 | ||
2570 | if (ref && ref->seq && | |
41d0bd3b | 2571 | btrfs_check_delayed_seq(fs_info, ref->seq)) { |
d7df2c79 | 2572 | spin_unlock(&locked_ref->lock); |
2eadaa22 | 2573 | unselect_delayed_ref_head(delayed_refs, locked_ref); |
d7df2c79 | 2574 | locked_ref = NULL; |
d1270cd9 | 2575 | cond_resched(); |
27a377db | 2576 | count++; |
d1270cd9 AJ |
2577 | continue; |
2578 | } | |
2579 | ||
c1103f7a JB |
2580 | /* |
2581 | * We're done processing refs in this ref_head, clean everything | |
2582 | * up and move on to the next ref_head. | |
2583 | */ | |
56bec294 | 2584 | if (!ref) { |
f9871edd | 2585 | ret = cleanup_ref_head(trans, locked_ref); |
194ab0bc | 2586 | if (ret > 0 ) { |
b00e6250 JB |
2587 | /* We dropped our lock, we need to loop. */ |
2588 | ret = 0; | |
d7df2c79 | 2589 | continue; |
194ab0bc JB |
2590 | } else if (ret) { |
2591 | return ret; | |
5d4f98a2 | 2592 | } |
c1103f7a JB |
2593 | locked_ref = NULL; |
2594 | count++; | |
2595 | continue; | |
2596 | } | |
02217ed2 | 2597 | |
c1103f7a JB |
2598 | actual_count++; |
2599 | ref->in_tree = 0; | |
0e0adbcf JB |
2600 | rb_erase(&ref->ref_node, &locked_ref->ref_tree); |
2601 | RB_CLEAR_NODE(&ref->ref_node); | |
c1103f7a JB |
2602 | if (!list_empty(&ref->add_list)) |
2603 | list_del(&ref->add_list); | |
2604 | /* | |
2605 | * When we play the delayed ref, also correct the ref_mod on | |
2606 | * head | |
2607 | */ | |
2608 | switch (ref->action) { | |
2609 | case BTRFS_ADD_DELAYED_REF: | |
2610 | case BTRFS_ADD_DELAYED_EXTENT: | |
d278850e | 2611 | locked_ref->ref_mod -= ref->ref_mod; |
c1103f7a JB |
2612 | break; |
2613 | case BTRFS_DROP_DELAYED_REF: | |
d278850e | 2614 | locked_ref->ref_mod += ref->ref_mod; |
c1103f7a JB |
2615 | break; |
2616 | default: | |
2617 | WARN_ON(1); | |
22cd2e7d | 2618 | } |
1ce7a5ec JB |
2619 | atomic_dec(&delayed_refs->num_entries); |
2620 | ||
b00e6250 JB |
2621 | /* |
2622 | * Record the must-insert_reserved flag before we drop the spin | |
2623 | * lock. | |
2624 | */ | |
2625 | must_insert_reserved = locked_ref->must_insert_reserved; | |
2626 | locked_ref->must_insert_reserved = 0; | |
2627 | ||
2628 | extent_op = locked_ref->extent_op; | |
2629 | locked_ref->extent_op = NULL; | |
d7df2c79 | 2630 | spin_unlock(&locked_ref->lock); |
925baedd | 2631 | |
5fac7f9e | 2632 | ret = run_one_delayed_ref(trans, ref, extent_op, |
56bec294 | 2633 | must_insert_reserved); |
eb099670 | 2634 | |
78a6184a | 2635 | btrfs_free_delayed_extent_op(extent_op); |
79787eaa | 2636 | if (ret) { |
2eadaa22 | 2637 | unselect_delayed_ref_head(delayed_refs, locked_ref); |
093486c4 | 2638 | btrfs_put_delayed_ref(ref); |
5d163e0e JM |
2639 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", |
2640 | ret); | |
79787eaa JM |
2641 | return ret; |
2642 | } | |
2643 | ||
093486c4 MX |
2644 | btrfs_put_delayed_ref(ref); |
2645 | count++; | |
c3e69d58 | 2646 | cond_resched(); |
c3e69d58 | 2647 | } |
0a2b2a84 JB |
2648 | |
2649 | /* | |
2650 | * We don't want to include ref heads since we can have empty ref heads | |
2651 | * and those will drastically skew our runtime down since we just do | |
2652 | * accounting, no actual extent tree updates. | |
2653 | */ | |
2654 | if (actual_count > 0) { | |
2655 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); | |
2656 | u64 avg; | |
2657 | ||
2658 | /* | |
2659 | * We weigh the current average higher than our current runtime | |
2660 | * to avoid large swings in the average. | |
2661 | */ | |
2662 | spin_lock(&delayed_refs->lock); | |
2663 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; | |
f8c269d7 | 2664 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
0a2b2a84 JB |
2665 | spin_unlock(&delayed_refs->lock); |
2666 | } | |
d7df2c79 | 2667 | return 0; |
c3e69d58 CM |
2668 | } |
2669 | ||
709c0486 AJ |
2670 | #ifdef SCRAMBLE_DELAYED_REFS |
2671 | /* | |
2672 | * Normally delayed refs get processed in ascending bytenr order. This | |
2673 | * correlates in most cases to the order added. To expose dependencies on this | |
2674 | * order, we start to process the tree in the middle instead of the beginning | |
2675 | */ | |
2676 | static u64 find_middle(struct rb_root *root) | |
2677 | { | |
2678 | struct rb_node *n = root->rb_node; | |
2679 | struct btrfs_delayed_ref_node *entry; | |
2680 | int alt = 1; | |
2681 | u64 middle; | |
2682 | u64 first = 0, last = 0; | |
2683 | ||
2684 | n = rb_first(root); | |
2685 | if (n) { | |
2686 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2687 | first = entry->bytenr; | |
2688 | } | |
2689 | n = rb_last(root); | |
2690 | if (n) { | |
2691 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2692 | last = entry->bytenr; | |
2693 | } | |
2694 | n = root->rb_node; | |
2695 | ||
2696 | while (n) { | |
2697 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); | |
2698 | WARN_ON(!entry->in_tree); | |
2699 | ||
2700 | middle = entry->bytenr; | |
2701 | ||
2702 | if (alt) | |
2703 | n = n->rb_left; | |
2704 | else | |
2705 | n = n->rb_right; | |
2706 | ||
2707 | alt = 1 - alt; | |
2708 | } | |
2709 | return middle; | |
2710 | } | |
2711 | #endif | |
2712 | ||
2ff7e61e | 2713 | static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads) |
1be41b78 JB |
2714 | { |
2715 | u64 num_bytes; | |
2716 | ||
2717 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + | |
2718 | sizeof(struct btrfs_extent_inline_ref)); | |
0b246afa | 2719 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
1be41b78 JB |
2720 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
2721 | ||
2722 | /* | |
2723 | * We don't ever fill up leaves all the way so multiply by 2 just to be | |
01327610 | 2724 | * closer to what we're really going to want to use. |
1be41b78 | 2725 | */ |
0b246afa | 2726 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info)); |
1be41b78 JB |
2727 | } |
2728 | ||
1262133b JB |
2729 | /* |
2730 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it | |
2731 | * would require to store the csums for that many bytes. | |
2732 | */ | |
2ff7e61e | 2733 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
1262133b JB |
2734 | { |
2735 | u64 csum_size; | |
2736 | u64 num_csums_per_leaf; | |
2737 | u64 num_csums; | |
2738 | ||
0b246afa | 2739 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
1262133b | 2740 | num_csums_per_leaf = div64_u64(csum_size, |
0b246afa JM |
2741 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
2742 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); | |
1262133b JB |
2743 | num_csums += num_csums_per_leaf - 1; |
2744 | num_csums = div64_u64(num_csums, num_csums_per_leaf); | |
2745 | return num_csums; | |
2746 | } | |
2747 | ||
0a2b2a84 | 2748 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2749 | struct btrfs_fs_info *fs_info) |
1be41b78 JB |
2750 | { |
2751 | struct btrfs_block_rsv *global_rsv; | |
2752 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; | |
1262133b | 2753 | u64 csum_bytes = trans->transaction->delayed_refs.pending_csums; |
165c8b02 | 2754 | unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs; |
cb723e49 | 2755 | u64 num_bytes, num_dirty_bgs_bytes; |
1be41b78 JB |
2756 | int ret = 0; |
2757 | ||
0b246afa | 2758 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
2ff7e61e | 2759 | num_heads = heads_to_leaves(fs_info, num_heads); |
1be41b78 | 2760 | if (num_heads > 1) |
0b246afa | 2761 | num_bytes += (num_heads - 1) * fs_info->nodesize; |
1be41b78 | 2762 | num_bytes <<= 1; |
2ff7e61e JM |
2763 | num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) * |
2764 | fs_info->nodesize; | |
0b246afa | 2765 | num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info, |
cb723e49 | 2766 | num_dirty_bgs); |
0b246afa | 2767 | global_rsv = &fs_info->global_block_rsv; |
1be41b78 JB |
2768 | |
2769 | /* | |
2770 | * If we can't allocate any more chunks lets make sure we have _lots_ of | |
2771 | * wiggle room since running delayed refs can create more delayed refs. | |
2772 | */ | |
cb723e49 JB |
2773 | if (global_rsv->space_info->full) { |
2774 | num_dirty_bgs_bytes <<= 1; | |
1be41b78 | 2775 | num_bytes <<= 1; |
cb723e49 | 2776 | } |
1be41b78 JB |
2777 | |
2778 | spin_lock(&global_rsv->lock); | |
cb723e49 | 2779 | if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes) |
1be41b78 JB |
2780 | ret = 1; |
2781 | spin_unlock(&global_rsv->lock); | |
2782 | return ret; | |
2783 | } | |
2784 | ||
0a2b2a84 | 2785 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, |
2ff7e61e | 2786 | struct btrfs_fs_info *fs_info) |
0a2b2a84 | 2787 | { |
0a2b2a84 JB |
2788 | u64 num_entries = |
2789 | atomic_read(&trans->transaction->delayed_refs.num_entries); | |
2790 | u64 avg_runtime; | |
a79b7d4b | 2791 | u64 val; |
0a2b2a84 JB |
2792 | |
2793 | smp_mb(); | |
2794 | avg_runtime = fs_info->avg_delayed_ref_runtime; | |
a79b7d4b | 2795 | val = num_entries * avg_runtime; |
dc1a90c6 | 2796 | if (val >= NSEC_PER_SEC) |
0a2b2a84 | 2797 | return 1; |
a79b7d4b CM |
2798 | if (val >= NSEC_PER_SEC / 2) |
2799 | return 2; | |
0a2b2a84 | 2800 | |
2ff7e61e | 2801 | return btrfs_check_space_for_delayed_refs(trans, fs_info); |
0a2b2a84 JB |
2802 | } |
2803 | ||
a79b7d4b CM |
2804 | struct async_delayed_refs { |
2805 | struct btrfs_root *root; | |
31b9655f | 2806 | u64 transid; |
a79b7d4b CM |
2807 | int count; |
2808 | int error; | |
2809 | int sync; | |
2810 | struct completion wait; | |
2811 | struct btrfs_work work; | |
2812 | }; | |
2813 | ||
2ff7e61e JM |
2814 | static inline struct async_delayed_refs * |
2815 | to_async_delayed_refs(struct btrfs_work *work) | |
2816 | { | |
2817 | return container_of(work, struct async_delayed_refs, work); | |
2818 | } | |
2819 | ||
a79b7d4b CM |
2820 | static void delayed_ref_async_start(struct btrfs_work *work) |
2821 | { | |
2ff7e61e | 2822 | struct async_delayed_refs *async = to_async_delayed_refs(work); |
a79b7d4b | 2823 | struct btrfs_trans_handle *trans; |
2ff7e61e | 2824 | struct btrfs_fs_info *fs_info = async->root->fs_info; |
a79b7d4b CM |
2825 | int ret; |
2826 | ||
0f873eca | 2827 | /* if the commit is already started, we don't need to wait here */ |
2ff7e61e | 2828 | if (btrfs_transaction_blocked(fs_info)) |
31b9655f | 2829 | goto done; |
31b9655f | 2830 | |
0f873eca CM |
2831 | trans = btrfs_join_transaction(async->root); |
2832 | if (IS_ERR(trans)) { | |
2833 | async->error = PTR_ERR(trans); | |
a79b7d4b CM |
2834 | goto done; |
2835 | } | |
2836 | ||
2837 | /* | |
01327610 | 2838 | * trans->sync means that when we call end_transaction, we won't |
a79b7d4b CM |
2839 | * wait on delayed refs |
2840 | */ | |
2841 | trans->sync = true; | |
0f873eca CM |
2842 | |
2843 | /* Don't bother flushing if we got into a different transaction */ | |
2844 | if (trans->transid > async->transid) | |
2845 | goto end; | |
2846 | ||
c79a70b1 | 2847 | ret = btrfs_run_delayed_refs(trans, async->count); |
a79b7d4b CM |
2848 | if (ret) |
2849 | async->error = ret; | |
0f873eca | 2850 | end: |
3a45bb20 | 2851 | ret = btrfs_end_transaction(trans); |
a79b7d4b CM |
2852 | if (ret && !async->error) |
2853 | async->error = ret; | |
2854 | done: | |
2855 | if (async->sync) | |
2856 | complete(&async->wait); | |
2857 | else | |
2858 | kfree(async); | |
2859 | } | |
2860 | ||
2ff7e61e | 2861 | int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info, |
31b9655f | 2862 | unsigned long count, u64 transid, int wait) |
a79b7d4b CM |
2863 | { |
2864 | struct async_delayed_refs *async; | |
2865 | int ret; | |
2866 | ||
2867 | async = kmalloc(sizeof(*async), GFP_NOFS); | |
2868 | if (!async) | |
2869 | return -ENOMEM; | |
2870 | ||
0b246afa | 2871 | async->root = fs_info->tree_root; |
a79b7d4b CM |
2872 | async->count = count; |
2873 | async->error = 0; | |
31b9655f | 2874 | async->transid = transid; |
a79b7d4b CM |
2875 | if (wait) |
2876 | async->sync = 1; | |
2877 | else | |
2878 | async->sync = 0; | |
2879 | init_completion(&async->wait); | |
2880 | ||
9e0af237 LB |
2881 | btrfs_init_work(&async->work, btrfs_extent_refs_helper, |
2882 | delayed_ref_async_start, NULL, NULL); | |
a79b7d4b | 2883 | |
0b246afa | 2884 | btrfs_queue_work(fs_info->extent_workers, &async->work); |
a79b7d4b CM |
2885 | |
2886 | if (wait) { | |
2887 | wait_for_completion(&async->wait); | |
2888 | ret = async->error; | |
2889 | kfree(async); | |
2890 | return ret; | |
2891 | } | |
2892 | return 0; | |
2893 | } | |
2894 | ||
c3e69d58 CM |
2895 | /* |
2896 | * this starts processing the delayed reference count updates and | |
2897 | * extent insertions we have queued up so far. count can be | |
2898 | * 0, which means to process everything in the tree at the start | |
2899 | * of the run (but not newly added entries), or it can be some target | |
2900 | * number you'd like to process. | |
79787eaa JM |
2901 | * |
2902 | * Returns 0 on success or if called with an aborted transaction | |
2903 | * Returns <0 on error and aborts the transaction | |
c3e69d58 CM |
2904 | */ |
2905 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, | |
c79a70b1 | 2906 | unsigned long count) |
c3e69d58 | 2907 | { |
c79a70b1 | 2908 | struct btrfs_fs_info *fs_info = trans->fs_info; |
c3e69d58 CM |
2909 | struct rb_node *node; |
2910 | struct btrfs_delayed_ref_root *delayed_refs; | |
c46effa6 | 2911 | struct btrfs_delayed_ref_head *head; |
c3e69d58 CM |
2912 | int ret; |
2913 | int run_all = count == (unsigned long)-1; | |
d9a0540a | 2914 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
c3e69d58 | 2915 | |
79787eaa JM |
2916 | /* We'll clean this up in btrfs_cleanup_transaction */ |
2917 | if (trans->aborted) | |
2918 | return 0; | |
2919 | ||
0b246afa | 2920 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
511711af CM |
2921 | return 0; |
2922 | ||
c3e69d58 | 2923 | delayed_refs = &trans->transaction->delayed_refs; |
26455d33 | 2924 | if (count == 0) |
d7df2c79 | 2925 | count = atomic_read(&delayed_refs->num_entries) * 2; |
bb721703 | 2926 | |
c3e69d58 | 2927 | again: |
709c0486 AJ |
2928 | #ifdef SCRAMBLE_DELAYED_REFS |
2929 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); | |
2930 | #endif | |
d9a0540a | 2931 | trans->can_flush_pending_bgs = false; |
0a1e458a | 2932 | ret = __btrfs_run_delayed_refs(trans, count); |
d7df2c79 | 2933 | if (ret < 0) { |
66642832 | 2934 | btrfs_abort_transaction(trans, ret); |
d7df2c79 | 2935 | return ret; |
eb099670 | 2936 | } |
c3e69d58 | 2937 | |
56bec294 | 2938 | if (run_all) { |
d7df2c79 | 2939 | if (!list_empty(&trans->new_bgs)) |
6c686b35 | 2940 | btrfs_create_pending_block_groups(trans); |
ea658bad | 2941 | |
d7df2c79 | 2942 | spin_lock(&delayed_refs->lock); |
c46effa6 | 2943 | node = rb_first(&delayed_refs->href_root); |
d7df2c79 JB |
2944 | if (!node) { |
2945 | spin_unlock(&delayed_refs->lock); | |
56bec294 | 2946 | goto out; |
d7df2c79 | 2947 | } |
d278850e JB |
2948 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
2949 | href_node); | |
2950 | refcount_inc(&head->refs); | |
2951 | spin_unlock(&delayed_refs->lock); | |
e9d0b13b | 2952 | |
d278850e JB |
2953 | /* Mutex was contended, block until it's released and retry. */ |
2954 | mutex_lock(&head->mutex); | |
2955 | mutex_unlock(&head->mutex); | |
56bec294 | 2956 | |
d278850e | 2957 | btrfs_put_delayed_ref_head(head); |
d7df2c79 | 2958 | cond_resched(); |
56bec294 | 2959 | goto again; |
5f39d397 | 2960 | } |
54aa1f4d | 2961 | out: |
d9a0540a | 2962 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
a28ec197 CM |
2963 | return 0; |
2964 | } | |
2965 | ||
5d4f98a2 | 2966 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
2ff7e61e | 2967 | struct btrfs_fs_info *fs_info, |
5d4f98a2 | 2968 | u64 bytenr, u64 num_bytes, u64 flags, |
b1c79e09 | 2969 | int level, int is_data) |
5d4f98a2 YZ |
2970 | { |
2971 | struct btrfs_delayed_extent_op *extent_op; | |
2972 | int ret; | |
2973 | ||
78a6184a | 2974 | extent_op = btrfs_alloc_delayed_extent_op(); |
5d4f98a2 YZ |
2975 | if (!extent_op) |
2976 | return -ENOMEM; | |
2977 | ||
2978 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
2979 | extent_op->update_flags = true; |
2980 | extent_op->update_key = false; | |
2981 | extent_op->is_data = is_data ? true : false; | |
b1c79e09 | 2982 | extent_op->level = level; |
5d4f98a2 | 2983 | |
0b246afa | 2984 | ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr, |
66d7e7f0 | 2985 | num_bytes, extent_op); |
5d4f98a2 | 2986 | if (ret) |
78a6184a | 2987 | btrfs_free_delayed_extent_op(extent_op); |
5d4f98a2 YZ |
2988 | return ret; |
2989 | } | |
2990 | ||
e4c3b2dc | 2991 | static noinline int check_delayed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
2992 | struct btrfs_path *path, |
2993 | u64 objectid, u64 offset, u64 bytenr) | |
2994 | { | |
2995 | struct btrfs_delayed_ref_head *head; | |
2996 | struct btrfs_delayed_ref_node *ref; | |
2997 | struct btrfs_delayed_data_ref *data_ref; | |
2998 | struct btrfs_delayed_ref_root *delayed_refs; | |
e4c3b2dc | 2999 | struct btrfs_transaction *cur_trans; |
0e0adbcf | 3000 | struct rb_node *node; |
5d4f98a2 YZ |
3001 | int ret = 0; |
3002 | ||
998ac6d2 | 3003 | spin_lock(&root->fs_info->trans_lock); |
e4c3b2dc | 3004 | cur_trans = root->fs_info->running_transaction; |
998ac6d2 | 3005 | if (cur_trans) |
3006 | refcount_inc(&cur_trans->use_count); | |
3007 | spin_unlock(&root->fs_info->trans_lock); | |
e4c3b2dc LB |
3008 | if (!cur_trans) |
3009 | return 0; | |
3010 | ||
3011 | delayed_refs = &cur_trans->delayed_refs; | |
5d4f98a2 | 3012 | spin_lock(&delayed_refs->lock); |
f72ad18e | 3013 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
d7df2c79 JB |
3014 | if (!head) { |
3015 | spin_unlock(&delayed_refs->lock); | |
998ac6d2 | 3016 | btrfs_put_transaction(cur_trans); |
d7df2c79 JB |
3017 | return 0; |
3018 | } | |
5d4f98a2 YZ |
3019 | |
3020 | if (!mutex_trylock(&head->mutex)) { | |
d278850e | 3021 | refcount_inc(&head->refs); |
5d4f98a2 YZ |
3022 | spin_unlock(&delayed_refs->lock); |
3023 | ||
b3b4aa74 | 3024 | btrfs_release_path(path); |
5d4f98a2 | 3025 | |
8cc33e5c DS |
3026 | /* |
3027 | * Mutex was contended, block until it's released and let | |
3028 | * caller try again | |
3029 | */ | |
5d4f98a2 YZ |
3030 | mutex_lock(&head->mutex); |
3031 | mutex_unlock(&head->mutex); | |
d278850e | 3032 | btrfs_put_delayed_ref_head(head); |
998ac6d2 | 3033 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
3034 | return -EAGAIN; |
3035 | } | |
d7df2c79 | 3036 | spin_unlock(&delayed_refs->lock); |
5d4f98a2 | 3037 | |
d7df2c79 | 3038 | spin_lock(&head->lock); |
0e0adbcf JB |
3039 | /* |
3040 | * XXX: We should replace this with a proper search function in the | |
3041 | * future. | |
3042 | */ | |
3043 | for (node = rb_first(&head->ref_tree); node; node = rb_next(node)) { | |
3044 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); | |
d7df2c79 JB |
3045 | /* If it's a shared ref we know a cross reference exists */ |
3046 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { | |
3047 | ret = 1; | |
3048 | break; | |
3049 | } | |
5d4f98a2 | 3050 | |
d7df2c79 | 3051 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
5d4f98a2 | 3052 | |
d7df2c79 JB |
3053 | /* |
3054 | * If our ref doesn't match the one we're currently looking at | |
3055 | * then we have a cross reference. | |
3056 | */ | |
3057 | if (data_ref->root != root->root_key.objectid || | |
3058 | data_ref->objectid != objectid || | |
3059 | data_ref->offset != offset) { | |
3060 | ret = 1; | |
3061 | break; | |
3062 | } | |
5d4f98a2 | 3063 | } |
d7df2c79 | 3064 | spin_unlock(&head->lock); |
5d4f98a2 | 3065 | mutex_unlock(&head->mutex); |
998ac6d2 | 3066 | btrfs_put_transaction(cur_trans); |
5d4f98a2 YZ |
3067 | return ret; |
3068 | } | |
3069 | ||
e4c3b2dc | 3070 | static noinline int check_committed_ref(struct btrfs_root *root, |
5d4f98a2 YZ |
3071 | struct btrfs_path *path, |
3072 | u64 objectid, u64 offset, u64 bytenr) | |
be20aa9d | 3073 | { |
0b246afa JM |
3074 | struct btrfs_fs_info *fs_info = root->fs_info; |
3075 | struct btrfs_root *extent_root = fs_info->extent_root; | |
f321e491 | 3076 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
3077 | struct btrfs_extent_data_ref *ref; |
3078 | struct btrfs_extent_inline_ref *iref; | |
3079 | struct btrfs_extent_item *ei; | |
f321e491 | 3080 | struct btrfs_key key; |
5d4f98a2 | 3081 | u32 item_size; |
3de28d57 | 3082 | int type; |
be20aa9d | 3083 | int ret; |
925baedd | 3084 | |
be20aa9d | 3085 | key.objectid = bytenr; |
31840ae1 | 3086 | key.offset = (u64)-1; |
f321e491 | 3087 | key.type = BTRFS_EXTENT_ITEM_KEY; |
be20aa9d | 3088 | |
be20aa9d CM |
3089 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
3090 | if (ret < 0) | |
3091 | goto out; | |
79787eaa | 3092 | BUG_ON(ret == 0); /* Corruption */ |
80ff3856 YZ |
3093 | |
3094 | ret = -ENOENT; | |
3095 | if (path->slots[0] == 0) | |
31840ae1 | 3096 | goto out; |
be20aa9d | 3097 | |
31840ae1 | 3098 | path->slots[0]--; |
f321e491 | 3099 | leaf = path->nodes[0]; |
5d4f98a2 | 3100 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
be20aa9d | 3101 | |
5d4f98a2 | 3102 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
be20aa9d | 3103 | goto out; |
f321e491 | 3104 | |
5d4f98a2 YZ |
3105 | ret = 1; |
3106 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); | |
5d4f98a2 | 3107 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
bd09835d | 3108 | |
5d4f98a2 YZ |
3109 | if (item_size != sizeof(*ei) + |
3110 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) | |
3111 | goto out; | |
be20aa9d | 3112 | |
5d4f98a2 YZ |
3113 | if (btrfs_extent_generation(leaf, ei) <= |
3114 | btrfs_root_last_snapshot(&root->root_item)) | |
3115 | goto out; | |
3116 | ||
3117 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); | |
3de28d57 LB |
3118 | |
3119 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); | |
3120 | if (type != BTRFS_EXTENT_DATA_REF_KEY) | |
5d4f98a2 YZ |
3121 | goto out; |
3122 | ||
3123 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
3124 | if (btrfs_extent_refs(leaf, ei) != | |
3125 | btrfs_extent_data_ref_count(leaf, ref) || | |
3126 | btrfs_extent_data_ref_root(leaf, ref) != | |
3127 | root->root_key.objectid || | |
3128 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || | |
3129 | btrfs_extent_data_ref_offset(leaf, ref) != offset) | |
3130 | goto out; | |
3131 | ||
3132 | ret = 0; | |
3133 | out: | |
3134 | return ret; | |
3135 | } | |
3136 | ||
e4c3b2dc LB |
3137 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
3138 | u64 bytenr) | |
5d4f98a2 YZ |
3139 | { |
3140 | struct btrfs_path *path; | |
3141 | int ret; | |
3142 | int ret2; | |
3143 | ||
3144 | path = btrfs_alloc_path(); | |
3145 | if (!path) | |
9132c4ff | 3146 | return -ENOMEM; |
5d4f98a2 YZ |
3147 | |
3148 | do { | |
e4c3b2dc | 3149 | ret = check_committed_ref(root, path, objectid, |
5d4f98a2 YZ |
3150 | offset, bytenr); |
3151 | if (ret && ret != -ENOENT) | |
f321e491 | 3152 | goto out; |
80ff3856 | 3153 | |
e4c3b2dc | 3154 | ret2 = check_delayed_ref(root, path, objectid, |
5d4f98a2 YZ |
3155 | offset, bytenr); |
3156 | } while (ret2 == -EAGAIN); | |
3157 | ||
3158 | if (ret2 && ret2 != -ENOENT) { | |
3159 | ret = ret2; | |
3160 | goto out; | |
f321e491 | 3161 | } |
5d4f98a2 YZ |
3162 | |
3163 | if (ret != -ENOENT || ret2 != -ENOENT) | |
3164 | ret = 0; | |
be20aa9d | 3165 | out: |
80ff3856 | 3166 | btrfs_free_path(path); |
f0486c68 YZ |
3167 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
3168 | WARN_ON(ret > 0); | |
f321e491 | 3169 | return ret; |
be20aa9d | 3170 | } |
c5739bba | 3171 | |
5d4f98a2 | 3172 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
b7a9f29f | 3173 | struct btrfs_root *root, |
5d4f98a2 | 3174 | struct extent_buffer *buf, |
e339a6b0 | 3175 | int full_backref, int inc) |
31840ae1 | 3176 | { |
0b246afa | 3177 | struct btrfs_fs_info *fs_info = root->fs_info; |
31840ae1 | 3178 | u64 bytenr; |
5d4f98a2 YZ |
3179 | u64 num_bytes; |
3180 | u64 parent; | |
31840ae1 | 3181 | u64 ref_root; |
31840ae1 | 3182 | u32 nritems; |
31840ae1 ZY |
3183 | struct btrfs_key key; |
3184 | struct btrfs_file_extent_item *fi; | |
3185 | int i; | |
3186 | int level; | |
3187 | int ret = 0; | |
2ff7e61e | 3188 | int (*process_func)(struct btrfs_trans_handle *, |
84f7d8e6 | 3189 | struct btrfs_root *, |
b06c4bf5 | 3190 | u64, u64, u64, u64, u64, u64); |
31840ae1 | 3191 | |
fccb84c9 | 3192 | |
0b246afa | 3193 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 3194 | return 0; |
fccb84c9 | 3195 | |
31840ae1 | 3196 | ref_root = btrfs_header_owner(buf); |
31840ae1 ZY |
3197 | nritems = btrfs_header_nritems(buf); |
3198 | level = btrfs_header_level(buf); | |
3199 | ||
27cdeb70 | 3200 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
5d4f98a2 | 3201 | return 0; |
31840ae1 | 3202 | |
5d4f98a2 YZ |
3203 | if (inc) |
3204 | process_func = btrfs_inc_extent_ref; | |
3205 | else | |
3206 | process_func = btrfs_free_extent; | |
31840ae1 | 3207 | |
5d4f98a2 YZ |
3208 | if (full_backref) |
3209 | parent = buf->start; | |
3210 | else | |
3211 | parent = 0; | |
3212 | ||
3213 | for (i = 0; i < nritems; i++) { | |
31840ae1 | 3214 | if (level == 0) { |
5d4f98a2 | 3215 | btrfs_item_key_to_cpu(buf, &key, i); |
962a298f | 3216 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
31840ae1 | 3217 | continue; |
5d4f98a2 | 3218 | fi = btrfs_item_ptr(buf, i, |
31840ae1 ZY |
3219 | struct btrfs_file_extent_item); |
3220 | if (btrfs_file_extent_type(buf, fi) == | |
3221 | BTRFS_FILE_EXTENT_INLINE) | |
3222 | continue; | |
3223 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); | |
3224 | if (bytenr == 0) | |
3225 | continue; | |
5d4f98a2 YZ |
3226 | |
3227 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); | |
3228 | key.offset -= btrfs_file_extent_offset(buf, fi); | |
84f7d8e6 | 3229 | ret = process_func(trans, root, bytenr, num_bytes, |
5d4f98a2 | 3230 | parent, ref_root, key.objectid, |
b06c4bf5 | 3231 | key.offset); |
31840ae1 ZY |
3232 | if (ret) |
3233 | goto fail; | |
3234 | } else { | |
5d4f98a2 | 3235 | bytenr = btrfs_node_blockptr(buf, i); |
0b246afa | 3236 | num_bytes = fs_info->nodesize; |
84f7d8e6 | 3237 | ret = process_func(trans, root, bytenr, num_bytes, |
b06c4bf5 | 3238 | parent, ref_root, level - 1, 0); |
31840ae1 ZY |
3239 | if (ret) |
3240 | goto fail; | |
3241 | } | |
3242 | } | |
3243 | return 0; | |
3244 | fail: | |
5d4f98a2 YZ |
3245 | return ret; |
3246 | } | |
3247 | ||
3248 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3249 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3250 | { |
e339a6b0 | 3251 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
5d4f98a2 YZ |
3252 | } |
3253 | ||
3254 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
e339a6b0 | 3255 | struct extent_buffer *buf, int full_backref) |
5d4f98a2 | 3256 | { |
e339a6b0 | 3257 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
31840ae1 ZY |
3258 | } |
3259 | ||
9078a3e1 | 3260 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
2ff7e61e | 3261 | struct btrfs_fs_info *fs_info, |
9078a3e1 CM |
3262 | struct btrfs_path *path, |
3263 | struct btrfs_block_group_cache *cache) | |
3264 | { | |
3265 | int ret; | |
0b246afa | 3266 | struct btrfs_root *extent_root = fs_info->extent_root; |
5f39d397 CM |
3267 | unsigned long bi; |
3268 | struct extent_buffer *leaf; | |
9078a3e1 | 3269 | |
9078a3e1 | 3270 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
df95e7f0 JB |
3271 | if (ret) { |
3272 | if (ret > 0) | |
3273 | ret = -ENOENT; | |
54aa1f4d | 3274 | goto fail; |
df95e7f0 | 3275 | } |
5f39d397 CM |
3276 | |
3277 | leaf = path->nodes[0]; | |
3278 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); | |
3279 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); | |
3280 | btrfs_mark_buffer_dirty(leaf); | |
54aa1f4d | 3281 | fail: |
24b89d08 | 3282 | btrfs_release_path(path); |
df95e7f0 | 3283 | return ret; |
9078a3e1 CM |
3284 | |
3285 | } | |
3286 | ||
4a8c9a62 | 3287 | static struct btrfs_block_group_cache * |
2ff7e61e | 3288 | next_block_group(struct btrfs_fs_info *fs_info, |
4a8c9a62 YZ |
3289 | struct btrfs_block_group_cache *cache) |
3290 | { | |
3291 | struct rb_node *node; | |
292cbd51 | 3292 | |
0b246afa | 3293 | spin_lock(&fs_info->block_group_cache_lock); |
292cbd51 FM |
3294 | |
3295 | /* If our block group was removed, we need a full search. */ | |
3296 | if (RB_EMPTY_NODE(&cache->cache_node)) { | |
3297 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; | |
3298 | ||
0b246afa | 3299 | spin_unlock(&fs_info->block_group_cache_lock); |
292cbd51 | 3300 | btrfs_put_block_group(cache); |
0b246afa | 3301 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; |
292cbd51 | 3302 | } |
4a8c9a62 YZ |
3303 | node = rb_next(&cache->cache_node); |
3304 | btrfs_put_block_group(cache); | |
3305 | if (node) { | |
3306 | cache = rb_entry(node, struct btrfs_block_group_cache, | |
3307 | cache_node); | |
11dfe35a | 3308 | btrfs_get_block_group(cache); |
4a8c9a62 YZ |
3309 | } else |
3310 | cache = NULL; | |
0b246afa | 3311 | spin_unlock(&fs_info->block_group_cache_lock); |
4a8c9a62 YZ |
3312 | return cache; |
3313 | } | |
3314 | ||
0af3d00b JB |
3315 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
3316 | struct btrfs_trans_handle *trans, | |
3317 | struct btrfs_path *path) | |
3318 | { | |
0b246afa JM |
3319 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
3320 | struct btrfs_root *root = fs_info->tree_root; | |
0af3d00b | 3321 | struct inode *inode = NULL; |
364ecf36 | 3322 | struct extent_changeset *data_reserved = NULL; |
0af3d00b | 3323 | u64 alloc_hint = 0; |
2b20982e | 3324 | int dcs = BTRFS_DC_ERROR; |
f8c269d7 | 3325 | u64 num_pages = 0; |
0af3d00b JB |
3326 | int retries = 0; |
3327 | int ret = 0; | |
3328 | ||
3329 | /* | |
3330 | * If this block group is smaller than 100 megs don't bother caching the | |
3331 | * block group. | |
3332 | */ | |
ee22184b | 3333 | if (block_group->key.offset < (100 * SZ_1M)) { |
0af3d00b JB |
3334 | spin_lock(&block_group->lock); |
3335 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; | |
3336 | spin_unlock(&block_group->lock); | |
3337 | return 0; | |
3338 | } | |
3339 | ||
0c0ef4bc JB |
3340 | if (trans->aborted) |
3341 | return 0; | |
0af3d00b | 3342 | again: |
77ab86bf | 3343 | inode = lookup_free_space_inode(fs_info, block_group, path); |
0af3d00b JB |
3344 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { |
3345 | ret = PTR_ERR(inode); | |
b3b4aa74 | 3346 | btrfs_release_path(path); |
0af3d00b JB |
3347 | goto out; |
3348 | } | |
3349 | ||
3350 | if (IS_ERR(inode)) { | |
3351 | BUG_ON(retries); | |
3352 | retries++; | |
3353 | ||
3354 | if (block_group->ro) | |
3355 | goto out_free; | |
3356 | ||
77ab86bf JM |
3357 | ret = create_free_space_inode(fs_info, trans, block_group, |
3358 | path); | |
0af3d00b JB |
3359 | if (ret) |
3360 | goto out_free; | |
3361 | goto again; | |
3362 | } | |
3363 | ||
3364 | /* | |
3365 | * We want to set the generation to 0, that way if anything goes wrong | |
3366 | * from here on out we know not to trust this cache when we load up next | |
3367 | * time. | |
3368 | */ | |
3369 | BTRFS_I(inode)->generation = 0; | |
3370 | ret = btrfs_update_inode(trans, root, inode); | |
0c0ef4bc JB |
3371 | if (ret) { |
3372 | /* | |
3373 | * So theoretically we could recover from this, simply set the | |
3374 | * super cache generation to 0 so we know to invalidate the | |
3375 | * cache, but then we'd have to keep track of the block groups | |
3376 | * that fail this way so we know we _have_ to reset this cache | |
3377 | * before the next commit or risk reading stale cache. So to | |
3378 | * limit our exposure to horrible edge cases lets just abort the | |
3379 | * transaction, this only happens in really bad situations | |
3380 | * anyway. | |
3381 | */ | |
66642832 | 3382 | btrfs_abort_transaction(trans, ret); |
0c0ef4bc JB |
3383 | goto out_put; |
3384 | } | |
0af3d00b JB |
3385 | WARN_ON(ret); |
3386 | ||
8e138e0d JB |
3387 | /* We've already setup this transaction, go ahead and exit */ |
3388 | if (block_group->cache_generation == trans->transid && | |
3389 | i_size_read(inode)) { | |
3390 | dcs = BTRFS_DC_SETUP; | |
3391 | goto out_put; | |
3392 | } | |
3393 | ||
0af3d00b | 3394 | if (i_size_read(inode) > 0) { |
2ff7e61e | 3395 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
0b246afa | 3396 | &fs_info->global_block_rsv); |
7b61cd92 MX |
3397 | if (ret) |
3398 | goto out_put; | |
3399 | ||
77ab86bf | 3400 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); |
0af3d00b JB |
3401 | if (ret) |
3402 | goto out_put; | |
3403 | } | |
3404 | ||
3405 | spin_lock(&block_group->lock); | |
cf7c1ef6 | 3406 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
0b246afa | 3407 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { |
cf7c1ef6 LB |
3408 | /* |
3409 | * don't bother trying to write stuff out _if_ | |
3410 | * a) we're not cached, | |
1a79c1f2 LB |
3411 | * b) we're with nospace_cache mount option, |
3412 | * c) we're with v2 space_cache (FREE_SPACE_TREE). | |
cf7c1ef6 | 3413 | */ |
2b20982e | 3414 | dcs = BTRFS_DC_WRITTEN; |
0af3d00b JB |
3415 | spin_unlock(&block_group->lock); |
3416 | goto out_put; | |
3417 | } | |
3418 | spin_unlock(&block_group->lock); | |
3419 | ||
2968b1f4 JB |
3420 | /* |
3421 | * We hit an ENOSPC when setting up the cache in this transaction, just | |
3422 | * skip doing the setup, we've already cleared the cache so we're safe. | |
3423 | */ | |
3424 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { | |
3425 | ret = -ENOSPC; | |
3426 | goto out_put; | |
3427 | } | |
3428 | ||
6fc823b1 JB |
3429 | /* |
3430 | * Try to preallocate enough space based on how big the block group is. | |
3431 | * Keep in mind this has to include any pinned space which could end up | |
3432 | * taking up quite a bit since it's not folded into the other space | |
3433 | * cache. | |
3434 | */ | |
ee22184b | 3435 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
0af3d00b JB |
3436 | if (!num_pages) |
3437 | num_pages = 1; | |
3438 | ||
0af3d00b | 3439 | num_pages *= 16; |
09cbfeaf | 3440 | num_pages *= PAGE_SIZE; |
0af3d00b | 3441 | |
364ecf36 | 3442 | ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages); |
0af3d00b JB |
3443 | if (ret) |
3444 | goto out_put; | |
3445 | ||
3446 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, | |
3447 | num_pages, num_pages, | |
3448 | &alloc_hint); | |
2968b1f4 JB |
3449 | /* |
3450 | * Our cache requires contiguous chunks so that we don't modify a bunch | |
3451 | * of metadata or split extents when writing the cache out, which means | |
3452 | * we can enospc if we are heavily fragmented in addition to just normal | |
3453 | * out of space conditions. So if we hit this just skip setting up any | |
3454 | * other block groups for this transaction, maybe we'll unpin enough | |
3455 | * space the next time around. | |
3456 | */ | |
2b20982e JB |
3457 | if (!ret) |
3458 | dcs = BTRFS_DC_SETUP; | |
2968b1f4 JB |
3459 | else if (ret == -ENOSPC) |
3460 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); | |
c09544e0 | 3461 | |
0af3d00b JB |
3462 | out_put: |
3463 | iput(inode); | |
3464 | out_free: | |
b3b4aa74 | 3465 | btrfs_release_path(path); |
0af3d00b JB |
3466 | out: |
3467 | spin_lock(&block_group->lock); | |
e65cbb94 | 3468 | if (!ret && dcs == BTRFS_DC_SETUP) |
5b0e95bf | 3469 | block_group->cache_generation = trans->transid; |
2b20982e | 3470 | block_group->disk_cache_state = dcs; |
0af3d00b JB |
3471 | spin_unlock(&block_group->lock); |
3472 | ||
364ecf36 | 3473 | extent_changeset_free(data_reserved); |
0af3d00b JB |
3474 | return ret; |
3475 | } | |
3476 | ||
dcdf7f6d | 3477 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans, |
2ff7e61e | 3478 | struct btrfs_fs_info *fs_info) |
dcdf7f6d JB |
3479 | { |
3480 | struct btrfs_block_group_cache *cache, *tmp; | |
3481 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3482 | struct btrfs_path *path; | |
3483 | ||
3484 | if (list_empty(&cur_trans->dirty_bgs) || | |
0b246afa | 3485 | !btrfs_test_opt(fs_info, SPACE_CACHE)) |
dcdf7f6d JB |
3486 | return 0; |
3487 | ||
3488 | path = btrfs_alloc_path(); | |
3489 | if (!path) | |
3490 | return -ENOMEM; | |
3491 | ||
3492 | /* Could add new block groups, use _safe just in case */ | |
3493 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, | |
3494 | dirty_list) { | |
3495 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) | |
3496 | cache_save_setup(cache, trans, path); | |
3497 | } | |
3498 | ||
3499 | btrfs_free_path(path); | |
3500 | return 0; | |
3501 | } | |
3502 | ||
1bbc621e CM |
3503 | /* |
3504 | * transaction commit does final block group cache writeback during a | |
3505 | * critical section where nothing is allowed to change the FS. This is | |
3506 | * required in order for the cache to actually match the block group, | |
3507 | * but can introduce a lot of latency into the commit. | |
3508 | * | |
3509 | * So, btrfs_start_dirty_block_groups is here to kick off block group | |
3510 | * cache IO. There's a chance we'll have to redo some of it if the | |
3511 | * block group changes again during the commit, but it greatly reduces | |
3512 | * the commit latency by getting rid of the easy block groups while | |
3513 | * we're still allowing others to join the commit. | |
3514 | */ | |
21217054 | 3515 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) |
9078a3e1 | 3516 | { |
21217054 | 3517 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4a8c9a62 | 3518 | struct btrfs_block_group_cache *cache; |
ce93ec54 JB |
3519 | struct btrfs_transaction *cur_trans = trans->transaction; |
3520 | int ret = 0; | |
c9dc4c65 | 3521 | int should_put; |
1bbc621e CM |
3522 | struct btrfs_path *path = NULL; |
3523 | LIST_HEAD(dirty); | |
3524 | struct list_head *io = &cur_trans->io_bgs; | |
c9dc4c65 | 3525 | int num_started = 0; |
1bbc621e CM |
3526 | int loops = 0; |
3527 | ||
3528 | spin_lock(&cur_trans->dirty_bgs_lock); | |
b58d1a9e FM |
3529 | if (list_empty(&cur_trans->dirty_bgs)) { |
3530 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3531 | return 0; | |
1bbc621e | 3532 | } |
b58d1a9e | 3533 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
1bbc621e | 3534 | spin_unlock(&cur_trans->dirty_bgs_lock); |
ce93ec54 | 3535 | |
1bbc621e | 3536 | again: |
1bbc621e CM |
3537 | /* |
3538 | * make sure all the block groups on our dirty list actually | |
3539 | * exist | |
3540 | */ | |
6c686b35 | 3541 | btrfs_create_pending_block_groups(trans); |
1bbc621e CM |
3542 | |
3543 | if (!path) { | |
3544 | path = btrfs_alloc_path(); | |
3545 | if (!path) | |
3546 | return -ENOMEM; | |
3547 | } | |
3548 | ||
b58d1a9e FM |
3549 | /* |
3550 | * cache_write_mutex is here only to save us from balance or automatic | |
3551 | * removal of empty block groups deleting this block group while we are | |
3552 | * writing out the cache | |
3553 | */ | |
3554 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e CM |
3555 | while (!list_empty(&dirty)) { |
3556 | cache = list_first_entry(&dirty, | |
3557 | struct btrfs_block_group_cache, | |
3558 | dirty_list); | |
1bbc621e CM |
3559 | /* |
3560 | * this can happen if something re-dirties a block | |
3561 | * group that is already under IO. Just wait for it to | |
3562 | * finish and then do it all again | |
3563 | */ | |
3564 | if (!list_empty(&cache->io_list)) { | |
3565 | list_del_init(&cache->io_list); | |
afdb5718 | 3566 | btrfs_wait_cache_io(trans, cache, path); |
1bbc621e CM |
3567 | btrfs_put_block_group(cache); |
3568 | } | |
3569 | ||
3570 | ||
3571 | /* | |
3572 | * btrfs_wait_cache_io uses the cache->dirty_list to decide | |
3573 | * if it should update the cache_state. Don't delete | |
3574 | * until after we wait. | |
3575 | * | |
3576 | * Since we're not running in the commit critical section | |
3577 | * we need the dirty_bgs_lock to protect from update_block_group | |
3578 | */ | |
3579 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3580 | list_del_init(&cache->dirty_list); | |
3581 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3582 | ||
3583 | should_put = 1; | |
3584 | ||
3585 | cache_save_setup(cache, trans, path); | |
3586 | ||
3587 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3588 | cache->io_ctl.inode = NULL; | |
0b246afa | 3589 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3590 | cache, path); |
1bbc621e CM |
3591 | if (ret == 0 && cache->io_ctl.inode) { |
3592 | num_started++; | |
3593 | should_put = 0; | |
3594 | ||
3595 | /* | |
45ae2c18 NB |
3596 | * The cache_write_mutex is protecting the |
3597 | * io_list, also refer to the definition of | |
3598 | * btrfs_transaction::io_bgs for more details | |
1bbc621e CM |
3599 | */ |
3600 | list_add_tail(&cache->io_list, io); | |
3601 | } else { | |
3602 | /* | |
3603 | * if we failed to write the cache, the | |
3604 | * generation will be bad and life goes on | |
3605 | */ | |
3606 | ret = 0; | |
3607 | } | |
3608 | } | |
ff1f8250 | 3609 | if (!ret) { |
2ff7e61e JM |
3610 | ret = write_one_cache_group(trans, fs_info, |
3611 | path, cache); | |
ff1f8250 FM |
3612 | /* |
3613 | * Our block group might still be attached to the list | |
3614 | * of new block groups in the transaction handle of some | |
3615 | * other task (struct btrfs_trans_handle->new_bgs). This | |
3616 | * means its block group item isn't yet in the extent | |
3617 | * tree. If this happens ignore the error, as we will | |
3618 | * try again later in the critical section of the | |
3619 | * transaction commit. | |
3620 | */ | |
3621 | if (ret == -ENOENT) { | |
3622 | ret = 0; | |
3623 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3624 | if (list_empty(&cache->dirty_list)) { | |
3625 | list_add_tail(&cache->dirty_list, | |
3626 | &cur_trans->dirty_bgs); | |
3627 | btrfs_get_block_group(cache); | |
3628 | } | |
3629 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3630 | } else if (ret) { | |
66642832 | 3631 | btrfs_abort_transaction(trans, ret); |
ff1f8250 FM |
3632 | } |
3633 | } | |
1bbc621e CM |
3634 | |
3635 | /* if its not on the io list, we need to put the block group */ | |
3636 | if (should_put) | |
3637 | btrfs_put_block_group(cache); | |
3638 | ||
3639 | if (ret) | |
3640 | break; | |
b58d1a9e FM |
3641 | |
3642 | /* | |
3643 | * Avoid blocking other tasks for too long. It might even save | |
3644 | * us from writing caches for block groups that are going to be | |
3645 | * removed. | |
3646 | */ | |
3647 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
3648 | mutex_lock(&trans->transaction->cache_write_mutex); | |
1bbc621e | 3649 | } |
b58d1a9e | 3650 | mutex_unlock(&trans->transaction->cache_write_mutex); |
1bbc621e CM |
3651 | |
3652 | /* | |
3653 | * go through delayed refs for all the stuff we've just kicked off | |
3654 | * and then loop back (just once) | |
3655 | */ | |
c79a70b1 | 3656 | ret = btrfs_run_delayed_refs(trans, 0); |
1bbc621e CM |
3657 | if (!ret && loops == 0) { |
3658 | loops++; | |
3659 | spin_lock(&cur_trans->dirty_bgs_lock); | |
3660 | list_splice_init(&cur_trans->dirty_bgs, &dirty); | |
b58d1a9e FM |
3661 | /* |
3662 | * dirty_bgs_lock protects us from concurrent block group | |
3663 | * deletes too (not just cache_write_mutex). | |
3664 | */ | |
3665 | if (!list_empty(&dirty)) { | |
3666 | spin_unlock(&cur_trans->dirty_bgs_lock); | |
3667 | goto again; | |
3668 | } | |
1bbc621e | 3669 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c79a1751 | 3670 | } else if (ret < 0) { |
2ff7e61e | 3671 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
1bbc621e CM |
3672 | } |
3673 | ||
3674 | btrfs_free_path(path); | |
3675 | return ret; | |
3676 | } | |
3677 | ||
3678 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, | |
2ff7e61e | 3679 | struct btrfs_fs_info *fs_info) |
1bbc621e CM |
3680 | { |
3681 | struct btrfs_block_group_cache *cache; | |
3682 | struct btrfs_transaction *cur_trans = trans->transaction; | |
3683 | int ret = 0; | |
3684 | int should_put; | |
3685 | struct btrfs_path *path; | |
3686 | struct list_head *io = &cur_trans->io_bgs; | |
3687 | int num_started = 0; | |
9078a3e1 CM |
3688 | |
3689 | path = btrfs_alloc_path(); | |
3690 | if (!path) | |
3691 | return -ENOMEM; | |
3692 | ||
ce93ec54 | 3693 | /* |
e44081ef FM |
3694 | * Even though we are in the critical section of the transaction commit, |
3695 | * we can still have concurrent tasks adding elements to this | |
3696 | * transaction's list of dirty block groups. These tasks correspond to | |
3697 | * endio free space workers started when writeback finishes for a | |
3698 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can | |
3699 | * allocate new block groups as a result of COWing nodes of the root | |
3700 | * tree when updating the free space inode. The writeback for the space | |
3701 | * caches is triggered by an earlier call to | |
3702 | * btrfs_start_dirty_block_groups() and iterations of the following | |
3703 | * loop. | |
3704 | * Also we want to do the cache_save_setup first and then run the | |
ce93ec54 JB |
3705 | * delayed refs to make sure we have the best chance at doing this all |
3706 | * in one shot. | |
3707 | */ | |
e44081ef | 3708 | spin_lock(&cur_trans->dirty_bgs_lock); |
ce93ec54 JB |
3709 | while (!list_empty(&cur_trans->dirty_bgs)) { |
3710 | cache = list_first_entry(&cur_trans->dirty_bgs, | |
3711 | struct btrfs_block_group_cache, | |
3712 | dirty_list); | |
c9dc4c65 CM |
3713 | |
3714 | /* | |
3715 | * this can happen if cache_save_setup re-dirties a block | |
3716 | * group that is already under IO. Just wait for it to | |
3717 | * finish and then do it all again | |
3718 | */ | |
3719 | if (!list_empty(&cache->io_list)) { | |
e44081ef | 3720 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3721 | list_del_init(&cache->io_list); |
afdb5718 | 3722 | btrfs_wait_cache_io(trans, cache, path); |
c9dc4c65 | 3723 | btrfs_put_block_group(cache); |
e44081ef | 3724 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3725 | } |
3726 | ||
1bbc621e CM |
3727 | /* |
3728 | * don't remove from the dirty list until after we've waited | |
3729 | * on any pending IO | |
3730 | */ | |
ce93ec54 | 3731 | list_del_init(&cache->dirty_list); |
e44081ef | 3732 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 CM |
3733 | should_put = 1; |
3734 | ||
1bbc621e | 3735 | cache_save_setup(cache, trans, path); |
c9dc4c65 | 3736 | |
ce93ec54 | 3737 | if (!ret) |
c79a70b1 | 3738 | ret = btrfs_run_delayed_refs(trans, |
2ff7e61e | 3739 | (unsigned long) -1); |
c9dc4c65 CM |
3740 | |
3741 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { | |
3742 | cache->io_ctl.inode = NULL; | |
0b246afa | 3743 | ret = btrfs_write_out_cache(fs_info, trans, |
5b4aacef | 3744 | cache, path); |
c9dc4c65 CM |
3745 | if (ret == 0 && cache->io_ctl.inode) { |
3746 | num_started++; | |
3747 | should_put = 0; | |
1bbc621e | 3748 | list_add_tail(&cache->io_list, io); |
c9dc4c65 CM |
3749 | } else { |
3750 | /* | |
3751 | * if we failed to write the cache, the | |
3752 | * generation will be bad and life goes on | |
3753 | */ | |
3754 | ret = 0; | |
3755 | } | |
3756 | } | |
ff1f8250 | 3757 | if (!ret) { |
2ff7e61e JM |
3758 | ret = write_one_cache_group(trans, fs_info, |
3759 | path, cache); | |
2bc0bb5f FM |
3760 | /* |
3761 | * One of the free space endio workers might have | |
3762 | * created a new block group while updating a free space | |
3763 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) | |
3764 | * and hasn't released its transaction handle yet, in | |
3765 | * which case the new block group is still attached to | |
3766 | * its transaction handle and its creation has not | |
3767 | * finished yet (no block group item in the extent tree | |
3768 | * yet, etc). If this is the case, wait for all free | |
3769 | * space endio workers to finish and retry. This is a | |
3770 | * a very rare case so no need for a more efficient and | |
3771 | * complex approach. | |
3772 | */ | |
3773 | if (ret == -ENOENT) { | |
3774 | wait_event(cur_trans->writer_wait, | |
3775 | atomic_read(&cur_trans->num_writers) == 1); | |
2ff7e61e JM |
3776 | ret = write_one_cache_group(trans, fs_info, |
3777 | path, cache); | |
2bc0bb5f | 3778 | } |
ff1f8250 | 3779 | if (ret) |
66642832 | 3780 | btrfs_abort_transaction(trans, ret); |
ff1f8250 | 3781 | } |
c9dc4c65 CM |
3782 | |
3783 | /* if its not on the io list, we need to put the block group */ | |
3784 | if (should_put) | |
3785 | btrfs_put_block_group(cache); | |
e44081ef | 3786 | spin_lock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3787 | } |
e44081ef | 3788 | spin_unlock(&cur_trans->dirty_bgs_lock); |
c9dc4c65 | 3789 | |
45ae2c18 NB |
3790 | /* |
3791 | * Refer to the definition of io_bgs member for details why it's safe | |
3792 | * to use it without any locking | |
3793 | */ | |
1bbc621e CM |
3794 | while (!list_empty(io)) { |
3795 | cache = list_first_entry(io, struct btrfs_block_group_cache, | |
c9dc4c65 CM |
3796 | io_list); |
3797 | list_del_init(&cache->io_list); | |
afdb5718 | 3798 | btrfs_wait_cache_io(trans, cache, path); |
0cb59c99 JB |
3799 | btrfs_put_block_group(cache); |
3800 | } | |
3801 | ||
9078a3e1 | 3802 | btrfs_free_path(path); |
ce93ec54 | 3803 | return ret; |
9078a3e1 CM |
3804 | } |
3805 | ||
2ff7e61e | 3806 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
d2fb3437 YZ |
3807 | { |
3808 | struct btrfs_block_group_cache *block_group; | |
3809 | int readonly = 0; | |
3810 | ||
0b246afa | 3811 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
d2fb3437 YZ |
3812 | if (!block_group || block_group->ro) |
3813 | readonly = 1; | |
3814 | if (block_group) | |
fa9c0d79 | 3815 | btrfs_put_block_group(block_group); |
d2fb3437 YZ |
3816 | return readonly; |
3817 | } | |
3818 | ||
f78c436c FM |
3819 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
3820 | { | |
3821 | struct btrfs_block_group_cache *bg; | |
3822 | bool ret = true; | |
3823 | ||
3824 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3825 | if (!bg) | |
3826 | return false; | |
3827 | ||
3828 | spin_lock(&bg->lock); | |
3829 | if (bg->ro) | |
3830 | ret = false; | |
3831 | else | |
3832 | atomic_inc(&bg->nocow_writers); | |
3833 | spin_unlock(&bg->lock); | |
3834 | ||
3835 | /* no put on block group, done by btrfs_dec_nocow_writers */ | |
3836 | if (!ret) | |
3837 | btrfs_put_block_group(bg); | |
3838 | ||
3839 | return ret; | |
3840 | ||
3841 | } | |
3842 | ||
3843 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) | |
3844 | { | |
3845 | struct btrfs_block_group_cache *bg; | |
3846 | ||
3847 | bg = btrfs_lookup_block_group(fs_info, bytenr); | |
3848 | ASSERT(bg); | |
3849 | if (atomic_dec_and_test(&bg->nocow_writers)) | |
4625956a | 3850 | wake_up_var(&bg->nocow_writers); |
f78c436c FM |
3851 | /* |
3852 | * Once for our lookup and once for the lookup done by a previous call | |
3853 | * to btrfs_inc_nocow_writers() | |
3854 | */ | |
3855 | btrfs_put_block_group(bg); | |
3856 | btrfs_put_block_group(bg); | |
3857 | } | |
3858 | ||
f78c436c FM |
3859 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) |
3860 | { | |
4625956a | 3861 | wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers)); |
f78c436c FM |
3862 | } |
3863 | ||
6ab0a202 JM |
3864 | static const char *alloc_name(u64 flags) |
3865 | { | |
3866 | switch (flags) { | |
3867 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: | |
3868 | return "mixed"; | |
3869 | case BTRFS_BLOCK_GROUP_METADATA: | |
3870 | return "metadata"; | |
3871 | case BTRFS_BLOCK_GROUP_DATA: | |
3872 | return "data"; | |
3873 | case BTRFS_BLOCK_GROUP_SYSTEM: | |
3874 | return "system"; | |
3875 | default: | |
3876 | WARN_ON(1); | |
3877 | return "invalid-combination"; | |
3878 | }; | |
3879 | } | |
3880 | ||
4ca61683 | 3881 | static int create_space_info(struct btrfs_fs_info *info, u64 flags) |
2be12ef7 NB |
3882 | { |
3883 | ||
3884 | struct btrfs_space_info *space_info; | |
3885 | int i; | |
3886 | int ret; | |
3887 | ||
3888 | space_info = kzalloc(sizeof(*space_info), GFP_NOFS); | |
3889 | if (!space_info) | |
3890 | return -ENOMEM; | |
3891 | ||
3892 | ret = percpu_counter_init(&space_info->total_bytes_pinned, 0, | |
3893 | GFP_KERNEL); | |
3894 | if (ret) { | |
3895 | kfree(space_info); | |
3896 | return ret; | |
3897 | } | |
3898 | ||
3899 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) | |
3900 | INIT_LIST_HEAD(&space_info->block_groups[i]); | |
3901 | init_rwsem(&space_info->groups_sem); | |
3902 | spin_lock_init(&space_info->lock); | |
3903 | space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3904 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; | |
3905 | init_waitqueue_head(&space_info->wait); | |
3906 | INIT_LIST_HEAD(&space_info->ro_bgs); | |
3907 | INIT_LIST_HEAD(&space_info->tickets); | |
3908 | INIT_LIST_HEAD(&space_info->priority_tickets); | |
3909 | ||
3910 | ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype, | |
3911 | info->space_info_kobj, "%s", | |
3912 | alloc_name(space_info->flags)); | |
3913 | if (ret) { | |
3914 | percpu_counter_destroy(&space_info->total_bytes_pinned); | |
3915 | kfree(space_info); | |
3916 | return ret; | |
3917 | } | |
3918 | ||
2be12ef7 NB |
3919 | list_add_rcu(&space_info->list, &info->space_info); |
3920 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
3921 | info->data_sinfo = space_info; | |
3922 | ||
3923 | return ret; | |
3924 | } | |
3925 | ||
d2006e6d | 3926 | static void update_space_info(struct btrfs_fs_info *info, u64 flags, |
593060d7 | 3927 | u64 total_bytes, u64 bytes_used, |
e40edf2d | 3928 | u64 bytes_readonly, |
593060d7 CM |
3929 | struct btrfs_space_info **space_info) |
3930 | { | |
3931 | struct btrfs_space_info *found; | |
b742bb82 YZ |
3932 | int factor; |
3933 | ||
46df06b8 | 3934 | factor = btrfs_bg_type_to_factor(flags); |
593060d7 CM |
3935 | |
3936 | found = __find_space_info(info, flags); | |
d2006e6d NB |
3937 | ASSERT(found); |
3938 | spin_lock(&found->lock); | |
3939 | found->total_bytes += total_bytes; | |
3940 | found->disk_total += total_bytes * factor; | |
3941 | found->bytes_used += bytes_used; | |
3942 | found->disk_used += bytes_used * factor; | |
3943 | found->bytes_readonly += bytes_readonly; | |
3944 | if (total_bytes > 0) | |
3945 | found->full = 0; | |
3946 | space_info_add_new_bytes(info, found, total_bytes - | |
3947 | bytes_used - bytes_readonly); | |
3948 | spin_unlock(&found->lock); | |
3949 | *space_info = found; | |
593060d7 CM |
3950 | } |
3951 | ||
8790d502 CM |
3952 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
3953 | { | |
899c81ea ID |
3954 | u64 extra_flags = chunk_to_extended(flags) & |
3955 | BTRFS_EXTENDED_PROFILE_MASK; | |
a46d11a8 | 3956 | |
de98ced9 | 3957 | write_seqlock(&fs_info->profiles_lock); |
a46d11a8 ID |
3958 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
3959 | fs_info->avail_data_alloc_bits |= extra_flags; | |
3960 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
3961 | fs_info->avail_metadata_alloc_bits |= extra_flags; | |
3962 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
3963 | fs_info->avail_system_alloc_bits |= extra_flags; | |
de98ced9 | 3964 | write_sequnlock(&fs_info->profiles_lock); |
8790d502 | 3965 | } |
593060d7 | 3966 | |
fc67c450 ID |
3967 | /* |
3968 | * returns target flags in extended format or 0 if restripe for this | |
3969 | * chunk_type is not in progress | |
c6664b42 | 3970 | * |
dccdb07b | 3971 | * should be called with balance_lock held |
fc67c450 ID |
3972 | */ |
3973 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) | |
3974 | { | |
3975 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3976 | u64 target = 0; | |
3977 | ||
fc67c450 ID |
3978 | if (!bctl) |
3979 | return 0; | |
3980 | ||
3981 | if (flags & BTRFS_BLOCK_GROUP_DATA && | |
3982 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3983 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; | |
3984 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && | |
3985 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3986 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; | |
3987 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && | |
3988 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3989 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; | |
3990 | } | |
3991 | ||
3992 | return target; | |
3993 | } | |
3994 | ||
a46d11a8 ID |
3995 | /* |
3996 | * @flags: available profiles in extended format (see ctree.h) | |
3997 | * | |
e4d8ec0f ID |
3998 | * Returns reduced profile in chunk format. If profile changing is in |
3999 | * progress (either running or paused) picks the target profile (if it's | |
4000 | * already available), otherwise falls back to plain reducing. | |
a46d11a8 | 4001 | */ |
2ff7e61e | 4002 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c | 4003 | { |
0b246afa | 4004 | u64 num_devices = fs_info->fs_devices->rw_devices; |
fc67c450 | 4005 | u64 target; |
9c170b26 ZL |
4006 | u64 raid_type; |
4007 | u64 allowed = 0; | |
a061fc8d | 4008 | |
fc67c450 ID |
4009 | /* |
4010 | * see if restripe for this chunk_type is in progress, if so | |
4011 | * try to reduce to the target profile | |
4012 | */ | |
0b246afa JM |
4013 | spin_lock(&fs_info->balance_lock); |
4014 | target = get_restripe_target(fs_info, flags); | |
fc67c450 ID |
4015 | if (target) { |
4016 | /* pick target profile only if it's already available */ | |
4017 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { | |
0b246afa | 4018 | spin_unlock(&fs_info->balance_lock); |
fc67c450 | 4019 | return extended_to_chunk(target); |
e4d8ec0f ID |
4020 | } |
4021 | } | |
0b246afa | 4022 | spin_unlock(&fs_info->balance_lock); |
e4d8ec0f | 4023 | |
53b381b3 | 4024 | /* First, mask out the RAID levels which aren't possible */ |
9c170b26 ZL |
4025 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
4026 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) | |
41a6e891 | 4027 | allowed |= btrfs_raid_array[raid_type].bg_flag; |
9c170b26 ZL |
4028 | } |
4029 | allowed &= flags; | |
4030 | ||
4031 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) | |
4032 | allowed = BTRFS_BLOCK_GROUP_RAID6; | |
4033 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) | |
4034 | allowed = BTRFS_BLOCK_GROUP_RAID5; | |
4035 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) | |
4036 | allowed = BTRFS_BLOCK_GROUP_RAID10; | |
4037 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) | |
4038 | allowed = BTRFS_BLOCK_GROUP_RAID1; | |
4039 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) | |
4040 | allowed = BTRFS_BLOCK_GROUP_RAID0; | |
4041 | ||
4042 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
4043 | ||
4044 | return extended_to_chunk(flags | allowed); | |
ec44a35c CM |
4045 | } |
4046 | ||
2ff7e61e | 4047 | static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
6a63209f | 4048 | { |
de98ced9 | 4049 | unsigned seq; |
f8213bdc | 4050 | u64 flags; |
de98ced9 MX |
4051 | |
4052 | do { | |
f8213bdc | 4053 | flags = orig_flags; |
0b246afa | 4054 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 MX |
4055 | |
4056 | if (flags & BTRFS_BLOCK_GROUP_DATA) | |
0b246afa | 4057 | flags |= fs_info->avail_data_alloc_bits; |
de98ced9 | 4058 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
0b246afa | 4059 | flags |= fs_info->avail_system_alloc_bits; |
de98ced9 | 4060 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
0b246afa JM |
4061 | flags |= fs_info->avail_metadata_alloc_bits; |
4062 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
6fef8df1 | 4063 | |
2ff7e61e | 4064 | return btrfs_reduce_alloc_profile(fs_info, flags); |
6a63209f JB |
4065 | } |
4066 | ||
1b86826d | 4067 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
9ed74f2d | 4068 | { |
0b246afa | 4069 | struct btrfs_fs_info *fs_info = root->fs_info; |
b742bb82 | 4070 | u64 flags; |
53b381b3 | 4071 | u64 ret; |
9ed74f2d | 4072 | |
b742bb82 YZ |
4073 | if (data) |
4074 | flags = BTRFS_BLOCK_GROUP_DATA; | |
0b246afa | 4075 | else if (root == fs_info->chunk_root) |
b742bb82 | 4076 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
9ed74f2d | 4077 | else |
b742bb82 | 4078 | flags = BTRFS_BLOCK_GROUP_METADATA; |
9ed74f2d | 4079 | |
2ff7e61e | 4080 | ret = get_alloc_profile(fs_info, flags); |
53b381b3 | 4081 | return ret; |
6a63209f | 4082 | } |
9ed74f2d | 4083 | |
1b86826d JM |
4084 | u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info) |
4085 | { | |
4086 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA); | |
4087 | } | |
4088 | ||
4089 | u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info) | |
4090 | { | |
4091 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4092 | } | |
4093 | ||
4094 | u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info) | |
4095 | { | |
4096 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); | |
4097 | } | |
4098 | ||
4136135b LB |
4099 | static u64 btrfs_space_info_used(struct btrfs_space_info *s_info, |
4100 | bool may_use_included) | |
4101 | { | |
4102 | ASSERT(s_info); | |
4103 | return s_info->bytes_used + s_info->bytes_reserved + | |
4104 | s_info->bytes_pinned + s_info->bytes_readonly + | |
4105 | (may_use_included ? s_info->bytes_may_use : 0); | |
4106 | } | |
4107 | ||
04f4f916 | 4108 | int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes) |
6a63209f | 4109 | { |
04f4f916 | 4110 | struct btrfs_root *root = inode->root; |
b4d7c3c9 | 4111 | struct btrfs_fs_info *fs_info = root->fs_info; |
1174cade | 4112 | struct btrfs_space_info *data_sinfo = fs_info->data_sinfo; |
ab6e2410 | 4113 | u64 used; |
94b947b2 | 4114 | int ret = 0; |
c99f1b0c ZL |
4115 | int need_commit = 2; |
4116 | int have_pinned_space; | |
6a63209f | 4117 | |
6a63209f | 4118 | /* make sure bytes are sectorsize aligned */ |
0b246afa | 4119 | bytes = ALIGN(bytes, fs_info->sectorsize); |
6a63209f | 4120 | |
9dced186 | 4121 | if (btrfs_is_free_space_inode(inode)) { |
c99f1b0c | 4122 | need_commit = 0; |
9dced186 | 4123 | ASSERT(current->journal_info); |
0af3d00b JB |
4124 | } |
4125 | ||
6a63209f JB |
4126 | again: |
4127 | /* make sure we have enough space to handle the data first */ | |
4128 | spin_lock(&data_sinfo->lock); | |
4136135b | 4129 | used = btrfs_space_info_used(data_sinfo, true); |
ab6e2410 JB |
4130 | |
4131 | if (used + bytes > data_sinfo->total_bytes) { | |
4e06bdd6 | 4132 | struct btrfs_trans_handle *trans; |
9ed74f2d | 4133 | |
6a63209f JB |
4134 | /* |
4135 | * if we don't have enough free bytes in this space then we need | |
4136 | * to alloc a new chunk. | |
4137 | */ | |
b9fd47cd | 4138 | if (!data_sinfo->full) { |
6a63209f | 4139 | u64 alloc_target; |
9ed74f2d | 4140 | |
0e4f8f88 | 4141 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
6a63209f | 4142 | spin_unlock(&data_sinfo->lock); |
1174cade | 4143 | |
1b86826d | 4144 | alloc_target = btrfs_data_alloc_profile(fs_info); |
9dced186 MX |
4145 | /* |
4146 | * It is ugly that we don't call nolock join | |
4147 | * transaction for the free space inode case here. | |
4148 | * But it is safe because we only do the data space | |
4149 | * reservation for the free space cache in the | |
4150 | * transaction context, the common join transaction | |
4151 | * just increase the counter of the current transaction | |
4152 | * handler, doesn't try to acquire the trans_lock of | |
4153 | * the fs. | |
4154 | */ | |
7a7eaa40 | 4155 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4156 | if (IS_ERR(trans)) |
4157 | return PTR_ERR(trans); | |
9ed74f2d | 4158 | |
01458828 | 4159 | ret = do_chunk_alloc(trans, alloc_target, |
0e4f8f88 | 4160 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 4161 | btrfs_end_transaction(trans); |
d52a5b5f MX |
4162 | if (ret < 0) { |
4163 | if (ret != -ENOSPC) | |
4164 | return ret; | |
c99f1b0c ZL |
4165 | else { |
4166 | have_pinned_space = 1; | |
d52a5b5f | 4167 | goto commit_trans; |
c99f1b0c | 4168 | } |
d52a5b5f | 4169 | } |
9ed74f2d | 4170 | |
6a63209f JB |
4171 | goto again; |
4172 | } | |
f2bb8f5c JB |
4173 | |
4174 | /* | |
b150a4f1 | 4175 | * If we don't have enough pinned space to deal with this |
94b947b2 ZL |
4176 | * allocation, and no removed chunk in current transaction, |
4177 | * don't bother committing the transaction. | |
f2bb8f5c | 4178 | */ |
dec59fa3 | 4179 | have_pinned_space = __percpu_counter_compare( |
c99f1b0c | 4180 | &data_sinfo->total_bytes_pinned, |
dec59fa3 EL |
4181 | used + bytes - data_sinfo->total_bytes, |
4182 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
6a63209f | 4183 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4184 | |
4e06bdd6 | 4185 | /* commit the current transaction and try again */ |
d52a5b5f | 4186 | commit_trans: |
92e2f7e3 | 4187 | if (need_commit) { |
c99f1b0c | 4188 | need_commit--; |
b150a4f1 | 4189 | |
e1746e83 | 4190 | if (need_commit > 0) { |
82b3e53b | 4191 | btrfs_start_delalloc_roots(fs_info, -1); |
6374e57a | 4192 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, |
0b246afa | 4193 | (u64)-1); |
e1746e83 | 4194 | } |
9a4e7276 | 4195 | |
7a7eaa40 | 4196 | trans = btrfs_join_transaction(root); |
a22285a6 YZ |
4197 | if (IS_ERR(trans)) |
4198 | return PTR_ERR(trans); | |
c99f1b0c | 4199 | if (have_pinned_space >= 0 || |
3204d33c JB |
4200 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
4201 | &trans->transaction->flags) || | |
c99f1b0c | 4202 | need_commit > 0) { |
3a45bb20 | 4203 | ret = btrfs_commit_transaction(trans); |
94b947b2 ZL |
4204 | if (ret) |
4205 | return ret; | |
d7c15171 | 4206 | /* |
c2d6cb16 FM |
4207 | * The cleaner kthread might still be doing iput |
4208 | * operations. Wait for it to finish so that | |
4209 | * more space is released. | |
d7c15171 | 4210 | */ |
0b246afa JM |
4211 | mutex_lock(&fs_info->cleaner_delayed_iput_mutex); |
4212 | mutex_unlock(&fs_info->cleaner_delayed_iput_mutex); | |
94b947b2 ZL |
4213 | goto again; |
4214 | } else { | |
3a45bb20 | 4215 | btrfs_end_transaction(trans); |
94b947b2 | 4216 | } |
4e06bdd6 | 4217 | } |
9ed74f2d | 4218 | |
0b246afa | 4219 | trace_btrfs_space_reservation(fs_info, |
cab45e22 JM |
4220 | "space_info:enospc", |
4221 | data_sinfo->flags, bytes, 1); | |
6a63209f JB |
4222 | return -ENOSPC; |
4223 | } | |
4224 | data_sinfo->bytes_may_use += bytes; | |
0b246afa | 4225 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 4226 | data_sinfo->flags, bytes, 1); |
6a63209f | 4227 | spin_unlock(&data_sinfo->lock); |
6a63209f | 4228 | |
4559b0a7 | 4229 | return 0; |
9ed74f2d | 4230 | } |
6a63209f | 4231 | |
364ecf36 QW |
4232 | int btrfs_check_data_free_space(struct inode *inode, |
4233 | struct extent_changeset **reserved, u64 start, u64 len) | |
4ceff079 | 4234 | { |
0b246afa | 4235 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4236 | int ret; |
4237 | ||
4238 | /* align the range */ | |
0b246afa JM |
4239 | len = round_up(start + len, fs_info->sectorsize) - |
4240 | round_down(start, fs_info->sectorsize); | |
4241 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4242 | |
04f4f916 | 4243 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len); |
4ceff079 QW |
4244 | if (ret < 0) |
4245 | return ret; | |
4246 | ||
1e5ec2e7 | 4247 | /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */ |
364ecf36 | 4248 | ret = btrfs_qgroup_reserve_data(inode, reserved, start, len); |
7bc329c1 | 4249 | if (ret < 0) |
1e5ec2e7 | 4250 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
364ecf36 QW |
4251 | else |
4252 | ret = 0; | |
4ceff079 QW |
4253 | return ret; |
4254 | } | |
4255 | ||
4ceff079 QW |
4256 | /* |
4257 | * Called if we need to clear a data reservation for this inode | |
4258 | * Normally in a error case. | |
4259 | * | |
51773bec QW |
4260 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
4261 | * which we can't sleep and is sure it won't affect qgroup reserved space. | |
4262 | * Like clear_bit_hook(). | |
4ceff079 | 4263 | */ |
51773bec QW |
4264 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
4265 | u64 len) | |
4ceff079 | 4266 | { |
0b246afa | 4267 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
4ceff079 QW |
4268 | struct btrfs_space_info *data_sinfo; |
4269 | ||
4270 | /* Make sure the range is aligned to sectorsize */ | |
0b246afa JM |
4271 | len = round_up(start + len, fs_info->sectorsize) - |
4272 | round_down(start, fs_info->sectorsize); | |
4273 | start = round_down(start, fs_info->sectorsize); | |
4ceff079 | 4274 | |
0b246afa | 4275 | data_sinfo = fs_info->data_sinfo; |
4ceff079 QW |
4276 | spin_lock(&data_sinfo->lock); |
4277 | if (WARN_ON(data_sinfo->bytes_may_use < len)) | |
4278 | data_sinfo->bytes_may_use = 0; | |
4279 | else | |
4280 | data_sinfo->bytes_may_use -= len; | |
0b246afa | 4281 | trace_btrfs_space_reservation(fs_info, "space_info", |
4ceff079 QW |
4282 | data_sinfo->flags, len, 0); |
4283 | spin_unlock(&data_sinfo->lock); | |
4284 | } | |
4285 | ||
51773bec QW |
4286 | /* |
4287 | * Called if we need to clear a data reservation for this inode | |
4288 | * Normally in a error case. | |
4289 | * | |
01327610 | 4290 | * This one will handle the per-inode data rsv map for accurate reserved |
51773bec QW |
4291 | * space framework. |
4292 | */ | |
bc42bda2 QW |
4293 | void btrfs_free_reserved_data_space(struct inode *inode, |
4294 | struct extent_changeset *reserved, u64 start, u64 len) | |
51773bec | 4295 | { |
0c476a5d JM |
4296 | struct btrfs_root *root = BTRFS_I(inode)->root; |
4297 | ||
4298 | /* Make sure the range is aligned to sectorsize */ | |
da17066c JM |
4299 | len = round_up(start + len, root->fs_info->sectorsize) - |
4300 | round_down(start, root->fs_info->sectorsize); | |
4301 | start = round_down(start, root->fs_info->sectorsize); | |
0c476a5d | 4302 | |
51773bec | 4303 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
bc42bda2 | 4304 | btrfs_qgroup_free_data(inode, reserved, start, len); |
51773bec QW |
4305 | } |
4306 | ||
97e728d4 | 4307 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
e3ccfa98 | 4308 | { |
97e728d4 JB |
4309 | struct list_head *head = &info->space_info; |
4310 | struct btrfs_space_info *found; | |
e3ccfa98 | 4311 | |
97e728d4 JB |
4312 | rcu_read_lock(); |
4313 | list_for_each_entry_rcu(found, head, list) { | |
4314 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) | |
0e4f8f88 | 4315 | found->force_alloc = CHUNK_ALLOC_FORCE; |
e3ccfa98 | 4316 | } |
97e728d4 | 4317 | rcu_read_unlock(); |
e3ccfa98 JB |
4318 | } |
4319 | ||
3c76cd84 MX |
4320 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
4321 | { | |
4322 | return (global->size << 1); | |
4323 | } | |
4324 | ||
2ff7e61e | 4325 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, |
698d0082 | 4326 | struct btrfs_space_info *sinfo, int force) |
32c00aff | 4327 | { |
0b246afa | 4328 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
8d8aafee | 4329 | u64 bytes_used = btrfs_space_info_used(sinfo, false); |
e5bc2458 | 4330 | u64 thresh; |
e3ccfa98 | 4331 | |
0e4f8f88 CM |
4332 | if (force == CHUNK_ALLOC_FORCE) |
4333 | return 1; | |
4334 | ||
fb25e914 JB |
4335 | /* |
4336 | * We need to take into account the global rsv because for all intents | |
4337 | * and purposes it's used space. Don't worry about locking the | |
4338 | * global_rsv, it doesn't change except when the transaction commits. | |
4339 | */ | |
54338b5c | 4340 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
8d8aafee | 4341 | bytes_used += calc_global_rsv_need_space(global_rsv); |
fb25e914 | 4342 | |
0e4f8f88 CM |
4343 | /* |
4344 | * in limited mode, we want to have some free space up to | |
4345 | * about 1% of the FS size. | |
4346 | */ | |
4347 | if (force == CHUNK_ALLOC_LIMITED) { | |
0b246afa | 4348 | thresh = btrfs_super_total_bytes(fs_info->super_copy); |
ee22184b | 4349 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
0e4f8f88 | 4350 | |
8d8aafee | 4351 | if (sinfo->total_bytes - bytes_used < thresh) |
0e4f8f88 CM |
4352 | return 1; |
4353 | } | |
0e4f8f88 | 4354 | |
8d8aafee | 4355 | if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) |
14ed0ca6 | 4356 | return 0; |
424499db | 4357 | return 1; |
32c00aff JB |
4358 | } |
4359 | ||
2ff7e61e | 4360 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) |
15d1ff81 LB |
4361 | { |
4362 | u64 num_dev; | |
4363 | ||
53b381b3 DW |
4364 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
4365 | BTRFS_BLOCK_GROUP_RAID0 | | |
4366 | BTRFS_BLOCK_GROUP_RAID5 | | |
4367 | BTRFS_BLOCK_GROUP_RAID6)) | |
0b246afa | 4368 | num_dev = fs_info->fs_devices->rw_devices; |
15d1ff81 LB |
4369 | else if (type & BTRFS_BLOCK_GROUP_RAID1) |
4370 | num_dev = 2; | |
4371 | else | |
4372 | num_dev = 1; /* DUP or single */ | |
4373 | ||
39c2d7fa | 4374 | return num_dev; |
15d1ff81 LB |
4375 | } |
4376 | ||
39c2d7fa FM |
4377 | /* |
4378 | * If @is_allocation is true, reserve space in the system space info necessary | |
4379 | * for allocating a chunk, otherwise if it's false, reserve space necessary for | |
4380 | * removing a chunk. | |
4381 | */ | |
451a2c13 | 4382 | void check_system_chunk(struct btrfs_trans_handle *trans, u64 type) |
15d1ff81 | 4383 | { |
451a2c13 | 4384 | struct btrfs_fs_info *fs_info = trans->fs_info; |
15d1ff81 LB |
4385 | struct btrfs_space_info *info; |
4386 | u64 left; | |
4387 | u64 thresh; | |
4fbcdf66 | 4388 | int ret = 0; |
39c2d7fa | 4389 | u64 num_devs; |
4fbcdf66 FM |
4390 | |
4391 | /* | |
4392 | * Needed because we can end up allocating a system chunk and for an | |
4393 | * atomic and race free space reservation in the chunk block reserve. | |
4394 | */ | |
a32bf9a3 | 4395 | lockdep_assert_held(&fs_info->chunk_mutex); |
15d1ff81 | 4396 | |
0b246afa | 4397 | info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
15d1ff81 | 4398 | spin_lock(&info->lock); |
4136135b | 4399 | left = info->total_bytes - btrfs_space_info_used(info, true); |
15d1ff81 LB |
4400 | spin_unlock(&info->lock); |
4401 | ||
2ff7e61e | 4402 | num_devs = get_profile_num_devs(fs_info, type); |
39c2d7fa FM |
4403 | |
4404 | /* num_devs device items to update and 1 chunk item to add or remove */ | |
0b246afa JM |
4405 | thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) + |
4406 | btrfs_calc_trans_metadata_size(fs_info, 1); | |
39c2d7fa | 4407 | |
0b246afa JM |
4408 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
4409 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", | |
4410 | left, thresh, type); | |
4411 | dump_space_info(fs_info, info, 0, 0); | |
15d1ff81 LB |
4412 | } |
4413 | ||
4414 | if (left < thresh) { | |
1b86826d | 4415 | u64 flags = btrfs_system_alloc_profile(fs_info); |
15d1ff81 | 4416 | |
4fbcdf66 FM |
4417 | /* |
4418 | * Ignore failure to create system chunk. We might end up not | |
4419 | * needing it, as we might not need to COW all nodes/leafs from | |
4420 | * the paths we visit in the chunk tree (they were already COWed | |
4421 | * or created in the current transaction for example). | |
4422 | */ | |
c216b203 | 4423 | ret = btrfs_alloc_chunk(trans, flags); |
4fbcdf66 FM |
4424 | } |
4425 | ||
4426 | if (!ret) { | |
0b246afa JM |
4427 | ret = btrfs_block_rsv_add(fs_info->chunk_root, |
4428 | &fs_info->chunk_block_rsv, | |
4fbcdf66 FM |
4429 | thresh, BTRFS_RESERVE_NO_FLUSH); |
4430 | if (!ret) | |
4431 | trans->chunk_bytes_reserved += thresh; | |
15d1ff81 LB |
4432 | } |
4433 | } | |
4434 | ||
28b737f6 LB |
4435 | /* |
4436 | * If force is CHUNK_ALLOC_FORCE: | |
4437 | * - return 1 if it successfully allocates a chunk, | |
4438 | * - return errors including -ENOSPC otherwise. | |
4439 | * If force is NOT CHUNK_ALLOC_FORCE: | |
4440 | * - return 0 if it doesn't need to allocate a new chunk, | |
4441 | * - return 1 if it successfully allocates a chunk, | |
4442 | * - return errors including -ENOSPC otherwise. | |
4443 | */ | |
01458828 NB |
4444 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, u64 flags, |
4445 | int force) | |
9ed74f2d | 4446 | { |
01458828 | 4447 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6324fbf3 | 4448 | struct btrfs_space_info *space_info; |
2556fbb0 NB |
4449 | bool wait_for_alloc = false; |
4450 | bool should_alloc = false; | |
9ed74f2d | 4451 | int ret = 0; |
9ed74f2d | 4452 | |
c6b305a8 JB |
4453 | /* Don't re-enter if we're already allocating a chunk */ |
4454 | if (trans->allocating_chunk) | |
4455 | return -ENOSPC; | |
4456 | ||
0b246afa | 4457 | space_info = __find_space_info(fs_info, flags); |
dc2d3005 | 4458 | ASSERT(space_info); |
9ed74f2d | 4459 | |
2556fbb0 NB |
4460 | do { |
4461 | spin_lock(&space_info->lock); | |
4462 | if (force < space_info->force_alloc) | |
4463 | force = space_info->force_alloc; | |
4464 | should_alloc = should_alloc_chunk(fs_info, space_info, force); | |
4465 | if (space_info->full) { | |
4466 | /* No more free physical space */ | |
4467 | if (should_alloc) | |
4468 | ret = -ENOSPC; | |
4469 | else | |
4470 | ret = 0; | |
4471 | spin_unlock(&space_info->lock); | |
4472 | return ret; | |
4473 | } else if (!should_alloc) { | |
4474 | spin_unlock(&space_info->lock); | |
4475 | return 0; | |
4476 | } else if (space_info->chunk_alloc) { | |
4477 | /* | |
4478 | * Someone is already allocating, so we need to block | |
4479 | * until this someone is finished and then loop to | |
4480 | * recheck if we should continue with our allocation | |
4481 | * attempt. | |
4482 | */ | |
4483 | wait_for_alloc = true; | |
4484 | spin_unlock(&space_info->lock); | |
4485 | mutex_lock(&fs_info->chunk_mutex); | |
4486 | mutex_unlock(&fs_info->chunk_mutex); | |
4487 | } else { | |
4488 | /* Proceed with allocation */ | |
4489 | space_info->chunk_alloc = 1; | |
4490 | wait_for_alloc = false; | |
4491 | spin_unlock(&space_info->lock); | |
4492 | } | |
6d74119f | 4493 | |
1e1c50a9 | 4494 | cond_resched(); |
2556fbb0 | 4495 | } while (wait_for_alloc); |
6d74119f | 4496 | |
2556fbb0 | 4497 | mutex_lock(&fs_info->chunk_mutex); |
c6b305a8 JB |
4498 | trans->allocating_chunk = true; |
4499 | ||
67377734 JB |
4500 | /* |
4501 | * If we have mixed data/metadata chunks we want to make sure we keep | |
4502 | * allocating mixed chunks instead of individual chunks. | |
4503 | */ | |
4504 | if (btrfs_mixed_space_info(space_info)) | |
4505 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); | |
4506 | ||
97e728d4 JB |
4507 | /* |
4508 | * if we're doing a data chunk, go ahead and make sure that | |
4509 | * we keep a reasonable number of metadata chunks allocated in the | |
4510 | * FS as well. | |
4511 | */ | |
9ed74f2d | 4512 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
97e728d4 JB |
4513 | fs_info->data_chunk_allocations++; |
4514 | if (!(fs_info->data_chunk_allocations % | |
4515 | fs_info->metadata_ratio)) | |
4516 | force_metadata_allocation(fs_info); | |
9ed74f2d JB |
4517 | } |
4518 | ||
15d1ff81 LB |
4519 | /* |
4520 | * Check if we have enough space in SYSTEM chunk because we may need | |
4521 | * to update devices. | |
4522 | */ | |
451a2c13 | 4523 | check_system_chunk(trans, flags); |
15d1ff81 | 4524 | |
c216b203 | 4525 | ret = btrfs_alloc_chunk(trans, flags); |
c6b305a8 | 4526 | trans->allocating_chunk = false; |
92b8e897 | 4527 | |
9ed74f2d | 4528 | spin_lock(&space_info->lock); |
57f1642e NB |
4529 | if (ret < 0) { |
4530 | if (ret == -ENOSPC) | |
4531 | space_info->full = 1; | |
4532 | else | |
4533 | goto out; | |
4534 | } else { | |
424499db | 4535 | ret = 1; |
57f1642e | 4536 | } |
6d74119f | 4537 | |
0e4f8f88 | 4538 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
a81cb9a2 | 4539 | out: |
6d74119f | 4540 | space_info->chunk_alloc = 0; |
9ed74f2d | 4541 | spin_unlock(&space_info->lock); |
a25c75d5 | 4542 | mutex_unlock(&fs_info->chunk_mutex); |
00d80e34 FM |
4543 | /* |
4544 | * When we allocate a new chunk we reserve space in the chunk block | |
4545 | * reserve to make sure we can COW nodes/leafs in the chunk tree or | |
4546 | * add new nodes/leafs to it if we end up needing to do it when | |
4547 | * inserting the chunk item and updating device items as part of the | |
4548 | * second phase of chunk allocation, performed by | |
4549 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a | |
4550 | * large number of new block groups to create in our transaction | |
4551 | * handle's new_bgs list to avoid exhausting the chunk block reserve | |
4552 | * in extreme cases - like having a single transaction create many new | |
4553 | * block groups when starting to write out the free space caches of all | |
4554 | * the block groups that were made dirty during the lifetime of the | |
4555 | * transaction. | |
4556 | */ | |
d9a0540a | 4557 | if (trans->can_flush_pending_bgs && |
ee22184b | 4558 | trans->chunk_bytes_reserved >= (u64)SZ_2M) { |
6c686b35 | 4559 | btrfs_create_pending_block_groups(trans); |
00d80e34 FM |
4560 | btrfs_trans_release_chunk_metadata(trans); |
4561 | } | |
0f9dd46c | 4562 | return ret; |
6324fbf3 | 4563 | } |
9ed74f2d | 4564 | |
c1c4919b | 4565 | static int can_overcommit(struct btrfs_fs_info *fs_info, |
a80c8dcf | 4566 | struct btrfs_space_info *space_info, u64 bytes, |
c1c4919b JM |
4567 | enum btrfs_reserve_flush_enum flush, |
4568 | bool system_chunk) | |
a80c8dcf | 4569 | { |
0b246afa | 4570 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
957780eb | 4571 | u64 profile; |
3c76cd84 | 4572 | u64 space_size; |
a80c8dcf JB |
4573 | u64 avail; |
4574 | u64 used; | |
46df06b8 | 4575 | int factor; |
a80c8dcf | 4576 | |
957780eb JB |
4577 | /* Don't overcommit when in mixed mode. */ |
4578 | if (space_info->flags & BTRFS_BLOCK_GROUP_DATA) | |
4579 | return 0; | |
4580 | ||
c1c4919b JM |
4581 | if (system_chunk) |
4582 | profile = btrfs_system_alloc_profile(fs_info); | |
4583 | else | |
4584 | profile = btrfs_metadata_alloc_profile(fs_info); | |
4585 | ||
4136135b | 4586 | used = btrfs_space_info_used(space_info, false); |
96f1bb57 | 4587 | |
96f1bb57 JB |
4588 | /* |
4589 | * We only want to allow over committing if we have lots of actual space | |
4590 | * free, but if we don't have enough space to handle the global reserve | |
4591 | * space then we could end up having a real enospc problem when trying | |
4592 | * to allocate a chunk or some other such important allocation. | |
4593 | */ | |
3c76cd84 MX |
4594 | spin_lock(&global_rsv->lock); |
4595 | space_size = calc_global_rsv_need_space(global_rsv); | |
4596 | spin_unlock(&global_rsv->lock); | |
4597 | if (used + space_size >= space_info->total_bytes) | |
96f1bb57 JB |
4598 | return 0; |
4599 | ||
4600 | used += space_info->bytes_may_use; | |
a80c8dcf | 4601 | |
a5ed45f8 | 4602 | avail = atomic64_read(&fs_info->free_chunk_space); |
a80c8dcf JB |
4603 | |
4604 | /* | |
4605 | * If we have dup, raid1 or raid10 then only half of the free | |
53b381b3 DW |
4606 | * space is actually useable. For raid56, the space info used |
4607 | * doesn't include the parity drive, so we don't have to | |
4608 | * change the math | |
a80c8dcf | 4609 | */ |
46df06b8 DS |
4610 | factor = btrfs_bg_type_to_factor(profile); |
4611 | avail = div_u64(avail, factor); | |
a80c8dcf JB |
4612 | |
4613 | /* | |
561c294d MX |
4614 | * If we aren't flushing all things, let us overcommit up to |
4615 | * 1/2th of the space. If we can flush, don't let us overcommit | |
4616 | * too much, let it overcommit up to 1/8 of the space. | |
a80c8dcf | 4617 | */ |
08e007d2 | 4618 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
14575aef | 4619 | avail >>= 3; |
a80c8dcf | 4620 | else |
14575aef | 4621 | avail >>= 1; |
a80c8dcf | 4622 | |
14575aef | 4623 | if (used + bytes < space_info->total_bytes + avail) |
a80c8dcf JB |
4624 | return 1; |
4625 | return 0; | |
4626 | } | |
4627 | ||
2ff7e61e | 4628 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info, |
6c255e67 | 4629 | unsigned long nr_pages, int nr_items) |
da633a42 | 4630 | { |
0b246afa | 4631 | struct super_block *sb = fs_info->sb; |
da633a42 | 4632 | |
925a6efb JB |
4633 | if (down_read_trylock(&sb->s_umount)) { |
4634 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); | |
4635 | up_read(&sb->s_umount); | |
4636 | } else { | |
da633a42 MX |
4637 | /* |
4638 | * We needn't worry the filesystem going from r/w to r/o though | |
4639 | * we don't acquire ->s_umount mutex, because the filesystem | |
4640 | * should guarantee the delalloc inodes list be empty after | |
4641 | * the filesystem is readonly(all dirty pages are written to | |
4642 | * the disk). | |
4643 | */ | |
82b3e53b | 4644 | btrfs_start_delalloc_roots(fs_info, nr_items); |
98ad69cf | 4645 | if (!current->journal_info) |
0b246afa | 4646 | btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1); |
da633a42 MX |
4647 | } |
4648 | } | |
4649 | ||
6374e57a | 4650 | static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info, |
2ff7e61e | 4651 | u64 to_reclaim) |
18cd8ea6 MX |
4652 | { |
4653 | u64 bytes; | |
6374e57a | 4654 | u64 nr; |
18cd8ea6 | 4655 | |
2ff7e61e | 4656 | bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
6374e57a | 4657 | nr = div64_u64(to_reclaim, bytes); |
18cd8ea6 MX |
4658 | if (!nr) |
4659 | nr = 1; | |
4660 | return nr; | |
4661 | } | |
4662 | ||
ee22184b | 4663 | #define EXTENT_SIZE_PER_ITEM SZ_256K |
c61a16a7 | 4664 | |
9ed74f2d | 4665 | /* |
5da9d01b | 4666 | * shrink metadata reservation for delalloc |
9ed74f2d | 4667 | */ |
c1c4919b JM |
4668 | static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim, |
4669 | u64 orig, bool wait_ordered) | |
5da9d01b | 4670 | { |
0019f10d | 4671 | struct btrfs_space_info *space_info; |
663350ac | 4672 | struct btrfs_trans_handle *trans; |
f4c738c2 | 4673 | u64 delalloc_bytes; |
5da9d01b | 4674 | u64 max_reclaim; |
6374e57a | 4675 | u64 items; |
b1953bce | 4676 | long time_left; |
d3ee29e3 MX |
4677 | unsigned long nr_pages; |
4678 | int loops; | |
5da9d01b | 4679 | |
c61a16a7 | 4680 | /* Calc the number of the pages we need flush for space reservation */ |
2ff7e61e | 4681 | items = calc_reclaim_items_nr(fs_info, to_reclaim); |
6374e57a | 4682 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; |
c61a16a7 | 4683 | |
663350ac | 4684 | trans = (struct btrfs_trans_handle *)current->journal_info; |
69fe2d75 | 4685 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
bf9022e0 | 4686 | |
963d678b | 4687 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4688 | &fs_info->delalloc_bytes); |
f4c738c2 | 4689 | if (delalloc_bytes == 0) { |
fdb5effd | 4690 | if (trans) |
f4c738c2 | 4691 | return; |
38c135af | 4692 | if (wait_ordered) |
0b246afa | 4693 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f4c738c2 | 4694 | return; |
fdb5effd JB |
4695 | } |
4696 | ||
d3ee29e3 | 4697 | loops = 0; |
f4c738c2 JB |
4698 | while (delalloc_bytes && loops < 3) { |
4699 | max_reclaim = min(delalloc_bytes, to_reclaim); | |
09cbfeaf | 4700 | nr_pages = max_reclaim >> PAGE_SHIFT; |
2ff7e61e | 4701 | btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items); |
dea31f52 JB |
4702 | /* |
4703 | * We need to wait for the async pages to actually start before | |
4704 | * we do anything. | |
4705 | */ | |
0b246afa | 4706 | max_reclaim = atomic_read(&fs_info->async_delalloc_pages); |
9f3a074d MX |
4707 | if (!max_reclaim) |
4708 | goto skip_async; | |
4709 | ||
4710 | if (max_reclaim <= nr_pages) | |
4711 | max_reclaim = 0; | |
4712 | else | |
4713 | max_reclaim -= nr_pages; | |
dea31f52 | 4714 | |
0b246afa JM |
4715 | wait_event(fs_info->async_submit_wait, |
4716 | atomic_read(&fs_info->async_delalloc_pages) <= | |
9f3a074d MX |
4717 | (int)max_reclaim); |
4718 | skip_async: | |
0019f10d | 4719 | spin_lock(&space_info->lock); |
957780eb JB |
4720 | if (list_empty(&space_info->tickets) && |
4721 | list_empty(&space_info->priority_tickets)) { | |
4722 | spin_unlock(&space_info->lock); | |
4723 | break; | |
4724 | } | |
0019f10d | 4725 | spin_unlock(&space_info->lock); |
5da9d01b | 4726 | |
36e39c40 | 4727 | loops++; |
f104d044 | 4728 | if (wait_ordered && !trans) { |
0b246afa | 4729 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
f104d044 | 4730 | } else { |
f4c738c2 | 4731 | time_left = schedule_timeout_killable(1); |
f104d044 JB |
4732 | if (time_left) |
4733 | break; | |
4734 | } | |
963d678b | 4735 | delalloc_bytes = percpu_counter_sum_positive( |
0b246afa | 4736 | &fs_info->delalloc_bytes); |
5da9d01b | 4737 | } |
5da9d01b YZ |
4738 | } |
4739 | ||
996478ca JB |
4740 | struct reserve_ticket { |
4741 | u64 bytes; | |
4742 | int error; | |
4743 | struct list_head list; | |
4744 | wait_queue_head_t wait; | |
4745 | }; | |
4746 | ||
663350ac JB |
4747 | /** |
4748 | * maybe_commit_transaction - possibly commit the transaction if its ok to | |
4749 | * @root - the root we're allocating for | |
4750 | * @bytes - the number of bytes we want to reserve | |
4751 | * @force - force the commit | |
8bb8ab2e | 4752 | * |
663350ac JB |
4753 | * This will check to make sure that committing the transaction will actually |
4754 | * get us somewhere and then commit the transaction if it does. Otherwise it | |
4755 | * will return -ENOSPC. | |
8bb8ab2e | 4756 | */ |
0c9ab349 | 4757 | static int may_commit_transaction(struct btrfs_fs_info *fs_info, |
996478ca | 4758 | struct btrfs_space_info *space_info) |
663350ac | 4759 | { |
996478ca | 4760 | struct reserve_ticket *ticket = NULL; |
0b246afa | 4761 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv; |
663350ac | 4762 | struct btrfs_trans_handle *trans; |
996478ca | 4763 | u64 bytes; |
663350ac JB |
4764 | |
4765 | trans = (struct btrfs_trans_handle *)current->journal_info; | |
4766 | if (trans) | |
4767 | return -EAGAIN; | |
4768 | ||
996478ca JB |
4769 | spin_lock(&space_info->lock); |
4770 | if (!list_empty(&space_info->priority_tickets)) | |
4771 | ticket = list_first_entry(&space_info->priority_tickets, | |
4772 | struct reserve_ticket, list); | |
4773 | else if (!list_empty(&space_info->tickets)) | |
4774 | ticket = list_first_entry(&space_info->tickets, | |
4775 | struct reserve_ticket, list); | |
4776 | bytes = (ticket) ? ticket->bytes : 0; | |
4777 | spin_unlock(&space_info->lock); | |
4778 | ||
4779 | if (!bytes) | |
4780 | return 0; | |
663350ac JB |
4781 | |
4782 | /* See if there is enough pinned space to make this reservation */ | |
dec59fa3 EL |
4783 | if (__percpu_counter_compare(&space_info->total_bytes_pinned, |
4784 | bytes, | |
4785 | BTRFS_TOTAL_BYTES_PINNED_BATCH) >= 0) | |
663350ac | 4786 | goto commit; |
663350ac JB |
4787 | |
4788 | /* | |
4789 | * See if there is some space in the delayed insertion reservation for | |
4790 | * this reservation. | |
4791 | */ | |
4792 | if (space_info != delayed_rsv->space_info) | |
4793 | return -ENOSPC; | |
4794 | ||
4795 | spin_lock(&delayed_rsv->lock); | |
996478ca JB |
4796 | if (delayed_rsv->size > bytes) |
4797 | bytes = 0; | |
4798 | else | |
4799 | bytes -= delayed_rsv->size; | |
057aac3e NB |
4800 | spin_unlock(&delayed_rsv->lock); |
4801 | ||
dec59fa3 EL |
4802 | if (__percpu_counter_compare(&space_info->total_bytes_pinned, |
4803 | bytes, | |
4804 | BTRFS_TOTAL_BYTES_PINNED_BATCH) < 0) { | |
663350ac JB |
4805 | return -ENOSPC; |
4806 | } | |
663350ac JB |
4807 | |
4808 | commit: | |
a9b3311e | 4809 | trans = btrfs_join_transaction(fs_info->extent_root); |
663350ac JB |
4810 | if (IS_ERR(trans)) |
4811 | return -ENOSPC; | |
4812 | ||
3a45bb20 | 4813 | return btrfs_commit_transaction(trans); |
663350ac JB |
4814 | } |
4815 | ||
e38ae7a0 NB |
4816 | /* |
4817 | * Try to flush some data based on policy set by @state. This is only advisory | |
4818 | * and may fail for various reasons. The caller is supposed to examine the | |
4819 | * state of @space_info to detect the outcome. | |
4820 | */ | |
4821 | static void flush_space(struct btrfs_fs_info *fs_info, | |
96c3f433 | 4822 | struct btrfs_space_info *space_info, u64 num_bytes, |
7bdd6277 | 4823 | int state) |
96c3f433 | 4824 | { |
a9b3311e | 4825 | struct btrfs_root *root = fs_info->extent_root; |
96c3f433 JB |
4826 | struct btrfs_trans_handle *trans; |
4827 | int nr; | |
f4c738c2 | 4828 | int ret = 0; |
96c3f433 JB |
4829 | |
4830 | switch (state) { | |
96c3f433 JB |
4831 | case FLUSH_DELAYED_ITEMS_NR: |
4832 | case FLUSH_DELAYED_ITEMS: | |
18cd8ea6 | 4833 | if (state == FLUSH_DELAYED_ITEMS_NR) |
2ff7e61e | 4834 | nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2; |
18cd8ea6 | 4835 | else |
96c3f433 | 4836 | nr = -1; |
18cd8ea6 | 4837 | |
96c3f433 JB |
4838 | trans = btrfs_join_transaction(root); |
4839 | if (IS_ERR(trans)) { | |
4840 | ret = PTR_ERR(trans); | |
4841 | break; | |
4842 | } | |
e5c304e6 | 4843 | ret = btrfs_run_delayed_items_nr(trans, nr); |
3a45bb20 | 4844 | btrfs_end_transaction(trans); |
96c3f433 | 4845 | break; |
67b0fd63 JB |
4846 | case FLUSH_DELALLOC: |
4847 | case FLUSH_DELALLOC_WAIT: | |
7bdd6277 | 4848 | shrink_delalloc(fs_info, num_bytes * 2, num_bytes, |
67b0fd63 JB |
4849 | state == FLUSH_DELALLOC_WAIT); |
4850 | break; | |
ea658bad JB |
4851 | case ALLOC_CHUNK: |
4852 | trans = btrfs_join_transaction(root); | |
4853 | if (IS_ERR(trans)) { | |
4854 | ret = PTR_ERR(trans); | |
4855 | break; | |
4856 | } | |
01458828 | 4857 | ret = do_chunk_alloc(trans, |
1b86826d | 4858 | btrfs_metadata_alloc_profile(fs_info), |
ea658bad | 4859 | CHUNK_ALLOC_NO_FORCE); |
3a45bb20 | 4860 | btrfs_end_transaction(trans); |
eecba891 | 4861 | if (ret > 0 || ret == -ENOSPC) |
ea658bad JB |
4862 | ret = 0; |
4863 | break; | |
96c3f433 | 4864 | case COMMIT_TRANS: |
996478ca | 4865 | ret = may_commit_transaction(fs_info, space_info); |
96c3f433 JB |
4866 | break; |
4867 | default: | |
4868 | ret = -ENOSPC; | |
4869 | break; | |
4870 | } | |
4871 | ||
7bdd6277 NB |
4872 | trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state, |
4873 | ret); | |
e38ae7a0 | 4874 | return; |
96c3f433 | 4875 | } |
21c7e756 MX |
4876 | |
4877 | static inline u64 | |
c1c4919b JM |
4878 | btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info, |
4879 | struct btrfs_space_info *space_info, | |
4880 | bool system_chunk) | |
21c7e756 | 4881 | { |
957780eb | 4882 | struct reserve_ticket *ticket; |
21c7e756 MX |
4883 | u64 used; |
4884 | u64 expected; | |
957780eb | 4885 | u64 to_reclaim = 0; |
21c7e756 | 4886 | |
957780eb JB |
4887 | list_for_each_entry(ticket, &space_info->tickets, list) |
4888 | to_reclaim += ticket->bytes; | |
4889 | list_for_each_entry(ticket, &space_info->priority_tickets, list) | |
4890 | to_reclaim += ticket->bytes; | |
4891 | if (to_reclaim) | |
4892 | return to_reclaim; | |
21c7e756 | 4893 | |
e0af2484 | 4894 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); |
c1c4919b JM |
4895 | if (can_overcommit(fs_info, space_info, to_reclaim, |
4896 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
e0af2484 WX |
4897 | return 0; |
4898 | ||
0eee8a49 NB |
4899 | used = btrfs_space_info_used(space_info, true); |
4900 | ||
c1c4919b JM |
4901 | if (can_overcommit(fs_info, space_info, SZ_1M, |
4902 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) | |
21c7e756 MX |
4903 | expected = div_factor_fine(space_info->total_bytes, 95); |
4904 | else | |
4905 | expected = div_factor_fine(space_info->total_bytes, 90); | |
4906 | ||
4907 | if (used > expected) | |
4908 | to_reclaim = used - expected; | |
4909 | else | |
4910 | to_reclaim = 0; | |
4911 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + | |
4912 | space_info->bytes_reserved); | |
21c7e756 MX |
4913 | return to_reclaim; |
4914 | } | |
4915 | ||
c1c4919b JM |
4916 | static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info, |
4917 | struct btrfs_space_info *space_info, | |
4918 | u64 used, bool system_chunk) | |
21c7e756 | 4919 | { |
365c5313 JB |
4920 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
4921 | ||
4922 | /* If we're just plain full then async reclaim just slows us down. */ | |
baee8790 | 4923 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) |
365c5313 JB |
4924 | return 0; |
4925 | ||
c1c4919b JM |
4926 | if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
4927 | system_chunk)) | |
d38b349c JB |
4928 | return 0; |
4929 | ||
0b246afa JM |
4930 | return (used >= thresh && !btrfs_fs_closing(fs_info) && |
4931 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); | |
21c7e756 MX |
4932 | } |
4933 | ||
957780eb | 4934 | static void wake_all_tickets(struct list_head *head) |
21c7e756 | 4935 | { |
957780eb | 4936 | struct reserve_ticket *ticket; |
25ce459c | 4937 | |
957780eb JB |
4938 | while (!list_empty(head)) { |
4939 | ticket = list_first_entry(head, struct reserve_ticket, list); | |
4940 | list_del_init(&ticket->list); | |
4941 | ticket->error = -ENOSPC; | |
4942 | wake_up(&ticket->wait); | |
21c7e756 | 4943 | } |
21c7e756 MX |
4944 | } |
4945 | ||
957780eb JB |
4946 | /* |
4947 | * This is for normal flushers, we can wait all goddamned day if we want to. We | |
4948 | * will loop and continuously try to flush as long as we are making progress. | |
4949 | * We count progress as clearing off tickets each time we have to loop. | |
4950 | */ | |
21c7e756 MX |
4951 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) |
4952 | { | |
4953 | struct btrfs_fs_info *fs_info; | |
4954 | struct btrfs_space_info *space_info; | |
4955 | u64 to_reclaim; | |
4956 | int flush_state; | |
957780eb | 4957 | int commit_cycles = 0; |
ce129655 | 4958 | u64 last_tickets_id; |
21c7e756 MX |
4959 | |
4960 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); | |
4961 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); | |
4962 | ||
957780eb | 4963 | spin_lock(&space_info->lock); |
c1c4919b JM |
4964 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
4965 | false); | |
957780eb JB |
4966 | if (!to_reclaim) { |
4967 | space_info->flush = 0; | |
4968 | spin_unlock(&space_info->lock); | |
21c7e756 | 4969 | return; |
957780eb | 4970 | } |
ce129655 | 4971 | last_tickets_id = space_info->tickets_id; |
957780eb | 4972 | spin_unlock(&space_info->lock); |
21c7e756 MX |
4973 | |
4974 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
957780eb | 4975 | do { |
e38ae7a0 | 4976 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
957780eb JB |
4977 | spin_lock(&space_info->lock); |
4978 | if (list_empty(&space_info->tickets)) { | |
4979 | space_info->flush = 0; | |
4980 | spin_unlock(&space_info->lock); | |
4981 | return; | |
4982 | } | |
c1c4919b JM |
4983 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, |
4984 | space_info, | |
4985 | false); | |
ce129655 | 4986 | if (last_tickets_id == space_info->tickets_id) { |
957780eb JB |
4987 | flush_state++; |
4988 | } else { | |
ce129655 | 4989 | last_tickets_id = space_info->tickets_id; |
957780eb JB |
4990 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
4991 | if (commit_cycles) | |
4992 | commit_cycles--; | |
4993 | } | |
4994 | ||
4995 | if (flush_state > COMMIT_TRANS) { | |
4996 | commit_cycles++; | |
4997 | if (commit_cycles > 2) { | |
4998 | wake_all_tickets(&space_info->tickets); | |
4999 | space_info->flush = 0; | |
5000 | } else { | |
5001 | flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5002 | } | |
5003 | } | |
5004 | spin_unlock(&space_info->lock); | |
5005 | } while (flush_state <= COMMIT_TRANS); | |
5006 | } | |
5007 | ||
5008 | void btrfs_init_async_reclaim_work(struct work_struct *work) | |
5009 | { | |
5010 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); | |
5011 | } | |
5012 | ||
5013 | static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, | |
5014 | struct btrfs_space_info *space_info, | |
5015 | struct reserve_ticket *ticket) | |
5016 | { | |
5017 | u64 to_reclaim; | |
5018 | int flush_state = FLUSH_DELAYED_ITEMS_NR; | |
5019 | ||
5020 | spin_lock(&space_info->lock); | |
c1c4919b JM |
5021 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
5022 | false); | |
957780eb JB |
5023 | if (!to_reclaim) { |
5024 | spin_unlock(&space_info->lock); | |
5025 | return; | |
5026 | } | |
5027 | spin_unlock(&space_info->lock); | |
5028 | ||
21c7e756 | 5029 | do { |
7bdd6277 | 5030 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
21c7e756 | 5031 | flush_state++; |
957780eb JB |
5032 | spin_lock(&space_info->lock); |
5033 | if (ticket->bytes == 0) { | |
5034 | spin_unlock(&space_info->lock); | |
21c7e756 | 5035 | return; |
957780eb JB |
5036 | } |
5037 | spin_unlock(&space_info->lock); | |
5038 | ||
5039 | /* | |
5040 | * Priority flushers can't wait on delalloc without | |
5041 | * deadlocking. | |
5042 | */ | |
5043 | if (flush_state == FLUSH_DELALLOC || | |
5044 | flush_state == FLUSH_DELALLOC_WAIT) | |
5045 | flush_state = ALLOC_CHUNK; | |
365c5313 | 5046 | } while (flush_state < COMMIT_TRANS); |
21c7e756 MX |
5047 | } |
5048 | ||
957780eb JB |
5049 | static int wait_reserve_ticket(struct btrfs_fs_info *fs_info, |
5050 | struct btrfs_space_info *space_info, | |
5051 | struct reserve_ticket *ticket, u64 orig_bytes) | |
5052 | ||
21c7e756 | 5053 | { |
957780eb JB |
5054 | DEFINE_WAIT(wait); |
5055 | int ret = 0; | |
5056 | ||
5057 | spin_lock(&space_info->lock); | |
5058 | while (ticket->bytes > 0 && ticket->error == 0) { | |
5059 | ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE); | |
5060 | if (ret) { | |
5061 | ret = -EINTR; | |
5062 | break; | |
5063 | } | |
5064 | spin_unlock(&space_info->lock); | |
5065 | ||
5066 | schedule(); | |
5067 | ||
5068 | finish_wait(&ticket->wait, &wait); | |
5069 | spin_lock(&space_info->lock); | |
5070 | } | |
5071 | if (!ret) | |
5072 | ret = ticket->error; | |
5073 | if (!list_empty(&ticket->list)) | |
5074 | list_del_init(&ticket->list); | |
5075 | if (ticket->bytes && ticket->bytes < orig_bytes) { | |
5076 | u64 num_bytes = orig_bytes - ticket->bytes; | |
5077 | space_info->bytes_may_use -= num_bytes; | |
5078 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5079 | space_info->flags, num_bytes, 0); | |
5080 | } | |
5081 | spin_unlock(&space_info->lock); | |
5082 | ||
5083 | return ret; | |
21c7e756 MX |
5084 | } |
5085 | ||
4a92b1b8 JB |
5086 | /** |
5087 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5088 | * @root - the root we're allocating for | |
957780eb | 5089 | * @space_info - the space info we want to allocate from |
4a92b1b8 | 5090 | * @orig_bytes - the number of bytes we want |
48fc7f7e | 5091 | * @flush - whether or not we can flush to make our reservation |
8bb8ab2e | 5092 | * |
01327610 | 5093 | * This will reserve orig_bytes number of bytes from the space info associated |
4a92b1b8 JB |
5094 | * with the block_rsv. If there is not enough space it will make an attempt to |
5095 | * flush out space to make room. It will do this by flushing delalloc if | |
5096 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5097 | * regain reservations will be made and this will fail if there is not enough | |
5098 | * space already. | |
8bb8ab2e | 5099 | */ |
c1c4919b | 5100 | static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info, |
957780eb JB |
5101 | struct btrfs_space_info *space_info, |
5102 | u64 orig_bytes, | |
c1c4919b JM |
5103 | enum btrfs_reserve_flush_enum flush, |
5104 | bool system_chunk) | |
9ed74f2d | 5105 | { |
957780eb | 5106 | struct reserve_ticket ticket; |
2bf64758 | 5107 | u64 used; |
8bb8ab2e | 5108 | int ret = 0; |
9ed74f2d | 5109 | |
957780eb | 5110 | ASSERT(orig_bytes); |
8ca17f0f | 5111 | ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL); |
fdb5effd | 5112 | |
8bb8ab2e | 5113 | spin_lock(&space_info->lock); |
fdb5effd | 5114 | ret = -ENOSPC; |
4136135b | 5115 | used = btrfs_space_info_used(space_info, true); |
9ed74f2d | 5116 | |
8bb8ab2e | 5117 | /* |
957780eb JB |
5118 | * If we have enough space then hooray, make our reservation and carry |
5119 | * on. If not see if we can overcommit, and if we can, hooray carry on. | |
5120 | * If not things get more complicated. | |
8bb8ab2e | 5121 | */ |
957780eb JB |
5122 | if (used + orig_bytes <= space_info->total_bytes) { |
5123 | space_info->bytes_may_use += orig_bytes; | |
0b246afa JM |
5124 | trace_btrfs_space_reservation(fs_info, "space_info", |
5125 | space_info->flags, orig_bytes, 1); | |
957780eb | 5126 | ret = 0; |
c1c4919b JM |
5127 | } else if (can_overcommit(fs_info, space_info, orig_bytes, flush, |
5128 | system_chunk)) { | |
44734ed1 | 5129 | space_info->bytes_may_use += orig_bytes; |
0b246afa JM |
5130 | trace_btrfs_space_reservation(fs_info, "space_info", |
5131 | space_info->flags, orig_bytes, 1); | |
44734ed1 | 5132 | ret = 0; |
2bf64758 JB |
5133 | } |
5134 | ||
8bb8ab2e | 5135 | /* |
957780eb JB |
5136 | * If we couldn't make a reservation then setup our reservation ticket |
5137 | * and kick the async worker if it's not already running. | |
08e007d2 | 5138 | * |
957780eb JB |
5139 | * If we are a priority flusher then we just need to add our ticket to |
5140 | * the list and we will do our own flushing further down. | |
8bb8ab2e | 5141 | */ |
72bcd99d | 5142 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
957780eb JB |
5143 | ticket.bytes = orig_bytes; |
5144 | ticket.error = 0; | |
5145 | init_waitqueue_head(&ticket.wait); | |
5146 | if (flush == BTRFS_RESERVE_FLUSH_ALL) { | |
5147 | list_add_tail(&ticket.list, &space_info->tickets); | |
5148 | if (!space_info->flush) { | |
5149 | space_info->flush = 1; | |
0b246afa | 5150 | trace_btrfs_trigger_flush(fs_info, |
f376df2b JB |
5151 | space_info->flags, |
5152 | orig_bytes, flush, | |
5153 | "enospc"); | |
957780eb | 5154 | queue_work(system_unbound_wq, |
c1c4919b | 5155 | &fs_info->async_reclaim_work); |
957780eb JB |
5156 | } |
5157 | } else { | |
5158 | list_add_tail(&ticket.list, | |
5159 | &space_info->priority_tickets); | |
5160 | } | |
21c7e756 MX |
5161 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
5162 | used += orig_bytes; | |
f6acfd50 JB |
5163 | /* |
5164 | * We will do the space reservation dance during log replay, | |
5165 | * which means we won't have fs_info->fs_root set, so don't do | |
5166 | * the async reclaim as we will panic. | |
5167 | */ | |
0b246afa | 5168 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) && |
c1c4919b JM |
5169 | need_do_async_reclaim(fs_info, space_info, |
5170 | used, system_chunk) && | |
0b246afa JM |
5171 | !work_busy(&fs_info->async_reclaim_work)) { |
5172 | trace_btrfs_trigger_flush(fs_info, space_info->flags, | |
5173 | orig_bytes, flush, "preempt"); | |
21c7e756 | 5174 | queue_work(system_unbound_wq, |
0b246afa | 5175 | &fs_info->async_reclaim_work); |
f376df2b | 5176 | } |
8bb8ab2e | 5177 | } |
f0486c68 | 5178 | spin_unlock(&space_info->lock); |
08e007d2 | 5179 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
957780eb | 5180 | return ret; |
f0486c68 | 5181 | |
957780eb | 5182 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
0b246afa | 5183 | return wait_reserve_ticket(fs_info, space_info, &ticket, |
957780eb | 5184 | orig_bytes); |
08e007d2 | 5185 | |
957780eb | 5186 | ret = 0; |
0b246afa | 5187 | priority_reclaim_metadata_space(fs_info, space_info, &ticket); |
957780eb JB |
5188 | spin_lock(&space_info->lock); |
5189 | if (ticket.bytes) { | |
5190 | if (ticket.bytes < orig_bytes) { | |
5191 | u64 num_bytes = orig_bytes - ticket.bytes; | |
5192 | space_info->bytes_may_use -= num_bytes; | |
0b246afa JM |
5193 | trace_btrfs_space_reservation(fs_info, "space_info", |
5194 | space_info->flags, | |
5195 | num_bytes, 0); | |
08e007d2 | 5196 | |
957780eb JB |
5197 | } |
5198 | list_del_init(&ticket.list); | |
5199 | ret = -ENOSPC; | |
5200 | } | |
5201 | spin_unlock(&space_info->lock); | |
5202 | ASSERT(list_empty(&ticket.list)); | |
5203 | return ret; | |
5204 | } | |
8bb8ab2e | 5205 | |
957780eb JB |
5206 | /** |
5207 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space | |
5208 | * @root - the root we're allocating for | |
5209 | * @block_rsv - the block_rsv we're allocating for | |
5210 | * @orig_bytes - the number of bytes we want | |
5211 | * @flush - whether or not we can flush to make our reservation | |
5212 | * | |
5213 | * This will reserve orgi_bytes number of bytes from the space info associated | |
5214 | * with the block_rsv. If there is not enough space it will make an attempt to | |
5215 | * flush out space to make room. It will do this by flushing delalloc if | |
5216 | * possible or committing the transaction. If flush is 0 then no attempts to | |
5217 | * regain reservations will be made and this will fail if there is not enough | |
5218 | * space already. | |
5219 | */ | |
5220 | static int reserve_metadata_bytes(struct btrfs_root *root, | |
5221 | struct btrfs_block_rsv *block_rsv, | |
5222 | u64 orig_bytes, | |
5223 | enum btrfs_reserve_flush_enum flush) | |
5224 | { | |
0b246afa JM |
5225 | struct btrfs_fs_info *fs_info = root->fs_info; |
5226 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
957780eb | 5227 | int ret; |
c1c4919b | 5228 | bool system_chunk = (root == fs_info->chunk_root); |
957780eb | 5229 | |
c1c4919b JM |
5230 | ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info, |
5231 | orig_bytes, flush, system_chunk); | |
5d80366e JB |
5232 | if (ret == -ENOSPC && |
5233 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { | |
5d80366e JB |
5234 | if (block_rsv != global_rsv && |
5235 | !block_rsv_use_bytes(global_rsv, orig_bytes)) | |
5236 | ret = 0; | |
5237 | } | |
9a3daff3 | 5238 | if (ret == -ENOSPC) { |
0b246afa | 5239 | trace_btrfs_space_reservation(fs_info, "space_info:enospc", |
957780eb JB |
5240 | block_rsv->space_info->flags, |
5241 | orig_bytes, 1); | |
9a3daff3 NB |
5242 | |
5243 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) | |
5244 | dump_space_info(fs_info, block_rsv->space_info, | |
5245 | orig_bytes, 0); | |
5246 | } | |
f0486c68 YZ |
5247 | return ret; |
5248 | } | |
5249 | ||
79787eaa JM |
5250 | static struct btrfs_block_rsv *get_block_rsv( |
5251 | const struct btrfs_trans_handle *trans, | |
5252 | const struct btrfs_root *root) | |
f0486c68 | 5253 | { |
0b246afa | 5254 | struct btrfs_fs_info *fs_info = root->fs_info; |
4c13d758 JB |
5255 | struct btrfs_block_rsv *block_rsv = NULL; |
5256 | ||
e9cf439f | 5257 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
0b246afa JM |
5258 | (root == fs_info->csum_root && trans->adding_csums) || |
5259 | (root == fs_info->uuid_root)) | |
f7a81ea4 SB |
5260 | block_rsv = trans->block_rsv; |
5261 | ||
4c13d758 | 5262 | if (!block_rsv) |
f0486c68 YZ |
5263 | block_rsv = root->block_rsv; |
5264 | ||
5265 | if (!block_rsv) | |
0b246afa | 5266 | block_rsv = &fs_info->empty_block_rsv; |
f0486c68 YZ |
5267 | |
5268 | return block_rsv; | |
5269 | } | |
5270 | ||
5271 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, | |
5272 | u64 num_bytes) | |
5273 | { | |
5274 | int ret = -ENOSPC; | |
5275 | spin_lock(&block_rsv->lock); | |
5276 | if (block_rsv->reserved >= num_bytes) { | |
5277 | block_rsv->reserved -= num_bytes; | |
5278 | if (block_rsv->reserved < block_rsv->size) | |
5279 | block_rsv->full = 0; | |
5280 | ret = 0; | |
5281 | } | |
5282 | spin_unlock(&block_rsv->lock); | |
5283 | return ret; | |
5284 | } | |
5285 | ||
5286 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, | |
3a584174 | 5287 | u64 num_bytes, bool update_size) |
f0486c68 YZ |
5288 | { |
5289 | spin_lock(&block_rsv->lock); | |
5290 | block_rsv->reserved += num_bytes; | |
5291 | if (update_size) | |
5292 | block_rsv->size += num_bytes; | |
5293 | else if (block_rsv->reserved >= block_rsv->size) | |
5294 | block_rsv->full = 1; | |
5295 | spin_unlock(&block_rsv->lock); | |
5296 | } | |
5297 | ||
d52be818 JB |
5298 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
5299 | struct btrfs_block_rsv *dest, u64 num_bytes, | |
5300 | int min_factor) | |
5301 | { | |
5302 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5303 | u64 min_bytes; | |
5304 | ||
5305 | if (global_rsv->space_info != dest->space_info) | |
5306 | return -ENOSPC; | |
5307 | ||
5308 | spin_lock(&global_rsv->lock); | |
5309 | min_bytes = div_factor(global_rsv->size, min_factor); | |
5310 | if (global_rsv->reserved < min_bytes + num_bytes) { | |
5311 | spin_unlock(&global_rsv->lock); | |
5312 | return -ENOSPC; | |
5313 | } | |
5314 | global_rsv->reserved -= num_bytes; | |
5315 | if (global_rsv->reserved < global_rsv->size) | |
5316 | global_rsv->full = 0; | |
5317 | spin_unlock(&global_rsv->lock); | |
5318 | ||
3a584174 | 5319 | block_rsv_add_bytes(dest, num_bytes, true); |
d52be818 JB |
5320 | return 0; |
5321 | } | |
5322 | ||
957780eb JB |
5323 | /* |
5324 | * This is for space we already have accounted in space_info->bytes_may_use, so | |
5325 | * basically when we're returning space from block_rsv's. | |
5326 | */ | |
5327 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, | |
5328 | struct btrfs_space_info *space_info, | |
5329 | u64 num_bytes) | |
5330 | { | |
5331 | struct reserve_ticket *ticket; | |
5332 | struct list_head *head; | |
5333 | u64 used; | |
5334 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH; | |
5335 | bool check_overcommit = false; | |
5336 | ||
5337 | spin_lock(&space_info->lock); | |
5338 | head = &space_info->priority_tickets; | |
5339 | ||
5340 | /* | |
5341 | * If we are over our limit then we need to check and see if we can | |
5342 | * overcommit, and if we can't then we just need to free up our space | |
5343 | * and not satisfy any requests. | |
5344 | */ | |
0eee8a49 | 5345 | used = btrfs_space_info_used(space_info, true); |
957780eb JB |
5346 | if (used - num_bytes >= space_info->total_bytes) |
5347 | check_overcommit = true; | |
5348 | again: | |
5349 | while (!list_empty(head) && num_bytes) { | |
5350 | ticket = list_first_entry(head, struct reserve_ticket, | |
5351 | list); | |
5352 | /* | |
5353 | * We use 0 bytes because this space is already reserved, so | |
5354 | * adding the ticket space would be a double count. | |
5355 | */ | |
5356 | if (check_overcommit && | |
c1c4919b | 5357 | !can_overcommit(fs_info, space_info, 0, flush, false)) |
957780eb JB |
5358 | break; |
5359 | if (num_bytes >= ticket->bytes) { | |
5360 | list_del_init(&ticket->list); | |
5361 | num_bytes -= ticket->bytes; | |
5362 | ticket->bytes = 0; | |
ce129655 | 5363 | space_info->tickets_id++; |
957780eb JB |
5364 | wake_up(&ticket->wait); |
5365 | } else { | |
5366 | ticket->bytes -= num_bytes; | |
5367 | num_bytes = 0; | |
5368 | } | |
5369 | } | |
5370 | ||
5371 | if (num_bytes && head == &space_info->priority_tickets) { | |
5372 | head = &space_info->tickets; | |
5373 | flush = BTRFS_RESERVE_FLUSH_ALL; | |
5374 | goto again; | |
5375 | } | |
5376 | space_info->bytes_may_use -= num_bytes; | |
5377 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5378 | space_info->flags, num_bytes, 0); | |
5379 | spin_unlock(&space_info->lock); | |
5380 | } | |
5381 | ||
5382 | /* | |
5383 | * This is for newly allocated space that isn't accounted in | |
5384 | * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent | |
5385 | * we use this helper. | |
5386 | */ | |
5387 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, | |
5388 | struct btrfs_space_info *space_info, | |
5389 | u64 num_bytes) | |
5390 | { | |
5391 | struct reserve_ticket *ticket; | |
5392 | struct list_head *head = &space_info->priority_tickets; | |
5393 | ||
5394 | again: | |
5395 | while (!list_empty(head) && num_bytes) { | |
5396 | ticket = list_first_entry(head, struct reserve_ticket, | |
5397 | list); | |
5398 | if (num_bytes >= ticket->bytes) { | |
5399 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5400 | space_info->flags, | |
5401 | ticket->bytes, 1); | |
5402 | list_del_init(&ticket->list); | |
5403 | num_bytes -= ticket->bytes; | |
5404 | space_info->bytes_may_use += ticket->bytes; | |
5405 | ticket->bytes = 0; | |
ce129655 | 5406 | space_info->tickets_id++; |
957780eb JB |
5407 | wake_up(&ticket->wait); |
5408 | } else { | |
5409 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5410 | space_info->flags, | |
5411 | num_bytes, 1); | |
5412 | space_info->bytes_may_use += num_bytes; | |
5413 | ticket->bytes -= num_bytes; | |
5414 | num_bytes = 0; | |
5415 | } | |
5416 | } | |
5417 | ||
5418 | if (num_bytes && head == &space_info->priority_tickets) { | |
5419 | head = &space_info->tickets; | |
5420 | goto again; | |
5421 | } | |
5422 | } | |
5423 | ||
69fe2d75 | 5424 | static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
8c2a3ca2 | 5425 | struct btrfs_block_rsv *block_rsv, |
ff6bc37e QW |
5426 | struct btrfs_block_rsv *dest, u64 num_bytes, |
5427 | u64 *qgroup_to_release_ret) | |
f0486c68 YZ |
5428 | { |
5429 | struct btrfs_space_info *space_info = block_rsv->space_info; | |
ff6bc37e | 5430 | u64 qgroup_to_release = 0; |
69fe2d75 | 5431 | u64 ret; |
f0486c68 YZ |
5432 | |
5433 | spin_lock(&block_rsv->lock); | |
ff6bc37e | 5434 | if (num_bytes == (u64)-1) { |
f0486c68 | 5435 | num_bytes = block_rsv->size; |
ff6bc37e QW |
5436 | qgroup_to_release = block_rsv->qgroup_rsv_size; |
5437 | } | |
f0486c68 YZ |
5438 | block_rsv->size -= num_bytes; |
5439 | if (block_rsv->reserved >= block_rsv->size) { | |
5440 | num_bytes = block_rsv->reserved - block_rsv->size; | |
5441 | block_rsv->reserved = block_rsv->size; | |
5442 | block_rsv->full = 1; | |
5443 | } else { | |
5444 | num_bytes = 0; | |
5445 | } | |
ff6bc37e QW |
5446 | if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) { |
5447 | qgroup_to_release = block_rsv->qgroup_rsv_reserved - | |
5448 | block_rsv->qgroup_rsv_size; | |
5449 | block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size; | |
5450 | } else { | |
5451 | qgroup_to_release = 0; | |
5452 | } | |
f0486c68 YZ |
5453 | spin_unlock(&block_rsv->lock); |
5454 | ||
69fe2d75 | 5455 | ret = num_bytes; |
f0486c68 YZ |
5456 | if (num_bytes > 0) { |
5457 | if (dest) { | |
e9e22899 JB |
5458 | spin_lock(&dest->lock); |
5459 | if (!dest->full) { | |
5460 | u64 bytes_to_add; | |
5461 | ||
5462 | bytes_to_add = dest->size - dest->reserved; | |
5463 | bytes_to_add = min(num_bytes, bytes_to_add); | |
5464 | dest->reserved += bytes_to_add; | |
5465 | if (dest->reserved >= dest->size) | |
5466 | dest->full = 1; | |
5467 | num_bytes -= bytes_to_add; | |
5468 | } | |
5469 | spin_unlock(&dest->lock); | |
5470 | } | |
957780eb JB |
5471 | if (num_bytes) |
5472 | space_info_add_old_bytes(fs_info, space_info, | |
5473 | num_bytes); | |
9ed74f2d | 5474 | } |
ff6bc37e QW |
5475 | if (qgroup_to_release_ret) |
5476 | *qgroup_to_release_ret = qgroup_to_release; | |
69fe2d75 | 5477 | return ret; |
f0486c68 | 5478 | } |
4e06bdd6 | 5479 | |
25d609f8 JB |
5480 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src, |
5481 | struct btrfs_block_rsv *dst, u64 num_bytes, | |
3a584174 | 5482 | bool update_size) |
f0486c68 YZ |
5483 | { |
5484 | int ret; | |
9ed74f2d | 5485 | |
f0486c68 YZ |
5486 | ret = block_rsv_use_bytes(src, num_bytes); |
5487 | if (ret) | |
5488 | return ret; | |
9ed74f2d | 5489 | |
25d609f8 | 5490 | block_rsv_add_bytes(dst, num_bytes, update_size); |
9ed74f2d JB |
5491 | return 0; |
5492 | } | |
5493 | ||
66d8f3dd | 5494 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
9ed74f2d | 5495 | { |
f0486c68 YZ |
5496 | memset(rsv, 0, sizeof(*rsv)); |
5497 | spin_lock_init(&rsv->lock); | |
66d8f3dd | 5498 | rsv->type = type; |
f0486c68 YZ |
5499 | } |
5500 | ||
69fe2d75 JB |
5501 | void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info, |
5502 | struct btrfs_block_rsv *rsv, | |
5503 | unsigned short type) | |
5504 | { | |
5505 | btrfs_init_block_rsv(rsv, type); | |
5506 | rsv->space_info = __find_space_info(fs_info, | |
5507 | BTRFS_BLOCK_GROUP_METADATA); | |
5508 | } | |
5509 | ||
2ff7e61e | 5510 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info, |
66d8f3dd | 5511 | unsigned short type) |
f0486c68 YZ |
5512 | { |
5513 | struct btrfs_block_rsv *block_rsv; | |
9ed74f2d | 5514 | |
f0486c68 YZ |
5515 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
5516 | if (!block_rsv) | |
5517 | return NULL; | |
9ed74f2d | 5518 | |
69fe2d75 | 5519 | btrfs_init_metadata_block_rsv(fs_info, block_rsv, type); |
f0486c68 YZ |
5520 | return block_rsv; |
5521 | } | |
9ed74f2d | 5522 | |
2ff7e61e | 5523 | void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5524 | struct btrfs_block_rsv *rsv) |
5525 | { | |
2aaa6655 JB |
5526 | if (!rsv) |
5527 | return; | |
2ff7e61e | 5528 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
dabdb640 | 5529 | kfree(rsv); |
9ed74f2d JB |
5530 | } |
5531 | ||
08e007d2 MX |
5532 | int btrfs_block_rsv_add(struct btrfs_root *root, |
5533 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, | |
5534 | enum btrfs_reserve_flush_enum flush) | |
9ed74f2d | 5535 | { |
f0486c68 | 5536 | int ret; |
9ed74f2d | 5537 | |
f0486c68 YZ |
5538 | if (num_bytes == 0) |
5539 | return 0; | |
8bb8ab2e | 5540 | |
61b520a9 | 5541 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
5a2cb25a | 5542 | if (!ret) |
3a584174 | 5543 | block_rsv_add_bytes(block_rsv, num_bytes, true); |
9ed74f2d | 5544 | |
f0486c68 | 5545 | return ret; |
f0486c68 | 5546 | } |
9ed74f2d | 5547 | |
2ff7e61e | 5548 | int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor) |
f0486c68 YZ |
5549 | { |
5550 | u64 num_bytes = 0; | |
f0486c68 | 5551 | int ret = -ENOSPC; |
9ed74f2d | 5552 | |
f0486c68 YZ |
5553 | if (!block_rsv) |
5554 | return 0; | |
9ed74f2d | 5555 | |
f0486c68 | 5556 | spin_lock(&block_rsv->lock); |
36ba022a JB |
5557 | num_bytes = div_factor(block_rsv->size, min_factor); |
5558 | if (block_rsv->reserved >= num_bytes) | |
5559 | ret = 0; | |
5560 | spin_unlock(&block_rsv->lock); | |
9ed74f2d | 5561 | |
36ba022a JB |
5562 | return ret; |
5563 | } | |
5564 | ||
08e007d2 MX |
5565 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
5566 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, | |
5567 | enum btrfs_reserve_flush_enum flush) | |
36ba022a JB |
5568 | { |
5569 | u64 num_bytes = 0; | |
5570 | int ret = -ENOSPC; | |
5571 | ||
5572 | if (!block_rsv) | |
5573 | return 0; | |
5574 | ||
5575 | spin_lock(&block_rsv->lock); | |
5576 | num_bytes = min_reserved; | |
13553e52 | 5577 | if (block_rsv->reserved >= num_bytes) |
f0486c68 | 5578 | ret = 0; |
13553e52 | 5579 | else |
f0486c68 | 5580 | num_bytes -= block_rsv->reserved; |
f0486c68 | 5581 | spin_unlock(&block_rsv->lock); |
13553e52 | 5582 | |
f0486c68 YZ |
5583 | if (!ret) |
5584 | return 0; | |
5585 | ||
aa38a711 | 5586 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
dabdb640 | 5587 | if (!ret) { |
3a584174 | 5588 | block_rsv_add_bytes(block_rsv, num_bytes, false); |
f0486c68 | 5589 | return 0; |
6a63209f | 5590 | } |
9ed74f2d | 5591 | |
13553e52 | 5592 | return ret; |
f0486c68 YZ |
5593 | } |
5594 | ||
69fe2d75 JB |
5595 | /** |
5596 | * btrfs_inode_rsv_refill - refill the inode block rsv. | |
5597 | * @inode - the inode we are refilling. | |
5598 | * @flush - the flusing restriction. | |
5599 | * | |
5600 | * Essentially the same as btrfs_block_rsv_refill, except it uses the | |
5601 | * block_rsv->size as the minimum size. We'll either refill the missing amount | |
5602 | * or return if we already have enough space. This will also handle the resreve | |
5603 | * tracepoint for the reserved amount. | |
5604 | */ | |
3f2dd7a0 QW |
5605 | static int btrfs_inode_rsv_refill(struct btrfs_inode *inode, |
5606 | enum btrfs_reserve_flush_enum flush) | |
69fe2d75 JB |
5607 | { |
5608 | struct btrfs_root *root = inode->root; | |
5609 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; | |
5610 | u64 num_bytes = 0; | |
ff6bc37e | 5611 | u64 qgroup_num_bytes = 0; |
69fe2d75 JB |
5612 | int ret = -ENOSPC; |
5613 | ||
5614 | spin_lock(&block_rsv->lock); | |
5615 | if (block_rsv->reserved < block_rsv->size) | |
5616 | num_bytes = block_rsv->size - block_rsv->reserved; | |
ff6bc37e QW |
5617 | if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size) |
5618 | qgroup_num_bytes = block_rsv->qgroup_rsv_size - | |
5619 | block_rsv->qgroup_rsv_reserved; | |
69fe2d75 JB |
5620 | spin_unlock(&block_rsv->lock); |
5621 | ||
5622 | if (num_bytes == 0) | |
5623 | return 0; | |
5624 | ||
ff6bc37e | 5625 | ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes, true); |
43b18595 QW |
5626 | if (ret) |
5627 | return ret; | |
69fe2d75 JB |
5628 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
5629 | if (!ret) { | |
3a584174 | 5630 | block_rsv_add_bytes(block_rsv, num_bytes, false); |
69fe2d75 JB |
5631 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
5632 | btrfs_ino(inode), num_bytes, 1); | |
ff6bc37e QW |
5633 | |
5634 | /* Don't forget to increase qgroup_rsv_reserved */ | |
5635 | spin_lock(&block_rsv->lock); | |
5636 | block_rsv->qgroup_rsv_reserved += qgroup_num_bytes; | |
5637 | spin_unlock(&block_rsv->lock); | |
5638 | } else | |
5639 | btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes); | |
69fe2d75 JB |
5640 | return ret; |
5641 | } | |
5642 | ||
5643 | /** | |
5644 | * btrfs_inode_rsv_release - release any excessive reservation. | |
5645 | * @inode - the inode we need to release from. | |
43b18595 QW |
5646 | * @qgroup_free - free or convert qgroup meta. |
5647 | * Unlike normal operation, qgroup meta reservation needs to know if we are | |
5648 | * freeing qgroup reservation or just converting it into per-trans. Normally | |
5649 | * @qgroup_free is true for error handling, and false for normal release. | |
69fe2d75 JB |
5650 | * |
5651 | * This is the same as btrfs_block_rsv_release, except that it handles the | |
5652 | * tracepoint for the reservation. | |
5653 | */ | |
43b18595 | 5654 | static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free) |
69fe2d75 JB |
5655 | { |
5656 | struct btrfs_fs_info *fs_info = inode->root->fs_info; | |
5657 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
5658 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; | |
5659 | u64 released = 0; | |
ff6bc37e | 5660 | u64 qgroup_to_release = 0; |
69fe2d75 JB |
5661 | |
5662 | /* | |
5663 | * Since we statically set the block_rsv->size we just want to say we | |
5664 | * are releasing 0 bytes, and then we'll just get the reservation over | |
5665 | * the size free'd. | |
5666 | */ | |
ff6bc37e QW |
5667 | released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0, |
5668 | &qgroup_to_release); | |
69fe2d75 JB |
5669 | if (released > 0) |
5670 | trace_btrfs_space_reservation(fs_info, "delalloc", | |
5671 | btrfs_ino(inode), released, 0); | |
43b18595 | 5672 | if (qgroup_free) |
ff6bc37e | 5673 | btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release); |
43b18595 | 5674 | else |
ff6bc37e QW |
5675 | btrfs_qgroup_convert_reserved_meta(inode->root, |
5676 | qgroup_to_release); | |
69fe2d75 JB |
5677 | } |
5678 | ||
2ff7e61e | 5679 | void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
5680 | struct btrfs_block_rsv *block_rsv, |
5681 | u64 num_bytes) | |
5682 | { | |
0b246afa JM |
5683 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
5684 | ||
17504584 | 5685 | if (global_rsv == block_rsv || |
f0486c68 YZ |
5686 | block_rsv->space_info != global_rsv->space_info) |
5687 | global_rsv = NULL; | |
ff6bc37e | 5688 | block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes, NULL); |
6a63209f JB |
5689 | } |
5690 | ||
8929ecfa YZ |
5691 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
5692 | { | |
5693 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; | |
5694 | struct btrfs_space_info *sinfo = block_rsv->space_info; | |
5695 | u64 num_bytes; | |
6a63209f | 5696 | |
ae2e4728 JB |
5697 | /* |
5698 | * The global block rsv is based on the size of the extent tree, the | |
5699 | * checksum tree and the root tree. If the fs is empty we want to set | |
5700 | * it to a minimal amount for safety. | |
5701 | */ | |
5702 | num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) + | |
5703 | btrfs_root_used(&fs_info->csum_root->root_item) + | |
5704 | btrfs_root_used(&fs_info->tree_root->root_item); | |
5705 | num_bytes = max_t(u64, num_bytes, SZ_16M); | |
33b4d47f | 5706 | |
8929ecfa | 5707 | spin_lock(&sinfo->lock); |
1f699d38 | 5708 | spin_lock(&block_rsv->lock); |
4e06bdd6 | 5709 | |
ee22184b | 5710 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); |
4e06bdd6 | 5711 | |
fb4b10e5 | 5712 | if (block_rsv->reserved < block_rsv->size) { |
4136135b | 5713 | num_bytes = btrfs_space_info_used(sinfo, true); |
fb4b10e5 JB |
5714 | if (sinfo->total_bytes > num_bytes) { |
5715 | num_bytes = sinfo->total_bytes - num_bytes; | |
5716 | num_bytes = min(num_bytes, | |
5717 | block_rsv->size - block_rsv->reserved); | |
5718 | block_rsv->reserved += num_bytes; | |
5719 | sinfo->bytes_may_use += num_bytes; | |
5720 | trace_btrfs_space_reservation(fs_info, "space_info", | |
5721 | sinfo->flags, num_bytes, | |
5722 | 1); | |
5723 | } | |
5724 | } else if (block_rsv->reserved > block_rsv->size) { | |
8929ecfa | 5725 | num_bytes = block_rsv->reserved - block_rsv->size; |
fb25e914 | 5726 | sinfo->bytes_may_use -= num_bytes; |
8c2a3ca2 | 5727 | trace_btrfs_space_reservation(fs_info, "space_info", |
2bcc0328 | 5728 | sinfo->flags, num_bytes, 0); |
8929ecfa | 5729 | block_rsv->reserved = block_rsv->size; |
8929ecfa | 5730 | } |
182608c8 | 5731 | |
fb4b10e5 JB |
5732 | if (block_rsv->reserved == block_rsv->size) |
5733 | block_rsv->full = 1; | |
5734 | else | |
5735 | block_rsv->full = 0; | |
5736 | ||
8929ecfa | 5737 | spin_unlock(&block_rsv->lock); |
1f699d38 | 5738 | spin_unlock(&sinfo->lock); |
6a63209f JB |
5739 | } |
5740 | ||
f0486c68 | 5741 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5742 | { |
f0486c68 | 5743 | struct btrfs_space_info *space_info; |
6a63209f | 5744 | |
f0486c68 YZ |
5745 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
5746 | fs_info->chunk_block_rsv.space_info = space_info; | |
6a63209f | 5747 | |
f0486c68 | 5748 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
8929ecfa | 5749 | fs_info->global_block_rsv.space_info = space_info; |
f0486c68 YZ |
5750 | fs_info->trans_block_rsv.space_info = space_info; |
5751 | fs_info->empty_block_rsv.space_info = space_info; | |
6d668dda | 5752 | fs_info->delayed_block_rsv.space_info = space_info; |
f0486c68 | 5753 | |
8929ecfa YZ |
5754 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
5755 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; | |
5756 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; | |
5757 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; | |
3a6cad90 SB |
5758 | if (fs_info->quota_root) |
5759 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; | |
f0486c68 | 5760 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
8929ecfa | 5761 | |
8929ecfa | 5762 | update_global_block_rsv(fs_info); |
6a63209f JB |
5763 | } |
5764 | ||
8929ecfa | 5765 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
6a63209f | 5766 | { |
8c2a3ca2 | 5767 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
ff6bc37e | 5768 | (u64)-1, NULL); |
8929ecfa YZ |
5769 | WARN_ON(fs_info->trans_block_rsv.size > 0); |
5770 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); | |
5771 | WARN_ON(fs_info->chunk_block_rsv.size > 0); | |
5772 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); | |
6d668dda JB |
5773 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
5774 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); | |
fcb80c2a JB |
5775 | } |
5776 | ||
6a63209f | 5777 | |
4fbcdf66 FM |
5778 | /* |
5779 | * To be called after all the new block groups attached to the transaction | |
5780 | * handle have been created (btrfs_create_pending_block_groups()). | |
5781 | */ | |
5782 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) | |
5783 | { | |
64b63580 | 5784 | struct btrfs_fs_info *fs_info = trans->fs_info; |
4fbcdf66 FM |
5785 | |
5786 | if (!trans->chunk_bytes_reserved) | |
5787 | return; | |
5788 | ||
5789 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); | |
5790 | ||
5791 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, | |
ff6bc37e | 5792 | trans->chunk_bytes_reserved, NULL); |
4fbcdf66 FM |
5793 | trans->chunk_bytes_reserved = 0; |
5794 | } | |
5795 | ||
d5c12070 MX |
5796 | /* |
5797 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation | |
5798 | * root: the root of the parent directory | |
5799 | * rsv: block reservation | |
5800 | * items: the number of items that we need do reservation | |
a5b7f429 | 5801 | * use_global_rsv: allow fallback to the global block reservation |
d5c12070 MX |
5802 | * |
5803 | * This function is used to reserve the space for snapshot/subvolume | |
5804 | * creation and deletion. Those operations are different with the | |
5805 | * common file/directory operations, they change two fs/file trees | |
5806 | * and root tree, the number of items that the qgroup reserves is | |
5807 | * different with the free space reservation. So we can not use | |
01327610 | 5808 | * the space reservation mechanism in start_transaction(). |
d5c12070 MX |
5809 | */ |
5810 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, | |
a5b7f429 | 5811 | struct btrfs_block_rsv *rsv, int items, |
ee3441b4 | 5812 | bool use_global_rsv) |
a22285a6 | 5813 | { |
a5b7f429 | 5814 | u64 qgroup_num_bytes = 0; |
d5c12070 MX |
5815 | u64 num_bytes; |
5816 | int ret; | |
0b246afa JM |
5817 | struct btrfs_fs_info *fs_info = root->fs_info; |
5818 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
d5c12070 | 5819 | |
0b246afa | 5820 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
d5c12070 | 5821 | /* One for parent inode, two for dir entries */ |
a5b7f429 LF |
5822 | qgroup_num_bytes = 3 * fs_info->nodesize; |
5823 | ret = btrfs_qgroup_reserve_meta_prealloc(root, | |
5824 | qgroup_num_bytes, true); | |
d5c12070 MX |
5825 | if (ret) |
5826 | return ret; | |
d5c12070 MX |
5827 | } |
5828 | ||
0b246afa JM |
5829 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, items); |
5830 | rsv->space_info = __find_space_info(fs_info, | |
d5c12070 MX |
5831 | BTRFS_BLOCK_GROUP_METADATA); |
5832 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, | |
5833 | BTRFS_RESERVE_FLUSH_ALL); | |
ee3441b4 JM |
5834 | |
5835 | if (ret == -ENOSPC && use_global_rsv) | |
3a584174 | 5836 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, true); |
ee3441b4 | 5837 | |
a5b7f429 LF |
5838 | if (ret && qgroup_num_bytes) |
5839 | btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes); | |
d5c12070 MX |
5840 | |
5841 | return ret; | |
5842 | } | |
5843 | ||
2ff7e61e | 5844 | void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info, |
7775c818 | 5845 | struct btrfs_block_rsv *rsv) |
d5c12070 | 5846 | { |
2ff7e61e | 5847 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
97e728d4 JB |
5848 | } |
5849 | ||
69fe2d75 JB |
5850 | static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info, |
5851 | struct btrfs_inode *inode) | |
9e0baf60 | 5852 | { |
69fe2d75 JB |
5853 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
5854 | u64 reserve_size = 0; | |
ff6bc37e | 5855 | u64 qgroup_rsv_size = 0; |
69fe2d75 JB |
5856 | u64 csum_leaves; |
5857 | unsigned outstanding_extents; | |
9e0baf60 | 5858 | |
69fe2d75 JB |
5859 | lockdep_assert_held(&inode->lock); |
5860 | outstanding_extents = inode->outstanding_extents; | |
5861 | if (outstanding_extents) | |
5862 | reserve_size = btrfs_calc_trans_metadata_size(fs_info, | |
5863 | outstanding_extents + 1); | |
5864 | csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, | |
5865 | inode->csum_bytes); | |
5866 | reserve_size += btrfs_calc_trans_metadata_size(fs_info, | |
5867 | csum_leaves); | |
ff6bc37e QW |
5868 | /* |
5869 | * For qgroup rsv, the calculation is very simple: | |
5870 | * account one nodesize for each outstanding extent | |
5871 | * | |
5872 | * This is overestimating in most cases. | |
5873 | */ | |
5874 | qgroup_rsv_size = outstanding_extents * fs_info->nodesize; | |
9e0baf60 | 5875 | |
69fe2d75 JB |
5876 | spin_lock(&block_rsv->lock); |
5877 | block_rsv->size = reserve_size; | |
ff6bc37e | 5878 | block_rsv->qgroup_rsv_size = qgroup_rsv_size; |
69fe2d75 | 5879 | spin_unlock(&block_rsv->lock); |
0ca1f7ce | 5880 | } |
c146afad | 5881 | |
9f3db423 | 5882 | int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes) |
0ca1f7ce | 5883 | { |
3ffbd68c | 5884 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
69fe2d75 | 5885 | unsigned nr_extents; |
08e007d2 | 5886 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
eb6b88d9 | 5887 | int ret = 0; |
c64c2bd8 | 5888 | bool delalloc_lock = true; |
6324fbf3 | 5889 | |
c64c2bd8 JB |
5890 | /* If we are a free space inode we need to not flush since we will be in |
5891 | * the middle of a transaction commit. We also don't need the delalloc | |
5892 | * mutex since we won't race with anybody. We need this mostly to make | |
5893 | * lockdep shut its filthy mouth. | |
bac357dc JB |
5894 | * |
5895 | * If we have a transaction open (can happen if we call truncate_block | |
5896 | * from truncate), then we need FLUSH_LIMIT so we don't deadlock. | |
c64c2bd8 JB |
5897 | */ |
5898 | if (btrfs_is_free_space_inode(inode)) { | |
08e007d2 | 5899 | flush = BTRFS_RESERVE_NO_FLUSH; |
c64c2bd8 | 5900 | delalloc_lock = false; |
da07d4ab NB |
5901 | } else { |
5902 | if (current->journal_info) | |
5903 | flush = BTRFS_RESERVE_FLUSH_LIMIT; | |
c09544e0 | 5904 | |
da07d4ab NB |
5905 | if (btrfs_transaction_in_commit(fs_info)) |
5906 | schedule_timeout(1); | |
5907 | } | |
ec44a35c | 5908 | |
c64c2bd8 | 5909 | if (delalloc_lock) |
9f3db423 | 5910 | mutex_lock(&inode->delalloc_mutex); |
c64c2bd8 | 5911 | |
0b246afa | 5912 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
69fe2d75 JB |
5913 | |
5914 | /* Add our new extents and calculate the new rsv size. */ | |
9f3db423 | 5915 | spin_lock(&inode->lock); |
69fe2d75 | 5916 | nr_extents = count_max_extents(num_bytes); |
8b62f87b | 5917 | btrfs_mod_outstanding_extents(inode, nr_extents); |
69fe2d75 JB |
5918 | inode->csum_bytes += num_bytes; |
5919 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 5920 | spin_unlock(&inode->lock); |
57a45ced | 5921 | |
69fe2d75 | 5922 | ret = btrfs_inode_rsv_refill(inode, flush); |
43b18595 | 5923 | if (unlikely(ret)) |
88e081bf | 5924 | goto out_fail; |
25179201 | 5925 | |
c64c2bd8 | 5926 | if (delalloc_lock) |
9f3db423 | 5927 | mutex_unlock(&inode->delalloc_mutex); |
0ca1f7ce | 5928 | return 0; |
88e081bf WS |
5929 | |
5930 | out_fail: | |
9f3db423 | 5931 | spin_lock(&inode->lock); |
8b62f87b JB |
5932 | nr_extents = count_max_extents(num_bytes); |
5933 | btrfs_mod_outstanding_extents(inode, -nr_extents); | |
69fe2d75 JB |
5934 | inode->csum_bytes -= num_bytes; |
5935 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
9f3db423 | 5936 | spin_unlock(&inode->lock); |
88e081bf | 5937 | |
43b18595 | 5938 | btrfs_inode_rsv_release(inode, true); |
88e081bf | 5939 | if (delalloc_lock) |
9f3db423 | 5940 | mutex_unlock(&inode->delalloc_mutex); |
88e081bf | 5941 | return ret; |
0ca1f7ce YZ |
5942 | } |
5943 | ||
7709cde3 JB |
5944 | /** |
5945 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode | |
8b62f87b JB |
5946 | * @inode: the inode to release the reservation for. |
5947 | * @num_bytes: the number of bytes we are releasing. | |
43b18595 | 5948 | * @qgroup_free: free qgroup reservation or convert it to per-trans reservation |
7709cde3 JB |
5949 | * |
5950 | * This will release the metadata reservation for an inode. This can be called | |
5951 | * once we complete IO for a given set of bytes to release their metadata | |
8b62f87b | 5952 | * reservations, or on error for the same reason. |
7709cde3 | 5953 | */ |
43b18595 QW |
5954 | void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes, |
5955 | bool qgroup_free) | |
0ca1f7ce | 5956 | { |
3ffbd68c | 5957 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
0ca1f7ce | 5958 | |
0b246afa | 5959 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
691fa059 | 5960 | spin_lock(&inode->lock); |
69fe2d75 JB |
5961 | inode->csum_bytes -= num_bytes; |
5962 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); | |
691fa059 | 5963 | spin_unlock(&inode->lock); |
0ca1f7ce | 5964 | |
0b246afa | 5965 | if (btrfs_is_testing(fs_info)) |
6a3891c5 JB |
5966 | return; |
5967 | ||
43b18595 | 5968 | btrfs_inode_rsv_release(inode, qgroup_free); |
0ca1f7ce YZ |
5969 | } |
5970 | ||
8b62f87b JB |
5971 | /** |
5972 | * btrfs_delalloc_release_extents - release our outstanding_extents | |
5973 | * @inode: the inode to balance the reservation for. | |
5974 | * @num_bytes: the number of bytes we originally reserved with | |
43b18595 | 5975 | * @qgroup_free: do we need to free qgroup meta reservation or convert them. |
8b62f87b JB |
5976 | * |
5977 | * When we reserve space we increase outstanding_extents for the extents we may | |
5978 | * add. Once we've set the range as delalloc or created our ordered extents we | |
5979 | * have outstanding_extents to track the real usage, so we use this to free our | |
5980 | * temporarily tracked outstanding_extents. This _must_ be used in conjunction | |
5981 | * with btrfs_delalloc_reserve_metadata. | |
5982 | */ | |
43b18595 QW |
5983 | void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes, |
5984 | bool qgroup_free) | |
8b62f87b | 5985 | { |
3ffbd68c | 5986 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
8b62f87b | 5987 | unsigned num_extents; |
8b62f87b JB |
5988 | |
5989 | spin_lock(&inode->lock); | |
5990 | num_extents = count_max_extents(num_bytes); | |
5991 | btrfs_mod_outstanding_extents(inode, -num_extents); | |
69fe2d75 | 5992 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); |
8b62f87b JB |
5993 | spin_unlock(&inode->lock); |
5994 | ||
8b62f87b JB |
5995 | if (btrfs_is_testing(fs_info)) |
5996 | return; | |
5997 | ||
43b18595 | 5998 | btrfs_inode_rsv_release(inode, qgroup_free); |
8b62f87b JB |
5999 | } |
6000 | ||
1ada3a62 | 6001 | /** |
7cf5b976 | 6002 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
1ada3a62 QW |
6003 | * delalloc |
6004 | * @inode: inode we're writing to | |
6005 | * @start: start range we are writing to | |
6006 | * @len: how long the range we are writing to | |
364ecf36 QW |
6007 | * @reserved: mandatory parameter, record actually reserved qgroup ranges of |
6008 | * current reservation. | |
1ada3a62 | 6009 | * |
1ada3a62 QW |
6010 | * This will do the following things |
6011 | * | |
6012 | * o reserve space in data space info for num bytes | |
6013 | * and reserve precious corresponding qgroup space | |
6014 | * (Done in check_data_free_space) | |
6015 | * | |
6016 | * o reserve space for metadata space, based on the number of outstanding | |
6017 | * extents and how much csums will be needed | |
6018 | * also reserve metadata space in a per root over-reserve method. | |
6019 | * o add to the inodes->delalloc_bytes | |
6020 | * o add it to the fs_info's delalloc inodes list. | |
6021 | * (Above 3 all done in delalloc_reserve_metadata) | |
6022 | * | |
6023 | * Return 0 for success | |
6024 | * Return <0 for error(-ENOSPC or -EQUOT) | |
6025 | */ | |
364ecf36 QW |
6026 | int btrfs_delalloc_reserve_space(struct inode *inode, |
6027 | struct extent_changeset **reserved, u64 start, u64 len) | |
1ada3a62 QW |
6028 | { |
6029 | int ret; | |
6030 | ||
364ecf36 | 6031 | ret = btrfs_check_data_free_space(inode, reserved, start, len); |
1ada3a62 QW |
6032 | if (ret < 0) |
6033 | return ret; | |
9f3db423 | 6034 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len); |
1ada3a62 | 6035 | if (ret < 0) |
bc42bda2 | 6036 | btrfs_free_reserved_data_space(inode, *reserved, start, len); |
1ada3a62 QW |
6037 | return ret; |
6038 | } | |
6039 | ||
7709cde3 | 6040 | /** |
7cf5b976 | 6041 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
1ada3a62 QW |
6042 | * @inode: inode we're releasing space for |
6043 | * @start: start position of the space already reserved | |
6044 | * @len: the len of the space already reserved | |
8b62f87b | 6045 | * @release_bytes: the len of the space we consumed or didn't use |
1ada3a62 QW |
6046 | * |
6047 | * This function will release the metadata space that was not used and will | |
6048 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes | |
6049 | * list if there are no delalloc bytes left. | |
6050 | * Also it will handle the qgroup reserved space. | |
6051 | */ | |
bc42bda2 | 6052 | void btrfs_delalloc_release_space(struct inode *inode, |
8b62f87b | 6053 | struct extent_changeset *reserved, |
43b18595 | 6054 | u64 start, u64 len, bool qgroup_free) |
1ada3a62 | 6055 | { |
43b18595 | 6056 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free); |
bc42bda2 | 6057 | btrfs_free_reserved_data_space(inode, reserved, start, len); |
6324fbf3 CM |
6058 | } |
6059 | ||
ce93ec54 | 6060 | static int update_block_group(struct btrfs_trans_handle *trans, |
6202df69 | 6061 | struct btrfs_fs_info *info, u64 bytenr, |
ce93ec54 | 6062 | u64 num_bytes, int alloc) |
9078a3e1 | 6063 | { |
0af3d00b | 6064 | struct btrfs_block_group_cache *cache = NULL; |
db94535d | 6065 | u64 total = num_bytes; |
9078a3e1 | 6066 | u64 old_val; |
db94535d | 6067 | u64 byte_in_group; |
0af3d00b | 6068 | int factor; |
3e1ad54f | 6069 | |
5d4f98a2 | 6070 | /* block accounting for super block */ |
eb73c1b7 | 6071 | spin_lock(&info->delalloc_root_lock); |
6c41761f | 6072 | old_val = btrfs_super_bytes_used(info->super_copy); |
5d4f98a2 YZ |
6073 | if (alloc) |
6074 | old_val += num_bytes; | |
6075 | else | |
6076 | old_val -= num_bytes; | |
6c41761f | 6077 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
eb73c1b7 | 6078 | spin_unlock(&info->delalloc_root_lock); |
5d4f98a2 | 6079 | |
d397712b | 6080 | while (total) { |
db94535d | 6081 | cache = btrfs_lookup_block_group(info, bytenr); |
f3465ca4 | 6082 | if (!cache) |
79787eaa | 6083 | return -ENOENT; |
46df06b8 DS |
6084 | factor = btrfs_bg_type_to_factor(cache->flags); |
6085 | ||
9d66e233 JB |
6086 | /* |
6087 | * If this block group has free space cache written out, we | |
6088 | * need to make sure to load it if we are removing space. This | |
6089 | * is because we need the unpinning stage to actually add the | |
6090 | * space back to the block group, otherwise we will leak space. | |
6091 | */ | |
6092 | if (!alloc && cache->cached == BTRFS_CACHE_NO) | |
f6373bf3 | 6093 | cache_block_group(cache, 1); |
0af3d00b | 6094 | |
db94535d CM |
6095 | byte_in_group = bytenr - cache->key.objectid; |
6096 | WARN_ON(byte_in_group > cache->key.offset); | |
9078a3e1 | 6097 | |
25179201 | 6098 | spin_lock(&cache->space_info->lock); |
c286ac48 | 6099 | spin_lock(&cache->lock); |
0af3d00b | 6100 | |
6202df69 | 6101 | if (btrfs_test_opt(info, SPACE_CACHE) && |
0af3d00b JB |
6102 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
6103 | cache->disk_cache_state = BTRFS_DC_CLEAR; | |
6104 | ||
9078a3e1 | 6105 | old_val = btrfs_block_group_used(&cache->item); |
db94535d | 6106 | num_bytes = min(total, cache->key.offset - byte_in_group); |
cd1bc465 | 6107 | if (alloc) { |
db94535d | 6108 | old_val += num_bytes; |
11833d66 YZ |
6109 | btrfs_set_block_group_used(&cache->item, old_val); |
6110 | cache->reserved -= num_bytes; | |
11833d66 | 6111 | cache->space_info->bytes_reserved -= num_bytes; |
b742bb82 YZ |
6112 | cache->space_info->bytes_used += num_bytes; |
6113 | cache->space_info->disk_used += num_bytes * factor; | |
c286ac48 | 6114 | spin_unlock(&cache->lock); |
25179201 | 6115 | spin_unlock(&cache->space_info->lock); |
cd1bc465 | 6116 | } else { |
db94535d | 6117 | old_val -= num_bytes; |
ae0ab003 FM |
6118 | btrfs_set_block_group_used(&cache->item, old_val); |
6119 | cache->pinned += num_bytes; | |
6120 | cache->space_info->bytes_pinned += num_bytes; | |
6121 | cache->space_info->bytes_used -= num_bytes; | |
6122 | cache->space_info->disk_used -= num_bytes * factor; | |
6123 | spin_unlock(&cache->lock); | |
6124 | spin_unlock(&cache->space_info->lock); | |
47ab2a6c | 6125 | |
0b246afa | 6126 | trace_btrfs_space_reservation(info, "pinned", |
c51e7bb1 JB |
6127 | cache->space_info->flags, |
6128 | num_bytes, 1); | |
dec59fa3 EL |
6129 | percpu_counter_add_batch(&cache->space_info->total_bytes_pinned, |
6130 | num_bytes, | |
6131 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
ae0ab003 FM |
6132 | set_extent_dirty(info->pinned_extents, |
6133 | bytenr, bytenr + num_bytes - 1, | |
6134 | GFP_NOFS | __GFP_NOFAIL); | |
cd1bc465 | 6135 | } |
1bbc621e CM |
6136 | |
6137 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
6138 | if (list_empty(&cache->dirty_list)) { | |
6139 | list_add_tail(&cache->dirty_list, | |
6140 | &trans->transaction->dirty_bgs); | |
bece2e82 | 6141 | trans->transaction->num_dirty_bgs++; |
1bbc621e CM |
6142 | btrfs_get_block_group(cache); |
6143 | } | |
6144 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
6145 | ||
036a9348 FM |
6146 | /* |
6147 | * No longer have used bytes in this block group, queue it for | |
6148 | * deletion. We do this after adding the block group to the | |
6149 | * dirty list to avoid races between cleaner kthread and space | |
6150 | * cache writeout. | |
6151 | */ | |
031f24da QW |
6152 | if (!alloc && old_val == 0) |
6153 | btrfs_mark_bg_unused(cache); | |
036a9348 | 6154 | |
fa9c0d79 | 6155 | btrfs_put_block_group(cache); |
db94535d CM |
6156 | total -= num_bytes; |
6157 | bytenr += num_bytes; | |
9078a3e1 CM |
6158 | } |
6159 | return 0; | |
6160 | } | |
6324fbf3 | 6161 | |
2ff7e61e | 6162 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
a061fc8d | 6163 | { |
0f9dd46c | 6164 | struct btrfs_block_group_cache *cache; |
d2fb3437 | 6165 | u64 bytenr; |
0f9dd46c | 6166 | |
0b246afa JM |
6167 | spin_lock(&fs_info->block_group_cache_lock); |
6168 | bytenr = fs_info->first_logical_byte; | |
6169 | spin_unlock(&fs_info->block_group_cache_lock); | |
a1897fdd LB |
6170 | |
6171 | if (bytenr < (u64)-1) | |
6172 | return bytenr; | |
6173 | ||
0b246afa | 6174 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
0f9dd46c | 6175 | if (!cache) |
a061fc8d | 6176 | return 0; |
0f9dd46c | 6177 | |
d2fb3437 | 6178 | bytenr = cache->key.objectid; |
fa9c0d79 | 6179 | btrfs_put_block_group(cache); |
d2fb3437 YZ |
6180 | |
6181 | return bytenr; | |
a061fc8d CM |
6182 | } |
6183 | ||
2ff7e61e | 6184 | static int pin_down_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6185 | struct btrfs_block_group_cache *cache, |
6186 | u64 bytenr, u64 num_bytes, int reserved) | |
324ae4df | 6187 | { |
11833d66 YZ |
6188 | spin_lock(&cache->space_info->lock); |
6189 | spin_lock(&cache->lock); | |
6190 | cache->pinned += num_bytes; | |
6191 | cache->space_info->bytes_pinned += num_bytes; | |
6192 | if (reserved) { | |
6193 | cache->reserved -= num_bytes; | |
6194 | cache->space_info->bytes_reserved -= num_bytes; | |
6195 | } | |
6196 | spin_unlock(&cache->lock); | |
6197 | spin_unlock(&cache->space_info->lock); | |
68b38550 | 6198 | |
0b246afa | 6199 | trace_btrfs_space_reservation(fs_info, "pinned", |
c51e7bb1 | 6200 | cache->space_info->flags, num_bytes, 1); |
dec59fa3 EL |
6201 | percpu_counter_add_batch(&cache->space_info->total_bytes_pinned, |
6202 | num_bytes, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
0b246afa | 6203 | set_extent_dirty(fs_info->pinned_extents, bytenr, |
f0486c68 YZ |
6204 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
6205 | return 0; | |
6206 | } | |
68b38550 | 6207 | |
f0486c68 YZ |
6208 | /* |
6209 | * this function must be called within transaction | |
6210 | */ | |
2ff7e61e | 6211 | int btrfs_pin_extent(struct btrfs_fs_info *fs_info, |
f0486c68 YZ |
6212 | u64 bytenr, u64 num_bytes, int reserved) |
6213 | { | |
6214 | struct btrfs_block_group_cache *cache; | |
68b38550 | 6215 | |
0b246afa | 6216 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
79787eaa | 6217 | BUG_ON(!cache); /* Logic error */ |
f0486c68 | 6218 | |
2ff7e61e | 6219 | pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved); |
f0486c68 YZ |
6220 | |
6221 | btrfs_put_block_group(cache); | |
11833d66 YZ |
6222 | return 0; |
6223 | } | |
6224 | ||
f0486c68 | 6225 | /* |
e688b725 CM |
6226 | * this function must be called within transaction |
6227 | */ | |
2ff7e61e | 6228 | int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info, |
e688b725 CM |
6229 | u64 bytenr, u64 num_bytes) |
6230 | { | |
6231 | struct btrfs_block_group_cache *cache; | |
b50c6e25 | 6232 | int ret; |
e688b725 | 6233 | |
0b246afa | 6234 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
b50c6e25 JB |
6235 | if (!cache) |
6236 | return -EINVAL; | |
e688b725 CM |
6237 | |
6238 | /* | |
6239 | * pull in the free space cache (if any) so that our pin | |
6240 | * removes the free space from the cache. We have load_only set | |
6241 | * to one because the slow code to read in the free extents does check | |
6242 | * the pinned extents. | |
6243 | */ | |
f6373bf3 | 6244 | cache_block_group(cache, 1); |
e688b725 | 6245 | |
2ff7e61e | 6246 | pin_down_extent(fs_info, cache, bytenr, num_bytes, 0); |
e688b725 CM |
6247 | |
6248 | /* remove us from the free space cache (if we're there at all) */ | |
b50c6e25 | 6249 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
e688b725 | 6250 | btrfs_put_block_group(cache); |
b50c6e25 | 6251 | return ret; |
e688b725 CM |
6252 | } |
6253 | ||
2ff7e61e JM |
6254 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
6255 | u64 start, u64 num_bytes) | |
8c2a1a30 JB |
6256 | { |
6257 | int ret; | |
6258 | struct btrfs_block_group_cache *block_group; | |
6259 | struct btrfs_caching_control *caching_ctl; | |
6260 | ||
0b246afa | 6261 | block_group = btrfs_lookup_block_group(fs_info, start); |
8c2a1a30 JB |
6262 | if (!block_group) |
6263 | return -EINVAL; | |
6264 | ||
6265 | cache_block_group(block_group, 0); | |
6266 | caching_ctl = get_caching_control(block_group); | |
6267 | ||
6268 | if (!caching_ctl) { | |
6269 | /* Logic error */ | |
6270 | BUG_ON(!block_group_cache_done(block_group)); | |
6271 | ret = btrfs_remove_free_space(block_group, start, num_bytes); | |
6272 | } else { | |
6273 | mutex_lock(&caching_ctl->mutex); | |
6274 | ||
6275 | if (start >= caching_ctl->progress) { | |
2ff7e61e | 6276 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6277 | } else if (start + num_bytes <= caching_ctl->progress) { |
6278 | ret = btrfs_remove_free_space(block_group, | |
6279 | start, num_bytes); | |
6280 | } else { | |
6281 | num_bytes = caching_ctl->progress - start; | |
6282 | ret = btrfs_remove_free_space(block_group, | |
6283 | start, num_bytes); | |
6284 | if (ret) | |
6285 | goto out_lock; | |
6286 | ||
6287 | num_bytes = (start + num_bytes) - | |
6288 | caching_ctl->progress; | |
6289 | start = caching_ctl->progress; | |
2ff7e61e | 6290 | ret = add_excluded_extent(fs_info, start, num_bytes); |
8c2a1a30 JB |
6291 | } |
6292 | out_lock: | |
6293 | mutex_unlock(&caching_ctl->mutex); | |
6294 | put_caching_control(caching_ctl); | |
6295 | } | |
6296 | btrfs_put_block_group(block_group); | |
6297 | return ret; | |
6298 | } | |
6299 | ||
2ff7e61e | 6300 | int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info, |
8c2a1a30 JB |
6301 | struct extent_buffer *eb) |
6302 | { | |
6303 | struct btrfs_file_extent_item *item; | |
6304 | struct btrfs_key key; | |
6305 | int found_type; | |
6306 | int i; | |
b89311ef | 6307 | int ret = 0; |
8c2a1a30 | 6308 | |
2ff7e61e | 6309 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
8c2a1a30 JB |
6310 | return 0; |
6311 | ||
6312 | for (i = 0; i < btrfs_header_nritems(eb); i++) { | |
6313 | btrfs_item_key_to_cpu(eb, &key, i); | |
6314 | if (key.type != BTRFS_EXTENT_DATA_KEY) | |
6315 | continue; | |
6316 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); | |
6317 | found_type = btrfs_file_extent_type(eb, item); | |
6318 | if (found_type == BTRFS_FILE_EXTENT_INLINE) | |
6319 | continue; | |
6320 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) | |
6321 | continue; | |
6322 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); | |
6323 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); | |
b89311ef GJ |
6324 | ret = __exclude_logged_extent(fs_info, key.objectid, key.offset); |
6325 | if (ret) | |
6326 | break; | |
8c2a1a30 JB |
6327 | } |
6328 | ||
b89311ef | 6329 | return ret; |
8c2a1a30 JB |
6330 | } |
6331 | ||
9cfa3e34 FM |
6332 | static void |
6333 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) | |
6334 | { | |
6335 | atomic_inc(&bg->reservations); | |
6336 | } | |
6337 | ||
6338 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, | |
6339 | const u64 start) | |
6340 | { | |
6341 | struct btrfs_block_group_cache *bg; | |
6342 | ||
6343 | bg = btrfs_lookup_block_group(fs_info, start); | |
6344 | ASSERT(bg); | |
6345 | if (atomic_dec_and_test(&bg->reservations)) | |
4625956a | 6346 | wake_up_var(&bg->reservations); |
9cfa3e34 FM |
6347 | btrfs_put_block_group(bg); |
6348 | } | |
6349 | ||
9cfa3e34 FM |
6350 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) |
6351 | { | |
6352 | struct btrfs_space_info *space_info = bg->space_info; | |
6353 | ||
6354 | ASSERT(bg->ro); | |
6355 | ||
6356 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) | |
6357 | return; | |
6358 | ||
6359 | /* | |
6360 | * Our block group is read only but before we set it to read only, | |
6361 | * some task might have had allocated an extent from it already, but it | |
6362 | * has not yet created a respective ordered extent (and added it to a | |
6363 | * root's list of ordered extents). | |
6364 | * Therefore wait for any task currently allocating extents, since the | |
6365 | * block group's reservations counter is incremented while a read lock | |
6366 | * on the groups' semaphore is held and decremented after releasing | |
6367 | * the read access on that semaphore and creating the ordered extent. | |
6368 | */ | |
6369 | down_write(&space_info->groups_sem); | |
6370 | up_write(&space_info->groups_sem); | |
6371 | ||
4625956a | 6372 | wait_var_event(&bg->reservations, !atomic_read(&bg->reservations)); |
9cfa3e34 FM |
6373 | } |
6374 | ||
fb25e914 | 6375 | /** |
4824f1f4 | 6376 | * btrfs_add_reserved_bytes - update the block_group and space info counters |
fb25e914 | 6377 | * @cache: The cache we are manipulating |
18513091 WX |
6378 | * @ram_bytes: The number of bytes of file content, and will be same to |
6379 | * @num_bytes except for the compress path. | |
fb25e914 | 6380 | * @num_bytes: The number of bytes in question |
e570fd27 | 6381 | * @delalloc: The blocks are allocated for the delalloc write |
fb25e914 | 6382 | * |
745699ef XW |
6383 | * This is called by the allocator when it reserves space. If this is a |
6384 | * reservation and the block group has become read only we cannot make the | |
6385 | * reservation and return -EAGAIN, otherwise this function always succeeds. | |
f0486c68 | 6386 | */ |
4824f1f4 | 6387 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
18513091 | 6388 | u64 ram_bytes, u64 num_bytes, int delalloc) |
11833d66 | 6389 | { |
fb25e914 | 6390 | struct btrfs_space_info *space_info = cache->space_info; |
f0486c68 | 6391 | int ret = 0; |
79787eaa | 6392 | |
fb25e914 JB |
6393 | spin_lock(&space_info->lock); |
6394 | spin_lock(&cache->lock); | |
4824f1f4 WX |
6395 | if (cache->ro) { |
6396 | ret = -EAGAIN; | |
fb25e914 | 6397 | } else { |
4824f1f4 WX |
6398 | cache->reserved += num_bytes; |
6399 | space_info->bytes_reserved += num_bytes; | |
e570fd27 | 6400 | |
18513091 WX |
6401 | trace_btrfs_space_reservation(cache->fs_info, |
6402 | "space_info", space_info->flags, | |
6403 | ram_bytes, 0); | |
6404 | space_info->bytes_may_use -= ram_bytes; | |
e570fd27 | 6405 | if (delalloc) |
4824f1f4 | 6406 | cache->delalloc_bytes += num_bytes; |
324ae4df | 6407 | } |
fb25e914 JB |
6408 | spin_unlock(&cache->lock); |
6409 | spin_unlock(&space_info->lock); | |
f0486c68 | 6410 | return ret; |
324ae4df | 6411 | } |
9078a3e1 | 6412 | |
4824f1f4 WX |
6413 | /** |
6414 | * btrfs_free_reserved_bytes - update the block_group and space info counters | |
6415 | * @cache: The cache we are manipulating | |
6416 | * @num_bytes: The number of bytes in question | |
6417 | * @delalloc: The blocks are allocated for the delalloc write | |
6418 | * | |
6419 | * This is called by somebody who is freeing space that was never actually used | |
6420 | * on disk. For example if you reserve some space for a new leaf in transaction | |
6421 | * A and before transaction A commits you free that leaf, you call this with | |
6422 | * reserve set to 0 in order to clear the reservation. | |
6423 | */ | |
6424 | ||
556f3ca8 | 6425 | static void btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, |
6426 | u64 num_bytes, int delalloc) | |
4824f1f4 WX |
6427 | { |
6428 | struct btrfs_space_info *space_info = cache->space_info; | |
4824f1f4 WX |
6429 | |
6430 | spin_lock(&space_info->lock); | |
6431 | spin_lock(&cache->lock); | |
6432 | if (cache->ro) | |
6433 | space_info->bytes_readonly += num_bytes; | |
6434 | cache->reserved -= num_bytes; | |
6435 | space_info->bytes_reserved -= num_bytes; | |
6436 | ||
6437 | if (delalloc) | |
6438 | cache->delalloc_bytes -= num_bytes; | |
6439 | spin_unlock(&cache->lock); | |
6440 | spin_unlock(&space_info->lock); | |
4824f1f4 | 6441 | } |
8b74c03e | 6442 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
e8569813 | 6443 | { |
11833d66 YZ |
6444 | struct btrfs_caching_control *next; |
6445 | struct btrfs_caching_control *caching_ctl; | |
6446 | struct btrfs_block_group_cache *cache; | |
e8569813 | 6447 | |
9e351cc8 | 6448 | down_write(&fs_info->commit_root_sem); |
25179201 | 6449 | |
11833d66 YZ |
6450 | list_for_each_entry_safe(caching_ctl, next, |
6451 | &fs_info->caching_block_groups, list) { | |
6452 | cache = caching_ctl->block_group; | |
6453 | if (block_group_cache_done(cache)) { | |
6454 | cache->last_byte_to_unpin = (u64)-1; | |
6455 | list_del_init(&caching_ctl->list); | |
6456 | put_caching_control(caching_ctl); | |
e8569813 | 6457 | } else { |
11833d66 | 6458 | cache->last_byte_to_unpin = caching_ctl->progress; |
e8569813 | 6459 | } |
e8569813 | 6460 | } |
11833d66 YZ |
6461 | |
6462 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) | |
6463 | fs_info->pinned_extents = &fs_info->freed_extents[1]; | |
6464 | else | |
6465 | fs_info->pinned_extents = &fs_info->freed_extents[0]; | |
6466 | ||
9e351cc8 | 6467 | up_write(&fs_info->commit_root_sem); |
8929ecfa YZ |
6468 | |
6469 | update_global_block_rsv(fs_info); | |
e8569813 ZY |
6470 | } |
6471 | ||
c759c4e1 JB |
6472 | /* |
6473 | * Returns the free cluster for the given space info and sets empty_cluster to | |
6474 | * what it should be based on the mount options. | |
6475 | */ | |
6476 | static struct btrfs_free_cluster * | |
2ff7e61e JM |
6477 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
6478 | struct btrfs_space_info *space_info, u64 *empty_cluster) | |
c759c4e1 JB |
6479 | { |
6480 | struct btrfs_free_cluster *ret = NULL; | |
c759c4e1 JB |
6481 | |
6482 | *empty_cluster = 0; | |
6483 | if (btrfs_mixed_space_info(space_info)) | |
6484 | return ret; | |
6485 | ||
c759c4e1 | 6486 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
0b246afa | 6487 | ret = &fs_info->meta_alloc_cluster; |
583b7231 HK |
6488 | if (btrfs_test_opt(fs_info, SSD)) |
6489 | *empty_cluster = SZ_2M; | |
6490 | else | |
ee22184b | 6491 | *empty_cluster = SZ_64K; |
583b7231 HK |
6492 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
6493 | btrfs_test_opt(fs_info, SSD_SPREAD)) { | |
6494 | *empty_cluster = SZ_2M; | |
0b246afa | 6495 | ret = &fs_info->data_alloc_cluster; |
c759c4e1 JB |
6496 | } |
6497 | ||
6498 | return ret; | |
6499 | } | |
6500 | ||
2ff7e61e JM |
6501 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
6502 | u64 start, u64 end, | |
678886bd | 6503 | const bool return_free_space) |
ccd467d6 | 6504 | { |
11833d66 | 6505 | struct btrfs_block_group_cache *cache = NULL; |
7b398f8e JB |
6506 | struct btrfs_space_info *space_info; |
6507 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; | |
c759c4e1 | 6508 | struct btrfs_free_cluster *cluster = NULL; |
11833d66 | 6509 | u64 len; |
c759c4e1 JB |
6510 | u64 total_unpinned = 0; |
6511 | u64 empty_cluster = 0; | |
7b398f8e | 6512 | bool readonly; |
ccd467d6 | 6513 | |
11833d66 | 6514 | while (start <= end) { |
7b398f8e | 6515 | readonly = false; |
11833d66 YZ |
6516 | if (!cache || |
6517 | start >= cache->key.objectid + cache->key.offset) { | |
6518 | if (cache) | |
6519 | btrfs_put_block_group(cache); | |
c759c4e1 | 6520 | total_unpinned = 0; |
11833d66 | 6521 | cache = btrfs_lookup_block_group(fs_info, start); |
79787eaa | 6522 | BUG_ON(!cache); /* Logic error */ |
c759c4e1 | 6523 | |
2ff7e61e | 6524 | cluster = fetch_cluster_info(fs_info, |
c759c4e1 JB |
6525 | cache->space_info, |
6526 | &empty_cluster); | |
6527 | empty_cluster <<= 1; | |
11833d66 YZ |
6528 | } |
6529 | ||
6530 | len = cache->key.objectid + cache->key.offset - start; | |
6531 | len = min(len, end + 1 - start); | |
6532 | ||
6533 | if (start < cache->last_byte_to_unpin) { | |
6534 | len = min(len, cache->last_byte_to_unpin - start); | |
678886bd FM |
6535 | if (return_free_space) |
6536 | btrfs_add_free_space(cache, start, len); | |
11833d66 YZ |
6537 | } |
6538 | ||
f0486c68 | 6539 | start += len; |
c759c4e1 | 6540 | total_unpinned += len; |
7b398f8e | 6541 | space_info = cache->space_info; |
f0486c68 | 6542 | |
c759c4e1 JB |
6543 | /* |
6544 | * If this space cluster has been marked as fragmented and we've | |
6545 | * unpinned enough in this block group to potentially allow a | |
6546 | * cluster to be created inside of it go ahead and clear the | |
6547 | * fragmented check. | |
6548 | */ | |
6549 | if (cluster && cluster->fragmented && | |
6550 | total_unpinned > empty_cluster) { | |
6551 | spin_lock(&cluster->lock); | |
6552 | cluster->fragmented = 0; | |
6553 | spin_unlock(&cluster->lock); | |
6554 | } | |
6555 | ||
7b398f8e | 6556 | spin_lock(&space_info->lock); |
11833d66 YZ |
6557 | spin_lock(&cache->lock); |
6558 | cache->pinned -= len; | |
7b398f8e | 6559 | space_info->bytes_pinned -= len; |
c51e7bb1 JB |
6560 | |
6561 | trace_btrfs_space_reservation(fs_info, "pinned", | |
6562 | space_info->flags, len, 0); | |
4f4db217 | 6563 | space_info->max_extent_size = 0; |
dec59fa3 EL |
6564 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
6565 | -len, BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
7b398f8e JB |
6566 | if (cache->ro) { |
6567 | space_info->bytes_readonly += len; | |
6568 | readonly = true; | |
6569 | } | |
11833d66 | 6570 | spin_unlock(&cache->lock); |
957780eb JB |
6571 | if (!readonly && return_free_space && |
6572 | global_rsv->space_info == space_info) { | |
6573 | u64 to_add = len; | |
92ac58ec | 6574 | |
7b398f8e JB |
6575 | spin_lock(&global_rsv->lock); |
6576 | if (!global_rsv->full) { | |
957780eb JB |
6577 | to_add = min(len, global_rsv->size - |
6578 | global_rsv->reserved); | |
6579 | global_rsv->reserved += to_add; | |
6580 | space_info->bytes_may_use += to_add; | |
7b398f8e JB |
6581 | if (global_rsv->reserved >= global_rsv->size) |
6582 | global_rsv->full = 1; | |
957780eb JB |
6583 | trace_btrfs_space_reservation(fs_info, |
6584 | "space_info", | |
6585 | space_info->flags, | |
6586 | to_add, 1); | |
6587 | len -= to_add; | |
7b398f8e JB |
6588 | } |
6589 | spin_unlock(&global_rsv->lock); | |
957780eb JB |
6590 | /* Add to any tickets we may have */ |
6591 | if (len) | |
6592 | space_info_add_new_bytes(fs_info, space_info, | |
6593 | len); | |
7b398f8e JB |
6594 | } |
6595 | spin_unlock(&space_info->lock); | |
ccd467d6 | 6596 | } |
11833d66 YZ |
6597 | |
6598 | if (cache) | |
6599 | btrfs_put_block_group(cache); | |
ccd467d6 CM |
6600 | return 0; |
6601 | } | |
6602 | ||
5ead2dd0 | 6603 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
a28ec197 | 6604 | { |
5ead2dd0 | 6605 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e33e17ee JM |
6606 | struct btrfs_block_group_cache *block_group, *tmp; |
6607 | struct list_head *deleted_bgs; | |
11833d66 | 6608 | struct extent_io_tree *unpin; |
1a5bc167 CM |
6609 | u64 start; |
6610 | u64 end; | |
a28ec197 | 6611 | int ret; |
a28ec197 | 6612 | |
11833d66 YZ |
6613 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
6614 | unpin = &fs_info->freed_extents[1]; | |
6615 | else | |
6616 | unpin = &fs_info->freed_extents[0]; | |
6617 | ||
e33e17ee | 6618 | while (!trans->aborted) { |
d4b450cd | 6619 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
1a5bc167 | 6620 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
e6138876 | 6621 | EXTENT_DIRTY, NULL); |
d4b450cd FM |
6622 | if (ret) { |
6623 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); | |
a28ec197 | 6624 | break; |
d4b450cd | 6625 | } |
1f3c79a2 | 6626 | |
0b246afa | 6627 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 6628 | ret = btrfs_discard_extent(fs_info, start, |
5378e607 | 6629 | end + 1 - start, NULL); |
1f3c79a2 | 6630 | |
af6f8f60 | 6631 | clear_extent_dirty(unpin, start, end); |
2ff7e61e | 6632 | unpin_extent_range(fs_info, start, end, true); |
d4b450cd | 6633 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
b9473439 | 6634 | cond_resched(); |
a28ec197 | 6635 | } |
817d52f8 | 6636 | |
e33e17ee JM |
6637 | /* |
6638 | * Transaction is finished. We don't need the lock anymore. We | |
6639 | * do need to clean up the block groups in case of a transaction | |
6640 | * abort. | |
6641 | */ | |
6642 | deleted_bgs = &trans->transaction->deleted_bgs; | |
6643 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { | |
6644 | u64 trimmed = 0; | |
6645 | ||
6646 | ret = -EROFS; | |
6647 | if (!trans->aborted) | |
2ff7e61e | 6648 | ret = btrfs_discard_extent(fs_info, |
e33e17ee JM |
6649 | block_group->key.objectid, |
6650 | block_group->key.offset, | |
6651 | &trimmed); | |
6652 | ||
6653 | list_del_init(&block_group->bg_list); | |
6654 | btrfs_put_block_group_trimming(block_group); | |
6655 | btrfs_put_block_group(block_group); | |
6656 | ||
6657 | if (ret) { | |
6658 | const char *errstr = btrfs_decode_error(ret); | |
6659 | btrfs_warn(fs_info, | |
913e1535 | 6660 | "discard failed while removing blockgroup: errno=%d %s", |
e33e17ee JM |
6661 | ret, errstr); |
6662 | } | |
6663 | } | |
6664 | ||
e20d96d6 CM |
6665 | return 0; |
6666 | } | |
6667 | ||
5d4f98a2 | 6668 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
e72cb923 NB |
6669 | struct btrfs_delayed_ref_node *node, u64 parent, |
6670 | u64 root_objectid, u64 owner_objectid, | |
6671 | u64 owner_offset, int refs_to_drop, | |
6672 | struct btrfs_delayed_extent_op *extent_op) | |
a28ec197 | 6673 | { |
e72cb923 | 6674 | struct btrfs_fs_info *info = trans->fs_info; |
e2fa7227 | 6675 | struct btrfs_key key; |
5d4f98a2 | 6676 | struct btrfs_path *path; |
1261ec42 | 6677 | struct btrfs_root *extent_root = info->extent_root; |
5f39d397 | 6678 | struct extent_buffer *leaf; |
5d4f98a2 YZ |
6679 | struct btrfs_extent_item *ei; |
6680 | struct btrfs_extent_inline_ref *iref; | |
a28ec197 | 6681 | int ret; |
5d4f98a2 | 6682 | int is_data; |
952fccac CM |
6683 | int extent_slot = 0; |
6684 | int found_extent = 0; | |
6685 | int num_to_del = 1; | |
5d4f98a2 YZ |
6686 | u32 item_size; |
6687 | u64 refs; | |
c682f9b3 QW |
6688 | u64 bytenr = node->bytenr; |
6689 | u64 num_bytes = node->num_bytes; | |
fcebe456 | 6690 | int last_ref = 0; |
0b246afa | 6691 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
037e6390 | 6692 | |
5caf2a00 | 6693 | path = btrfs_alloc_path(); |
54aa1f4d CM |
6694 | if (!path) |
6695 | return -ENOMEM; | |
5f26f772 | 6696 | |
e4058b54 | 6697 | path->reada = READA_FORWARD; |
b9473439 | 6698 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6699 | |
6700 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; | |
6701 | BUG_ON(!is_data && refs_to_drop != 1); | |
6702 | ||
3173a18f | 6703 | if (is_data) |
897ca819 | 6704 | skinny_metadata = false; |
3173a18f | 6705 | |
fbe4801b NB |
6706 | ret = lookup_extent_backref(trans, path, &iref, bytenr, num_bytes, |
6707 | parent, root_objectid, owner_objectid, | |
5d4f98a2 | 6708 | owner_offset); |
7bb86316 | 6709 | if (ret == 0) { |
952fccac | 6710 | extent_slot = path->slots[0]; |
5d4f98a2 YZ |
6711 | while (extent_slot >= 0) { |
6712 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
952fccac | 6713 | extent_slot); |
5d4f98a2 | 6714 | if (key.objectid != bytenr) |
952fccac | 6715 | break; |
5d4f98a2 YZ |
6716 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
6717 | key.offset == num_bytes) { | |
952fccac CM |
6718 | found_extent = 1; |
6719 | break; | |
6720 | } | |
3173a18f JB |
6721 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
6722 | key.offset == owner_objectid) { | |
6723 | found_extent = 1; | |
6724 | break; | |
6725 | } | |
952fccac CM |
6726 | if (path->slots[0] - extent_slot > 5) |
6727 | break; | |
5d4f98a2 | 6728 | extent_slot--; |
952fccac | 6729 | } |
a79865c6 | 6730 | |
31840ae1 | 6731 | if (!found_extent) { |
5d4f98a2 | 6732 | BUG_ON(iref); |
87cc7a8a | 6733 | ret = remove_extent_backref(trans, path, NULL, |
87bde3cd | 6734 | refs_to_drop, |
fcebe456 | 6735 | is_data, &last_ref); |
005d6427 | 6736 | if (ret) { |
66642832 | 6737 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6738 | goto out; |
6739 | } | |
b3b4aa74 | 6740 | btrfs_release_path(path); |
b9473439 | 6741 | path->leave_spinning = 1; |
5d4f98a2 YZ |
6742 | |
6743 | key.objectid = bytenr; | |
6744 | key.type = BTRFS_EXTENT_ITEM_KEY; | |
6745 | key.offset = num_bytes; | |
6746 | ||
3173a18f JB |
6747 | if (!is_data && skinny_metadata) { |
6748 | key.type = BTRFS_METADATA_ITEM_KEY; | |
6749 | key.offset = owner_objectid; | |
6750 | } | |
6751 | ||
31840ae1 ZY |
6752 | ret = btrfs_search_slot(trans, extent_root, |
6753 | &key, path, -1, 1); | |
3173a18f JB |
6754 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
6755 | /* | |
6756 | * Couldn't find our skinny metadata item, | |
6757 | * see if we have ye olde extent item. | |
6758 | */ | |
6759 | path->slots[0]--; | |
6760 | btrfs_item_key_to_cpu(path->nodes[0], &key, | |
6761 | path->slots[0]); | |
6762 | if (key.objectid == bytenr && | |
6763 | key.type == BTRFS_EXTENT_ITEM_KEY && | |
6764 | key.offset == num_bytes) | |
6765 | ret = 0; | |
6766 | } | |
6767 | ||
6768 | if (ret > 0 && skinny_metadata) { | |
6769 | skinny_metadata = false; | |
9ce49a0b | 6770 | key.objectid = bytenr; |
3173a18f JB |
6771 | key.type = BTRFS_EXTENT_ITEM_KEY; |
6772 | key.offset = num_bytes; | |
6773 | btrfs_release_path(path); | |
6774 | ret = btrfs_search_slot(trans, extent_root, | |
6775 | &key, path, -1, 1); | |
6776 | } | |
6777 | ||
f3465ca4 | 6778 | if (ret) { |
5d163e0e JM |
6779 | btrfs_err(info, |
6780 | "umm, got %d back from search, was looking for %llu", | |
6781 | ret, bytenr); | |
b783e62d | 6782 | if (ret > 0) |
a4f78750 | 6783 | btrfs_print_leaf(path->nodes[0]); |
f3465ca4 | 6784 | } |
005d6427 | 6785 | if (ret < 0) { |
66642832 | 6786 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6787 | goto out; |
6788 | } | |
31840ae1 ZY |
6789 | extent_slot = path->slots[0]; |
6790 | } | |
fae7f21c | 6791 | } else if (WARN_ON(ret == -ENOENT)) { |
a4f78750 | 6792 | btrfs_print_leaf(path->nodes[0]); |
c2cf52eb SK |
6793 | btrfs_err(info, |
6794 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", | |
c1c9ff7c GU |
6795 | bytenr, parent, root_objectid, owner_objectid, |
6796 | owner_offset); | |
66642832 | 6797 | btrfs_abort_transaction(trans, ret); |
c4a050bb | 6798 | goto out; |
79787eaa | 6799 | } else { |
66642832 | 6800 | btrfs_abort_transaction(trans, ret); |
005d6427 | 6801 | goto out; |
7bb86316 | 6802 | } |
5f39d397 CM |
6803 | |
6804 | leaf = path->nodes[0]; | |
5d4f98a2 | 6805 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
6d8ff4e4 | 6806 | if (unlikely(item_size < sizeof(*ei))) { |
ba3c2b19 NB |
6807 | ret = -EINVAL; |
6808 | btrfs_print_v0_err(info); | |
6809 | btrfs_abort_transaction(trans, ret); | |
6810 | goto out; | |
6811 | } | |
952fccac | 6812 | ei = btrfs_item_ptr(leaf, extent_slot, |
123abc88 | 6813 | struct btrfs_extent_item); |
3173a18f JB |
6814 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
6815 | key.type == BTRFS_EXTENT_ITEM_KEY) { | |
5d4f98a2 YZ |
6816 | struct btrfs_tree_block_info *bi; |
6817 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); | |
6818 | bi = (struct btrfs_tree_block_info *)(ei + 1); | |
6819 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); | |
6820 | } | |
56bec294 | 6821 | |
5d4f98a2 | 6822 | refs = btrfs_extent_refs(leaf, ei); |
32b02538 | 6823 | if (refs < refs_to_drop) { |
5d163e0e JM |
6824 | btrfs_err(info, |
6825 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", | |
6826 | refs_to_drop, refs, bytenr); | |
32b02538 | 6827 | ret = -EINVAL; |
66642832 | 6828 | btrfs_abort_transaction(trans, ret); |
32b02538 JB |
6829 | goto out; |
6830 | } | |
56bec294 | 6831 | refs -= refs_to_drop; |
5f39d397 | 6832 | |
5d4f98a2 YZ |
6833 | if (refs > 0) { |
6834 | if (extent_op) | |
6835 | __run_delayed_extent_op(extent_op, leaf, ei); | |
6836 | /* | |
6837 | * In the case of inline back ref, reference count will | |
6838 | * be updated by remove_extent_backref | |
952fccac | 6839 | */ |
5d4f98a2 YZ |
6840 | if (iref) { |
6841 | BUG_ON(!found_extent); | |
6842 | } else { | |
6843 | btrfs_set_extent_refs(leaf, ei, refs); | |
6844 | btrfs_mark_buffer_dirty(leaf); | |
6845 | } | |
6846 | if (found_extent) { | |
87cc7a8a NB |
6847 | ret = remove_extent_backref(trans, path, iref, |
6848 | refs_to_drop, is_data, | |
6849 | &last_ref); | |
005d6427 | 6850 | if (ret) { |
66642832 | 6851 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6852 | goto out; |
6853 | } | |
952fccac | 6854 | } |
5d4f98a2 | 6855 | } else { |
5d4f98a2 YZ |
6856 | if (found_extent) { |
6857 | BUG_ON(is_data && refs_to_drop != | |
9ed0dea0 | 6858 | extent_data_ref_count(path, iref)); |
5d4f98a2 YZ |
6859 | if (iref) { |
6860 | BUG_ON(path->slots[0] != extent_slot); | |
6861 | } else { | |
6862 | BUG_ON(path->slots[0] != extent_slot + 1); | |
6863 | path->slots[0] = extent_slot; | |
6864 | num_to_del = 2; | |
6865 | } | |
78fae27e | 6866 | } |
b9473439 | 6867 | |
fcebe456 | 6868 | last_ref = 1; |
952fccac CM |
6869 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
6870 | num_to_del); | |
005d6427 | 6871 | if (ret) { |
66642832 | 6872 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6873 | goto out; |
6874 | } | |
b3b4aa74 | 6875 | btrfs_release_path(path); |
21af804c | 6876 | |
5d4f98a2 | 6877 | if (is_data) { |
5b4aacef | 6878 | ret = btrfs_del_csums(trans, info, bytenr, num_bytes); |
005d6427 | 6879 | if (ret) { |
66642832 | 6880 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6881 | goto out; |
6882 | } | |
459931ec CM |
6883 | } |
6884 | ||
e7355e50 | 6885 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
1e144fb8 | 6886 | if (ret) { |
66642832 | 6887 | btrfs_abort_transaction(trans, ret); |
1e144fb8 OS |
6888 | goto out; |
6889 | } | |
6890 | ||
0b246afa | 6891 | ret = update_block_group(trans, info, bytenr, num_bytes, 0); |
005d6427 | 6892 | if (ret) { |
66642832 | 6893 | btrfs_abort_transaction(trans, ret); |
005d6427 DS |
6894 | goto out; |
6895 | } | |
a28ec197 | 6896 | } |
fcebe456 JB |
6897 | btrfs_release_path(path); |
6898 | ||
79787eaa | 6899 | out: |
5caf2a00 | 6900 | btrfs_free_path(path); |
a28ec197 CM |
6901 | return ret; |
6902 | } | |
6903 | ||
1887be66 | 6904 | /* |
f0486c68 | 6905 | * when we free an block, it is possible (and likely) that we free the last |
1887be66 CM |
6906 | * delayed ref for that extent as well. This searches the delayed ref tree for |
6907 | * a given extent, and if there are no other delayed refs to be processed, it | |
6908 | * removes it from the tree. | |
6909 | */ | |
6910 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, | |
2ff7e61e | 6911 | u64 bytenr) |
1887be66 CM |
6912 | { |
6913 | struct btrfs_delayed_ref_head *head; | |
6914 | struct btrfs_delayed_ref_root *delayed_refs; | |
f0486c68 | 6915 | int ret = 0; |
1887be66 CM |
6916 | |
6917 | delayed_refs = &trans->transaction->delayed_refs; | |
6918 | spin_lock(&delayed_refs->lock); | |
f72ad18e | 6919 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
1887be66 | 6920 | if (!head) |
cf93da7b | 6921 | goto out_delayed_unlock; |
1887be66 | 6922 | |
d7df2c79 | 6923 | spin_lock(&head->lock); |
0e0adbcf | 6924 | if (!RB_EMPTY_ROOT(&head->ref_tree)) |
1887be66 CM |
6925 | goto out; |
6926 | ||
5d4f98a2 YZ |
6927 | if (head->extent_op) { |
6928 | if (!head->must_insert_reserved) | |
6929 | goto out; | |
78a6184a | 6930 | btrfs_free_delayed_extent_op(head->extent_op); |
5d4f98a2 YZ |
6931 | head->extent_op = NULL; |
6932 | } | |
6933 | ||
1887be66 CM |
6934 | /* |
6935 | * waiting for the lock here would deadlock. If someone else has it | |
6936 | * locked they are already in the process of dropping it anyway | |
6937 | */ | |
6938 | if (!mutex_trylock(&head->mutex)) | |
6939 | goto out; | |
6940 | ||
6941 | /* | |
6942 | * at this point we have a head with no other entries. Go | |
6943 | * ahead and process it. | |
6944 | */ | |
c46effa6 | 6945 | rb_erase(&head->href_node, &delayed_refs->href_root); |
d278850e | 6946 | RB_CLEAR_NODE(&head->href_node); |
d7df2c79 | 6947 | atomic_dec(&delayed_refs->num_entries); |
1887be66 CM |
6948 | |
6949 | /* | |
6950 | * we don't take a ref on the node because we're removing it from the | |
6951 | * tree, so we just steal the ref the tree was holding. | |
6952 | */ | |
c3e69d58 | 6953 | delayed_refs->num_heads--; |
d7df2c79 | 6954 | if (head->processing == 0) |
c3e69d58 | 6955 | delayed_refs->num_heads_ready--; |
d7df2c79 JB |
6956 | head->processing = 0; |
6957 | spin_unlock(&head->lock); | |
1887be66 CM |
6958 | spin_unlock(&delayed_refs->lock); |
6959 | ||
f0486c68 YZ |
6960 | BUG_ON(head->extent_op); |
6961 | if (head->must_insert_reserved) | |
6962 | ret = 1; | |
6963 | ||
6964 | mutex_unlock(&head->mutex); | |
d278850e | 6965 | btrfs_put_delayed_ref_head(head); |
f0486c68 | 6966 | return ret; |
1887be66 | 6967 | out: |
d7df2c79 | 6968 | spin_unlock(&head->lock); |
cf93da7b CM |
6969 | |
6970 | out_delayed_unlock: | |
1887be66 CM |
6971 | spin_unlock(&delayed_refs->lock); |
6972 | return 0; | |
6973 | } | |
6974 | ||
f0486c68 YZ |
6975 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
6976 | struct btrfs_root *root, | |
6977 | struct extent_buffer *buf, | |
5581a51a | 6978 | u64 parent, int last_ref) |
f0486c68 | 6979 | { |
0b246afa | 6980 | struct btrfs_fs_info *fs_info = root->fs_info; |
b150a4f1 | 6981 | int pin = 1; |
f0486c68 YZ |
6982 | int ret; |
6983 | ||
6984 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { | |
d7eae340 OS |
6985 | int old_ref_mod, new_ref_mod; |
6986 | ||
fd708b81 JB |
6987 | btrfs_ref_tree_mod(root, buf->start, buf->len, parent, |
6988 | root->root_key.objectid, | |
6989 | btrfs_header_level(buf), 0, | |
6990 | BTRFS_DROP_DELAYED_REF); | |
44e1c47d | 6991 | ret = btrfs_add_delayed_tree_ref(trans, buf->start, |
7be07912 | 6992 | buf->len, parent, |
0b246afa JM |
6993 | root->root_key.objectid, |
6994 | btrfs_header_level(buf), | |
7be07912 | 6995 | BTRFS_DROP_DELAYED_REF, NULL, |
d7eae340 | 6996 | &old_ref_mod, &new_ref_mod); |
79787eaa | 6997 | BUG_ON(ret); /* -ENOMEM */ |
d7eae340 | 6998 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
f0486c68 YZ |
6999 | } |
7000 | ||
0a16c7d7 | 7001 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
6219872d FM |
7002 | struct btrfs_block_group_cache *cache; |
7003 | ||
f0486c68 | 7004 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
2ff7e61e | 7005 | ret = check_ref_cleanup(trans, buf->start); |
f0486c68 | 7006 | if (!ret) |
37be25bc | 7007 | goto out; |
f0486c68 YZ |
7008 | } |
7009 | ||
4da8b76d | 7010 | pin = 0; |
0b246afa | 7011 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
6219872d | 7012 | |
f0486c68 | 7013 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
2ff7e61e JM |
7014 | pin_down_extent(fs_info, cache, buf->start, |
7015 | buf->len, 1); | |
6219872d | 7016 | btrfs_put_block_group(cache); |
37be25bc | 7017 | goto out; |
f0486c68 YZ |
7018 | } |
7019 | ||
7020 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); | |
7021 | ||
7022 | btrfs_add_free_space(cache, buf->start, buf->len); | |
4824f1f4 | 7023 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
6219872d | 7024 | btrfs_put_block_group(cache); |
71ff6437 | 7025 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
f0486c68 YZ |
7026 | } |
7027 | out: | |
b150a4f1 | 7028 | if (pin) |
29d2b84c | 7029 | add_pinned_bytes(fs_info, buf->len, true, |
b150a4f1 JB |
7030 | root->root_key.objectid); |
7031 | ||
0a16c7d7 OS |
7032 | if (last_ref) { |
7033 | /* | |
7034 | * Deleting the buffer, clear the corrupt flag since it doesn't | |
7035 | * matter anymore. | |
7036 | */ | |
7037 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); | |
7038 | } | |
f0486c68 YZ |
7039 | } |
7040 | ||
79787eaa | 7041 | /* Can return -ENOMEM */ |
2ff7e61e | 7042 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
84f7d8e6 | 7043 | struct btrfs_root *root, |
66d7e7f0 | 7044 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, |
b06c4bf5 | 7045 | u64 owner, u64 offset) |
925baedd | 7046 | { |
84f7d8e6 | 7047 | struct btrfs_fs_info *fs_info = root->fs_info; |
d7eae340 | 7048 | int old_ref_mod, new_ref_mod; |
925baedd CM |
7049 | int ret; |
7050 | ||
f5ee5c9a | 7051 | if (btrfs_is_testing(fs_info)) |
faa2dbf0 | 7052 | return 0; |
fccb84c9 | 7053 | |
fd708b81 JB |
7054 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) |
7055 | btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, | |
7056 | root_objectid, owner, offset, | |
7057 | BTRFS_DROP_DELAYED_REF); | |
7058 | ||
56bec294 CM |
7059 | /* |
7060 | * tree log blocks never actually go into the extent allocation | |
7061 | * tree, just update pinning info and exit early. | |
56bec294 | 7062 | */ |
5d4f98a2 YZ |
7063 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
7064 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); | |
b9473439 | 7065 | /* unlocks the pinned mutex */ |
2ff7e61e | 7066 | btrfs_pin_extent(fs_info, bytenr, num_bytes, 1); |
d7eae340 | 7067 | old_ref_mod = new_ref_mod = 0; |
56bec294 | 7068 | ret = 0; |
5d4f98a2 | 7069 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
44e1c47d | 7070 | ret = btrfs_add_delayed_tree_ref(trans, bytenr, |
7be07912 OS |
7071 | num_bytes, parent, |
7072 | root_objectid, (int)owner, | |
7073 | BTRFS_DROP_DELAYED_REF, NULL, | |
d7eae340 | 7074 | &old_ref_mod, &new_ref_mod); |
5d4f98a2 | 7075 | } else { |
88a979c6 | 7076 | ret = btrfs_add_delayed_data_ref(trans, bytenr, |
7be07912 OS |
7077 | num_bytes, parent, |
7078 | root_objectid, owner, offset, | |
7079 | 0, BTRFS_DROP_DELAYED_REF, | |
d7eae340 | 7080 | &old_ref_mod, &new_ref_mod); |
56bec294 | 7081 | } |
d7eae340 | 7082 | |
29d2b84c NB |
7083 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) { |
7084 | bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID; | |
7085 | ||
7086 | add_pinned_bytes(fs_info, num_bytes, metadata, root_objectid); | |
7087 | } | |
d7eae340 | 7088 | |
925baedd CM |
7089 | return ret; |
7090 | } | |
7091 | ||
817d52f8 JB |
7092 | /* |
7093 | * when we wait for progress in the block group caching, its because | |
7094 | * our allocation attempt failed at least once. So, we must sleep | |
7095 | * and let some progress happen before we try again. | |
7096 | * | |
7097 | * This function will sleep at least once waiting for new free space to | |
7098 | * show up, and then it will check the block group free space numbers | |
7099 | * for our min num_bytes. Another option is to have it go ahead | |
7100 | * and look in the rbtree for a free extent of a given size, but this | |
7101 | * is a good start. | |
36cce922 JB |
7102 | * |
7103 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using | |
7104 | * any of the information in this block group. | |
817d52f8 | 7105 | */ |
36cce922 | 7106 | static noinline void |
817d52f8 JB |
7107 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
7108 | u64 num_bytes) | |
7109 | { | |
11833d66 | 7110 | struct btrfs_caching_control *caching_ctl; |
817d52f8 | 7111 | |
11833d66 YZ |
7112 | caching_ctl = get_caching_control(cache); |
7113 | if (!caching_ctl) | |
36cce922 | 7114 | return; |
817d52f8 | 7115 | |
11833d66 | 7116 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
34d52cb6 | 7117 | (cache->free_space_ctl->free_space >= num_bytes)); |
11833d66 YZ |
7118 | |
7119 | put_caching_control(caching_ctl); | |
11833d66 YZ |
7120 | } |
7121 | ||
7122 | static noinline int | |
7123 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) | |
7124 | { | |
7125 | struct btrfs_caching_control *caching_ctl; | |
36cce922 | 7126 | int ret = 0; |
11833d66 YZ |
7127 | |
7128 | caching_ctl = get_caching_control(cache); | |
7129 | if (!caching_ctl) | |
36cce922 | 7130 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
11833d66 YZ |
7131 | |
7132 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); | |
36cce922 JB |
7133 | if (cache->cached == BTRFS_CACHE_ERROR) |
7134 | ret = -EIO; | |
11833d66 | 7135 | put_caching_control(caching_ctl); |
36cce922 | 7136 | return ret; |
817d52f8 JB |
7137 | } |
7138 | ||
7139 | enum btrfs_loop_type { | |
285ff5af JB |
7140 | LOOP_CACHING_NOWAIT = 0, |
7141 | LOOP_CACHING_WAIT = 1, | |
7142 | LOOP_ALLOC_CHUNK = 2, | |
7143 | LOOP_NO_EMPTY_SIZE = 3, | |
817d52f8 JB |
7144 | }; |
7145 | ||
e570fd27 MX |
7146 | static inline void |
7147 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, | |
7148 | int delalloc) | |
7149 | { | |
7150 | if (delalloc) | |
7151 | down_read(&cache->data_rwsem); | |
7152 | } | |
7153 | ||
7154 | static inline void | |
7155 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, | |
7156 | int delalloc) | |
7157 | { | |
7158 | btrfs_get_block_group(cache); | |
7159 | if (delalloc) | |
7160 | down_read(&cache->data_rwsem); | |
7161 | } | |
7162 | ||
7163 | static struct btrfs_block_group_cache * | |
7164 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, | |
7165 | struct btrfs_free_cluster *cluster, | |
7166 | int delalloc) | |
7167 | { | |
89771cc9 | 7168 | struct btrfs_block_group_cache *used_bg = NULL; |
6719afdc | 7169 | |
e570fd27 | 7170 | spin_lock(&cluster->refill_lock); |
6719afdc GU |
7171 | while (1) { |
7172 | used_bg = cluster->block_group; | |
7173 | if (!used_bg) | |
7174 | return NULL; | |
7175 | ||
7176 | if (used_bg == block_group) | |
e570fd27 MX |
7177 | return used_bg; |
7178 | ||
6719afdc | 7179 | btrfs_get_block_group(used_bg); |
e570fd27 | 7180 | |
6719afdc GU |
7181 | if (!delalloc) |
7182 | return used_bg; | |
e570fd27 | 7183 | |
6719afdc GU |
7184 | if (down_read_trylock(&used_bg->data_rwsem)) |
7185 | return used_bg; | |
e570fd27 | 7186 | |
6719afdc | 7187 | spin_unlock(&cluster->refill_lock); |
e570fd27 | 7188 | |
e321f8a8 LB |
7189 | /* We should only have one-level nested. */ |
7190 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); | |
e570fd27 | 7191 | |
6719afdc GU |
7192 | spin_lock(&cluster->refill_lock); |
7193 | if (used_bg == cluster->block_group) | |
7194 | return used_bg; | |
e570fd27 | 7195 | |
6719afdc GU |
7196 | up_read(&used_bg->data_rwsem); |
7197 | btrfs_put_block_group(used_bg); | |
7198 | } | |
e570fd27 MX |
7199 | } |
7200 | ||
7201 | static inline void | |
7202 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, | |
7203 | int delalloc) | |
7204 | { | |
7205 | if (delalloc) | |
7206 | up_read(&cache->data_rwsem); | |
7207 | btrfs_put_block_group(cache); | |
7208 | } | |
7209 | ||
fec577fb CM |
7210 | /* |
7211 | * walks the btree of allocated extents and find a hole of a given size. | |
7212 | * The key ins is changed to record the hole: | |
a4820398 | 7213 | * ins->objectid == start position |
62e2749e | 7214 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
a4820398 | 7215 | * ins->offset == the size of the hole. |
fec577fb | 7216 | * Any available blocks before search_start are skipped. |
a4820398 MX |
7217 | * |
7218 | * If there is no suitable free space, we will record the max size of | |
7219 | * the free space extent currently. | |
fec577fb | 7220 | */ |
87bde3cd | 7221 | static noinline int find_free_extent(struct btrfs_fs_info *fs_info, |
18513091 WX |
7222 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
7223 | u64 hint_byte, struct btrfs_key *ins, | |
7224 | u64 flags, int delalloc) | |
fec577fb | 7225 | { |
80eb234a | 7226 | int ret = 0; |
0b246afa | 7227 | struct btrfs_root *root = fs_info->extent_root; |
fa9c0d79 | 7228 | struct btrfs_free_cluster *last_ptr = NULL; |
80eb234a | 7229 | struct btrfs_block_group_cache *block_group = NULL; |
81c9ad23 | 7230 | u64 search_start = 0; |
a4820398 | 7231 | u64 max_extent_size = 0; |
c759c4e1 | 7232 | u64 empty_cluster = 0; |
80eb234a | 7233 | struct btrfs_space_info *space_info; |
fa9c0d79 | 7234 | int loop = 0; |
3e72ee88 | 7235 | int index = btrfs_bg_flags_to_raid_index(flags); |
0a24325e | 7236 | bool failed_cluster_refill = false; |
1cdda9b8 | 7237 | bool failed_alloc = false; |
67377734 | 7238 | bool use_cluster = true; |
60d2adbb | 7239 | bool have_caching_bg = false; |
13a0db5a | 7240 | bool orig_have_caching_bg = false; |
a5e681d9 | 7241 | bool full_search = false; |
fec577fb | 7242 | |
0b246afa | 7243 | WARN_ON(num_bytes < fs_info->sectorsize); |
962a298f | 7244 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
80eb234a JB |
7245 | ins->objectid = 0; |
7246 | ins->offset = 0; | |
b1a4d965 | 7247 | |
71ff6437 | 7248 | trace_find_free_extent(fs_info, num_bytes, empty_size, flags); |
3f7de037 | 7249 | |
0b246afa | 7250 | space_info = __find_space_info(fs_info, flags); |
1b1d1f66 | 7251 | if (!space_info) { |
0b246afa | 7252 | btrfs_err(fs_info, "No space info for %llu", flags); |
1b1d1f66 JB |
7253 | return -ENOSPC; |
7254 | } | |
2552d17e | 7255 | |
67377734 | 7256 | /* |
4f4db217 JB |
7257 | * If our free space is heavily fragmented we may not be able to make |
7258 | * big contiguous allocations, so instead of doing the expensive search | |
7259 | * for free space, simply return ENOSPC with our max_extent_size so we | |
7260 | * can go ahead and search for a more manageable chunk. | |
7261 | * | |
7262 | * If our max_extent_size is large enough for our allocation simply | |
7263 | * disable clustering since we will likely not be able to find enough | |
7264 | * space to create a cluster and induce latency trying. | |
67377734 | 7265 | */ |
4f4db217 JB |
7266 | if (unlikely(space_info->max_extent_size)) { |
7267 | spin_lock(&space_info->lock); | |
7268 | if (space_info->max_extent_size && | |
7269 | num_bytes > space_info->max_extent_size) { | |
7270 | ins->offset = space_info->max_extent_size; | |
7271 | spin_unlock(&space_info->lock); | |
7272 | return -ENOSPC; | |
7273 | } else if (space_info->max_extent_size) { | |
7274 | use_cluster = false; | |
7275 | } | |
7276 | spin_unlock(&space_info->lock); | |
fa9c0d79 | 7277 | } |
0f9dd46c | 7278 | |
2ff7e61e | 7279 | last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster); |
239b14b3 | 7280 | if (last_ptr) { |
fa9c0d79 CM |
7281 | spin_lock(&last_ptr->lock); |
7282 | if (last_ptr->block_group) | |
7283 | hint_byte = last_ptr->window_start; | |
c759c4e1 JB |
7284 | if (last_ptr->fragmented) { |
7285 | /* | |
7286 | * We still set window_start so we can keep track of the | |
7287 | * last place we found an allocation to try and save | |
7288 | * some time. | |
7289 | */ | |
7290 | hint_byte = last_ptr->window_start; | |
7291 | use_cluster = false; | |
7292 | } | |
fa9c0d79 | 7293 | spin_unlock(&last_ptr->lock); |
239b14b3 | 7294 | } |
fa9c0d79 | 7295 | |
2ff7e61e | 7296 | search_start = max(search_start, first_logical_byte(fs_info, 0)); |
239b14b3 | 7297 | search_start = max(search_start, hint_byte); |
2552d17e | 7298 | if (search_start == hint_byte) { |
0b246afa | 7299 | block_group = btrfs_lookup_block_group(fs_info, search_start); |
817d52f8 JB |
7300 | /* |
7301 | * we don't want to use the block group if it doesn't match our | |
7302 | * allocation bits, or if its not cached. | |
ccf0e725 JB |
7303 | * |
7304 | * However if we are re-searching with an ideal block group | |
7305 | * picked out then we don't care that the block group is cached. | |
817d52f8 | 7306 | */ |
b6919a58 | 7307 | if (block_group && block_group_bits(block_group, flags) && |
285ff5af | 7308 | block_group->cached != BTRFS_CACHE_NO) { |
2552d17e | 7309 | down_read(&space_info->groups_sem); |
44fb5511 CM |
7310 | if (list_empty(&block_group->list) || |
7311 | block_group->ro) { | |
7312 | /* | |
7313 | * someone is removing this block group, | |
7314 | * we can't jump into the have_block_group | |
7315 | * target because our list pointers are not | |
7316 | * valid | |
7317 | */ | |
7318 | btrfs_put_block_group(block_group); | |
7319 | up_read(&space_info->groups_sem); | |
ccf0e725 | 7320 | } else { |
3e72ee88 QW |
7321 | index = btrfs_bg_flags_to_raid_index( |
7322 | block_group->flags); | |
e570fd27 | 7323 | btrfs_lock_block_group(block_group, delalloc); |
44fb5511 | 7324 | goto have_block_group; |
ccf0e725 | 7325 | } |
2552d17e | 7326 | } else if (block_group) { |
fa9c0d79 | 7327 | btrfs_put_block_group(block_group); |
2552d17e | 7328 | } |
42e70e7a | 7329 | } |
2552d17e | 7330 | search: |
60d2adbb | 7331 | have_caching_bg = false; |
3e72ee88 | 7332 | if (index == 0 || index == btrfs_bg_flags_to_raid_index(flags)) |
a5e681d9 | 7333 | full_search = true; |
80eb234a | 7334 | down_read(&space_info->groups_sem); |
b742bb82 YZ |
7335 | list_for_each_entry(block_group, &space_info->block_groups[index], |
7336 | list) { | |
6226cb0a | 7337 | u64 offset; |
817d52f8 | 7338 | int cached; |
8a1413a2 | 7339 | |
14443937 JM |
7340 | /* If the block group is read-only, we can skip it entirely. */ |
7341 | if (unlikely(block_group->ro)) | |
7342 | continue; | |
7343 | ||
e570fd27 | 7344 | btrfs_grab_block_group(block_group, delalloc); |
2552d17e | 7345 | search_start = block_group->key.objectid; |
42e70e7a | 7346 | |
83a50de9 CM |
7347 | /* |
7348 | * this can happen if we end up cycling through all the | |
7349 | * raid types, but we want to make sure we only allocate | |
7350 | * for the proper type. | |
7351 | */ | |
b6919a58 | 7352 | if (!block_group_bits(block_group, flags)) { |
bece2e82 | 7353 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
83a50de9 | 7354 | BTRFS_BLOCK_GROUP_RAID1 | |
53b381b3 DW |
7355 | BTRFS_BLOCK_GROUP_RAID5 | |
7356 | BTRFS_BLOCK_GROUP_RAID6 | | |
83a50de9 CM |
7357 | BTRFS_BLOCK_GROUP_RAID10; |
7358 | ||
7359 | /* | |
7360 | * if they asked for extra copies and this block group | |
7361 | * doesn't provide them, bail. This does allow us to | |
7362 | * fill raid0 from raid1. | |
7363 | */ | |
b6919a58 | 7364 | if ((flags & extra) && !(block_group->flags & extra)) |
83a50de9 CM |
7365 | goto loop; |
7366 | } | |
7367 | ||
2552d17e | 7368 | have_block_group: |
291c7d2f JB |
7369 | cached = block_group_cache_done(block_group); |
7370 | if (unlikely(!cached)) { | |
a5e681d9 | 7371 | have_caching_bg = true; |
f6373bf3 | 7372 | ret = cache_block_group(block_group, 0); |
1d4284bd CM |
7373 | BUG_ON(ret < 0); |
7374 | ret = 0; | |
817d52f8 JB |
7375 | } |
7376 | ||
36cce922 JB |
7377 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
7378 | goto loop; | |
0f9dd46c | 7379 | |
0a24325e | 7380 | /* |
062c05c4 AO |
7381 | * Ok we want to try and use the cluster allocator, so |
7382 | * lets look there | |
0a24325e | 7383 | */ |
c759c4e1 | 7384 | if (last_ptr && use_cluster) { |
215a63d1 | 7385 | struct btrfs_block_group_cache *used_block_group; |
8de972b4 | 7386 | unsigned long aligned_cluster; |
fa9c0d79 CM |
7387 | /* |
7388 | * the refill lock keeps out other | |
7389 | * people trying to start a new cluster | |
7390 | */ | |
e570fd27 MX |
7391 | used_block_group = btrfs_lock_cluster(block_group, |
7392 | last_ptr, | |
7393 | delalloc); | |
7394 | if (!used_block_group) | |
44fb5511 | 7395 | goto refill_cluster; |
274bd4fb | 7396 | |
e570fd27 MX |
7397 | if (used_block_group != block_group && |
7398 | (used_block_group->ro || | |
7399 | !block_group_bits(used_block_group, flags))) | |
7400 | goto release_cluster; | |
44fb5511 | 7401 | |
274bd4fb | 7402 | offset = btrfs_alloc_from_cluster(used_block_group, |
a4820398 MX |
7403 | last_ptr, |
7404 | num_bytes, | |
7405 | used_block_group->key.objectid, | |
7406 | &max_extent_size); | |
fa9c0d79 CM |
7407 | if (offset) { |
7408 | /* we have a block, we're done */ | |
7409 | spin_unlock(&last_ptr->refill_lock); | |
3dca5c94 | 7410 | trace_btrfs_reserve_extent_cluster( |
89d4346a MX |
7411 | used_block_group, |
7412 | search_start, num_bytes); | |
215a63d1 | 7413 | if (used_block_group != block_group) { |
e570fd27 MX |
7414 | btrfs_release_block_group(block_group, |
7415 | delalloc); | |
215a63d1 MX |
7416 | block_group = used_block_group; |
7417 | } | |
fa9c0d79 CM |
7418 | goto checks; |
7419 | } | |
7420 | ||
274bd4fb | 7421 | WARN_ON(last_ptr->block_group != used_block_group); |
e570fd27 | 7422 | release_cluster: |
062c05c4 AO |
7423 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
7424 | * set up a new clusters, so lets just skip it | |
7425 | * and let the allocator find whatever block | |
7426 | * it can find. If we reach this point, we | |
7427 | * will have tried the cluster allocator | |
7428 | * plenty of times and not have found | |
7429 | * anything, so we are likely way too | |
7430 | * fragmented for the clustering stuff to find | |
a5f6f719 AO |
7431 | * anything. |
7432 | * | |
7433 | * However, if the cluster is taken from the | |
7434 | * current block group, release the cluster | |
7435 | * first, so that we stand a better chance of | |
7436 | * succeeding in the unclustered | |
7437 | * allocation. */ | |
7438 | if (loop >= LOOP_NO_EMPTY_SIZE && | |
e570fd27 | 7439 | used_block_group != block_group) { |
062c05c4 | 7440 | spin_unlock(&last_ptr->refill_lock); |
e570fd27 MX |
7441 | btrfs_release_block_group(used_block_group, |
7442 | delalloc); | |
062c05c4 AO |
7443 | goto unclustered_alloc; |
7444 | } | |
7445 | ||
fa9c0d79 CM |
7446 | /* |
7447 | * this cluster didn't work out, free it and | |
7448 | * start over | |
7449 | */ | |
7450 | btrfs_return_cluster_to_free_space(NULL, last_ptr); | |
7451 | ||
e570fd27 MX |
7452 | if (used_block_group != block_group) |
7453 | btrfs_release_block_group(used_block_group, | |
7454 | delalloc); | |
7455 | refill_cluster: | |
a5f6f719 AO |
7456 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
7457 | spin_unlock(&last_ptr->refill_lock); | |
7458 | goto unclustered_alloc; | |
7459 | } | |
7460 | ||
8de972b4 CM |
7461 | aligned_cluster = max_t(unsigned long, |
7462 | empty_cluster + empty_size, | |
7463 | block_group->full_stripe_len); | |
7464 | ||
fa9c0d79 | 7465 | /* allocate a cluster in this block group */ |
2ff7e61e | 7466 | ret = btrfs_find_space_cluster(fs_info, block_group, |
00361589 JB |
7467 | last_ptr, search_start, |
7468 | num_bytes, | |
7469 | aligned_cluster); | |
fa9c0d79 CM |
7470 | if (ret == 0) { |
7471 | /* | |
7472 | * now pull our allocation out of this | |
7473 | * cluster | |
7474 | */ | |
7475 | offset = btrfs_alloc_from_cluster(block_group, | |
a4820398 MX |
7476 | last_ptr, |
7477 | num_bytes, | |
7478 | search_start, | |
7479 | &max_extent_size); | |
fa9c0d79 CM |
7480 | if (offset) { |
7481 | /* we found one, proceed */ | |
7482 | spin_unlock(&last_ptr->refill_lock); | |
3dca5c94 | 7483 | trace_btrfs_reserve_extent_cluster( |
3f7de037 JB |
7484 | block_group, search_start, |
7485 | num_bytes); | |
fa9c0d79 CM |
7486 | goto checks; |
7487 | } | |
0a24325e JB |
7488 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
7489 | && !failed_cluster_refill) { | |
817d52f8 JB |
7490 | spin_unlock(&last_ptr->refill_lock); |
7491 | ||
0a24325e | 7492 | failed_cluster_refill = true; |
817d52f8 JB |
7493 | wait_block_group_cache_progress(block_group, |
7494 | num_bytes + empty_cluster + empty_size); | |
7495 | goto have_block_group; | |
fa9c0d79 | 7496 | } |
817d52f8 | 7497 | |
fa9c0d79 CM |
7498 | /* |
7499 | * at this point we either didn't find a cluster | |
7500 | * or we weren't able to allocate a block from our | |
7501 | * cluster. Free the cluster we've been trying | |
7502 | * to use, and go to the next block group | |
7503 | */ | |
0a24325e | 7504 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
fa9c0d79 | 7505 | spin_unlock(&last_ptr->refill_lock); |
0a24325e | 7506 | goto loop; |
fa9c0d79 CM |
7507 | } |
7508 | ||
062c05c4 | 7509 | unclustered_alloc: |
c759c4e1 JB |
7510 | /* |
7511 | * We are doing an unclustered alloc, set the fragmented flag so | |
7512 | * we don't bother trying to setup a cluster again until we get | |
7513 | * more space. | |
7514 | */ | |
7515 | if (unlikely(last_ptr)) { | |
7516 | spin_lock(&last_ptr->lock); | |
7517 | last_ptr->fragmented = 1; | |
7518 | spin_unlock(&last_ptr->lock); | |
7519 | } | |
0c9b36e0 LB |
7520 | if (cached) { |
7521 | struct btrfs_free_space_ctl *ctl = | |
7522 | block_group->free_space_ctl; | |
7523 | ||
7524 | spin_lock(&ctl->tree_lock); | |
7525 | if (ctl->free_space < | |
7526 | num_bytes + empty_cluster + empty_size) { | |
7527 | if (ctl->free_space > max_extent_size) | |
7528 | max_extent_size = ctl->free_space; | |
7529 | spin_unlock(&ctl->tree_lock); | |
7530 | goto loop; | |
7531 | } | |
7532 | spin_unlock(&ctl->tree_lock); | |
a5f6f719 | 7533 | } |
a5f6f719 | 7534 | |
6226cb0a | 7535 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
a4820398 MX |
7536 | num_bytes, empty_size, |
7537 | &max_extent_size); | |
1cdda9b8 JB |
7538 | /* |
7539 | * If we didn't find a chunk, and we haven't failed on this | |
7540 | * block group before, and this block group is in the middle of | |
7541 | * caching and we are ok with waiting, then go ahead and wait | |
7542 | * for progress to be made, and set failed_alloc to true. | |
7543 | * | |
7544 | * If failed_alloc is true then we've already waited on this | |
7545 | * block group once and should move on to the next block group. | |
7546 | */ | |
7547 | if (!offset && !failed_alloc && !cached && | |
7548 | loop > LOOP_CACHING_NOWAIT) { | |
817d52f8 | 7549 | wait_block_group_cache_progress(block_group, |
1cdda9b8 JB |
7550 | num_bytes + empty_size); |
7551 | failed_alloc = true; | |
817d52f8 | 7552 | goto have_block_group; |
1cdda9b8 JB |
7553 | } else if (!offset) { |
7554 | goto loop; | |
817d52f8 | 7555 | } |
fa9c0d79 | 7556 | checks: |
5e23a6fe | 7557 | search_start = round_up(offset, fs_info->stripesize); |
25179201 | 7558 | |
2552d17e JB |
7559 | /* move on to the next group */ |
7560 | if (search_start + num_bytes > | |
215a63d1 MX |
7561 | block_group->key.objectid + block_group->key.offset) { |
7562 | btrfs_add_free_space(block_group, offset, num_bytes); | |
2552d17e | 7563 | goto loop; |
6226cb0a | 7564 | } |
f5a31e16 | 7565 | |
f0486c68 | 7566 | if (offset < search_start) |
215a63d1 | 7567 | btrfs_add_free_space(block_group, offset, |
f0486c68 | 7568 | search_start - offset); |
2552d17e | 7569 | |
18513091 WX |
7570 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
7571 | num_bytes, delalloc); | |
f0486c68 | 7572 | if (ret == -EAGAIN) { |
215a63d1 | 7573 | btrfs_add_free_space(block_group, offset, num_bytes); |
2552d17e | 7574 | goto loop; |
0f9dd46c | 7575 | } |
9cfa3e34 | 7576 | btrfs_inc_block_group_reservations(block_group); |
0b86a832 | 7577 | |
f0486c68 | 7578 | /* we are all good, lets return */ |
2552d17e JB |
7579 | ins->objectid = search_start; |
7580 | ins->offset = num_bytes; | |
d2fb3437 | 7581 | |
3dca5c94 | 7582 | trace_btrfs_reserve_extent(block_group, search_start, num_bytes); |
e570fd27 | 7583 | btrfs_release_block_group(block_group, delalloc); |
2552d17e JB |
7584 | break; |
7585 | loop: | |
0a24325e | 7586 | failed_cluster_refill = false; |
1cdda9b8 | 7587 | failed_alloc = false; |
3e72ee88 QW |
7588 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != |
7589 | index); | |
e570fd27 | 7590 | btrfs_release_block_group(block_group, delalloc); |
14443937 | 7591 | cond_resched(); |
2552d17e JB |
7592 | } |
7593 | up_read(&space_info->groups_sem); | |
7594 | ||
13a0db5a | 7595 | if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg |
7596 | && !orig_have_caching_bg) | |
7597 | orig_have_caching_bg = true; | |
7598 | ||
60d2adbb MX |
7599 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
7600 | goto search; | |
7601 | ||
b742bb82 YZ |
7602 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
7603 | goto search; | |
7604 | ||
285ff5af | 7605 | /* |
ccf0e725 JB |
7606 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
7607 | * caching kthreads as we move along | |
817d52f8 JB |
7608 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
7609 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again | |
7610 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try | |
7611 | * again | |
fa9c0d79 | 7612 | */ |
723bda20 | 7613 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
b742bb82 | 7614 | index = 0; |
a5e681d9 JB |
7615 | if (loop == LOOP_CACHING_NOWAIT) { |
7616 | /* | |
7617 | * We want to skip the LOOP_CACHING_WAIT step if we | |
01327610 | 7618 | * don't have any uncached bgs and we've already done a |
a5e681d9 JB |
7619 | * full search through. |
7620 | */ | |
13a0db5a | 7621 | if (orig_have_caching_bg || !full_search) |
a5e681d9 JB |
7622 | loop = LOOP_CACHING_WAIT; |
7623 | else | |
7624 | loop = LOOP_ALLOC_CHUNK; | |
7625 | } else { | |
7626 | loop++; | |
7627 | } | |
7628 | ||
817d52f8 | 7629 | if (loop == LOOP_ALLOC_CHUNK) { |
00361589 | 7630 | struct btrfs_trans_handle *trans; |
f017f15f WS |
7631 | int exist = 0; |
7632 | ||
7633 | trans = current->journal_info; | |
7634 | if (trans) | |
7635 | exist = 1; | |
7636 | else | |
7637 | trans = btrfs_join_transaction(root); | |
00361589 | 7638 | |
00361589 JB |
7639 | if (IS_ERR(trans)) { |
7640 | ret = PTR_ERR(trans); | |
7641 | goto out; | |
7642 | } | |
7643 | ||
01458828 | 7644 | ret = do_chunk_alloc(trans, flags, CHUNK_ALLOC_FORCE); |
a5e681d9 JB |
7645 | |
7646 | /* | |
7647 | * If we can't allocate a new chunk we've already looped | |
7648 | * through at least once, move on to the NO_EMPTY_SIZE | |
7649 | * case. | |
7650 | */ | |
7651 | if (ret == -ENOSPC) | |
7652 | loop = LOOP_NO_EMPTY_SIZE; | |
7653 | ||
ea658bad JB |
7654 | /* |
7655 | * Do not bail out on ENOSPC since we | |
7656 | * can do more things. | |
7657 | */ | |
00361589 | 7658 | if (ret < 0 && ret != -ENOSPC) |
66642832 | 7659 | btrfs_abort_transaction(trans, ret); |
00361589 JB |
7660 | else |
7661 | ret = 0; | |
f017f15f | 7662 | if (!exist) |
3a45bb20 | 7663 | btrfs_end_transaction(trans); |
00361589 | 7664 | if (ret) |
ea658bad | 7665 | goto out; |
2552d17e JB |
7666 | } |
7667 | ||
723bda20 | 7668 | if (loop == LOOP_NO_EMPTY_SIZE) { |
a5e681d9 JB |
7669 | /* |
7670 | * Don't loop again if we already have no empty_size and | |
7671 | * no empty_cluster. | |
7672 | */ | |
7673 | if (empty_size == 0 && | |
7674 | empty_cluster == 0) { | |
7675 | ret = -ENOSPC; | |
7676 | goto out; | |
7677 | } | |
723bda20 JB |
7678 | empty_size = 0; |
7679 | empty_cluster = 0; | |
fa9c0d79 | 7680 | } |
723bda20 JB |
7681 | |
7682 | goto search; | |
2552d17e JB |
7683 | } else if (!ins->objectid) { |
7684 | ret = -ENOSPC; | |
d82a6f1d | 7685 | } else if (ins->objectid) { |
c759c4e1 JB |
7686 | if (!use_cluster && last_ptr) { |
7687 | spin_lock(&last_ptr->lock); | |
7688 | last_ptr->window_start = ins->objectid; | |
7689 | spin_unlock(&last_ptr->lock); | |
7690 | } | |
80eb234a | 7691 | ret = 0; |
be744175 | 7692 | } |
79787eaa | 7693 | out: |
4f4db217 JB |
7694 | if (ret == -ENOSPC) { |
7695 | spin_lock(&space_info->lock); | |
7696 | space_info->max_extent_size = max_extent_size; | |
7697 | spin_unlock(&space_info->lock); | |
a4820398 | 7698 | ins->offset = max_extent_size; |
4f4db217 | 7699 | } |
0f70abe2 | 7700 | return ret; |
fec577fb | 7701 | } |
ec44a35c | 7702 | |
ab8d0fc4 JM |
7703 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
7704 | struct btrfs_space_info *info, u64 bytes, | |
9ed74f2d | 7705 | int dump_block_groups) |
0f9dd46c JB |
7706 | { |
7707 | struct btrfs_block_group_cache *cache; | |
b742bb82 | 7708 | int index = 0; |
0f9dd46c | 7709 | |
9ed74f2d | 7710 | spin_lock(&info->lock); |
ab8d0fc4 JM |
7711 | btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull", |
7712 | info->flags, | |
4136135b LB |
7713 | info->total_bytes - btrfs_space_info_used(info, true), |
7714 | info->full ? "" : "not "); | |
ab8d0fc4 JM |
7715 | btrfs_info(fs_info, |
7716 | "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu", | |
7717 | info->total_bytes, info->bytes_used, info->bytes_pinned, | |
7718 | info->bytes_reserved, info->bytes_may_use, | |
7719 | info->bytes_readonly); | |
9ed74f2d JB |
7720 | spin_unlock(&info->lock); |
7721 | ||
7722 | if (!dump_block_groups) | |
7723 | return; | |
0f9dd46c | 7724 | |
80eb234a | 7725 | down_read(&info->groups_sem); |
b742bb82 YZ |
7726 | again: |
7727 | list_for_each_entry(cache, &info->block_groups[index], list) { | |
0f9dd46c | 7728 | spin_lock(&cache->lock); |
ab8d0fc4 JM |
7729 | btrfs_info(fs_info, |
7730 | "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s", | |
7731 | cache->key.objectid, cache->key.offset, | |
7732 | btrfs_block_group_used(&cache->item), cache->pinned, | |
7733 | cache->reserved, cache->ro ? "[readonly]" : ""); | |
0f9dd46c JB |
7734 | btrfs_dump_free_space(cache, bytes); |
7735 | spin_unlock(&cache->lock); | |
7736 | } | |
b742bb82 YZ |
7737 | if (++index < BTRFS_NR_RAID_TYPES) |
7738 | goto again; | |
80eb234a | 7739 | up_read(&info->groups_sem); |
0f9dd46c | 7740 | } |
e8569813 | 7741 | |
6f47c706 NB |
7742 | /* |
7743 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a | |
7744 | * hole that is at least as big as @num_bytes. | |
7745 | * | |
7746 | * @root - The root that will contain this extent | |
7747 | * | |
7748 | * @ram_bytes - The amount of space in ram that @num_bytes take. This | |
7749 | * is used for accounting purposes. This value differs | |
7750 | * from @num_bytes only in the case of compressed extents. | |
7751 | * | |
7752 | * @num_bytes - Number of bytes to allocate on-disk. | |
7753 | * | |
7754 | * @min_alloc_size - Indicates the minimum amount of space that the | |
7755 | * allocator should try to satisfy. In some cases | |
7756 | * @num_bytes may be larger than what is required and if | |
7757 | * the filesystem is fragmented then allocation fails. | |
7758 | * However, the presence of @min_alloc_size gives a | |
7759 | * chance to try and satisfy the smaller allocation. | |
7760 | * | |
7761 | * @empty_size - A hint that you plan on doing more COW. This is the | |
7762 | * size in bytes the allocator should try to find free | |
7763 | * next to the block it returns. This is just a hint and | |
7764 | * may be ignored by the allocator. | |
7765 | * | |
7766 | * @hint_byte - Hint to the allocator to start searching above the byte | |
7767 | * address passed. It might be ignored. | |
7768 | * | |
7769 | * @ins - This key is modified to record the found hole. It will | |
7770 | * have the following values: | |
7771 | * ins->objectid == start position | |
7772 | * ins->flags = BTRFS_EXTENT_ITEM_KEY | |
7773 | * ins->offset == the size of the hole. | |
7774 | * | |
7775 | * @is_data - Boolean flag indicating whether an extent is | |
7776 | * allocated for data (true) or metadata (false) | |
7777 | * | |
7778 | * @delalloc - Boolean flag indicating whether this allocation is for | |
7779 | * delalloc or not. If 'true' data_rwsem of block groups | |
7780 | * is going to be acquired. | |
7781 | * | |
7782 | * | |
7783 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In | |
7784 | * case -ENOSPC is returned then @ins->offset will contain the size of the | |
7785 | * largest available hole the allocator managed to find. | |
7786 | */ | |
18513091 | 7787 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
11833d66 YZ |
7788 | u64 num_bytes, u64 min_alloc_size, |
7789 | u64 empty_size, u64 hint_byte, | |
e570fd27 | 7790 | struct btrfs_key *ins, int is_data, int delalloc) |
fec577fb | 7791 | { |
ab8d0fc4 | 7792 | struct btrfs_fs_info *fs_info = root->fs_info; |
36af4e07 | 7793 | bool final_tried = num_bytes == min_alloc_size; |
b6919a58 | 7794 | u64 flags; |
fec577fb | 7795 | int ret; |
925baedd | 7796 | |
1b86826d | 7797 | flags = get_alloc_profile_by_root(root, is_data); |
98d20f67 | 7798 | again: |
0b246afa | 7799 | WARN_ON(num_bytes < fs_info->sectorsize); |
87bde3cd | 7800 | ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size, |
18513091 | 7801 | hint_byte, ins, flags, delalloc); |
9cfa3e34 | 7802 | if (!ret && !is_data) { |
ab8d0fc4 | 7803 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
9cfa3e34 | 7804 | } else if (ret == -ENOSPC) { |
a4820398 MX |
7805 | if (!final_tried && ins->offset) { |
7806 | num_bytes = min(num_bytes >> 1, ins->offset); | |
da17066c | 7807 | num_bytes = round_down(num_bytes, |
0b246afa | 7808 | fs_info->sectorsize); |
9e622d6b | 7809 | num_bytes = max(num_bytes, min_alloc_size); |
18513091 | 7810 | ram_bytes = num_bytes; |
9e622d6b MX |
7811 | if (num_bytes == min_alloc_size) |
7812 | final_tried = true; | |
7813 | goto again; | |
ab8d0fc4 | 7814 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
9e622d6b MX |
7815 | struct btrfs_space_info *sinfo; |
7816 | ||
ab8d0fc4 | 7817 | sinfo = __find_space_info(fs_info, flags); |
0b246afa | 7818 | btrfs_err(fs_info, |
5d163e0e JM |
7819 | "allocation failed flags %llu, wanted %llu", |
7820 | flags, num_bytes); | |
53804280 | 7821 | if (sinfo) |
ab8d0fc4 | 7822 | dump_space_info(fs_info, sinfo, num_bytes, 1); |
9e622d6b | 7823 | } |
925baedd | 7824 | } |
0f9dd46c JB |
7825 | |
7826 | return ret; | |
e6dcd2dc CM |
7827 | } |
7828 | ||
2ff7e61e | 7829 | static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 MX |
7830 | u64 start, u64 len, |
7831 | int pin, int delalloc) | |
65b51a00 | 7832 | { |
0f9dd46c | 7833 | struct btrfs_block_group_cache *cache; |
1f3c79a2 | 7834 | int ret = 0; |
0f9dd46c | 7835 | |
0b246afa | 7836 | cache = btrfs_lookup_block_group(fs_info, start); |
0f9dd46c | 7837 | if (!cache) { |
0b246afa JM |
7838 | btrfs_err(fs_info, "Unable to find block group for %llu", |
7839 | start); | |
0f9dd46c JB |
7840 | return -ENOSPC; |
7841 | } | |
1f3c79a2 | 7842 | |
e688b725 | 7843 | if (pin) |
2ff7e61e | 7844 | pin_down_extent(fs_info, cache, start, len, 1); |
e688b725 | 7845 | else { |
0b246afa | 7846 | if (btrfs_test_opt(fs_info, DISCARD)) |
2ff7e61e | 7847 | ret = btrfs_discard_extent(fs_info, start, len, NULL); |
e688b725 | 7848 | btrfs_add_free_space(cache, start, len); |
4824f1f4 | 7849 | btrfs_free_reserved_bytes(cache, len, delalloc); |
71ff6437 | 7850 | trace_btrfs_reserved_extent_free(fs_info, start, len); |
e688b725 | 7851 | } |
31193213 | 7852 | |
fa9c0d79 | 7853 | btrfs_put_block_group(cache); |
e6dcd2dc CM |
7854 | return ret; |
7855 | } | |
7856 | ||
2ff7e61e | 7857 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
e570fd27 | 7858 | u64 start, u64 len, int delalloc) |
e688b725 | 7859 | { |
2ff7e61e | 7860 | return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc); |
e688b725 CM |
7861 | } |
7862 | ||
2ff7e61e | 7863 | int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info, |
e688b725 CM |
7864 | u64 start, u64 len) |
7865 | { | |
2ff7e61e | 7866 | return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0); |
e688b725 CM |
7867 | } |
7868 | ||
5d4f98a2 | 7869 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
7870 | u64 parent, u64 root_objectid, |
7871 | u64 flags, u64 owner, u64 offset, | |
7872 | struct btrfs_key *ins, int ref_mod) | |
e6dcd2dc | 7873 | { |
ef89b824 | 7874 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 7875 | int ret; |
e6dcd2dc | 7876 | struct btrfs_extent_item *extent_item; |
5d4f98a2 | 7877 | struct btrfs_extent_inline_ref *iref; |
e6dcd2dc | 7878 | struct btrfs_path *path; |
5d4f98a2 YZ |
7879 | struct extent_buffer *leaf; |
7880 | int type; | |
7881 | u32 size; | |
26b8003f | 7882 | |
5d4f98a2 YZ |
7883 | if (parent > 0) |
7884 | type = BTRFS_SHARED_DATA_REF_KEY; | |
7885 | else | |
7886 | type = BTRFS_EXTENT_DATA_REF_KEY; | |
58176a96 | 7887 | |
5d4f98a2 | 7888 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
7bb86316 CM |
7889 | |
7890 | path = btrfs_alloc_path(); | |
db5b493a TI |
7891 | if (!path) |
7892 | return -ENOMEM; | |
47e4bb98 | 7893 | |
b9473439 | 7894 | path->leave_spinning = 1; |
5d4f98a2 YZ |
7895 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
7896 | ins, size); | |
79787eaa JM |
7897 | if (ret) { |
7898 | btrfs_free_path(path); | |
7899 | return ret; | |
7900 | } | |
0f9dd46c | 7901 | |
5d4f98a2 YZ |
7902 | leaf = path->nodes[0]; |
7903 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
47e4bb98 | 7904 | struct btrfs_extent_item); |
5d4f98a2 YZ |
7905 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
7906 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
7907 | btrfs_set_extent_flags(leaf, extent_item, | |
7908 | flags | BTRFS_EXTENT_FLAG_DATA); | |
7909 | ||
7910 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
7911 | btrfs_set_extent_inline_ref_type(leaf, iref, type); | |
7912 | if (parent > 0) { | |
7913 | struct btrfs_shared_data_ref *ref; | |
7914 | ref = (struct btrfs_shared_data_ref *)(iref + 1); | |
7915 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); | |
7916 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); | |
7917 | } else { | |
7918 | struct btrfs_extent_data_ref *ref; | |
7919 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); | |
7920 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); | |
7921 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); | |
7922 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); | |
7923 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); | |
7924 | } | |
47e4bb98 CM |
7925 | |
7926 | btrfs_mark_buffer_dirty(path->nodes[0]); | |
7bb86316 | 7927 | btrfs_free_path(path); |
f510cfec | 7928 | |
25a356d3 | 7929 | ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset); |
1e144fb8 OS |
7930 | if (ret) |
7931 | return ret; | |
7932 | ||
6202df69 | 7933 | ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1); |
79787eaa | 7934 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 7935 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
c1c9ff7c | 7936 | ins->objectid, ins->offset); |
f5947066 CM |
7937 | BUG(); |
7938 | } | |
71ff6437 | 7939 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
e6dcd2dc CM |
7940 | return ret; |
7941 | } | |
7942 | ||
5d4f98a2 | 7943 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
4e6bd4e0 | 7944 | struct btrfs_delayed_ref_node *node, |
21ebfbe7 | 7945 | struct btrfs_delayed_extent_op *extent_op) |
e6dcd2dc | 7946 | { |
9dcdbe01 | 7947 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e6dcd2dc | 7948 | int ret; |
5d4f98a2 | 7949 | struct btrfs_extent_item *extent_item; |
4e6bd4e0 | 7950 | struct btrfs_key extent_key; |
5d4f98a2 YZ |
7951 | struct btrfs_tree_block_info *block_info; |
7952 | struct btrfs_extent_inline_ref *iref; | |
7953 | struct btrfs_path *path; | |
7954 | struct extent_buffer *leaf; | |
4e6bd4e0 | 7955 | struct btrfs_delayed_tree_ref *ref; |
3173a18f | 7956 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
4e6bd4e0 | 7957 | u64 num_bytes; |
21ebfbe7 | 7958 | u64 flags = extent_op->flags_to_set; |
0b246afa | 7959 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
3173a18f | 7960 | |
4e6bd4e0 NB |
7961 | ref = btrfs_delayed_node_to_tree_ref(node); |
7962 | ||
4e6bd4e0 NB |
7963 | extent_key.objectid = node->bytenr; |
7964 | if (skinny_metadata) { | |
7965 | extent_key.offset = ref->level; | |
7966 | extent_key.type = BTRFS_METADATA_ITEM_KEY; | |
7967 | num_bytes = fs_info->nodesize; | |
7968 | } else { | |
7969 | extent_key.offset = node->num_bytes; | |
7970 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; | |
3173a18f | 7971 | size += sizeof(*block_info); |
4e6bd4e0 NB |
7972 | num_bytes = node->num_bytes; |
7973 | } | |
1c2308f8 | 7974 | |
5d4f98a2 | 7975 | path = btrfs_alloc_path(); |
857cc2fc | 7976 | if (!path) { |
4e6bd4e0 NB |
7977 | btrfs_free_and_pin_reserved_extent(fs_info, |
7978 | extent_key.objectid, | |
0b246afa | 7979 | fs_info->nodesize); |
d8926bb3 | 7980 | return -ENOMEM; |
857cc2fc | 7981 | } |
56bec294 | 7982 | |
5d4f98a2 YZ |
7983 | path->leave_spinning = 1; |
7984 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, | |
4e6bd4e0 | 7985 | &extent_key, size); |
79787eaa | 7986 | if (ret) { |
dd825259 | 7987 | btrfs_free_path(path); |
4e6bd4e0 NB |
7988 | btrfs_free_and_pin_reserved_extent(fs_info, |
7989 | extent_key.objectid, | |
0b246afa | 7990 | fs_info->nodesize); |
79787eaa JM |
7991 | return ret; |
7992 | } | |
5d4f98a2 YZ |
7993 | |
7994 | leaf = path->nodes[0]; | |
7995 | extent_item = btrfs_item_ptr(leaf, path->slots[0], | |
7996 | struct btrfs_extent_item); | |
7997 | btrfs_set_extent_refs(leaf, extent_item, 1); | |
7998 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); | |
7999 | btrfs_set_extent_flags(leaf, extent_item, | |
8000 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); | |
5d4f98a2 | 8001 | |
3173a18f JB |
8002 | if (skinny_metadata) { |
8003 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); | |
8004 | } else { | |
8005 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); | |
21ebfbe7 | 8006 | btrfs_set_tree_block_key(leaf, block_info, &extent_op->key); |
4e6bd4e0 | 8007 | btrfs_set_tree_block_level(leaf, block_info, ref->level); |
3173a18f JB |
8008 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
8009 | } | |
5d4f98a2 | 8010 | |
d4b20733 | 8011 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) { |
5d4f98a2 YZ |
8012 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); |
8013 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8014 | BTRFS_SHARED_BLOCK_REF_KEY); | |
d4b20733 | 8015 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent); |
5d4f98a2 YZ |
8016 | } else { |
8017 | btrfs_set_extent_inline_ref_type(leaf, iref, | |
8018 | BTRFS_TREE_BLOCK_REF_KEY); | |
4e6bd4e0 | 8019 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root); |
5d4f98a2 YZ |
8020 | } |
8021 | ||
8022 | btrfs_mark_buffer_dirty(leaf); | |
8023 | btrfs_free_path(path); | |
8024 | ||
4e6bd4e0 NB |
8025 | ret = remove_from_free_space_tree(trans, extent_key.objectid, |
8026 | num_bytes); | |
1e144fb8 OS |
8027 | if (ret) |
8028 | return ret; | |
8029 | ||
4e6bd4e0 | 8030 | ret = update_block_group(trans, fs_info, extent_key.objectid, |
6202df69 | 8031 | fs_info->nodesize, 1); |
79787eaa | 8032 | if (ret) { /* -ENOENT, logic error */ |
c2cf52eb | 8033 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
4e6bd4e0 | 8034 | extent_key.objectid, extent_key.offset); |
5d4f98a2 YZ |
8035 | BUG(); |
8036 | } | |
0be5dc67 | 8037 | |
4e6bd4e0 | 8038 | trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid, |
0b246afa | 8039 | fs_info->nodesize); |
5d4f98a2 YZ |
8040 | return ret; |
8041 | } | |
8042 | ||
8043 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, | |
84f7d8e6 | 8044 | struct btrfs_root *root, u64 owner, |
5846a3c2 QW |
8045 | u64 offset, u64 ram_bytes, |
8046 | struct btrfs_key *ins) | |
5d4f98a2 YZ |
8047 | { |
8048 | int ret; | |
8049 | ||
84f7d8e6 | 8050 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
5d4f98a2 | 8051 | |
fd708b81 JB |
8052 | btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0, |
8053 | root->root_key.objectid, owner, offset, | |
8054 | BTRFS_ADD_DELAYED_EXTENT); | |
8055 | ||
88a979c6 | 8056 | ret = btrfs_add_delayed_data_ref(trans, ins->objectid, |
84f7d8e6 JB |
8057 | ins->offset, 0, |
8058 | root->root_key.objectid, owner, | |
7be07912 OS |
8059 | offset, ram_bytes, |
8060 | BTRFS_ADD_DELAYED_EXTENT, NULL, NULL); | |
e6dcd2dc CM |
8061 | return ret; |
8062 | } | |
e02119d5 CM |
8063 | |
8064 | /* | |
8065 | * this is used by the tree logging recovery code. It records that | |
8066 | * an extent has been allocated and makes sure to clear the free | |
8067 | * space cache bits as well | |
8068 | */ | |
5d4f98a2 | 8069 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
5d4f98a2 YZ |
8070 | u64 root_objectid, u64 owner, u64 offset, |
8071 | struct btrfs_key *ins) | |
e02119d5 | 8072 | { |
61da2abf | 8073 | struct btrfs_fs_info *fs_info = trans->fs_info; |
e02119d5 CM |
8074 | int ret; |
8075 | struct btrfs_block_group_cache *block_group; | |
ed7a6948 | 8076 | struct btrfs_space_info *space_info; |
11833d66 | 8077 | |
8c2a1a30 JB |
8078 | /* |
8079 | * Mixed block groups will exclude before processing the log so we only | |
01327610 | 8080 | * need to do the exclude dance if this fs isn't mixed. |
8c2a1a30 | 8081 | */ |
0b246afa | 8082 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
2ff7e61e JM |
8083 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
8084 | ins->offset); | |
b50c6e25 | 8085 | if (ret) |
8c2a1a30 | 8086 | return ret; |
11833d66 YZ |
8087 | } |
8088 | ||
0b246afa | 8089 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
8c2a1a30 JB |
8090 | if (!block_group) |
8091 | return -EINVAL; | |
8092 | ||
ed7a6948 WX |
8093 | space_info = block_group->space_info; |
8094 | spin_lock(&space_info->lock); | |
8095 | spin_lock(&block_group->lock); | |
8096 | space_info->bytes_reserved += ins->offset; | |
8097 | block_group->reserved += ins->offset; | |
8098 | spin_unlock(&block_group->lock); | |
8099 | spin_unlock(&space_info->lock); | |
8100 | ||
ef89b824 NB |
8101 | ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner, |
8102 | offset, ins, 1); | |
b50c6e25 | 8103 | btrfs_put_block_group(block_group); |
e02119d5 CM |
8104 | return ret; |
8105 | } | |
8106 | ||
48a3b636 ES |
8107 | static struct extent_buffer * |
8108 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, | |
bc877d28 | 8109 | u64 bytenr, int level, u64 owner) |
65b51a00 | 8110 | { |
0b246afa | 8111 | struct btrfs_fs_info *fs_info = root->fs_info; |
65b51a00 CM |
8112 | struct extent_buffer *buf; |
8113 | ||
2ff7e61e | 8114 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8115 | if (IS_ERR(buf)) |
8116 | return buf; | |
8117 | ||
85d4e461 | 8118 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
65b51a00 | 8119 | btrfs_tree_lock(buf); |
7c302b49 | 8120 | clean_tree_block(fs_info, buf); |
3083ee2e | 8121 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
b4ce94de CM |
8122 | |
8123 | btrfs_set_lock_blocking(buf); | |
4db8c528 | 8124 | set_extent_buffer_uptodate(buf); |
b4ce94de | 8125 | |
bc877d28 NB |
8126 | memzero_extent_buffer(buf, 0, sizeof(struct btrfs_header)); |
8127 | btrfs_set_header_level(buf, level); | |
8128 | btrfs_set_header_bytenr(buf, buf->start); | |
8129 | btrfs_set_header_generation(buf, trans->transid); | |
8130 | btrfs_set_header_backref_rev(buf, BTRFS_MIXED_BACKREF_REV); | |
8131 | btrfs_set_header_owner(buf, owner); | |
8132 | write_extent_buffer_fsid(buf, fs_info->fsid); | |
8133 | write_extent_buffer_chunk_tree_uuid(buf, fs_info->chunk_tree_uuid); | |
d0c803c4 | 8134 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
656f30db | 8135 | buf->log_index = root->log_transid % 2; |
8cef4e16 YZ |
8136 | /* |
8137 | * we allow two log transactions at a time, use different | |
8138 | * EXENT bit to differentiate dirty pages. | |
8139 | */ | |
656f30db | 8140 | if (buf->log_index == 0) |
8cef4e16 YZ |
8141 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
8142 | buf->start + buf->len - 1, GFP_NOFS); | |
8143 | else | |
8144 | set_extent_new(&root->dirty_log_pages, buf->start, | |
3744dbeb | 8145 | buf->start + buf->len - 1); |
d0c803c4 | 8146 | } else { |
656f30db | 8147 | buf->log_index = -1; |
d0c803c4 | 8148 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
65b51a00 | 8149 | buf->start + buf->len - 1, GFP_NOFS); |
d0c803c4 | 8150 | } |
64c12921 | 8151 | trans->dirty = true; |
b4ce94de | 8152 | /* this returns a buffer locked for blocking */ |
65b51a00 CM |
8153 | return buf; |
8154 | } | |
8155 | ||
f0486c68 YZ |
8156 | static struct btrfs_block_rsv * |
8157 | use_block_rsv(struct btrfs_trans_handle *trans, | |
8158 | struct btrfs_root *root, u32 blocksize) | |
8159 | { | |
0b246afa | 8160 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8161 | struct btrfs_block_rsv *block_rsv; |
0b246afa | 8162 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
f0486c68 | 8163 | int ret; |
d88033db | 8164 | bool global_updated = false; |
f0486c68 YZ |
8165 | |
8166 | block_rsv = get_block_rsv(trans, root); | |
8167 | ||
b586b323 MX |
8168 | if (unlikely(block_rsv->size == 0)) |
8169 | goto try_reserve; | |
d88033db | 8170 | again: |
f0486c68 YZ |
8171 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
8172 | if (!ret) | |
8173 | return block_rsv; | |
8174 | ||
b586b323 MX |
8175 | if (block_rsv->failfast) |
8176 | return ERR_PTR(ret); | |
8177 | ||
d88033db MX |
8178 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
8179 | global_updated = true; | |
0b246afa | 8180 | update_global_block_rsv(fs_info); |
d88033db MX |
8181 | goto again; |
8182 | } | |
8183 | ||
0b246afa | 8184 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
b586b323 MX |
8185 | static DEFINE_RATELIMIT_STATE(_rs, |
8186 | DEFAULT_RATELIMIT_INTERVAL * 10, | |
8187 | /*DEFAULT_RATELIMIT_BURST*/ 1); | |
8188 | if (__ratelimit(&_rs)) | |
8189 | WARN(1, KERN_DEBUG | |
efe120a0 | 8190 | "BTRFS: block rsv returned %d\n", ret); |
b586b323 MX |
8191 | } |
8192 | try_reserve: | |
8193 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, | |
8194 | BTRFS_RESERVE_NO_FLUSH); | |
8195 | if (!ret) | |
8196 | return block_rsv; | |
8197 | /* | |
8198 | * If we couldn't reserve metadata bytes try and use some from | |
5881cfc9 MX |
8199 | * the global reserve if its space type is the same as the global |
8200 | * reservation. | |
b586b323 | 8201 | */ |
5881cfc9 MX |
8202 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
8203 | block_rsv->space_info == global_rsv->space_info) { | |
b586b323 MX |
8204 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
8205 | if (!ret) | |
8206 | return global_rsv; | |
8207 | } | |
8208 | return ERR_PTR(ret); | |
f0486c68 YZ |
8209 | } |
8210 | ||
8c2a3ca2 JB |
8211 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
8212 | struct btrfs_block_rsv *block_rsv, u32 blocksize) | |
f0486c68 | 8213 | { |
3a584174 | 8214 | block_rsv_add_bytes(block_rsv, blocksize, false); |
ff6bc37e | 8215 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL); |
f0486c68 YZ |
8216 | } |
8217 | ||
fec577fb | 8218 | /* |
f0486c68 | 8219 | * finds a free extent and does all the dirty work required for allocation |
67b7859e | 8220 | * returns the tree buffer or an ERR_PTR on error. |
fec577fb | 8221 | */ |
4d75f8a9 | 8222 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
310712b2 OS |
8223 | struct btrfs_root *root, |
8224 | u64 parent, u64 root_objectid, | |
8225 | const struct btrfs_disk_key *key, | |
8226 | int level, u64 hint, | |
8227 | u64 empty_size) | |
fec577fb | 8228 | { |
0b246afa | 8229 | struct btrfs_fs_info *fs_info = root->fs_info; |
e2fa7227 | 8230 | struct btrfs_key ins; |
f0486c68 | 8231 | struct btrfs_block_rsv *block_rsv; |
5f39d397 | 8232 | struct extent_buffer *buf; |
67b7859e | 8233 | struct btrfs_delayed_extent_op *extent_op; |
f0486c68 YZ |
8234 | u64 flags = 0; |
8235 | int ret; | |
0b246afa JM |
8236 | u32 blocksize = fs_info->nodesize; |
8237 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); | |
fec577fb | 8238 | |
05653ef3 | 8239 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
0b246afa | 8240 | if (btrfs_is_testing(fs_info)) { |
faa2dbf0 | 8241 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
bc877d28 | 8242 | level, root_objectid); |
faa2dbf0 JB |
8243 | if (!IS_ERR(buf)) |
8244 | root->alloc_bytenr += blocksize; | |
8245 | return buf; | |
8246 | } | |
05653ef3 | 8247 | #endif |
fccb84c9 | 8248 | |
f0486c68 YZ |
8249 | block_rsv = use_block_rsv(trans, root, blocksize); |
8250 | if (IS_ERR(block_rsv)) | |
8251 | return ERR_CAST(block_rsv); | |
8252 | ||
18513091 | 8253 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
e570fd27 | 8254 | empty_size, hint, &ins, 0, 0); |
67b7859e OS |
8255 | if (ret) |
8256 | goto out_unuse; | |
55c69072 | 8257 | |
bc877d28 NB |
8258 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level, |
8259 | root_objectid); | |
67b7859e OS |
8260 | if (IS_ERR(buf)) { |
8261 | ret = PTR_ERR(buf); | |
8262 | goto out_free_reserved; | |
8263 | } | |
f0486c68 YZ |
8264 | |
8265 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { | |
8266 | if (parent == 0) | |
8267 | parent = ins.objectid; | |
8268 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
8269 | } else | |
8270 | BUG_ON(parent > 0); | |
8271 | ||
8272 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { | |
78a6184a | 8273 | extent_op = btrfs_alloc_delayed_extent_op(); |
67b7859e OS |
8274 | if (!extent_op) { |
8275 | ret = -ENOMEM; | |
8276 | goto out_free_buf; | |
8277 | } | |
f0486c68 YZ |
8278 | if (key) |
8279 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); | |
8280 | else | |
8281 | memset(&extent_op->key, 0, sizeof(extent_op->key)); | |
8282 | extent_op->flags_to_set = flags; | |
35b3ad50 DS |
8283 | extent_op->update_key = skinny_metadata ? false : true; |
8284 | extent_op->update_flags = true; | |
8285 | extent_op->is_data = false; | |
b1c79e09 | 8286 | extent_op->level = level; |
f0486c68 | 8287 | |
fd708b81 JB |
8288 | btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent, |
8289 | root_objectid, level, 0, | |
8290 | BTRFS_ADD_DELAYED_EXTENT); | |
44e1c47d | 8291 | ret = btrfs_add_delayed_tree_ref(trans, ins.objectid, |
7be07912 OS |
8292 | ins.offset, parent, |
8293 | root_objectid, level, | |
67b7859e | 8294 | BTRFS_ADD_DELAYED_EXTENT, |
7be07912 | 8295 | extent_op, NULL, NULL); |
67b7859e OS |
8296 | if (ret) |
8297 | goto out_free_delayed; | |
f0486c68 | 8298 | } |
fec577fb | 8299 | return buf; |
67b7859e OS |
8300 | |
8301 | out_free_delayed: | |
8302 | btrfs_free_delayed_extent_op(extent_op); | |
8303 | out_free_buf: | |
8304 | free_extent_buffer(buf); | |
8305 | out_free_reserved: | |
2ff7e61e | 8306 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
67b7859e | 8307 | out_unuse: |
0b246afa | 8308 | unuse_block_rsv(fs_info, block_rsv, blocksize); |
67b7859e | 8309 | return ERR_PTR(ret); |
fec577fb | 8310 | } |
a28ec197 | 8311 | |
2c47e605 YZ |
8312 | struct walk_control { |
8313 | u64 refs[BTRFS_MAX_LEVEL]; | |
8314 | u64 flags[BTRFS_MAX_LEVEL]; | |
8315 | struct btrfs_key update_progress; | |
8316 | int stage; | |
8317 | int level; | |
8318 | int shared_level; | |
8319 | int update_ref; | |
8320 | int keep_locks; | |
1c4850e2 YZ |
8321 | int reada_slot; |
8322 | int reada_count; | |
2c47e605 YZ |
8323 | }; |
8324 | ||
8325 | #define DROP_REFERENCE 1 | |
8326 | #define UPDATE_BACKREF 2 | |
8327 | ||
1c4850e2 YZ |
8328 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
8329 | struct btrfs_root *root, | |
8330 | struct walk_control *wc, | |
8331 | struct btrfs_path *path) | |
6407bf6d | 8332 | { |
0b246afa | 8333 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8334 | u64 bytenr; |
8335 | u64 generation; | |
8336 | u64 refs; | |
94fcca9f | 8337 | u64 flags; |
5d4f98a2 | 8338 | u32 nritems; |
1c4850e2 YZ |
8339 | struct btrfs_key key; |
8340 | struct extent_buffer *eb; | |
6407bf6d | 8341 | int ret; |
1c4850e2 YZ |
8342 | int slot; |
8343 | int nread = 0; | |
6407bf6d | 8344 | |
1c4850e2 YZ |
8345 | if (path->slots[wc->level] < wc->reada_slot) { |
8346 | wc->reada_count = wc->reada_count * 2 / 3; | |
8347 | wc->reada_count = max(wc->reada_count, 2); | |
8348 | } else { | |
8349 | wc->reada_count = wc->reada_count * 3 / 2; | |
8350 | wc->reada_count = min_t(int, wc->reada_count, | |
0b246afa | 8351 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
1c4850e2 | 8352 | } |
7bb86316 | 8353 | |
1c4850e2 YZ |
8354 | eb = path->nodes[wc->level]; |
8355 | nritems = btrfs_header_nritems(eb); | |
bd56b302 | 8356 | |
1c4850e2 YZ |
8357 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
8358 | if (nread >= wc->reada_count) | |
8359 | break; | |
bd56b302 | 8360 | |
2dd3e67b | 8361 | cond_resched(); |
1c4850e2 YZ |
8362 | bytenr = btrfs_node_blockptr(eb, slot); |
8363 | generation = btrfs_node_ptr_generation(eb, slot); | |
2dd3e67b | 8364 | |
1c4850e2 YZ |
8365 | if (slot == path->slots[wc->level]) |
8366 | goto reada; | |
5d4f98a2 | 8367 | |
1c4850e2 YZ |
8368 | if (wc->stage == UPDATE_BACKREF && |
8369 | generation <= root->root_key.offset) | |
bd56b302 CM |
8370 | continue; |
8371 | ||
94fcca9f | 8372 | /* We don't lock the tree block, it's OK to be racy here */ |
2ff7e61e | 8373 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
3173a18f JB |
8374 | wc->level - 1, 1, &refs, |
8375 | &flags); | |
79787eaa JM |
8376 | /* We don't care about errors in readahead. */ |
8377 | if (ret < 0) | |
8378 | continue; | |
94fcca9f YZ |
8379 | BUG_ON(refs == 0); |
8380 | ||
1c4850e2 | 8381 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 YZ |
8382 | if (refs == 1) |
8383 | goto reada; | |
bd56b302 | 8384 | |
94fcca9f YZ |
8385 | if (wc->level == 1 && |
8386 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8387 | continue; | |
1c4850e2 YZ |
8388 | if (!wc->update_ref || |
8389 | generation <= root->root_key.offset) | |
8390 | continue; | |
8391 | btrfs_node_key_to_cpu(eb, &key, slot); | |
8392 | ret = btrfs_comp_cpu_keys(&key, | |
8393 | &wc->update_progress); | |
8394 | if (ret < 0) | |
8395 | continue; | |
94fcca9f YZ |
8396 | } else { |
8397 | if (wc->level == 1 && | |
8398 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8399 | continue; | |
6407bf6d | 8400 | } |
1c4850e2 | 8401 | reada: |
2ff7e61e | 8402 | readahead_tree_block(fs_info, bytenr); |
1c4850e2 | 8403 | nread++; |
20524f02 | 8404 | } |
1c4850e2 | 8405 | wc->reada_slot = slot; |
20524f02 | 8406 | } |
2c47e605 | 8407 | |
f82d02d9 | 8408 | /* |
2c016dc2 | 8409 | * helper to process tree block while walking down the tree. |
2c47e605 | 8410 | * |
2c47e605 YZ |
8411 | * when wc->stage == UPDATE_BACKREF, this function updates |
8412 | * back refs for pointers in the block. | |
8413 | * | |
8414 | * NOTE: return value 1 means we should stop walking down. | |
f82d02d9 | 8415 | */ |
2c47e605 | 8416 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
5d4f98a2 | 8417 | struct btrfs_root *root, |
2c47e605 | 8418 | struct btrfs_path *path, |
94fcca9f | 8419 | struct walk_control *wc, int lookup_info) |
f82d02d9 | 8420 | { |
2ff7e61e | 8421 | struct btrfs_fs_info *fs_info = root->fs_info; |
2c47e605 YZ |
8422 | int level = wc->level; |
8423 | struct extent_buffer *eb = path->nodes[level]; | |
2c47e605 | 8424 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
f82d02d9 YZ |
8425 | int ret; |
8426 | ||
2c47e605 YZ |
8427 | if (wc->stage == UPDATE_BACKREF && |
8428 | btrfs_header_owner(eb) != root->root_key.objectid) | |
8429 | return 1; | |
f82d02d9 | 8430 | |
2c47e605 YZ |
8431 | /* |
8432 | * when reference count of tree block is 1, it won't increase | |
8433 | * again. once full backref flag is set, we never clear it. | |
8434 | */ | |
94fcca9f YZ |
8435 | if (lookup_info && |
8436 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || | |
8437 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { | |
2c47e605 | 8438 | BUG_ON(!path->locks[level]); |
2ff7e61e | 8439 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8440 | eb->start, level, 1, |
2c47e605 YZ |
8441 | &wc->refs[level], |
8442 | &wc->flags[level]); | |
79787eaa JM |
8443 | BUG_ON(ret == -ENOMEM); |
8444 | if (ret) | |
8445 | return ret; | |
2c47e605 YZ |
8446 | BUG_ON(wc->refs[level] == 0); |
8447 | } | |
5d4f98a2 | 8448 | |
2c47e605 YZ |
8449 | if (wc->stage == DROP_REFERENCE) { |
8450 | if (wc->refs[level] > 1) | |
8451 | return 1; | |
f82d02d9 | 8452 | |
2c47e605 | 8453 | if (path->locks[level] && !wc->keep_locks) { |
bd681513 | 8454 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8455 | path->locks[level] = 0; |
8456 | } | |
8457 | return 0; | |
8458 | } | |
f82d02d9 | 8459 | |
2c47e605 YZ |
8460 | /* wc->stage == UPDATE_BACKREF */ |
8461 | if (!(wc->flags[level] & flag)) { | |
8462 | BUG_ON(!path->locks[level]); | |
e339a6b0 | 8463 | ret = btrfs_inc_ref(trans, root, eb, 1); |
79787eaa | 8464 | BUG_ON(ret); /* -ENOMEM */ |
e339a6b0 | 8465 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8466 | BUG_ON(ret); /* -ENOMEM */ |
2ff7e61e | 8467 | ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start, |
b1c79e09 JB |
8468 | eb->len, flag, |
8469 | btrfs_header_level(eb), 0); | |
79787eaa | 8470 | BUG_ON(ret); /* -ENOMEM */ |
2c47e605 YZ |
8471 | wc->flags[level] |= flag; |
8472 | } | |
8473 | ||
8474 | /* | |
8475 | * the block is shared by multiple trees, so it's not good to | |
8476 | * keep the tree lock | |
8477 | */ | |
8478 | if (path->locks[level] && level > 0) { | |
bd681513 | 8479 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
2c47e605 YZ |
8480 | path->locks[level] = 0; |
8481 | } | |
8482 | return 0; | |
8483 | } | |
8484 | ||
1c4850e2 | 8485 | /* |
2c016dc2 | 8486 | * helper to process tree block pointer. |
1c4850e2 YZ |
8487 | * |
8488 | * when wc->stage == DROP_REFERENCE, this function checks | |
8489 | * reference count of the block pointed to. if the block | |
8490 | * is shared and we need update back refs for the subtree | |
8491 | * rooted at the block, this function changes wc->stage to | |
8492 | * UPDATE_BACKREF. if the block is shared and there is no | |
8493 | * need to update back, this function drops the reference | |
8494 | * to the block. | |
8495 | * | |
8496 | * NOTE: return value 1 means we should stop walking down. | |
8497 | */ | |
8498 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, | |
8499 | struct btrfs_root *root, | |
8500 | struct btrfs_path *path, | |
94fcca9f | 8501 | struct walk_control *wc, int *lookup_info) |
1c4850e2 | 8502 | { |
0b246afa | 8503 | struct btrfs_fs_info *fs_info = root->fs_info; |
1c4850e2 YZ |
8504 | u64 bytenr; |
8505 | u64 generation; | |
8506 | u64 parent; | |
8507 | u32 blocksize; | |
8508 | struct btrfs_key key; | |
581c1760 | 8509 | struct btrfs_key first_key; |
1c4850e2 YZ |
8510 | struct extent_buffer *next; |
8511 | int level = wc->level; | |
8512 | int reada = 0; | |
8513 | int ret = 0; | |
1152651a | 8514 | bool need_account = false; |
1c4850e2 YZ |
8515 | |
8516 | generation = btrfs_node_ptr_generation(path->nodes[level], | |
8517 | path->slots[level]); | |
8518 | /* | |
8519 | * if the lower level block was created before the snapshot | |
8520 | * was created, we know there is no need to update back refs | |
8521 | * for the subtree | |
8522 | */ | |
8523 | if (wc->stage == UPDATE_BACKREF && | |
94fcca9f YZ |
8524 | generation <= root->root_key.offset) { |
8525 | *lookup_info = 1; | |
1c4850e2 | 8526 | return 1; |
94fcca9f | 8527 | } |
1c4850e2 YZ |
8528 | |
8529 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); | |
581c1760 QW |
8530 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
8531 | path->slots[level]); | |
0b246afa | 8532 | blocksize = fs_info->nodesize; |
1c4850e2 | 8533 | |
0b246afa | 8534 | next = find_extent_buffer(fs_info, bytenr); |
1c4850e2 | 8535 | if (!next) { |
2ff7e61e | 8536 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
c871b0f2 LB |
8537 | if (IS_ERR(next)) |
8538 | return PTR_ERR(next); | |
8539 | ||
b2aaaa3b JB |
8540 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
8541 | level - 1); | |
1c4850e2 YZ |
8542 | reada = 1; |
8543 | } | |
8544 | btrfs_tree_lock(next); | |
8545 | btrfs_set_lock_blocking(next); | |
8546 | ||
2ff7e61e | 8547 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
94fcca9f YZ |
8548 | &wc->refs[level - 1], |
8549 | &wc->flags[level - 1]); | |
4867268c JB |
8550 | if (ret < 0) |
8551 | goto out_unlock; | |
79787eaa | 8552 | |
c2cf52eb | 8553 | if (unlikely(wc->refs[level - 1] == 0)) { |
0b246afa | 8554 | btrfs_err(fs_info, "Missing references."); |
4867268c JB |
8555 | ret = -EIO; |
8556 | goto out_unlock; | |
c2cf52eb | 8557 | } |
94fcca9f | 8558 | *lookup_info = 0; |
1c4850e2 | 8559 | |
94fcca9f | 8560 | if (wc->stage == DROP_REFERENCE) { |
1c4850e2 | 8561 | if (wc->refs[level - 1] > 1) { |
1152651a | 8562 | need_account = true; |
94fcca9f YZ |
8563 | if (level == 1 && |
8564 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8565 | goto skip; | |
8566 | ||
1c4850e2 YZ |
8567 | if (!wc->update_ref || |
8568 | generation <= root->root_key.offset) | |
8569 | goto skip; | |
8570 | ||
8571 | btrfs_node_key_to_cpu(path->nodes[level], &key, | |
8572 | path->slots[level]); | |
8573 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); | |
8574 | if (ret < 0) | |
8575 | goto skip; | |
8576 | ||
8577 | wc->stage = UPDATE_BACKREF; | |
8578 | wc->shared_level = level - 1; | |
8579 | } | |
94fcca9f YZ |
8580 | } else { |
8581 | if (level == 1 && | |
8582 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) | |
8583 | goto skip; | |
1c4850e2 YZ |
8584 | } |
8585 | ||
b9fab919 | 8586 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
1c4850e2 YZ |
8587 | btrfs_tree_unlock(next); |
8588 | free_extent_buffer(next); | |
8589 | next = NULL; | |
94fcca9f | 8590 | *lookup_info = 1; |
1c4850e2 YZ |
8591 | } |
8592 | ||
8593 | if (!next) { | |
8594 | if (reada && level == 1) | |
8595 | reada_walk_down(trans, root, wc, path); | |
581c1760 QW |
8596 | next = read_tree_block(fs_info, bytenr, generation, level - 1, |
8597 | &first_key); | |
64c043de LB |
8598 | if (IS_ERR(next)) { |
8599 | return PTR_ERR(next); | |
8600 | } else if (!extent_buffer_uptodate(next)) { | |
416bc658 | 8601 | free_extent_buffer(next); |
97d9a8a4 | 8602 | return -EIO; |
416bc658 | 8603 | } |
1c4850e2 YZ |
8604 | btrfs_tree_lock(next); |
8605 | btrfs_set_lock_blocking(next); | |
8606 | } | |
8607 | ||
8608 | level--; | |
4867268c JB |
8609 | ASSERT(level == btrfs_header_level(next)); |
8610 | if (level != btrfs_header_level(next)) { | |
8611 | btrfs_err(root->fs_info, "mismatched level"); | |
8612 | ret = -EIO; | |
8613 | goto out_unlock; | |
8614 | } | |
1c4850e2 YZ |
8615 | path->nodes[level] = next; |
8616 | path->slots[level] = 0; | |
bd681513 | 8617 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
1c4850e2 YZ |
8618 | wc->level = level; |
8619 | if (wc->level == 1) | |
8620 | wc->reada_slot = 0; | |
8621 | return 0; | |
8622 | skip: | |
8623 | wc->refs[level - 1] = 0; | |
8624 | wc->flags[level - 1] = 0; | |
94fcca9f YZ |
8625 | if (wc->stage == DROP_REFERENCE) { |
8626 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { | |
8627 | parent = path->nodes[level]->start; | |
8628 | } else { | |
4867268c | 8629 | ASSERT(root->root_key.objectid == |
94fcca9f | 8630 | btrfs_header_owner(path->nodes[level])); |
4867268c JB |
8631 | if (root->root_key.objectid != |
8632 | btrfs_header_owner(path->nodes[level])) { | |
8633 | btrfs_err(root->fs_info, | |
8634 | "mismatched block owner"); | |
8635 | ret = -EIO; | |
8636 | goto out_unlock; | |
8637 | } | |
94fcca9f YZ |
8638 | parent = 0; |
8639 | } | |
1c4850e2 | 8640 | |
1152651a | 8641 | if (need_account) { |
deb40627 | 8642 | ret = btrfs_qgroup_trace_subtree(trans, next, |
33d1f05c | 8643 | generation, level - 1); |
1152651a | 8644 | if (ret) { |
0b246afa | 8645 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8646 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
8647 | ret); | |
1152651a MF |
8648 | } |
8649 | } | |
84f7d8e6 | 8650 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, |
2ff7e61e JM |
8651 | parent, root->root_key.objectid, |
8652 | level - 1, 0); | |
4867268c JB |
8653 | if (ret) |
8654 | goto out_unlock; | |
1c4850e2 | 8655 | } |
4867268c JB |
8656 | |
8657 | *lookup_info = 1; | |
8658 | ret = 1; | |
8659 | ||
8660 | out_unlock: | |
1c4850e2 YZ |
8661 | btrfs_tree_unlock(next); |
8662 | free_extent_buffer(next); | |
4867268c JB |
8663 | |
8664 | return ret; | |
1c4850e2 YZ |
8665 | } |
8666 | ||
2c47e605 | 8667 | /* |
2c016dc2 | 8668 | * helper to process tree block while walking up the tree. |
2c47e605 YZ |
8669 | * |
8670 | * when wc->stage == DROP_REFERENCE, this function drops | |
8671 | * reference count on the block. | |
8672 | * | |
8673 | * when wc->stage == UPDATE_BACKREF, this function changes | |
8674 | * wc->stage back to DROP_REFERENCE if we changed wc->stage | |
8675 | * to UPDATE_BACKREF previously while processing the block. | |
8676 | * | |
8677 | * NOTE: return value 1 means we should stop walking up. | |
8678 | */ | |
8679 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, | |
8680 | struct btrfs_root *root, | |
8681 | struct btrfs_path *path, | |
8682 | struct walk_control *wc) | |
8683 | { | |
0b246afa | 8684 | struct btrfs_fs_info *fs_info = root->fs_info; |
f0486c68 | 8685 | int ret; |
2c47e605 YZ |
8686 | int level = wc->level; |
8687 | struct extent_buffer *eb = path->nodes[level]; | |
8688 | u64 parent = 0; | |
8689 | ||
8690 | if (wc->stage == UPDATE_BACKREF) { | |
8691 | BUG_ON(wc->shared_level < level); | |
8692 | if (level < wc->shared_level) | |
8693 | goto out; | |
8694 | ||
2c47e605 YZ |
8695 | ret = find_next_key(path, level + 1, &wc->update_progress); |
8696 | if (ret > 0) | |
8697 | wc->update_ref = 0; | |
8698 | ||
8699 | wc->stage = DROP_REFERENCE; | |
8700 | wc->shared_level = -1; | |
8701 | path->slots[level] = 0; | |
8702 | ||
8703 | /* | |
8704 | * check reference count again if the block isn't locked. | |
8705 | * we should start walking down the tree again if reference | |
8706 | * count is one. | |
8707 | */ | |
8708 | if (!path->locks[level]) { | |
8709 | BUG_ON(level == 0); | |
8710 | btrfs_tree_lock(eb); | |
8711 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8712 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8713 | |
2ff7e61e | 8714 | ret = btrfs_lookup_extent_info(trans, fs_info, |
3173a18f | 8715 | eb->start, level, 1, |
2c47e605 YZ |
8716 | &wc->refs[level], |
8717 | &wc->flags[level]); | |
79787eaa JM |
8718 | if (ret < 0) { |
8719 | btrfs_tree_unlock_rw(eb, path->locks[level]); | |
3268a246 | 8720 | path->locks[level] = 0; |
79787eaa JM |
8721 | return ret; |
8722 | } | |
2c47e605 YZ |
8723 | BUG_ON(wc->refs[level] == 0); |
8724 | if (wc->refs[level] == 1) { | |
bd681513 | 8725 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
3268a246 | 8726 | path->locks[level] = 0; |
2c47e605 YZ |
8727 | return 1; |
8728 | } | |
f82d02d9 | 8729 | } |
2c47e605 | 8730 | } |
f82d02d9 | 8731 | |
2c47e605 YZ |
8732 | /* wc->stage == DROP_REFERENCE */ |
8733 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); | |
5d4f98a2 | 8734 | |
2c47e605 YZ |
8735 | if (wc->refs[level] == 1) { |
8736 | if (level == 0) { | |
8737 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
e339a6b0 | 8738 | ret = btrfs_dec_ref(trans, root, eb, 1); |
2c47e605 | 8739 | else |
e339a6b0 | 8740 | ret = btrfs_dec_ref(trans, root, eb, 0); |
79787eaa | 8741 | BUG_ON(ret); /* -ENOMEM */ |
8d38d7eb | 8742 | ret = btrfs_qgroup_trace_leaf_items(trans, eb); |
1152651a | 8743 | if (ret) { |
0b246afa | 8744 | btrfs_err_rl(fs_info, |
5d163e0e JM |
8745 | "error %d accounting leaf items. Quota is out of sync, rescan required.", |
8746 | ret); | |
1152651a | 8747 | } |
2c47e605 YZ |
8748 | } |
8749 | /* make block locked assertion in clean_tree_block happy */ | |
8750 | if (!path->locks[level] && | |
8751 | btrfs_header_generation(eb) == trans->transid) { | |
8752 | btrfs_tree_lock(eb); | |
8753 | btrfs_set_lock_blocking(eb); | |
bd681513 | 8754 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8755 | } |
7c302b49 | 8756 | clean_tree_block(fs_info, eb); |
2c47e605 YZ |
8757 | } |
8758 | ||
8759 | if (eb == root->node) { | |
8760 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8761 | parent = eb->start; | |
65c6e82b QW |
8762 | else if (root->root_key.objectid != btrfs_header_owner(eb)) |
8763 | goto owner_mismatch; | |
2c47e605 YZ |
8764 | } else { |
8765 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) | |
8766 | parent = path->nodes[level + 1]->start; | |
65c6e82b QW |
8767 | else if (root->root_key.objectid != |
8768 | btrfs_header_owner(path->nodes[level + 1])) | |
8769 | goto owner_mismatch; | |
f82d02d9 | 8770 | } |
f82d02d9 | 8771 | |
5581a51a | 8772 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
2c47e605 YZ |
8773 | out: |
8774 | wc->refs[level] = 0; | |
8775 | wc->flags[level] = 0; | |
f0486c68 | 8776 | return 0; |
65c6e82b QW |
8777 | |
8778 | owner_mismatch: | |
8779 | btrfs_err_rl(fs_info, "unexpected tree owner, have %llu expect %llu", | |
8780 | btrfs_header_owner(eb), root->root_key.objectid); | |
8781 | return -EUCLEAN; | |
2c47e605 YZ |
8782 | } |
8783 | ||
8784 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, | |
8785 | struct btrfs_root *root, | |
8786 | struct btrfs_path *path, | |
8787 | struct walk_control *wc) | |
8788 | { | |
2c47e605 | 8789 | int level = wc->level; |
94fcca9f | 8790 | int lookup_info = 1; |
2c47e605 YZ |
8791 | int ret; |
8792 | ||
8793 | while (level >= 0) { | |
94fcca9f | 8794 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
2c47e605 YZ |
8795 | if (ret > 0) |
8796 | break; | |
8797 | ||
8798 | if (level == 0) | |
8799 | break; | |
8800 | ||
7a7965f8 YZ |
8801 | if (path->slots[level] >= |
8802 | btrfs_header_nritems(path->nodes[level])) | |
8803 | break; | |
8804 | ||
94fcca9f | 8805 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
1c4850e2 YZ |
8806 | if (ret > 0) { |
8807 | path->slots[level]++; | |
8808 | continue; | |
90d2c51d MX |
8809 | } else if (ret < 0) |
8810 | return ret; | |
1c4850e2 | 8811 | level = wc->level; |
f82d02d9 | 8812 | } |
f82d02d9 YZ |
8813 | return 0; |
8814 | } | |
8815 | ||
d397712b | 8816 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
98ed5174 | 8817 | struct btrfs_root *root, |
f82d02d9 | 8818 | struct btrfs_path *path, |
2c47e605 | 8819 | struct walk_control *wc, int max_level) |
20524f02 | 8820 | { |
2c47e605 | 8821 | int level = wc->level; |
20524f02 | 8822 | int ret; |
9f3a7427 | 8823 | |
2c47e605 YZ |
8824 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
8825 | while (level < max_level && path->nodes[level]) { | |
8826 | wc->level = level; | |
8827 | if (path->slots[level] + 1 < | |
8828 | btrfs_header_nritems(path->nodes[level])) { | |
8829 | path->slots[level]++; | |
20524f02 CM |
8830 | return 0; |
8831 | } else { | |
2c47e605 YZ |
8832 | ret = walk_up_proc(trans, root, path, wc); |
8833 | if (ret > 0) | |
8834 | return 0; | |
65c6e82b QW |
8835 | if (ret < 0) |
8836 | return ret; | |
bd56b302 | 8837 | |
2c47e605 | 8838 | if (path->locks[level]) { |
bd681513 CM |
8839 | btrfs_tree_unlock_rw(path->nodes[level], |
8840 | path->locks[level]); | |
2c47e605 | 8841 | path->locks[level] = 0; |
f82d02d9 | 8842 | } |
2c47e605 YZ |
8843 | free_extent_buffer(path->nodes[level]); |
8844 | path->nodes[level] = NULL; | |
8845 | level++; | |
20524f02 CM |
8846 | } |
8847 | } | |
8848 | return 1; | |
8849 | } | |
8850 | ||
9aca1d51 | 8851 | /* |
2c47e605 YZ |
8852 | * drop a subvolume tree. |
8853 | * | |
8854 | * this function traverses the tree freeing any blocks that only | |
8855 | * referenced by the tree. | |
8856 | * | |
8857 | * when a shared tree block is found. this function decreases its | |
8858 | * reference count by one. if update_ref is true, this function | |
8859 | * also make sure backrefs for the shared block and all lower level | |
8860 | * blocks are properly updated. | |
9d1a2a3a DS |
8861 | * |
8862 | * If called with for_reloc == 0, may exit early with -EAGAIN | |
9aca1d51 | 8863 | */ |
2c536799 | 8864 | int btrfs_drop_snapshot(struct btrfs_root *root, |
66d7e7f0 AJ |
8865 | struct btrfs_block_rsv *block_rsv, int update_ref, |
8866 | int for_reloc) | |
20524f02 | 8867 | { |
ab8d0fc4 | 8868 | struct btrfs_fs_info *fs_info = root->fs_info; |
5caf2a00 | 8869 | struct btrfs_path *path; |
2c47e605 | 8870 | struct btrfs_trans_handle *trans; |
ab8d0fc4 | 8871 | struct btrfs_root *tree_root = fs_info->tree_root; |
9f3a7427 | 8872 | struct btrfs_root_item *root_item = &root->root_item; |
2c47e605 YZ |
8873 | struct walk_control *wc; |
8874 | struct btrfs_key key; | |
8875 | int err = 0; | |
8876 | int ret; | |
8877 | int level; | |
d29a9f62 | 8878 | bool root_dropped = false; |
20524f02 | 8879 | |
4fd786e6 | 8880 | btrfs_debug(fs_info, "Drop subvolume %llu", root->root_key.objectid); |
1152651a | 8881 | |
5caf2a00 | 8882 | path = btrfs_alloc_path(); |
cb1b69f4 TI |
8883 | if (!path) { |
8884 | err = -ENOMEM; | |
8885 | goto out; | |
8886 | } | |
20524f02 | 8887 | |
2c47e605 | 8888 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
38a1a919 MF |
8889 | if (!wc) { |
8890 | btrfs_free_path(path); | |
cb1b69f4 TI |
8891 | err = -ENOMEM; |
8892 | goto out; | |
38a1a919 | 8893 | } |
2c47e605 | 8894 | |
a22285a6 | 8895 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
8896 | if (IS_ERR(trans)) { |
8897 | err = PTR_ERR(trans); | |
8898 | goto out_free; | |
8899 | } | |
98d5dc13 | 8900 | |
3fd0a558 YZ |
8901 | if (block_rsv) |
8902 | trans->block_rsv = block_rsv; | |
2c47e605 | 8903 | |
9f3a7427 | 8904 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
2c47e605 | 8905 | level = btrfs_header_level(root->node); |
5d4f98a2 YZ |
8906 | path->nodes[level] = btrfs_lock_root_node(root); |
8907 | btrfs_set_lock_blocking(path->nodes[level]); | |
9f3a7427 | 8908 | path->slots[level] = 0; |
bd681513 | 8909 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
8910 | memset(&wc->update_progress, 0, |
8911 | sizeof(wc->update_progress)); | |
9f3a7427 | 8912 | } else { |
9f3a7427 | 8913 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
2c47e605 YZ |
8914 | memcpy(&wc->update_progress, &key, |
8915 | sizeof(wc->update_progress)); | |
8916 | ||
6702ed49 | 8917 | level = root_item->drop_level; |
2c47e605 | 8918 | BUG_ON(level == 0); |
6702ed49 | 8919 | path->lowest_level = level; |
2c47e605 YZ |
8920 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
8921 | path->lowest_level = 0; | |
8922 | if (ret < 0) { | |
8923 | err = ret; | |
79787eaa | 8924 | goto out_end_trans; |
9f3a7427 | 8925 | } |
1c4850e2 | 8926 | WARN_ON(ret > 0); |
2c47e605 | 8927 | |
7d9eb12c CM |
8928 | /* |
8929 | * unlock our path, this is safe because only this | |
8930 | * function is allowed to delete this snapshot | |
8931 | */ | |
5d4f98a2 | 8932 | btrfs_unlock_up_safe(path, 0); |
2c47e605 YZ |
8933 | |
8934 | level = btrfs_header_level(root->node); | |
8935 | while (1) { | |
8936 | btrfs_tree_lock(path->nodes[level]); | |
8937 | btrfs_set_lock_blocking(path->nodes[level]); | |
fec386ac | 8938 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 | 8939 | |
2ff7e61e | 8940 | ret = btrfs_lookup_extent_info(trans, fs_info, |
2c47e605 | 8941 | path->nodes[level]->start, |
3173a18f | 8942 | level, 1, &wc->refs[level], |
2c47e605 | 8943 | &wc->flags[level]); |
79787eaa JM |
8944 | if (ret < 0) { |
8945 | err = ret; | |
8946 | goto out_end_trans; | |
8947 | } | |
2c47e605 YZ |
8948 | BUG_ON(wc->refs[level] == 0); |
8949 | ||
8950 | if (level == root_item->drop_level) | |
8951 | break; | |
8952 | ||
8953 | btrfs_tree_unlock(path->nodes[level]); | |
fec386ac | 8954 | path->locks[level] = 0; |
2c47e605 YZ |
8955 | WARN_ON(wc->refs[level] != 1); |
8956 | level--; | |
8957 | } | |
9f3a7427 | 8958 | } |
2c47e605 YZ |
8959 | |
8960 | wc->level = level; | |
8961 | wc->shared_level = -1; | |
8962 | wc->stage = DROP_REFERENCE; | |
8963 | wc->update_ref = update_ref; | |
8964 | wc->keep_locks = 0; | |
0b246afa | 8965 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
2c47e605 | 8966 | |
d397712b | 8967 | while (1) { |
9d1a2a3a | 8968 | |
2c47e605 YZ |
8969 | ret = walk_down_tree(trans, root, path, wc); |
8970 | if (ret < 0) { | |
8971 | err = ret; | |
20524f02 | 8972 | break; |
2c47e605 | 8973 | } |
9aca1d51 | 8974 | |
2c47e605 YZ |
8975 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
8976 | if (ret < 0) { | |
8977 | err = ret; | |
20524f02 | 8978 | break; |
2c47e605 YZ |
8979 | } |
8980 | ||
8981 | if (ret > 0) { | |
8982 | BUG_ON(wc->stage != DROP_REFERENCE); | |
e7a84565 CM |
8983 | break; |
8984 | } | |
2c47e605 YZ |
8985 | |
8986 | if (wc->stage == DROP_REFERENCE) { | |
8987 | level = wc->level; | |
8988 | btrfs_node_key(path->nodes[level], | |
8989 | &root_item->drop_progress, | |
8990 | path->slots[level]); | |
8991 | root_item->drop_level = level; | |
8992 | } | |
8993 | ||
8994 | BUG_ON(wc->level == 0); | |
3a45bb20 | 8995 | if (btrfs_should_end_transaction(trans) || |
2ff7e61e | 8996 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
2c47e605 YZ |
8997 | ret = btrfs_update_root(trans, tree_root, |
8998 | &root->root_key, | |
8999 | root_item); | |
79787eaa | 9000 | if (ret) { |
66642832 | 9001 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9002 | err = ret; |
9003 | goto out_end_trans; | |
9004 | } | |
2c47e605 | 9005 | |
3a45bb20 | 9006 | btrfs_end_transaction_throttle(trans); |
2ff7e61e | 9007 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
ab8d0fc4 JM |
9008 | btrfs_debug(fs_info, |
9009 | "drop snapshot early exit"); | |
3c8f2422 JB |
9010 | err = -EAGAIN; |
9011 | goto out_free; | |
9012 | } | |
9013 | ||
a22285a6 | 9014 | trans = btrfs_start_transaction(tree_root, 0); |
79787eaa JM |
9015 | if (IS_ERR(trans)) { |
9016 | err = PTR_ERR(trans); | |
9017 | goto out_free; | |
9018 | } | |
3fd0a558 YZ |
9019 | if (block_rsv) |
9020 | trans->block_rsv = block_rsv; | |
c3e69d58 | 9021 | } |
20524f02 | 9022 | } |
b3b4aa74 | 9023 | btrfs_release_path(path); |
79787eaa JM |
9024 | if (err) |
9025 | goto out_end_trans; | |
2c47e605 | 9026 | |
ab9ce7d4 | 9027 | ret = btrfs_del_root(trans, &root->root_key); |
79787eaa | 9028 | if (ret) { |
66642832 | 9029 | btrfs_abort_transaction(trans, ret); |
e19182c0 | 9030 | err = ret; |
79787eaa JM |
9031 | goto out_end_trans; |
9032 | } | |
2c47e605 | 9033 | |
76dda93c | 9034 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
cb517eab MX |
9035 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
9036 | NULL, NULL); | |
79787eaa | 9037 | if (ret < 0) { |
66642832 | 9038 | btrfs_abort_transaction(trans, ret); |
79787eaa JM |
9039 | err = ret; |
9040 | goto out_end_trans; | |
9041 | } else if (ret > 0) { | |
84cd948c JB |
9042 | /* if we fail to delete the orphan item this time |
9043 | * around, it'll get picked up the next time. | |
9044 | * | |
9045 | * The most common failure here is just -ENOENT. | |
9046 | */ | |
9047 | btrfs_del_orphan_item(trans, tree_root, | |
9048 | root->root_key.objectid); | |
76dda93c YZ |
9049 | } |
9050 | } | |
9051 | ||
27cdeb70 | 9052 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
2b9dbef2 | 9053 | btrfs_add_dropped_root(trans, root); |
76dda93c YZ |
9054 | } else { |
9055 | free_extent_buffer(root->node); | |
9056 | free_extent_buffer(root->commit_root); | |
b0feb9d9 | 9057 | btrfs_put_fs_root(root); |
76dda93c | 9058 | } |
d29a9f62 | 9059 | root_dropped = true; |
79787eaa | 9060 | out_end_trans: |
3a45bb20 | 9061 | btrfs_end_transaction_throttle(trans); |
79787eaa | 9062 | out_free: |
2c47e605 | 9063 | kfree(wc); |
5caf2a00 | 9064 | btrfs_free_path(path); |
cb1b69f4 | 9065 | out: |
d29a9f62 JB |
9066 | /* |
9067 | * So if we need to stop dropping the snapshot for whatever reason we | |
9068 | * need to make sure to add it back to the dead root list so that we | |
9069 | * keep trying to do the work later. This also cleans up roots if we | |
9070 | * don't have it in the radix (like when we recover after a power fail | |
9071 | * or unmount) so we don't leak memory. | |
9072 | */ | |
897ca819 | 9073 | if (!for_reloc && !root_dropped) |
d29a9f62 | 9074 | btrfs_add_dead_root(root); |
90515e7f | 9075 | if (err && err != -EAGAIN) |
ab8d0fc4 | 9076 | btrfs_handle_fs_error(fs_info, err, NULL); |
2c536799 | 9077 | return err; |
20524f02 | 9078 | } |
9078a3e1 | 9079 | |
2c47e605 YZ |
9080 | /* |
9081 | * drop subtree rooted at tree block 'node'. | |
9082 | * | |
9083 | * NOTE: this function will unlock and release tree block 'node' | |
66d7e7f0 | 9084 | * only used by relocation code |
2c47e605 | 9085 | */ |
f82d02d9 YZ |
9086 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
9087 | struct btrfs_root *root, | |
9088 | struct extent_buffer *node, | |
9089 | struct extent_buffer *parent) | |
9090 | { | |
0b246afa | 9091 | struct btrfs_fs_info *fs_info = root->fs_info; |
f82d02d9 | 9092 | struct btrfs_path *path; |
2c47e605 | 9093 | struct walk_control *wc; |
f82d02d9 YZ |
9094 | int level; |
9095 | int parent_level; | |
9096 | int ret = 0; | |
9097 | int wret; | |
9098 | ||
2c47e605 YZ |
9099 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
9100 | ||
f82d02d9 | 9101 | path = btrfs_alloc_path(); |
db5b493a TI |
9102 | if (!path) |
9103 | return -ENOMEM; | |
f82d02d9 | 9104 | |
2c47e605 | 9105 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
db5b493a TI |
9106 | if (!wc) { |
9107 | btrfs_free_path(path); | |
9108 | return -ENOMEM; | |
9109 | } | |
2c47e605 | 9110 | |
b9447ef8 | 9111 | btrfs_assert_tree_locked(parent); |
f82d02d9 YZ |
9112 | parent_level = btrfs_header_level(parent); |
9113 | extent_buffer_get(parent); | |
9114 | path->nodes[parent_level] = parent; | |
9115 | path->slots[parent_level] = btrfs_header_nritems(parent); | |
9116 | ||
b9447ef8 | 9117 | btrfs_assert_tree_locked(node); |
f82d02d9 | 9118 | level = btrfs_header_level(node); |
f82d02d9 YZ |
9119 | path->nodes[level] = node; |
9120 | path->slots[level] = 0; | |
bd681513 | 9121 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
2c47e605 YZ |
9122 | |
9123 | wc->refs[parent_level] = 1; | |
9124 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; | |
9125 | wc->level = level; | |
9126 | wc->shared_level = -1; | |
9127 | wc->stage = DROP_REFERENCE; | |
9128 | wc->update_ref = 0; | |
9129 | wc->keep_locks = 1; | |
0b246afa | 9130 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
f82d02d9 YZ |
9131 | |
9132 | while (1) { | |
2c47e605 YZ |
9133 | wret = walk_down_tree(trans, root, path, wc); |
9134 | if (wret < 0) { | |
f82d02d9 | 9135 | ret = wret; |
f82d02d9 | 9136 | break; |
2c47e605 | 9137 | } |
f82d02d9 | 9138 | |
2c47e605 | 9139 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
f82d02d9 YZ |
9140 | if (wret < 0) |
9141 | ret = wret; | |
9142 | if (wret != 0) | |
9143 | break; | |
9144 | } | |
9145 | ||
2c47e605 | 9146 | kfree(wc); |
f82d02d9 YZ |
9147 | btrfs_free_path(path); |
9148 | return ret; | |
9149 | } | |
9150 | ||
6202df69 | 9151 | static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags) |
ec44a35c CM |
9152 | { |
9153 | u64 num_devices; | |
fc67c450 | 9154 | u64 stripped; |
e4d8ec0f | 9155 | |
fc67c450 ID |
9156 | /* |
9157 | * if restripe for this chunk_type is on pick target profile and | |
9158 | * return, otherwise do the usual balance | |
9159 | */ | |
6202df69 | 9160 | stripped = get_restripe_target(fs_info, flags); |
fc67c450 ID |
9161 | if (stripped) |
9162 | return extended_to_chunk(stripped); | |
e4d8ec0f | 9163 | |
6202df69 | 9164 | num_devices = fs_info->fs_devices->rw_devices; |
cd02dca5 | 9165 | |
fc67c450 | 9166 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
53b381b3 | 9167 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
fc67c450 ID |
9168 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
9169 | ||
ec44a35c CM |
9170 | if (num_devices == 1) { |
9171 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9172 | stripped = flags & ~stripped; | |
9173 | ||
9174 | /* turn raid0 into single device chunks */ | |
9175 | if (flags & BTRFS_BLOCK_GROUP_RAID0) | |
9176 | return stripped; | |
9177 | ||
9178 | /* turn mirroring into duplication */ | |
9179 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | | |
9180 | BTRFS_BLOCK_GROUP_RAID10)) | |
9181 | return stripped | BTRFS_BLOCK_GROUP_DUP; | |
ec44a35c CM |
9182 | } else { |
9183 | /* they already had raid on here, just return */ | |
ec44a35c CM |
9184 | if (flags & stripped) |
9185 | return flags; | |
9186 | ||
9187 | stripped |= BTRFS_BLOCK_GROUP_DUP; | |
9188 | stripped = flags & ~stripped; | |
9189 | ||
9190 | /* switch duplicated blocks with raid1 */ | |
9191 | if (flags & BTRFS_BLOCK_GROUP_DUP) | |
9192 | return stripped | BTRFS_BLOCK_GROUP_RAID1; | |
9193 | ||
e3176ca2 | 9194 | /* this is drive concat, leave it alone */ |
ec44a35c | 9195 | } |
e3176ca2 | 9196 | |
ec44a35c CM |
9197 | return flags; |
9198 | } | |
9199 | ||
868f401a | 9200 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
0ef3e66b | 9201 | { |
f0486c68 YZ |
9202 | struct btrfs_space_info *sinfo = cache->space_info; |
9203 | u64 num_bytes; | |
199c36ea | 9204 | u64 min_allocable_bytes; |
f0486c68 | 9205 | int ret = -ENOSPC; |
0ef3e66b | 9206 | |
199c36ea MX |
9207 | /* |
9208 | * We need some metadata space and system metadata space for | |
9209 | * allocating chunks in some corner cases until we force to set | |
9210 | * it to be readonly. | |
9211 | */ | |
9212 | if ((sinfo->flags & | |
9213 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && | |
9214 | !force) | |
ee22184b | 9215 | min_allocable_bytes = SZ_1M; |
199c36ea MX |
9216 | else |
9217 | min_allocable_bytes = 0; | |
9218 | ||
f0486c68 YZ |
9219 | spin_lock(&sinfo->lock); |
9220 | spin_lock(&cache->lock); | |
61cfea9b W |
9221 | |
9222 | if (cache->ro) { | |
868f401a | 9223 | cache->ro++; |
61cfea9b W |
9224 | ret = 0; |
9225 | goto out; | |
9226 | } | |
9227 | ||
f0486c68 YZ |
9228 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
9229 | cache->bytes_super - btrfs_block_group_used(&cache->item); | |
9230 | ||
4136135b | 9231 | if (btrfs_space_info_used(sinfo, true) + num_bytes + |
37be25bc | 9232 | min_allocable_bytes <= sinfo->total_bytes) { |
f0486c68 | 9233 | sinfo->bytes_readonly += num_bytes; |
868f401a | 9234 | cache->ro++; |
633c0aad | 9235 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
f0486c68 YZ |
9236 | ret = 0; |
9237 | } | |
61cfea9b | 9238 | out: |
f0486c68 YZ |
9239 | spin_unlock(&cache->lock); |
9240 | spin_unlock(&sinfo->lock); | |
9241 | return ret; | |
9242 | } | |
7d9eb12c | 9243 | |
c83488af | 9244 | int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache) |
c286ac48 | 9245 | |
f0486c68 | 9246 | { |
c83488af | 9247 | struct btrfs_fs_info *fs_info = cache->fs_info; |
f0486c68 YZ |
9248 | struct btrfs_trans_handle *trans; |
9249 | u64 alloc_flags; | |
9250 | int ret; | |
7d9eb12c | 9251 | |
1bbc621e | 9252 | again: |
5e00f193 | 9253 | trans = btrfs_join_transaction(fs_info->extent_root); |
79787eaa JM |
9254 | if (IS_ERR(trans)) |
9255 | return PTR_ERR(trans); | |
5d4f98a2 | 9256 | |
1bbc621e CM |
9257 | /* |
9258 | * we're not allowed to set block groups readonly after the dirty | |
9259 | * block groups cache has started writing. If it already started, | |
9260 | * back off and let this transaction commit | |
9261 | */ | |
0b246afa | 9262 | mutex_lock(&fs_info->ro_block_group_mutex); |
3204d33c | 9263 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
1bbc621e CM |
9264 | u64 transid = trans->transid; |
9265 | ||
0b246afa | 9266 | mutex_unlock(&fs_info->ro_block_group_mutex); |
3a45bb20 | 9267 | btrfs_end_transaction(trans); |
1bbc621e | 9268 | |
2ff7e61e | 9269 | ret = btrfs_wait_for_commit(fs_info, transid); |
1bbc621e CM |
9270 | if (ret) |
9271 | return ret; | |
9272 | goto again; | |
9273 | } | |
9274 | ||
153c35b6 CM |
9275 | /* |
9276 | * if we are changing raid levels, try to allocate a corresponding | |
9277 | * block group with the new raid level. | |
9278 | */ | |
0b246afa | 9279 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
153c35b6 | 9280 | if (alloc_flags != cache->flags) { |
01458828 | 9281 | ret = do_chunk_alloc(trans, alloc_flags, |
153c35b6 CM |
9282 | CHUNK_ALLOC_FORCE); |
9283 | /* | |
9284 | * ENOSPC is allowed here, we may have enough space | |
9285 | * already allocated at the new raid level to | |
9286 | * carry on | |
9287 | */ | |
9288 | if (ret == -ENOSPC) | |
9289 | ret = 0; | |
9290 | if (ret < 0) | |
9291 | goto out; | |
9292 | } | |
1bbc621e | 9293 | |
868f401a | 9294 | ret = inc_block_group_ro(cache, 0); |
f0486c68 YZ |
9295 | if (!ret) |
9296 | goto out; | |
2ff7e61e | 9297 | alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags); |
01458828 | 9298 | ret = do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); |
f0486c68 YZ |
9299 | if (ret < 0) |
9300 | goto out; | |
868f401a | 9301 | ret = inc_block_group_ro(cache, 0); |
f0486c68 | 9302 | out: |
2f081088 | 9303 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 9304 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
34441361 | 9305 | mutex_lock(&fs_info->chunk_mutex); |
451a2c13 | 9306 | check_system_chunk(trans, alloc_flags); |
34441361 | 9307 | mutex_unlock(&fs_info->chunk_mutex); |
2f081088 | 9308 | } |
0b246afa | 9309 | mutex_unlock(&fs_info->ro_block_group_mutex); |
2f081088 | 9310 | |
3a45bb20 | 9311 | btrfs_end_transaction(trans); |
f0486c68 YZ |
9312 | return ret; |
9313 | } | |
5d4f98a2 | 9314 | |
43a7e99d | 9315 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, u64 type) |
c87f08ca | 9316 | { |
43a7e99d | 9317 | u64 alloc_flags = get_alloc_profile(trans->fs_info, type); |
2ff7e61e | 9318 | |
01458828 | 9319 | return do_chunk_alloc(trans, alloc_flags, CHUNK_ALLOC_FORCE); |
c87f08ca CM |
9320 | } |
9321 | ||
6d07bcec MX |
9322 | /* |
9323 | * helper to account the unused space of all the readonly block group in the | |
633c0aad | 9324 | * space_info. takes mirrors into account. |
6d07bcec | 9325 | */ |
633c0aad | 9326 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
6d07bcec MX |
9327 | { |
9328 | struct btrfs_block_group_cache *block_group; | |
9329 | u64 free_bytes = 0; | |
9330 | int factor; | |
9331 | ||
01327610 | 9332 | /* It's df, we don't care if it's racy */ |
633c0aad JB |
9333 | if (list_empty(&sinfo->ro_bgs)) |
9334 | return 0; | |
9335 | ||
9336 | spin_lock(&sinfo->lock); | |
9337 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { | |
6d07bcec MX |
9338 | spin_lock(&block_group->lock); |
9339 | ||
9340 | if (!block_group->ro) { | |
9341 | spin_unlock(&block_group->lock); | |
9342 | continue; | |
9343 | } | |
9344 | ||
46df06b8 | 9345 | factor = btrfs_bg_type_to_factor(block_group->flags); |
6d07bcec MX |
9346 | free_bytes += (block_group->key.offset - |
9347 | btrfs_block_group_used(&block_group->item)) * | |
9348 | factor; | |
9349 | ||
9350 | spin_unlock(&block_group->lock); | |
9351 | } | |
6d07bcec MX |
9352 | spin_unlock(&sinfo->lock); |
9353 | ||
9354 | return free_bytes; | |
9355 | } | |
9356 | ||
2ff7e61e | 9357 | void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache) |
5d4f98a2 | 9358 | { |
f0486c68 YZ |
9359 | struct btrfs_space_info *sinfo = cache->space_info; |
9360 | u64 num_bytes; | |
9361 | ||
9362 | BUG_ON(!cache->ro); | |
9363 | ||
9364 | spin_lock(&sinfo->lock); | |
9365 | spin_lock(&cache->lock); | |
868f401a Z |
9366 | if (!--cache->ro) { |
9367 | num_bytes = cache->key.offset - cache->reserved - | |
9368 | cache->pinned - cache->bytes_super - | |
9369 | btrfs_block_group_used(&cache->item); | |
9370 | sinfo->bytes_readonly -= num_bytes; | |
9371 | list_del_init(&cache->ro_list); | |
9372 | } | |
f0486c68 YZ |
9373 | spin_unlock(&cache->lock); |
9374 | spin_unlock(&sinfo->lock); | |
5d4f98a2 YZ |
9375 | } |
9376 | ||
ba1bf481 JB |
9377 | /* |
9378 | * checks to see if its even possible to relocate this block group. | |
9379 | * | |
9380 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its | |
9381 | * ok to go ahead and try. | |
9382 | */ | |
6bccf3ab | 9383 | int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr) |
1a40e23b | 9384 | { |
6bccf3ab | 9385 | struct btrfs_root *root = fs_info->extent_root; |
ba1bf481 JB |
9386 | struct btrfs_block_group_cache *block_group; |
9387 | struct btrfs_space_info *space_info; | |
0b246afa | 9388 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
ba1bf481 | 9389 | struct btrfs_device *device; |
6df9a95e | 9390 | struct btrfs_trans_handle *trans; |
cdcb725c | 9391 | u64 min_free; |
6719db6a JB |
9392 | u64 dev_min = 1; |
9393 | u64 dev_nr = 0; | |
4a5e98f5 | 9394 | u64 target; |
0305bc27 | 9395 | int debug; |
cdcb725c | 9396 | int index; |
ba1bf481 JB |
9397 | int full = 0; |
9398 | int ret = 0; | |
1a40e23b | 9399 | |
0b246afa | 9400 | debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG); |
0305bc27 | 9401 | |
0b246afa | 9402 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
1a40e23b | 9403 | |
ba1bf481 | 9404 | /* odd, couldn't find the block group, leave it alone */ |
0305bc27 QW |
9405 | if (!block_group) { |
9406 | if (debug) | |
0b246afa | 9407 | btrfs_warn(fs_info, |
0305bc27 QW |
9408 | "can't find block group for bytenr %llu", |
9409 | bytenr); | |
ba1bf481 | 9410 | return -1; |
0305bc27 | 9411 | } |
1a40e23b | 9412 | |
cdcb725c | 9413 | min_free = btrfs_block_group_used(&block_group->item); |
9414 | ||
ba1bf481 | 9415 | /* no bytes used, we're good */ |
cdcb725c | 9416 | if (!min_free) |
1a40e23b ZY |
9417 | goto out; |
9418 | ||
ba1bf481 JB |
9419 | space_info = block_group->space_info; |
9420 | spin_lock(&space_info->lock); | |
17d217fe | 9421 | |
ba1bf481 | 9422 | full = space_info->full; |
17d217fe | 9423 | |
ba1bf481 JB |
9424 | /* |
9425 | * if this is the last block group we have in this space, we can't | |
7ce618db CM |
9426 | * relocate it unless we're able to allocate a new chunk below. |
9427 | * | |
9428 | * Otherwise, we need to make sure we have room in the space to handle | |
9429 | * all of the extents from this block group. If we can, we're good | |
ba1bf481 | 9430 | */ |
7ce618db | 9431 | if ((space_info->total_bytes != block_group->key.offset) && |
4136135b LB |
9432 | (btrfs_space_info_used(space_info, false) + min_free < |
9433 | space_info->total_bytes)) { | |
ba1bf481 JB |
9434 | spin_unlock(&space_info->lock); |
9435 | goto out; | |
17d217fe | 9436 | } |
ba1bf481 | 9437 | spin_unlock(&space_info->lock); |
ea8c2819 | 9438 | |
ba1bf481 JB |
9439 | /* |
9440 | * ok we don't have enough space, but maybe we have free space on our | |
9441 | * devices to allocate new chunks for relocation, so loop through our | |
4a5e98f5 ID |
9442 | * alloc devices and guess if we have enough space. if this block |
9443 | * group is going to be restriped, run checks against the target | |
9444 | * profile instead of the current one. | |
ba1bf481 JB |
9445 | */ |
9446 | ret = -1; | |
ea8c2819 | 9447 | |
cdcb725c | 9448 | /* |
9449 | * index: | |
9450 | * 0: raid10 | |
9451 | * 1: raid1 | |
9452 | * 2: dup | |
9453 | * 3: raid0 | |
9454 | * 4: single | |
9455 | */ | |
0b246afa | 9456 | target = get_restripe_target(fs_info, block_group->flags); |
4a5e98f5 | 9457 | if (target) { |
3e72ee88 | 9458 | index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target)); |
4a5e98f5 ID |
9459 | } else { |
9460 | /* | |
9461 | * this is just a balance, so if we were marked as full | |
9462 | * we know there is no space for a new chunk | |
9463 | */ | |
0305bc27 QW |
9464 | if (full) { |
9465 | if (debug) | |
0b246afa JM |
9466 | btrfs_warn(fs_info, |
9467 | "no space to alloc new chunk for block group %llu", | |
9468 | block_group->key.objectid); | |
4a5e98f5 | 9469 | goto out; |
0305bc27 | 9470 | } |
4a5e98f5 | 9471 | |
3e72ee88 | 9472 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
4a5e98f5 ID |
9473 | } |
9474 | ||
e6ec716f | 9475 | if (index == BTRFS_RAID_RAID10) { |
cdcb725c | 9476 | dev_min = 4; |
6719db6a JB |
9477 | /* Divide by 2 */ |
9478 | min_free >>= 1; | |
e6ec716f | 9479 | } else if (index == BTRFS_RAID_RAID1) { |
cdcb725c | 9480 | dev_min = 2; |
e6ec716f | 9481 | } else if (index == BTRFS_RAID_DUP) { |
6719db6a JB |
9482 | /* Multiply by 2 */ |
9483 | min_free <<= 1; | |
e6ec716f | 9484 | } else if (index == BTRFS_RAID_RAID0) { |
cdcb725c | 9485 | dev_min = fs_devices->rw_devices; |
47c5713f | 9486 | min_free = div64_u64(min_free, dev_min); |
cdcb725c | 9487 | } |
9488 | ||
6df9a95e JB |
9489 | /* We need to do this so that we can look at pending chunks */ |
9490 | trans = btrfs_join_transaction(root); | |
9491 | if (IS_ERR(trans)) { | |
9492 | ret = PTR_ERR(trans); | |
9493 | goto out; | |
9494 | } | |
9495 | ||
0b246afa | 9496 | mutex_lock(&fs_info->chunk_mutex); |
ba1bf481 | 9497 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
7bfc837d | 9498 | u64 dev_offset; |
56bec294 | 9499 | |
ba1bf481 JB |
9500 | /* |
9501 | * check to make sure we can actually find a chunk with enough | |
9502 | * space to fit our block group in. | |
9503 | */ | |
63a212ab | 9504 | if (device->total_bytes > device->bytes_used + min_free && |
401e29c1 | 9505 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
6df9a95e | 9506 | ret = find_free_dev_extent(trans, device, min_free, |
7bfc837d | 9507 | &dev_offset, NULL); |
ba1bf481 | 9508 | if (!ret) |
cdcb725c | 9509 | dev_nr++; |
9510 | ||
9511 | if (dev_nr >= dev_min) | |
73e48b27 | 9512 | break; |
cdcb725c | 9513 | |
ba1bf481 | 9514 | ret = -1; |
725c8463 | 9515 | } |
edbd8d4e | 9516 | } |
0305bc27 | 9517 | if (debug && ret == -1) |
0b246afa JM |
9518 | btrfs_warn(fs_info, |
9519 | "no space to allocate a new chunk for block group %llu", | |
9520 | block_group->key.objectid); | |
9521 | mutex_unlock(&fs_info->chunk_mutex); | |
3a45bb20 | 9522 | btrfs_end_transaction(trans); |
edbd8d4e | 9523 | out: |
ba1bf481 | 9524 | btrfs_put_block_group(block_group); |
edbd8d4e CM |
9525 | return ret; |
9526 | } | |
9527 | ||
6bccf3ab JM |
9528 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
9529 | struct btrfs_path *path, | |
9530 | struct btrfs_key *key) | |
0b86a832 | 9531 | { |
6bccf3ab | 9532 | struct btrfs_root *root = fs_info->extent_root; |
925baedd | 9533 | int ret = 0; |
0b86a832 CM |
9534 | struct btrfs_key found_key; |
9535 | struct extent_buffer *leaf; | |
514c7dca QW |
9536 | struct btrfs_block_group_item bg; |
9537 | u64 flags; | |
0b86a832 | 9538 | int slot; |
edbd8d4e | 9539 | |
0b86a832 CM |
9540 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
9541 | if (ret < 0) | |
925baedd CM |
9542 | goto out; |
9543 | ||
d397712b | 9544 | while (1) { |
0b86a832 | 9545 | slot = path->slots[0]; |
edbd8d4e | 9546 | leaf = path->nodes[0]; |
0b86a832 CM |
9547 | if (slot >= btrfs_header_nritems(leaf)) { |
9548 | ret = btrfs_next_leaf(root, path); | |
9549 | if (ret == 0) | |
9550 | continue; | |
9551 | if (ret < 0) | |
925baedd | 9552 | goto out; |
0b86a832 | 9553 | break; |
edbd8d4e | 9554 | } |
0b86a832 | 9555 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
edbd8d4e | 9556 | |
0b86a832 | 9557 | if (found_key.objectid >= key->objectid && |
925baedd | 9558 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
6fb37b75 LB |
9559 | struct extent_map_tree *em_tree; |
9560 | struct extent_map *em; | |
9561 | ||
9562 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
9563 | read_lock(&em_tree->lock); | |
9564 | em = lookup_extent_mapping(em_tree, found_key.objectid, | |
9565 | found_key.offset); | |
9566 | read_unlock(&em_tree->lock); | |
9567 | if (!em) { | |
0b246afa | 9568 | btrfs_err(fs_info, |
6fb37b75 LB |
9569 | "logical %llu len %llu found bg but no related chunk", |
9570 | found_key.objectid, found_key.offset); | |
9571 | ret = -ENOENT; | |
514c7dca QW |
9572 | } else if (em->start != found_key.objectid || |
9573 | em->len != found_key.offset) { | |
9574 | btrfs_err(fs_info, | |
9575 | "block group %llu len %llu mismatch with chunk %llu len %llu", | |
9576 | found_key.objectid, found_key.offset, | |
9577 | em->start, em->len); | |
9578 | ret = -EUCLEAN; | |
6fb37b75 | 9579 | } else { |
514c7dca QW |
9580 | read_extent_buffer(leaf, &bg, |
9581 | btrfs_item_ptr_offset(leaf, slot), | |
9582 | sizeof(bg)); | |
9583 | flags = btrfs_block_group_flags(&bg) & | |
9584 | BTRFS_BLOCK_GROUP_TYPE_MASK; | |
9585 | ||
9586 | if (flags != (em->map_lookup->type & | |
9587 | BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
9588 | btrfs_err(fs_info, | |
9589 | "block group %llu len %llu type flags 0x%llx mismatch with chunk type flags 0x%llx", | |
9590 | found_key.objectid, | |
9591 | found_key.offset, flags, | |
9592 | (BTRFS_BLOCK_GROUP_TYPE_MASK & | |
9593 | em->map_lookup->type)); | |
9594 | ret = -EUCLEAN; | |
9595 | } else { | |
9596 | ret = 0; | |
9597 | } | |
6fb37b75 | 9598 | } |
187ee58c | 9599 | free_extent_map(em); |
925baedd CM |
9600 | goto out; |
9601 | } | |
0b86a832 | 9602 | path->slots[0]++; |
edbd8d4e | 9603 | } |
925baedd | 9604 | out: |
0b86a832 | 9605 | return ret; |
edbd8d4e CM |
9606 | } |
9607 | ||
0af3d00b JB |
9608 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
9609 | { | |
9610 | struct btrfs_block_group_cache *block_group; | |
9611 | u64 last = 0; | |
9612 | ||
9613 | while (1) { | |
9614 | struct inode *inode; | |
9615 | ||
9616 | block_group = btrfs_lookup_first_block_group(info, last); | |
9617 | while (block_group) { | |
9618 | spin_lock(&block_group->lock); | |
9619 | if (block_group->iref) | |
9620 | break; | |
9621 | spin_unlock(&block_group->lock); | |
2ff7e61e | 9622 | block_group = next_block_group(info, block_group); |
0af3d00b JB |
9623 | } |
9624 | if (!block_group) { | |
9625 | if (last == 0) | |
9626 | break; | |
9627 | last = 0; | |
9628 | continue; | |
9629 | } | |
9630 | ||
9631 | inode = block_group->inode; | |
9632 | block_group->iref = 0; | |
9633 | block_group->inode = NULL; | |
9634 | spin_unlock(&block_group->lock); | |
f3bca802 | 9635 | ASSERT(block_group->io_ctl.inode == NULL); |
0af3d00b JB |
9636 | iput(inode); |
9637 | last = block_group->key.objectid + block_group->key.offset; | |
9638 | btrfs_put_block_group(block_group); | |
9639 | } | |
9640 | } | |
9641 | ||
5cdd7db6 FM |
9642 | /* |
9643 | * Must be called only after stopping all workers, since we could have block | |
9644 | * group caching kthreads running, and therefore they could race with us if we | |
9645 | * freed the block groups before stopping them. | |
9646 | */ | |
1a40e23b ZY |
9647 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
9648 | { | |
9649 | struct btrfs_block_group_cache *block_group; | |
4184ea7f | 9650 | struct btrfs_space_info *space_info; |
11833d66 | 9651 | struct btrfs_caching_control *caching_ctl; |
1a40e23b ZY |
9652 | struct rb_node *n; |
9653 | ||
9e351cc8 | 9654 | down_write(&info->commit_root_sem); |
11833d66 YZ |
9655 | while (!list_empty(&info->caching_block_groups)) { |
9656 | caching_ctl = list_entry(info->caching_block_groups.next, | |
9657 | struct btrfs_caching_control, list); | |
9658 | list_del(&caching_ctl->list); | |
9659 | put_caching_control(caching_ctl); | |
9660 | } | |
9e351cc8 | 9661 | up_write(&info->commit_root_sem); |
11833d66 | 9662 | |
47ab2a6c JB |
9663 | spin_lock(&info->unused_bgs_lock); |
9664 | while (!list_empty(&info->unused_bgs)) { | |
9665 | block_group = list_first_entry(&info->unused_bgs, | |
9666 | struct btrfs_block_group_cache, | |
9667 | bg_list); | |
9668 | list_del_init(&block_group->bg_list); | |
9669 | btrfs_put_block_group(block_group); | |
9670 | } | |
9671 | spin_unlock(&info->unused_bgs_lock); | |
9672 | ||
1a40e23b ZY |
9673 | spin_lock(&info->block_group_cache_lock); |
9674 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { | |
9675 | block_group = rb_entry(n, struct btrfs_block_group_cache, | |
9676 | cache_node); | |
1a40e23b ZY |
9677 | rb_erase(&block_group->cache_node, |
9678 | &info->block_group_cache_tree); | |
01eacb27 | 9679 | RB_CLEAR_NODE(&block_group->cache_node); |
d899e052 YZ |
9680 | spin_unlock(&info->block_group_cache_lock); |
9681 | ||
80eb234a | 9682 | down_write(&block_group->space_info->groups_sem); |
1a40e23b | 9683 | list_del(&block_group->list); |
80eb234a | 9684 | up_write(&block_group->space_info->groups_sem); |
d2fb3437 | 9685 | |
3c14874a JB |
9686 | /* |
9687 | * We haven't cached this block group, which means we could | |
9688 | * possibly have excluded extents on this block group. | |
9689 | */ | |
36cce922 JB |
9690 | if (block_group->cached == BTRFS_CACHE_NO || |
9691 | block_group->cached == BTRFS_CACHE_ERROR) | |
9e715da8 | 9692 | free_excluded_extents(block_group); |
3c14874a | 9693 | |
817d52f8 | 9694 | btrfs_remove_free_space_cache(block_group); |
5cdd7db6 | 9695 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); |
f3bca802 LB |
9696 | ASSERT(list_empty(&block_group->dirty_list)); |
9697 | ASSERT(list_empty(&block_group->io_list)); | |
9698 | ASSERT(list_empty(&block_group->bg_list)); | |
9699 | ASSERT(atomic_read(&block_group->count) == 1); | |
11dfe35a | 9700 | btrfs_put_block_group(block_group); |
d899e052 YZ |
9701 | |
9702 | spin_lock(&info->block_group_cache_lock); | |
1a40e23b ZY |
9703 | } |
9704 | spin_unlock(&info->block_group_cache_lock); | |
4184ea7f CM |
9705 | |
9706 | /* now that all the block groups are freed, go through and | |
9707 | * free all the space_info structs. This is only called during | |
9708 | * the final stages of unmount, and so we know nobody is | |
9709 | * using them. We call synchronize_rcu() once before we start, | |
9710 | * just to be on the safe side. | |
9711 | */ | |
9712 | synchronize_rcu(); | |
9713 | ||
8929ecfa YZ |
9714 | release_global_block_rsv(info); |
9715 | ||
67871254 | 9716 | while (!list_empty(&info->space_info)) { |
6ab0a202 JM |
9717 | int i; |
9718 | ||
4184ea7f CM |
9719 | space_info = list_entry(info->space_info.next, |
9720 | struct btrfs_space_info, | |
9721 | list); | |
d555b6c3 JB |
9722 | |
9723 | /* | |
9724 | * Do not hide this behind enospc_debug, this is actually | |
9725 | * important and indicates a real bug if this happens. | |
9726 | */ | |
9727 | if (WARN_ON(space_info->bytes_pinned > 0 || | |
b069e0c3 | 9728 | space_info->bytes_reserved > 0 || |
d555b6c3 | 9729 | space_info->bytes_may_use > 0)) |
ab8d0fc4 | 9730 | dump_space_info(info, space_info, 0, 0); |
4184ea7f | 9731 | list_del(&space_info->list); |
6ab0a202 JM |
9732 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
9733 | struct kobject *kobj; | |
c1895442 JM |
9734 | kobj = space_info->block_group_kobjs[i]; |
9735 | space_info->block_group_kobjs[i] = NULL; | |
9736 | if (kobj) { | |
6ab0a202 JM |
9737 | kobject_del(kobj); |
9738 | kobject_put(kobj); | |
9739 | } | |
9740 | } | |
9741 | kobject_del(&space_info->kobj); | |
9742 | kobject_put(&space_info->kobj); | |
4184ea7f | 9743 | } |
1a40e23b ZY |
9744 | return 0; |
9745 | } | |
9746 | ||
75cb379d JM |
9747 | /* link_block_group will queue up kobjects to add when we're reclaim-safe */ |
9748 | void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info) | |
9749 | { | |
9750 | struct btrfs_space_info *space_info; | |
9751 | struct raid_kobject *rkobj; | |
9752 | LIST_HEAD(list); | |
9753 | int index; | |
9754 | int ret = 0; | |
9755 | ||
9756 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
9757 | list_splice_init(&fs_info->pending_raid_kobjs, &list); | |
9758 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
9759 | ||
9760 | list_for_each_entry(rkobj, &list, list) { | |
9761 | space_info = __find_space_info(fs_info, rkobj->flags); | |
9762 | index = btrfs_bg_flags_to_raid_index(rkobj->flags); | |
9763 | ||
9764 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, | |
9765 | "%s", get_raid_name(index)); | |
9766 | if (ret) { | |
9767 | kobject_put(&rkobj->kobj); | |
9768 | break; | |
9769 | } | |
9770 | } | |
9771 | if (ret) | |
9772 | btrfs_warn(fs_info, | |
9773 | "failed to add kobject for block cache, ignoring"); | |
9774 | } | |
9775 | ||
c434d21c | 9776 | static void link_block_group(struct btrfs_block_group_cache *cache) |
b742bb82 | 9777 | { |
c434d21c | 9778 | struct btrfs_space_info *space_info = cache->space_info; |
75cb379d | 9779 | struct btrfs_fs_info *fs_info = cache->fs_info; |
3e72ee88 | 9780 | int index = btrfs_bg_flags_to_raid_index(cache->flags); |
ed55b6ac | 9781 | bool first = false; |
b742bb82 YZ |
9782 | |
9783 | down_write(&space_info->groups_sem); | |
ed55b6ac JM |
9784 | if (list_empty(&space_info->block_groups[index])) |
9785 | first = true; | |
9786 | list_add_tail(&cache->list, &space_info->block_groups[index]); | |
9787 | up_write(&space_info->groups_sem); | |
9788 | ||
9789 | if (first) { | |
75cb379d JM |
9790 | struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
9791 | if (!rkobj) { | |
9792 | btrfs_warn(cache->fs_info, | |
9793 | "couldn't alloc memory for raid level kobject"); | |
9794 | return; | |
6ab0a202 | 9795 | } |
75cb379d JM |
9796 | rkobj->flags = cache->flags; |
9797 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); | |
9798 | ||
9799 | spin_lock(&fs_info->pending_raid_kobjs_lock); | |
9800 | list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs); | |
9801 | spin_unlock(&fs_info->pending_raid_kobjs_lock); | |
c1895442 | 9802 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
6ab0a202 | 9803 | } |
b742bb82 YZ |
9804 | } |
9805 | ||
920e4a58 | 9806 | static struct btrfs_block_group_cache * |
2ff7e61e JM |
9807 | btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info, |
9808 | u64 start, u64 size) | |
920e4a58 MX |
9809 | { |
9810 | struct btrfs_block_group_cache *cache; | |
9811 | ||
9812 | cache = kzalloc(sizeof(*cache), GFP_NOFS); | |
9813 | if (!cache) | |
9814 | return NULL; | |
9815 | ||
9816 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), | |
9817 | GFP_NOFS); | |
9818 | if (!cache->free_space_ctl) { | |
9819 | kfree(cache); | |
9820 | return NULL; | |
9821 | } | |
9822 | ||
9823 | cache->key.objectid = start; | |
9824 | cache->key.offset = size; | |
9825 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; | |
9826 | ||
0b246afa | 9827 | cache->fs_info = fs_info; |
e4ff5fb5 | 9828 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); |
1e144fb8 OS |
9829 | set_free_space_tree_thresholds(cache); |
9830 | ||
920e4a58 MX |
9831 | atomic_set(&cache->count, 1); |
9832 | spin_lock_init(&cache->lock); | |
e570fd27 | 9833 | init_rwsem(&cache->data_rwsem); |
920e4a58 MX |
9834 | INIT_LIST_HEAD(&cache->list); |
9835 | INIT_LIST_HEAD(&cache->cluster_list); | |
47ab2a6c | 9836 | INIT_LIST_HEAD(&cache->bg_list); |
633c0aad | 9837 | INIT_LIST_HEAD(&cache->ro_list); |
ce93ec54 | 9838 | INIT_LIST_HEAD(&cache->dirty_list); |
c9dc4c65 | 9839 | INIT_LIST_HEAD(&cache->io_list); |
920e4a58 | 9840 | btrfs_init_free_space_ctl(cache); |
04216820 | 9841 | atomic_set(&cache->trimming, 0); |
a5ed9182 | 9842 | mutex_init(&cache->free_space_lock); |
0966a7b1 | 9843 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); |
920e4a58 MX |
9844 | |
9845 | return cache; | |
9846 | } | |
9847 | ||
7ef49515 QW |
9848 | |
9849 | /* | |
9850 | * Iterate all chunks and verify that each of them has the corresponding block | |
9851 | * group | |
9852 | */ | |
9853 | static int check_chunk_block_group_mappings(struct btrfs_fs_info *fs_info) | |
9854 | { | |
9855 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
9856 | struct extent_map *em; | |
9857 | struct btrfs_block_group_cache *bg; | |
9858 | u64 start = 0; | |
9859 | int ret = 0; | |
9860 | ||
9861 | while (1) { | |
9862 | read_lock(&map_tree->map_tree.lock); | |
9863 | /* | |
9864 | * lookup_extent_mapping will return the first extent map | |
9865 | * intersecting the range, so setting @len to 1 is enough to | |
9866 | * get the first chunk. | |
9867 | */ | |
9868 | em = lookup_extent_mapping(&map_tree->map_tree, start, 1); | |
9869 | read_unlock(&map_tree->map_tree.lock); | |
9870 | if (!em) | |
9871 | break; | |
9872 | ||
9873 | bg = btrfs_lookup_block_group(fs_info, em->start); | |
9874 | if (!bg) { | |
9875 | btrfs_err(fs_info, | |
9876 | "chunk start=%llu len=%llu doesn't have corresponding block group", | |
9877 | em->start, em->len); | |
9878 | ret = -EUCLEAN; | |
9879 | free_extent_map(em); | |
9880 | break; | |
9881 | } | |
9882 | if (bg->key.objectid != em->start || | |
9883 | bg->key.offset != em->len || | |
9884 | (bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) != | |
9885 | (em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK)) { | |
9886 | btrfs_err(fs_info, | |
9887 | "chunk start=%llu len=%llu flags=0x%llx doesn't match block group start=%llu len=%llu flags=0x%llx", | |
9888 | em->start, em->len, | |
9889 | em->map_lookup->type & BTRFS_BLOCK_GROUP_TYPE_MASK, | |
9890 | bg->key.objectid, bg->key.offset, | |
9891 | bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK); | |
9892 | ret = -EUCLEAN; | |
9893 | free_extent_map(em); | |
9894 | btrfs_put_block_group(bg); | |
9895 | break; | |
9896 | } | |
9897 | start = em->start + em->len; | |
9898 | free_extent_map(em); | |
9899 | btrfs_put_block_group(bg); | |
9900 | } | |
9901 | return ret; | |
9902 | } | |
9903 | ||
5b4aacef | 9904 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
9078a3e1 CM |
9905 | { |
9906 | struct btrfs_path *path; | |
9907 | int ret; | |
9078a3e1 | 9908 | struct btrfs_block_group_cache *cache; |
6324fbf3 | 9909 | struct btrfs_space_info *space_info; |
9078a3e1 CM |
9910 | struct btrfs_key key; |
9911 | struct btrfs_key found_key; | |
5f39d397 | 9912 | struct extent_buffer *leaf; |
0af3d00b JB |
9913 | int need_clear = 0; |
9914 | u64 cache_gen; | |
49303381 LB |
9915 | u64 feature; |
9916 | int mixed; | |
9917 | ||
9918 | feature = btrfs_super_incompat_flags(info->super_copy); | |
9919 | mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS); | |
96b5179d | 9920 | |
9078a3e1 | 9921 | key.objectid = 0; |
0b86a832 | 9922 | key.offset = 0; |
962a298f | 9923 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
9078a3e1 CM |
9924 | path = btrfs_alloc_path(); |
9925 | if (!path) | |
9926 | return -ENOMEM; | |
e4058b54 | 9927 | path->reada = READA_FORWARD; |
9078a3e1 | 9928 | |
0b246afa JM |
9929 | cache_gen = btrfs_super_cache_generation(info->super_copy); |
9930 | if (btrfs_test_opt(info, SPACE_CACHE) && | |
9931 | btrfs_super_generation(info->super_copy) != cache_gen) | |
0af3d00b | 9932 | need_clear = 1; |
0b246afa | 9933 | if (btrfs_test_opt(info, CLEAR_CACHE)) |
88c2ba3b | 9934 | need_clear = 1; |
0af3d00b | 9935 | |
d397712b | 9936 | while (1) { |
6bccf3ab | 9937 | ret = find_first_block_group(info, path, &key); |
b742bb82 YZ |
9938 | if (ret > 0) |
9939 | break; | |
0b86a832 CM |
9940 | if (ret != 0) |
9941 | goto error; | |
920e4a58 | 9942 | |
5f39d397 CM |
9943 | leaf = path->nodes[0]; |
9944 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
920e4a58 | 9945 | |
2ff7e61e | 9946 | cache = btrfs_create_block_group_cache(info, found_key.objectid, |
920e4a58 | 9947 | found_key.offset); |
9078a3e1 | 9948 | if (!cache) { |
0b86a832 | 9949 | ret = -ENOMEM; |
f0486c68 | 9950 | goto error; |
9078a3e1 | 9951 | } |
96303081 | 9952 | |
cf7c1ef6 LB |
9953 | if (need_clear) { |
9954 | /* | |
9955 | * When we mount with old space cache, we need to | |
9956 | * set BTRFS_DC_CLEAR and set dirty flag. | |
9957 | * | |
9958 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we | |
9959 | * truncate the old free space cache inode and | |
9960 | * setup a new one. | |
9961 | * b) Setting 'dirty flag' makes sure that we flush | |
9962 | * the new space cache info onto disk. | |
9963 | */ | |
0b246afa | 9964 | if (btrfs_test_opt(info, SPACE_CACHE)) |
ce93ec54 | 9965 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
cf7c1ef6 | 9966 | } |
0af3d00b | 9967 | |
5f39d397 CM |
9968 | read_extent_buffer(leaf, &cache->item, |
9969 | btrfs_item_ptr_offset(leaf, path->slots[0]), | |
9970 | sizeof(cache->item)); | |
920e4a58 | 9971 | cache->flags = btrfs_block_group_flags(&cache->item); |
49303381 LB |
9972 | if (!mixed && |
9973 | ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && | |
9974 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { | |
9975 | btrfs_err(info, | |
9976 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", | |
9977 | cache->key.objectid); | |
9978 | ret = -EINVAL; | |
9979 | goto error; | |
9980 | } | |
0b86a832 | 9981 | |
9078a3e1 | 9982 | key.objectid = found_key.objectid + found_key.offset; |
b3b4aa74 | 9983 | btrfs_release_path(path); |
34d52cb6 | 9984 | |
3c14874a JB |
9985 | /* |
9986 | * We need to exclude the super stripes now so that the space | |
9987 | * info has super bytes accounted for, otherwise we'll think | |
9988 | * we have more space than we actually do. | |
9989 | */ | |
3c4da657 | 9990 | ret = exclude_super_stripes(cache); |
835d974f JB |
9991 | if (ret) { |
9992 | /* | |
9993 | * We may have excluded something, so call this just in | |
9994 | * case. | |
9995 | */ | |
9e715da8 | 9996 | free_excluded_extents(cache); |
920e4a58 | 9997 | btrfs_put_block_group(cache); |
835d974f JB |
9998 | goto error; |
9999 | } | |
3c14874a | 10000 | |
817d52f8 JB |
10001 | /* |
10002 | * check for two cases, either we are full, and therefore | |
10003 | * don't need to bother with the caching work since we won't | |
10004 | * find any space, or we are empty, and we can just add all | |
10005 | * the space in and be done with it. This saves us _alot_ of | |
10006 | * time, particularly in the full case. | |
10007 | */ | |
10008 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { | |
11833d66 | 10009 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10010 | cache->cached = BTRFS_CACHE_FINISHED; |
9e715da8 | 10011 | free_excluded_extents(cache); |
817d52f8 | 10012 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
11833d66 | 10013 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10014 | cache->cached = BTRFS_CACHE_FINISHED; |
4457c1c7 | 10015 | add_new_free_space(cache, found_key.objectid, |
817d52f8 JB |
10016 | found_key.objectid + |
10017 | found_key.offset); | |
9e715da8 | 10018 | free_excluded_extents(cache); |
817d52f8 | 10019 | } |
96b5179d | 10020 | |
0b246afa | 10021 | ret = btrfs_add_block_group_cache(info, cache); |
8c579fe7 JB |
10022 | if (ret) { |
10023 | btrfs_remove_free_space_cache(cache); | |
10024 | btrfs_put_block_group(cache); | |
10025 | goto error; | |
10026 | } | |
10027 | ||
0b246afa | 10028 | trace_btrfs_add_block_group(info, cache, 0); |
d2006e6d NB |
10029 | update_space_info(info, cache->flags, found_key.offset, |
10030 | btrfs_block_group_used(&cache->item), | |
10031 | cache->bytes_super, &space_info); | |
8c579fe7 | 10032 | |
6324fbf3 | 10033 | cache->space_info = space_info; |
1b2da372 | 10034 | |
c434d21c | 10035 | link_block_group(cache); |
0f9dd46c | 10036 | |
0b246afa | 10037 | set_avail_alloc_bits(info, cache->flags); |
2ff7e61e | 10038 | if (btrfs_chunk_readonly(info, cache->key.objectid)) { |
868f401a | 10039 | inc_block_group_ro(cache, 1); |
47ab2a6c | 10040 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
031f24da QW |
10041 | ASSERT(list_empty(&cache->bg_list)); |
10042 | btrfs_mark_bg_unused(cache); | |
47ab2a6c | 10043 | } |
9078a3e1 | 10044 | } |
b742bb82 | 10045 | |
0b246afa | 10046 | list_for_each_entry_rcu(space_info, &info->space_info, list) { |
2ff7e61e | 10047 | if (!(get_alloc_profile(info, space_info->flags) & |
b742bb82 YZ |
10048 | (BTRFS_BLOCK_GROUP_RAID10 | |
10049 | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
10050 | BTRFS_BLOCK_GROUP_RAID5 | |
10051 | BTRFS_BLOCK_GROUP_RAID6 | | |
b742bb82 YZ |
10052 | BTRFS_BLOCK_GROUP_DUP))) |
10053 | continue; | |
10054 | /* | |
10055 | * avoid allocating from un-mirrored block group if there are | |
10056 | * mirrored block groups. | |
10057 | */ | |
1095cc0d | 10058 | list_for_each_entry(cache, |
10059 | &space_info->block_groups[BTRFS_RAID_RAID0], | |
10060 | list) | |
868f401a | 10061 | inc_block_group_ro(cache, 1); |
1095cc0d | 10062 | list_for_each_entry(cache, |
10063 | &space_info->block_groups[BTRFS_RAID_SINGLE], | |
10064 | list) | |
868f401a | 10065 | inc_block_group_ro(cache, 1); |
9078a3e1 | 10066 | } |
f0486c68 | 10067 | |
75cb379d | 10068 | btrfs_add_raid_kobjects(info); |
f0486c68 | 10069 | init_global_block_rsv(info); |
7ef49515 | 10070 | ret = check_chunk_block_group_mappings(info); |
0b86a832 | 10071 | error: |
9078a3e1 | 10072 | btrfs_free_path(path); |
0b86a832 | 10073 | return ret; |
9078a3e1 | 10074 | } |
6324fbf3 | 10075 | |
6c686b35 | 10076 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) |
ea658bad | 10077 | { |
6c686b35 | 10078 | struct btrfs_fs_info *fs_info = trans->fs_info; |
ea658bad | 10079 | struct btrfs_block_group_cache *block_group, *tmp; |
0b246afa | 10080 | struct btrfs_root *extent_root = fs_info->extent_root; |
ea658bad JB |
10081 | struct btrfs_block_group_item item; |
10082 | struct btrfs_key key; | |
10083 | int ret = 0; | |
d9a0540a | 10084 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
ea658bad | 10085 | |
d9a0540a | 10086 | trans->can_flush_pending_bgs = false; |
47ab2a6c | 10087 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { |
ea658bad | 10088 | if (ret) |
c92f6be3 | 10089 | goto next; |
ea658bad JB |
10090 | |
10091 | spin_lock(&block_group->lock); | |
10092 | memcpy(&item, &block_group->item, sizeof(item)); | |
10093 | memcpy(&key, &block_group->key, sizeof(key)); | |
10094 | spin_unlock(&block_group->lock); | |
10095 | ||
10096 | ret = btrfs_insert_item(trans, extent_root, &key, &item, | |
10097 | sizeof(item)); | |
10098 | if (ret) | |
66642832 | 10099 | btrfs_abort_transaction(trans, ret); |
97aff912 | 10100 | ret = btrfs_finish_chunk_alloc(trans, key.objectid, key.offset); |
6df9a95e | 10101 | if (ret) |
66642832 | 10102 | btrfs_abort_transaction(trans, ret); |
e4e0711c | 10103 | add_block_group_free_space(trans, block_group); |
1e144fb8 | 10104 | /* already aborted the transaction if it failed. */ |
c92f6be3 FM |
10105 | next: |
10106 | list_del_init(&block_group->bg_list); | |
ea658bad | 10107 | } |
d9a0540a | 10108 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
ea658bad JB |
10109 | } |
10110 | ||
e7e02096 | 10111 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, u64 bytes_used, |
0174484d | 10112 | u64 type, u64 chunk_offset, u64 size) |
6324fbf3 | 10113 | { |
e7e02096 | 10114 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6324fbf3 | 10115 | struct btrfs_block_group_cache *cache; |
0b246afa | 10116 | int ret; |
6324fbf3 | 10117 | |
0b246afa | 10118 | btrfs_set_log_full_commit(fs_info, trans); |
e02119d5 | 10119 | |
2ff7e61e | 10120 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size); |
0f9dd46c JB |
10121 | if (!cache) |
10122 | return -ENOMEM; | |
34d52cb6 | 10123 | |
6324fbf3 | 10124 | btrfs_set_block_group_used(&cache->item, bytes_used); |
0174484d NB |
10125 | btrfs_set_block_group_chunk_objectid(&cache->item, |
10126 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); | |
6324fbf3 CM |
10127 | btrfs_set_block_group_flags(&cache->item, type); |
10128 | ||
920e4a58 | 10129 | cache->flags = type; |
11833d66 | 10130 | cache->last_byte_to_unpin = (u64)-1; |
817d52f8 | 10131 | cache->cached = BTRFS_CACHE_FINISHED; |
1e144fb8 | 10132 | cache->needs_free_space = 1; |
3c4da657 | 10133 | ret = exclude_super_stripes(cache); |
835d974f JB |
10134 | if (ret) { |
10135 | /* | |
10136 | * We may have excluded something, so call this just in | |
10137 | * case. | |
10138 | */ | |
9e715da8 | 10139 | free_excluded_extents(cache); |
920e4a58 | 10140 | btrfs_put_block_group(cache); |
835d974f JB |
10141 | return ret; |
10142 | } | |
96303081 | 10143 | |
4457c1c7 | 10144 | add_new_free_space(cache, chunk_offset, chunk_offset + size); |
817d52f8 | 10145 | |
9e715da8 | 10146 | free_excluded_extents(cache); |
11833d66 | 10147 | |
d0bd4560 | 10148 | #ifdef CONFIG_BTRFS_DEBUG |
2ff7e61e | 10149 | if (btrfs_should_fragment_free_space(cache)) { |
d0bd4560 JB |
10150 | u64 new_bytes_used = size - bytes_used; |
10151 | ||
10152 | bytes_used += new_bytes_used >> 1; | |
2ff7e61e | 10153 | fragment_free_space(cache); |
d0bd4560 JB |
10154 | } |
10155 | #endif | |
2e6e5183 | 10156 | /* |
2be12ef7 NB |
10157 | * Ensure the corresponding space_info object is created and |
10158 | * assigned to our block group. We want our bg to be added to the rbtree | |
10159 | * with its ->space_info set. | |
2e6e5183 | 10160 | */ |
2be12ef7 | 10161 | cache->space_info = __find_space_info(fs_info, cache->flags); |
dc2d3005 | 10162 | ASSERT(cache->space_info); |
2e6e5183 | 10163 | |
0b246afa | 10164 | ret = btrfs_add_block_group_cache(fs_info, cache); |
8c579fe7 JB |
10165 | if (ret) { |
10166 | btrfs_remove_free_space_cache(cache); | |
10167 | btrfs_put_block_group(cache); | |
10168 | return ret; | |
10169 | } | |
10170 | ||
2e6e5183 FM |
10171 | /* |
10172 | * Now that our block group has its ->space_info set and is inserted in | |
10173 | * the rbtree, update the space info's counters. | |
10174 | */ | |
0b246afa | 10175 | trace_btrfs_add_block_group(fs_info, cache, 1); |
d2006e6d | 10176 | update_space_info(fs_info, cache->flags, size, bytes_used, |
e40edf2d | 10177 | cache->bytes_super, &cache->space_info); |
0b246afa | 10178 | update_global_block_rsv(fs_info); |
1b2da372 | 10179 | |
c434d21c | 10180 | link_block_group(cache); |
6324fbf3 | 10181 | |
47ab2a6c | 10182 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
6324fbf3 | 10183 | |
0b246afa | 10184 | set_avail_alloc_bits(fs_info, type); |
6324fbf3 CM |
10185 | return 0; |
10186 | } | |
1a40e23b | 10187 | |
10ea00f5 ID |
10188 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
10189 | { | |
899c81ea ID |
10190 | u64 extra_flags = chunk_to_extended(flags) & |
10191 | BTRFS_EXTENDED_PROFILE_MASK; | |
10ea00f5 | 10192 | |
de98ced9 | 10193 | write_seqlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10194 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
10195 | fs_info->avail_data_alloc_bits &= ~extra_flags; | |
10196 | if (flags & BTRFS_BLOCK_GROUP_METADATA) | |
10197 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; | |
10198 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) | |
10199 | fs_info->avail_system_alloc_bits &= ~extra_flags; | |
de98ced9 | 10200 | write_sequnlock(&fs_info->profiles_lock); |
10ea00f5 ID |
10201 | } |
10202 | ||
1a40e23b | 10203 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
5a98ec01 | 10204 | u64 group_start, struct extent_map *em) |
1a40e23b | 10205 | { |
5a98ec01 | 10206 | struct btrfs_fs_info *fs_info = trans->fs_info; |
6bccf3ab | 10207 | struct btrfs_root *root = fs_info->extent_root; |
1a40e23b ZY |
10208 | struct btrfs_path *path; |
10209 | struct btrfs_block_group_cache *block_group; | |
44fb5511 | 10210 | struct btrfs_free_cluster *cluster; |
0b246afa | 10211 | struct btrfs_root *tree_root = fs_info->tree_root; |
1a40e23b | 10212 | struct btrfs_key key; |
0af3d00b | 10213 | struct inode *inode; |
c1895442 | 10214 | struct kobject *kobj = NULL; |
1a40e23b | 10215 | int ret; |
10ea00f5 | 10216 | int index; |
89a55897 | 10217 | int factor; |
4f69cb98 | 10218 | struct btrfs_caching_control *caching_ctl = NULL; |
04216820 | 10219 | bool remove_em; |
1a40e23b | 10220 | |
6bccf3ab | 10221 | block_group = btrfs_lookup_block_group(fs_info, group_start); |
1a40e23b | 10222 | BUG_ON(!block_group); |
c146afad | 10223 | BUG_ON(!block_group->ro); |
1a40e23b | 10224 | |
4ed0a7a3 | 10225 | trace_btrfs_remove_block_group(block_group); |
9f7c43c9 | 10226 | /* |
10227 | * Free the reserved super bytes from this block group before | |
10228 | * remove it. | |
10229 | */ | |
9e715da8 | 10230 | free_excluded_extents(block_group); |
fd708b81 JB |
10231 | btrfs_free_ref_tree_range(fs_info, block_group->key.objectid, |
10232 | block_group->key.offset); | |
9f7c43c9 | 10233 | |
1a40e23b | 10234 | memcpy(&key, &block_group->key, sizeof(key)); |
3e72ee88 | 10235 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
46df06b8 | 10236 | factor = btrfs_bg_type_to_factor(block_group->flags); |
1a40e23b | 10237 | |
44fb5511 | 10238 | /* make sure this block group isn't part of an allocation cluster */ |
0b246afa | 10239 | cluster = &fs_info->data_alloc_cluster; |
44fb5511 CM |
10240 | spin_lock(&cluster->refill_lock); |
10241 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10242 | spin_unlock(&cluster->refill_lock); | |
10243 | ||
10244 | /* | |
10245 | * make sure this block group isn't part of a metadata | |
10246 | * allocation cluster | |
10247 | */ | |
0b246afa | 10248 | cluster = &fs_info->meta_alloc_cluster; |
44fb5511 CM |
10249 | spin_lock(&cluster->refill_lock); |
10250 | btrfs_return_cluster_to_free_space(block_group, cluster); | |
10251 | spin_unlock(&cluster->refill_lock); | |
10252 | ||
1a40e23b | 10253 | path = btrfs_alloc_path(); |
d8926bb3 MF |
10254 | if (!path) { |
10255 | ret = -ENOMEM; | |
10256 | goto out; | |
10257 | } | |
1a40e23b | 10258 | |
1bbc621e CM |
10259 | /* |
10260 | * get the inode first so any iput calls done for the io_list | |
10261 | * aren't the final iput (no unlinks allowed now) | |
10262 | */ | |
77ab86bf | 10263 | inode = lookup_free_space_inode(fs_info, block_group, path); |
1bbc621e CM |
10264 | |
10265 | mutex_lock(&trans->transaction->cache_write_mutex); | |
10266 | /* | |
10267 | * make sure our free spache cache IO is done before remove the | |
10268 | * free space inode | |
10269 | */ | |
10270 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10271 | if (!list_empty(&block_group->io_list)) { | |
10272 | list_del_init(&block_group->io_list); | |
10273 | ||
10274 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); | |
10275 | ||
10276 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
afdb5718 | 10277 | btrfs_wait_cache_io(trans, block_group, path); |
1bbc621e CM |
10278 | btrfs_put_block_group(block_group); |
10279 | spin_lock(&trans->transaction->dirty_bgs_lock); | |
10280 | } | |
10281 | ||
10282 | if (!list_empty(&block_group->dirty_list)) { | |
10283 | list_del_init(&block_group->dirty_list); | |
10284 | btrfs_put_block_group(block_group); | |
10285 | } | |
10286 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
10287 | mutex_unlock(&trans->transaction->cache_write_mutex); | |
10288 | ||
0af3d00b | 10289 | if (!IS_ERR(inode)) { |
73f2e545 | 10290 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
79787eaa JM |
10291 | if (ret) { |
10292 | btrfs_add_delayed_iput(inode); | |
10293 | goto out; | |
10294 | } | |
0af3d00b JB |
10295 | clear_nlink(inode); |
10296 | /* One for the block groups ref */ | |
10297 | spin_lock(&block_group->lock); | |
10298 | if (block_group->iref) { | |
10299 | block_group->iref = 0; | |
10300 | block_group->inode = NULL; | |
10301 | spin_unlock(&block_group->lock); | |
10302 | iput(inode); | |
10303 | } else { | |
10304 | spin_unlock(&block_group->lock); | |
10305 | } | |
10306 | /* One for our lookup ref */ | |
455757c3 | 10307 | btrfs_add_delayed_iput(inode); |
0af3d00b JB |
10308 | } |
10309 | ||
10310 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; | |
10311 | key.offset = block_group->key.objectid; | |
10312 | key.type = 0; | |
10313 | ||
10314 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); | |
10315 | if (ret < 0) | |
10316 | goto out; | |
10317 | if (ret > 0) | |
b3b4aa74 | 10318 | btrfs_release_path(path); |
0af3d00b JB |
10319 | if (ret == 0) { |
10320 | ret = btrfs_del_item(trans, tree_root, path); | |
10321 | if (ret) | |
10322 | goto out; | |
b3b4aa74 | 10323 | btrfs_release_path(path); |
0af3d00b JB |
10324 | } |
10325 | ||
0b246afa | 10326 | spin_lock(&fs_info->block_group_cache_lock); |
1a40e23b | 10327 | rb_erase(&block_group->cache_node, |
0b246afa | 10328 | &fs_info->block_group_cache_tree); |
292cbd51 | 10329 | RB_CLEAR_NODE(&block_group->cache_node); |
a1897fdd | 10330 | |
0b246afa JM |
10331 | if (fs_info->first_logical_byte == block_group->key.objectid) |
10332 | fs_info->first_logical_byte = (u64)-1; | |
10333 | spin_unlock(&fs_info->block_group_cache_lock); | |
817d52f8 | 10334 | |
80eb234a | 10335 | down_write(&block_group->space_info->groups_sem); |
44fb5511 CM |
10336 | /* |
10337 | * we must use list_del_init so people can check to see if they | |
10338 | * are still on the list after taking the semaphore | |
10339 | */ | |
10340 | list_del_init(&block_group->list); | |
6ab0a202 | 10341 | if (list_empty(&block_group->space_info->block_groups[index])) { |
c1895442 JM |
10342 | kobj = block_group->space_info->block_group_kobjs[index]; |
10343 | block_group->space_info->block_group_kobjs[index] = NULL; | |
0b246afa | 10344 | clear_avail_alloc_bits(fs_info, block_group->flags); |
6ab0a202 | 10345 | } |
80eb234a | 10346 | up_write(&block_group->space_info->groups_sem); |
c1895442 JM |
10347 | if (kobj) { |
10348 | kobject_del(kobj); | |
10349 | kobject_put(kobj); | |
10350 | } | |
1a40e23b | 10351 | |
4f69cb98 FM |
10352 | if (block_group->has_caching_ctl) |
10353 | caching_ctl = get_caching_control(block_group); | |
817d52f8 | 10354 | if (block_group->cached == BTRFS_CACHE_STARTED) |
11833d66 | 10355 | wait_block_group_cache_done(block_group); |
4f69cb98 | 10356 | if (block_group->has_caching_ctl) { |
0b246afa | 10357 | down_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10358 | if (!caching_ctl) { |
10359 | struct btrfs_caching_control *ctl; | |
10360 | ||
10361 | list_for_each_entry(ctl, | |
0b246afa | 10362 | &fs_info->caching_block_groups, list) |
4f69cb98 FM |
10363 | if (ctl->block_group == block_group) { |
10364 | caching_ctl = ctl; | |
1e4f4714 | 10365 | refcount_inc(&caching_ctl->count); |
4f69cb98 FM |
10366 | break; |
10367 | } | |
10368 | } | |
10369 | if (caching_ctl) | |
10370 | list_del_init(&caching_ctl->list); | |
0b246afa | 10371 | up_write(&fs_info->commit_root_sem); |
4f69cb98 FM |
10372 | if (caching_ctl) { |
10373 | /* Once for the caching bgs list and once for us. */ | |
10374 | put_caching_control(caching_ctl); | |
10375 | put_caching_control(caching_ctl); | |
10376 | } | |
10377 | } | |
817d52f8 | 10378 | |
ce93ec54 JB |
10379 | spin_lock(&trans->transaction->dirty_bgs_lock); |
10380 | if (!list_empty(&block_group->dirty_list)) { | |
1bbc621e CM |
10381 | WARN_ON(1); |
10382 | } | |
10383 | if (!list_empty(&block_group->io_list)) { | |
10384 | WARN_ON(1); | |
ce93ec54 JB |
10385 | } |
10386 | spin_unlock(&trans->transaction->dirty_bgs_lock); | |
817d52f8 JB |
10387 | btrfs_remove_free_space_cache(block_group); |
10388 | ||
c146afad | 10389 | spin_lock(&block_group->space_info->lock); |
75c68e9f | 10390 | list_del_init(&block_group->ro_list); |
18d018ad | 10391 | |
0b246afa | 10392 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
18d018ad ZL |
10393 | WARN_ON(block_group->space_info->total_bytes |
10394 | < block_group->key.offset); | |
10395 | WARN_ON(block_group->space_info->bytes_readonly | |
10396 | < block_group->key.offset); | |
10397 | WARN_ON(block_group->space_info->disk_total | |
10398 | < block_group->key.offset * factor); | |
10399 | } | |
c146afad YZ |
10400 | block_group->space_info->total_bytes -= block_group->key.offset; |
10401 | block_group->space_info->bytes_readonly -= block_group->key.offset; | |
89a55897 | 10402 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
18d018ad | 10403 | |
c146afad | 10404 | spin_unlock(&block_group->space_info->lock); |
283bb197 | 10405 | |
0af3d00b JB |
10406 | memcpy(&key, &block_group->key, sizeof(key)); |
10407 | ||
34441361 | 10408 | mutex_lock(&fs_info->chunk_mutex); |
495e64f4 FM |
10409 | if (!list_empty(&em->list)) { |
10410 | /* We're in the transaction->pending_chunks list. */ | |
10411 | free_extent_map(em); | |
10412 | } | |
04216820 FM |
10413 | spin_lock(&block_group->lock); |
10414 | block_group->removed = 1; | |
10415 | /* | |
10416 | * At this point trimming can't start on this block group, because we | |
10417 | * removed the block group from the tree fs_info->block_group_cache_tree | |
10418 | * so no one can't find it anymore and even if someone already got this | |
10419 | * block group before we removed it from the rbtree, they have already | |
10420 | * incremented block_group->trimming - if they didn't, they won't find | |
10421 | * any free space entries because we already removed them all when we | |
10422 | * called btrfs_remove_free_space_cache(). | |
10423 | * | |
10424 | * And we must not remove the extent map from the fs_info->mapping_tree | |
10425 | * to prevent the same logical address range and physical device space | |
10426 | * ranges from being reused for a new block group. This is because our | |
10427 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is | |
10428 | * completely transactionless, so while it is trimming a range the | |
10429 | * currently running transaction might finish and a new one start, | |
10430 | * allowing for new block groups to be created that can reuse the same | |
10431 | * physical device locations unless we take this special care. | |
e33e17ee JM |
10432 | * |
10433 | * There may also be an implicit trim operation if the file system | |
10434 | * is mounted with -odiscard. The same protections must remain | |
10435 | * in place until the extents have been discarded completely when | |
10436 | * the transaction commit has completed. | |
04216820 FM |
10437 | */ |
10438 | remove_em = (atomic_read(&block_group->trimming) == 0); | |
10439 | /* | |
10440 | * Make sure a trimmer task always sees the em in the pinned_chunks list | |
10441 | * if it sees block_group->removed == 1 (needs to lock block_group->lock | |
10442 | * before checking block_group->removed). | |
10443 | */ | |
10444 | if (!remove_em) { | |
10445 | /* | |
10446 | * Our em might be in trans->transaction->pending_chunks which | |
10447 | * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks), | |
10448 | * and so is the fs_info->pinned_chunks list. | |
10449 | * | |
10450 | * So at this point we must be holding the chunk_mutex to avoid | |
10451 | * any races with chunk allocation (more specifically at | |
10452 | * volumes.c:contains_pending_extent()), to ensure it always | |
10453 | * sees the em, either in the pending_chunks list or in the | |
10454 | * pinned_chunks list. | |
10455 | */ | |
0b246afa | 10456 | list_move_tail(&em->list, &fs_info->pinned_chunks); |
04216820 FM |
10457 | } |
10458 | spin_unlock(&block_group->lock); | |
04216820 FM |
10459 | |
10460 | if (remove_em) { | |
10461 | struct extent_map_tree *em_tree; | |
10462 | ||
0b246afa | 10463 | em_tree = &fs_info->mapping_tree.map_tree; |
04216820 | 10464 | write_lock(&em_tree->lock); |
8dbcd10f FM |
10465 | /* |
10466 | * The em might be in the pending_chunks list, so make sure the | |
10467 | * chunk mutex is locked, since remove_extent_mapping() will | |
10468 | * delete us from that list. | |
10469 | */ | |
04216820 FM |
10470 | remove_extent_mapping(em_tree, em); |
10471 | write_unlock(&em_tree->lock); | |
10472 | /* once for the tree */ | |
10473 | free_extent_map(em); | |
10474 | } | |
10475 | ||
34441361 | 10476 | mutex_unlock(&fs_info->chunk_mutex); |
8dbcd10f | 10477 | |
f3f72779 | 10478 | ret = remove_block_group_free_space(trans, block_group); |
1e144fb8 OS |
10479 | if (ret) |
10480 | goto out; | |
10481 | ||
fa9c0d79 CM |
10482 | btrfs_put_block_group(block_group); |
10483 | btrfs_put_block_group(block_group); | |
1a40e23b ZY |
10484 | |
10485 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
10486 | if (ret > 0) | |
10487 | ret = -EIO; | |
10488 | if (ret < 0) | |
10489 | goto out; | |
10490 | ||
10491 | ret = btrfs_del_item(trans, root, path); | |
10492 | out: | |
10493 | btrfs_free_path(path); | |
10494 | return ret; | |
10495 | } | |
acce952b | 10496 | |
8eab77ff | 10497 | struct btrfs_trans_handle * |
7fd01182 FM |
10498 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
10499 | const u64 chunk_offset) | |
8eab77ff | 10500 | { |
7fd01182 FM |
10501 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; |
10502 | struct extent_map *em; | |
10503 | struct map_lookup *map; | |
10504 | unsigned int num_items; | |
10505 | ||
10506 | read_lock(&em_tree->lock); | |
10507 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
10508 | read_unlock(&em_tree->lock); | |
10509 | ASSERT(em && em->start == chunk_offset); | |
10510 | ||
8eab77ff | 10511 | /* |
7fd01182 FM |
10512 | * We need to reserve 3 + N units from the metadata space info in order |
10513 | * to remove a block group (done at btrfs_remove_chunk() and at | |
10514 | * btrfs_remove_block_group()), which are used for: | |
10515 | * | |
8eab77ff FM |
10516 | * 1 unit for adding the free space inode's orphan (located in the tree |
10517 | * of tree roots). | |
7fd01182 FM |
10518 | * 1 unit for deleting the block group item (located in the extent |
10519 | * tree). | |
10520 | * 1 unit for deleting the free space item (located in tree of tree | |
10521 | * roots). | |
10522 | * N units for deleting N device extent items corresponding to each | |
10523 | * stripe (located in the device tree). | |
10524 | * | |
10525 | * In order to remove a block group we also need to reserve units in the | |
10526 | * system space info in order to update the chunk tree (update one or | |
10527 | * more device items and remove one chunk item), but this is done at | |
10528 | * btrfs_remove_chunk() through a call to check_system_chunk(). | |
8eab77ff | 10529 | */ |
95617d69 | 10530 | map = em->map_lookup; |
7fd01182 FM |
10531 | num_items = 3 + map->num_stripes; |
10532 | free_extent_map(em); | |
10533 | ||
8eab77ff | 10534 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
7fd01182 | 10535 | num_items, 1); |
8eab77ff FM |
10536 | } |
10537 | ||
47ab2a6c JB |
10538 | /* |
10539 | * Process the unused_bgs list and remove any that don't have any allocated | |
10540 | * space inside of them. | |
10541 | */ | |
10542 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) | |
10543 | { | |
10544 | struct btrfs_block_group_cache *block_group; | |
10545 | struct btrfs_space_info *space_info; | |
47ab2a6c JB |
10546 | struct btrfs_trans_handle *trans; |
10547 | int ret = 0; | |
10548 | ||
afcdd129 | 10549 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) |
47ab2a6c JB |
10550 | return; |
10551 | ||
10552 | spin_lock(&fs_info->unused_bgs_lock); | |
10553 | while (!list_empty(&fs_info->unused_bgs)) { | |
10554 | u64 start, end; | |
e33e17ee | 10555 | int trimming; |
47ab2a6c JB |
10556 | |
10557 | block_group = list_first_entry(&fs_info->unused_bgs, | |
10558 | struct btrfs_block_group_cache, | |
10559 | bg_list); | |
47ab2a6c | 10560 | list_del_init(&block_group->bg_list); |
aefbe9a6 ZL |
10561 | |
10562 | space_info = block_group->space_info; | |
10563 | ||
47ab2a6c JB |
10564 | if (ret || btrfs_mixed_space_info(space_info)) { |
10565 | btrfs_put_block_group(block_group); | |
10566 | continue; | |
10567 | } | |
10568 | spin_unlock(&fs_info->unused_bgs_lock); | |
10569 | ||
d5f2e33b | 10570 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
67c5e7d4 | 10571 | |
47ab2a6c JB |
10572 | /* Don't want to race with allocators so take the groups_sem */ |
10573 | down_write(&space_info->groups_sem); | |
10574 | spin_lock(&block_group->lock); | |
43794446 | 10575 | if (block_group->reserved || block_group->pinned || |
47ab2a6c | 10576 | btrfs_block_group_used(&block_group->item) || |
19c4d2f9 | 10577 | block_group->ro || |
aefbe9a6 | 10578 | list_is_singular(&block_group->list)) { |
47ab2a6c JB |
10579 | /* |
10580 | * We want to bail if we made new allocations or have | |
10581 | * outstanding allocations in this block group. We do | |
10582 | * the ro check in case balance is currently acting on | |
10583 | * this block group. | |
10584 | */ | |
4ed0a7a3 | 10585 | trace_btrfs_skip_unused_block_group(block_group); |
47ab2a6c JB |
10586 | spin_unlock(&block_group->lock); |
10587 | up_write(&space_info->groups_sem); | |
10588 | goto next; | |
10589 | } | |
10590 | spin_unlock(&block_group->lock); | |
10591 | ||
10592 | /* We don't want to force the issue, only flip if it's ok. */ | |
868f401a | 10593 | ret = inc_block_group_ro(block_group, 0); |
47ab2a6c JB |
10594 | up_write(&space_info->groups_sem); |
10595 | if (ret < 0) { | |
10596 | ret = 0; | |
10597 | goto next; | |
10598 | } | |
10599 | ||
10600 | /* | |
10601 | * Want to do this before we do anything else so we can recover | |
10602 | * properly if we fail to join the transaction. | |
10603 | */ | |
7fd01182 FM |
10604 | trans = btrfs_start_trans_remove_block_group(fs_info, |
10605 | block_group->key.objectid); | |
47ab2a6c | 10606 | if (IS_ERR(trans)) { |
2ff7e61e | 10607 | btrfs_dec_block_group_ro(block_group); |
47ab2a6c JB |
10608 | ret = PTR_ERR(trans); |
10609 | goto next; | |
10610 | } | |
10611 | ||
10612 | /* | |
10613 | * We could have pending pinned extents for this block group, | |
10614 | * just delete them, we don't care about them anymore. | |
10615 | */ | |
10616 | start = block_group->key.objectid; | |
10617 | end = start + block_group->key.offset - 1; | |
d4b450cd FM |
10618 | /* |
10619 | * Hold the unused_bg_unpin_mutex lock to avoid racing with | |
10620 | * btrfs_finish_extent_commit(). If we are at transaction N, | |
10621 | * another task might be running finish_extent_commit() for the | |
10622 | * previous transaction N - 1, and have seen a range belonging | |
10623 | * to the block group in freed_extents[] before we were able to | |
10624 | * clear the whole block group range from freed_extents[]. This | |
10625 | * means that task can lookup for the block group after we | |
10626 | * unpinned it from freed_extents[] and removed it, leading to | |
10627 | * a BUG_ON() at btrfs_unpin_extent_range(). | |
10628 | */ | |
10629 | mutex_lock(&fs_info->unused_bg_unpin_mutex); | |
758eb51e | 10630 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
91166212 | 10631 | EXTENT_DIRTY); |
758eb51e | 10632 | if (ret) { |
d4b450cd | 10633 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10634 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10635 | goto end_trans; |
10636 | } | |
10637 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, | |
91166212 | 10638 | EXTENT_DIRTY); |
758eb51e | 10639 | if (ret) { |
d4b450cd | 10640 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
2ff7e61e | 10641 | btrfs_dec_block_group_ro(block_group); |
758eb51e FM |
10642 | goto end_trans; |
10643 | } | |
d4b450cd | 10644 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
47ab2a6c JB |
10645 | |
10646 | /* Reset pinned so btrfs_put_block_group doesn't complain */ | |
c30666d4 ZL |
10647 | spin_lock(&space_info->lock); |
10648 | spin_lock(&block_group->lock); | |
10649 | ||
10650 | space_info->bytes_pinned -= block_group->pinned; | |
10651 | space_info->bytes_readonly += block_group->pinned; | |
dec59fa3 EL |
10652 | percpu_counter_add_batch(&space_info->total_bytes_pinned, |
10653 | -block_group->pinned, | |
10654 | BTRFS_TOTAL_BYTES_PINNED_BATCH); | |
47ab2a6c JB |
10655 | block_group->pinned = 0; |
10656 | ||
c30666d4 ZL |
10657 | spin_unlock(&block_group->lock); |
10658 | spin_unlock(&space_info->lock); | |
10659 | ||
e33e17ee | 10660 | /* DISCARD can flip during remount */ |
0b246afa | 10661 | trimming = btrfs_test_opt(fs_info, DISCARD); |
e33e17ee JM |
10662 | |
10663 | /* Implicit trim during transaction commit. */ | |
10664 | if (trimming) | |
10665 | btrfs_get_block_group_trimming(block_group); | |
10666 | ||
47ab2a6c JB |
10667 | /* |
10668 | * Btrfs_remove_chunk will abort the transaction if things go | |
10669 | * horribly wrong. | |
10670 | */ | |
97aff912 | 10671 | ret = btrfs_remove_chunk(trans, block_group->key.objectid); |
e33e17ee JM |
10672 | |
10673 | if (ret) { | |
10674 | if (trimming) | |
10675 | btrfs_put_block_group_trimming(block_group); | |
10676 | goto end_trans; | |
10677 | } | |
10678 | ||
10679 | /* | |
10680 | * If we're not mounted with -odiscard, we can just forget | |
10681 | * about this block group. Otherwise we'll need to wait | |
10682 | * until transaction commit to do the actual discard. | |
10683 | */ | |
10684 | if (trimming) { | |
348a0013 FM |
10685 | spin_lock(&fs_info->unused_bgs_lock); |
10686 | /* | |
10687 | * A concurrent scrub might have added us to the list | |
10688 | * fs_info->unused_bgs, so use a list_move operation | |
10689 | * to add the block group to the deleted_bgs list. | |
10690 | */ | |
e33e17ee JM |
10691 | list_move(&block_group->bg_list, |
10692 | &trans->transaction->deleted_bgs); | |
348a0013 | 10693 | spin_unlock(&fs_info->unused_bgs_lock); |
e33e17ee JM |
10694 | btrfs_get_block_group(block_group); |
10695 | } | |
758eb51e | 10696 | end_trans: |
3a45bb20 | 10697 | btrfs_end_transaction(trans); |
47ab2a6c | 10698 | next: |
d5f2e33b | 10699 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
47ab2a6c JB |
10700 | btrfs_put_block_group(block_group); |
10701 | spin_lock(&fs_info->unused_bgs_lock); | |
10702 | } | |
10703 | spin_unlock(&fs_info->unused_bgs_lock); | |
10704 | } | |
10705 | ||
c59021f8 | 10706 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
10707 | { | |
1aba86d6 | 10708 | struct btrfs_super_block *disk_super; |
10709 | u64 features; | |
10710 | u64 flags; | |
10711 | int mixed = 0; | |
c59021f8 | 10712 | int ret; |
10713 | ||
6c41761f | 10714 | disk_super = fs_info->super_copy; |
1aba86d6 | 10715 | if (!btrfs_super_root(disk_super)) |
0dc924c5 | 10716 | return -EINVAL; |
c59021f8 | 10717 | |
1aba86d6 | 10718 | features = btrfs_super_incompat_flags(disk_super); |
10719 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
10720 | mixed = 1; | |
c59021f8 | 10721 | |
1aba86d6 | 10722 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
4ca61683 | 10723 | ret = create_space_info(fs_info, flags); |
c59021f8 | 10724 | if (ret) |
1aba86d6 | 10725 | goto out; |
c59021f8 | 10726 | |
1aba86d6 | 10727 | if (mixed) { |
10728 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; | |
4ca61683 | 10729 | ret = create_space_info(fs_info, flags); |
1aba86d6 | 10730 | } else { |
10731 | flags = BTRFS_BLOCK_GROUP_METADATA; | |
4ca61683 | 10732 | ret = create_space_info(fs_info, flags); |
1aba86d6 | 10733 | if (ret) |
10734 | goto out; | |
10735 | ||
10736 | flags = BTRFS_BLOCK_GROUP_DATA; | |
4ca61683 | 10737 | ret = create_space_info(fs_info, flags); |
1aba86d6 | 10738 | } |
10739 | out: | |
c59021f8 | 10740 | return ret; |
10741 | } | |
10742 | ||
2ff7e61e JM |
10743 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
10744 | u64 start, u64 end) | |
acce952b | 10745 | { |
2ff7e61e | 10746 | return unpin_extent_range(fs_info, start, end, false); |
acce952b | 10747 | } |
10748 | ||
499f377f JM |
10749 | /* |
10750 | * It used to be that old block groups would be left around forever. | |
10751 | * Iterating over them would be enough to trim unused space. Since we | |
10752 | * now automatically remove them, we also need to iterate over unallocated | |
10753 | * space. | |
10754 | * | |
10755 | * We don't want a transaction for this since the discard may take a | |
10756 | * substantial amount of time. We don't require that a transaction be | |
10757 | * running, but we do need to take a running transaction into account | |
10758 | * to ensure that we're not discarding chunks that were released in | |
10759 | * the current transaction. | |
10760 | * | |
10761 | * Holding the chunks lock will prevent other threads from allocating | |
10762 | * or releasing chunks, but it won't prevent a running transaction | |
10763 | * from committing and releasing the memory that the pending chunks | |
10764 | * list head uses. For that, we need to take a reference to the | |
10765 | * transaction. | |
10766 | */ | |
10767 | static int btrfs_trim_free_extents(struct btrfs_device *device, | |
10768 | u64 minlen, u64 *trimmed) | |
10769 | { | |
10770 | u64 start = 0, len = 0; | |
10771 | int ret; | |
10772 | ||
10773 | *trimmed = 0; | |
10774 | ||
10775 | /* Not writeable = nothing to do. */ | |
ebbede42 | 10776 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
499f377f JM |
10777 | return 0; |
10778 | ||
10779 | /* No free space = nothing to do. */ | |
10780 | if (device->total_bytes <= device->bytes_used) | |
10781 | return 0; | |
10782 | ||
10783 | ret = 0; | |
10784 | ||
10785 | while (1) { | |
fb456252 | 10786 | struct btrfs_fs_info *fs_info = device->fs_info; |
499f377f JM |
10787 | struct btrfs_transaction *trans; |
10788 | u64 bytes; | |
10789 | ||
10790 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); | |
10791 | if (ret) | |
10792 | return ret; | |
10793 | ||
10794 | down_read(&fs_info->commit_root_sem); | |
10795 | ||
10796 | spin_lock(&fs_info->trans_lock); | |
10797 | trans = fs_info->running_transaction; | |
10798 | if (trans) | |
9b64f57d | 10799 | refcount_inc(&trans->use_count); |
499f377f JM |
10800 | spin_unlock(&fs_info->trans_lock); |
10801 | ||
10802 | ret = find_free_dev_extent_start(trans, device, minlen, start, | |
10803 | &start, &len); | |
10804 | if (trans) | |
10805 | btrfs_put_transaction(trans); | |
10806 | ||
10807 | if (ret) { | |
10808 | up_read(&fs_info->commit_root_sem); | |
10809 | mutex_unlock(&fs_info->chunk_mutex); | |
10810 | if (ret == -ENOSPC) | |
10811 | ret = 0; | |
10812 | break; | |
10813 | } | |
10814 | ||
10815 | ret = btrfs_issue_discard(device->bdev, start, len, &bytes); | |
10816 | up_read(&fs_info->commit_root_sem); | |
10817 | mutex_unlock(&fs_info->chunk_mutex); | |
10818 | ||
10819 | if (ret) | |
10820 | break; | |
10821 | ||
10822 | start += len; | |
10823 | *trimmed += bytes; | |
10824 | ||
10825 | if (fatal_signal_pending(current)) { | |
10826 | ret = -ERESTARTSYS; | |
10827 | break; | |
10828 | } | |
10829 | ||
10830 | cond_resched(); | |
10831 | } | |
10832 | ||
10833 | return ret; | |
10834 | } | |
10835 | ||
2ff7e61e | 10836 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
f7039b1d | 10837 | { |
f7039b1d | 10838 | struct btrfs_block_group_cache *cache = NULL; |
499f377f JM |
10839 | struct btrfs_device *device; |
10840 | struct list_head *devices; | |
f7039b1d LD |
10841 | u64 group_trimmed; |
10842 | u64 start; | |
10843 | u64 end; | |
10844 | u64 trimmed = 0; | |
2cac13e4 | 10845 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
f7039b1d LD |
10846 | int ret = 0; |
10847 | ||
2cac13e4 LB |
10848 | /* |
10849 | * try to trim all FS space, our block group may start from non-zero. | |
10850 | */ | |
10851 | if (range->len == total_bytes) | |
10852 | cache = btrfs_lookup_first_block_group(fs_info, range->start); | |
10853 | else | |
10854 | cache = btrfs_lookup_block_group(fs_info, range->start); | |
f7039b1d LD |
10855 | |
10856 | while (cache) { | |
10857 | if (cache->key.objectid >= (range->start + range->len)) { | |
10858 | btrfs_put_block_group(cache); | |
10859 | break; | |
10860 | } | |
10861 | ||
10862 | start = max(range->start, cache->key.objectid); | |
10863 | end = min(range->start + range->len, | |
10864 | cache->key.objectid + cache->key.offset); | |
10865 | ||
10866 | if (end - start >= range->minlen) { | |
10867 | if (!block_group_cache_done(cache)) { | |
f6373bf3 | 10868 | ret = cache_block_group(cache, 0); |
1be41b78 JB |
10869 | if (ret) { |
10870 | btrfs_put_block_group(cache); | |
10871 | break; | |
10872 | } | |
10873 | ret = wait_block_group_cache_done(cache); | |
10874 | if (ret) { | |
10875 | btrfs_put_block_group(cache); | |
10876 | break; | |
10877 | } | |
f7039b1d LD |
10878 | } |
10879 | ret = btrfs_trim_block_group(cache, | |
10880 | &group_trimmed, | |
10881 | start, | |
10882 | end, | |
10883 | range->minlen); | |
10884 | ||
10885 | trimmed += group_trimmed; | |
10886 | if (ret) { | |
10887 | btrfs_put_block_group(cache); | |
10888 | break; | |
10889 | } | |
10890 | } | |
10891 | ||
2ff7e61e | 10892 | cache = next_block_group(fs_info, cache); |
f7039b1d LD |
10893 | } |
10894 | ||
0b246afa JM |
10895 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
10896 | devices = &fs_info->fs_devices->alloc_list; | |
499f377f JM |
10897 | list_for_each_entry(device, devices, dev_alloc_list) { |
10898 | ret = btrfs_trim_free_extents(device, range->minlen, | |
10899 | &group_trimmed); | |
10900 | if (ret) | |
10901 | break; | |
10902 | ||
10903 | trimmed += group_trimmed; | |
10904 | } | |
0b246afa | 10905 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
499f377f | 10906 | |
f7039b1d LD |
10907 | range->len = trimmed; |
10908 | return ret; | |
10909 | } | |
8257b2dc MX |
10910 | |
10911 | /* | |
ea14b57f | 10912 | * btrfs_{start,end}_write_no_snapshotting() are similar to |
9ea24bbe FM |
10913 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing |
10914 | * data into the page cache through nocow before the subvolume is snapshoted, | |
10915 | * but flush the data into disk after the snapshot creation, or to prevent | |
ea14b57f | 10916 | * operations while snapshotting is ongoing and that cause the snapshot to be |
9ea24bbe | 10917 | * inconsistent (writes followed by expanding truncates for example). |
8257b2dc | 10918 | */ |
ea14b57f | 10919 | void btrfs_end_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc MX |
10920 | { |
10921 | percpu_counter_dec(&root->subv_writers->counter); | |
093258e6 | 10922 | cond_wake_up(&root->subv_writers->wait); |
8257b2dc MX |
10923 | } |
10924 | ||
ea14b57f | 10925 | int btrfs_start_write_no_snapshotting(struct btrfs_root *root) |
8257b2dc | 10926 | { |
ea14b57f | 10927 | if (atomic_read(&root->will_be_snapshotted)) |
8257b2dc MX |
10928 | return 0; |
10929 | ||
10930 | percpu_counter_inc(&root->subv_writers->counter); | |
10931 | /* | |
10932 | * Make sure counter is updated before we check for snapshot creation. | |
10933 | */ | |
10934 | smp_mb(); | |
ea14b57f DS |
10935 | if (atomic_read(&root->will_be_snapshotted)) { |
10936 | btrfs_end_write_no_snapshotting(root); | |
8257b2dc MX |
10937 | return 0; |
10938 | } | |
10939 | return 1; | |
10940 | } | |
0bc19f90 | 10941 | |
0bc19f90 ZL |
10942 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) |
10943 | { | |
10944 | while (true) { | |
10945 | int ret; | |
10946 | ||
ea14b57f | 10947 | ret = btrfs_start_write_no_snapshotting(root); |
0bc19f90 ZL |
10948 | if (ret) |
10949 | break; | |
4625956a PZ |
10950 | wait_var_event(&root->will_be_snapshotted, |
10951 | !atomic_read(&root->will_be_snapshotted)); | |
0bc19f90 ZL |
10952 | } |
10953 | } | |
031f24da QW |
10954 | |
10955 | void btrfs_mark_bg_unused(struct btrfs_block_group_cache *bg) | |
10956 | { | |
10957 | struct btrfs_fs_info *fs_info = bg->fs_info; | |
10958 | ||
10959 | spin_lock(&fs_info->unused_bgs_lock); | |
10960 | if (list_empty(&bg->bg_list)) { | |
10961 | btrfs_get_block_group(bg); | |
10962 | trace_btrfs_add_unused_block_group(bg); | |
10963 | list_add_tail(&bg->bg_list, &fs_info->unused_bgs); | |
10964 | } | |
10965 | spin_unlock(&fs_info->unused_bgs_lock); | |
10966 | } |